0% found this document useful (0 votes)

152 views

Spru187u Optimizingcompiler

Uploaded by

AbraCadaver

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

152 views

Spru187u Optimizingcompiler

Uploaded by

AbraCadaver

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 278

TMS320C6000 Optimizing Compiler v7.

User's Guide

Literature Number: SPRU187U

July 2012
Contents

Preface...................................................................................................................................... 11
1 Introduction to the Software Development Tools ................................................................... 14
1.1 Software Development Tools Overview ................................................................................ 15
1.2 C/C++ Compiler Overview ................................................................................................ 16
1.2.1 ANSI/ISO Standard ............................................................................................... 16
1.2.2 Output Files ....................................................................................................... 17
1.2.3 Compiler Interface ................................................................................................ 17
1.2.4 Utilities ............................................................................................................. 17
2 Using the C/C++ Compiler .................................................................................................. 18
2.1 About the Compiler ........................................................................................................ 19
2.2 Invoking the C/C++ Compiler ............................................................................................ 19
2.3 Changing the Compiler's Behavior With Options ...................................................................... 20
2.3.1 Frequently Used Options ........................................................................................ 30
2.3.2 Miscellaneous Useful Options .................................................................................. 31
2.3.3 Run-Time Model Options ........................................................................................ 32
2.3.4 Selecting Target CPU Version (--silicon_version Option) ................................................... 33
2.3.5 Symbolic Debugging and Profiling Options ................................................................... 34
2.3.6 Specifying Filenames ............................................................................................ 35
2.3.7 Changing How the Compiler Interprets Filenames ........................................................... 35
2.3.8 Changing How the Compiler Processes C Files ............................................................. 36
2.3.9 Changing How the Compiler Interprets and Names Extensions ........................................... 36
2.3.10 Specifying Directories ........................................................................................... 36
2.3.11 Assembler Options .............................................................................................. 37
2.3.12 Dynamic Linking ................................................................................................. 38
2.3.13 Deprecated Options ............................................................................................. 39
2.4 Controlling the Compiler Through Environment Variables ........................................................... 39
2.4.1 Setting Default Compiler Options (C6X_C_OPTION) ....................................................... 39
2.4.2 Naming an Alternate Directory (C6X_C_DIR) ................................................................ 40
2.5 Precompiled Header Support ............................................................................................ 41
2.5.1 Automatic Precompiled Header ................................................................................. 41
2.5.2 Manual Precompiled Header .................................................................................... 41
2.5.3 Additional Precompiled Header Options ....................................................................... 41
2.6 Controlling the Preprocessor ............................................................................................. 42
2.6.1 Predefined Macro Names ....................................................................................... 42
2.6.2 The Search Path for #include Files ............................................................................ 43
2.6.3 Generating a Preprocessed Listing File (--preproc_only Option) .......................................... 44
2.6.4 Continuing Compilation After Preprocessing (--preproc_with_compile Option) .......................... 44
2.6.5 Generating a Preprocessed Listing File With Comments (--preproc_with_comment Option) .......... 44
2.6.6 Generating a Preprocessed Listing File With Line-Control Information (--preproc_with_line
Option) ............................................................................................................. 44
2.6.7 Generating Preprocessed Output for a Make Utility (--preproc_dependency Option) ................... 45
2.6.8 Generating a List of Files Included With the #include Directive (--preproc_includes Option) .......... 45
2.6.9 Generating a List of Macros in a File (--preproc_macros Option) .......................................... 45
2.7 Understanding Diagnostic Messages ................................................................................... 45
2.7.1 Controlling Diagnostics .......................................................................................... 46

2 Contents SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

2.7.2 How You Can Use Diagnostic Suppression Options ........................................................ 47

2.8 Other Messages ........................................................................................................... 48
2.9 Generating Cross-Reference Listing Information (--gen_acp_xref Option) ........................................ 48
2.10 Generating a Raw Listing File (--gen_acp_raw Option) .............................................................. 48
2.11 Using Inline Function Expansion ........................................................................................ 50
2.11.1 Inlining Intrinsic Operators ..................................................................................... 50
2.11.2 Automatic Inlining ................................................................................................ 50
2.11.3 Unguarded Definition-Controlled Inlining ..................................................................... 50
2.11.4 Guarded Inlining and the _INLINE Preprocessor Symbol .................................................. 51
2.11.5 Inlining Restrictions ............................................................................................. 52
2.12 Interrupt Flexibility Options (--interrupt_threshold Option) ........................................................... 53
2.13 Linking C6400 Code With C6200/C6700/Older C6400 Object Code ............................................... 54
2.14 Using Interlist ............................................................................................................... 54
2.15 Generating and Using Performance Advice ........................................................................... 55
2.16 Controlling Application Binary Interface ................................................................................ 55
2.17 Enabling Entry Hook and Exit Hook Functions ........................................................................ 57
3 Optimizing Your Code ........................................................................................................ 58
3.1 Invoking Optimization ..................................................................................................... 59
3.2 Optimizing Software Pipelining .......................................................................................... 60
3.2.1 Turn Off Software Pipelining (--disable_software_pipelining Option) ...................................... 61
3.2.2 Software Pipelining Information ................................................................................. 61
3.2.3 Collapsing Prologs and Epilogs for Improved Performance and Code Size .............................. 65
3.3 Redundant Loops .......................................................................................................... 67
3.4 Utilizing the Loop Buffer Using SPLOOP on C6400+, C6740, and C6600 ........................................ 68
3.5 Reducing Code Size (--opt_for_space (or -ms) Option) .............................................................. 68
3.6 Performing File-Level Optimization (--opt_level=3 option) ........................................................... 70
3.6.1 Controlling File-Level Optimization (--std_lib_func_def Options) ........................................... 70
3.6.2 Creating an Optimization Information File (--gen_opt_info Option) ........................................ 70
3.7 Performing Program-Level Optimization (--program_level_compile and --opt_level=3 options) ................ 71
3.7.1 Controlling Program-Level Optimization (--call_assumptions Option) ..................................... 71
3.7.2 Optimization Considerations When Mixing C/C++ and Assembly ......................................... 72
3.8 Using Feedback Directed Optimization ................................................................................. 73
3.8.1 Feedback Directed Optimization ............................................................................... 73
3.8.2 Profile Data Decoder ............................................................................................. 75
3.8.3 Feedback Directed Optimization API .......................................................................... 76
3.8.4 Feedback Directed Optimization Summary ................................................................... 76
3.9 Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage ................. 77
3.9.1 Background and Motivation ..................................................................................... 77
3.9.2 Code Coverage ................................................................................................... 78
3.9.3 What Performance Improvements Can You Expect to See? ............................................... 79
3.9.4 Program Cache Layout Related Features and Capabilities ................................................ 79
3.9.5 Program Instruction Cache Layout Development Flow ...................................................... 80
3.9.6 Comma-Separated Values (CSV) Files with Weighted Call Graph (WCG) Information ................ 83
3.9.7 Linker Command File Operator - unordered() ................................................................ 83
3.9.8 Things To Be Aware Of .......................................................................................... 86
3.10 Indicating Whether Certain Aliasing Techniques Are Used .......................................................... 87
3.10.1 Use the --aliased_variables Option When Certain Aliases are Used ..................................... 87
3.10.2 Use the --no_bad_aliases Option to Indicate That These Techniques Are Not Used .................. 87
3.10.3 Using the --no_bad_aliases Option With the Assembly Optimizer ........................................ 88
3.11 Prevent Reordering of Associative Floating-Point Operations ....................................................... 88
3.12 Use Caution With asm Statements in Optimized Code .............................................................. 89
3.13 Automatic Inline Expansion (--auto_inline Option) .................................................................... 89
3.14 Using Performance Advice to Optimize your Code ................................................................... 90

SPRU187U – July 2012 Contents 3

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

3.14.1 Types of Advice .................................................................................................. 91

3.14.2 Using the Interlist Feature With Optimization ................................................................ 96
3.14.3 Debugging and Profiling Optimized Code .................................................................... 99
3.14.4 Controlling Code Size Versus Speed ....................................................................... 100
3.14.5 What Kind of Optimization Is Being Performed? .......................................................... 101
4 Using the Assembly Optimizer .......................................................................................... 104
4.1 Code Development Flow to Increase Performance ................................................................. 105
4.2 About the Assembly Optimizer ......................................................................................... 106
4.3 What You Need to Know to Write Linear Assembly ................................................................. 107
4.3.1 Linear Assembly Source Statement Format ................................................................. 108
4.3.2 Register Specification for Linear Assembly .................................................................. 109
4.3.3 Functional Unit Specification for Linear Assembly .......................................................... 111
4.3.4 Using Linear Assembly Source Comments .................................................................. 111
4.3.5 Assembly File Retains Your Symbolic Register Names ................................................... 112
4.4 Assembly Optimizer Directives ......................................................................................... 113
4.4.1 Instructions That Are Not Allowed in Procedures ........................................................... 126
4.5 Avoiding Memory Bank Conflicts With the Assembly Optimizer ................................................... 127
4.5.1 Preventing Memory Bank Conflicts ........................................................................... 128
4.5.2 A Dot Product Example That Avoids Memory Bank Conflicts ............................................ 129
4.5.3 Memory Bank Conflicts for Indexed Pointers ................................................................ 132
4.5.4 Memory Bank Conflict Algorithm .............................................................................. 133
4.6 Memory Alias Disambiguation .......................................................................................... 133
4.6.1 How the Assembly Optimizer Handles Memory References (Default) ................................... 133
4.6.2 Using the --no_bad_aliases Option to Handle Memory References ..................................... 133
4.6.3 Using the .no_mdep Directive ................................................................................. 133
4.6.4 Using the .mdep Directive to Identify Specific Memory Dependencies .................................. 134
4.6.5 Memory Alias Examples ....................................................................................... 135
5 Linking C/C++ Code ......................................................................................................... 137
5.1 Invoking the Linker Through the Compiler (-z Option) .............................................................. 138
5.1.1 Invoking the Linker Separately ................................................................................ 138
5.1.2 Invoking the Linker as Part of the Compile Step ............................................................ 139
5.1.3 Disabling the Linker (--compile_only Compiler Option) .................................................... 139
5.2 Linker Code Optimizations .............................................................................................. 140
5.2.1 Generating Function Subsections (--gen_func_subsections Compiler Option) ......................... 140
5.2.2 Conditional Linking .............................................................................................. 140
5.3 Controlling the Linking Process ........................................................................................ 141
5.3.1 Including the Run-Time-Support Library ..................................................................... 141
5.3.2 Run-Time Initialization .......................................................................................... 142
5.3.3 Global Object Constructors .................................................................................... 142
5.3.4 Specifying the Type of Global Variable Initialization ....................................................... 143
5.3.5 Specifying Where to Allocate Sections in Memory ......................................................... 143
5.3.6 A Sample Linker Command File .............................................................................. 145
6 TMS320C6000C/C++ Language Implementation ................................................................... 146
6.1 Characteristics of TMS320C6000 C ................................................................................... 147
6.2 Characteristics of TMS320C6000 C++ ................................................................................ 147
6.3 Using MISRA-C:2004 .................................................................................................... 148
6.4 Data Types ................................................................................................................ 149
6.5 Keywords .................................................................................................................. 150
6.5.1 The const Keyword ............................................................................................. 150
6.5.2 The cregister Keyword ......................................................................................... 150
6.5.3 The interrupt Keyword .......................................................................................... 152
6.5.4 The near and far Keywords .................................................................................... 152
6.5.5 The restrict Keyword ............................................................................................ 153

4 Contents SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

6.5.6 The volatile Keyword ........................................................................................... 154

6.6 C++ Exception Handling ................................................................................................ 155
6.7 Register Variables and Parameters ................................................................................... 155
6.8 The asm Statement ...................................................................................................... 156
6.9 Pragma Directives ....................................................................................................... 157
6.9.1 The CHECK_MISRA Pragma ................................................................................. 158
6.9.2 The CLINK Pragma ............................................................................................. 158
6.9.3 The CODE_SECTION Pragma ............................................................................... 158
6.9.4 The DATA_ALIGN Pragma .................................................................................... 159
6.9.5 The DATA_MEM_BANK Pragma ............................................................................. 160
6.9.6 The DATA_SECTION Pragma ................................................................................ 161
6.9.7 The Diagnostic Message Pragmas ........................................................................... 162
6.9.8 The FUNC_ALWAYS_INLINE Pragma ...................................................................... 162
6.9.9 The FUNC_CANNOT_INLINE Pragma ...................................................................... 163
6.9.10 The FUNC_EXT_CALLED Pragma ......................................................................... 163
6.9.11 The FUNC_INTERRUPT_THRESHOLD Pragma ......................................................... 164
6.9.12 The FUNC_IS_PURE Pragma ............................................................................... 164
6.9.13 The FUNC_IS_SYSTEM Pragma ............................................................................ 165
6.9.14 The FUNC_NEVER_RETURNS Pragma ................................................................... 165
6.9.15 The FUNC_NO_GLOBAL_ASG Pragma ................................................................... 165
6.9.16 The FUNC_NO_IND_ASG Pragma ......................................................................... 166
6.9.17 The FUNCTION_OPTIONS Pragma ........................................................................ 166
6.9.18 The INTERRUPT Pragma .................................................................................... 166
6.9.19 The LOCATION Pragma ...................................................................................... 167
6.9.20 The MUST_ITERATE Pragma ............................................................................... 167
6.9.21 The NMI_INTERRUPT Pragma .............................................................................. 168
6.9.22 The NOINIT and PERSISTENT Pragmas .................................................................. 169
6.9.23 The NO_HOOKS Pragma .................................................................................... 169
6.9.24 The PROB_ITERATE Pragma ............................................................................... 170
6.9.25 The RESET_MISRA Pragma ................................................................................. 170
6.9.26 The RETAIN Pragma .......................................................................................... 170
6.9.27 The SET_CODE_SECTION and SET_DATA_SECTION Pragmas ..................................... 171
6.9.28 The STRUCT_ALIGN Pragma ............................................................................... 172
6.9.29 The UNROLL Pragma ......................................................................................... 172
6.10 The _Pragma Operator .................................................................................................. 173
6.11 Application Binary Interface ............................................................................................ 174
6.11.1 COFF ABI ....................................................................................................... 174
6.11.2 EABI ............................................................................................................. 174
6.12 Object File Symbol Naming Conventions (Linknames) ............................................................. 174
6.13 Initializing Static and Global Variables in COFF ABI Mode ........................................................ 175
6.13.1 Initializing Static and Global Variables With the Linker ................................................... 175
6.13.2 Initializing Static and Global Variables With the const Type Qualifier .................................. 175
6.14 Changing the ANSI/ISO C Language Mode .......................................................................... 176
6.14.1 Compatibility With K&R C (--kr_compatible Option) ....................................................... 176
6.14.2 Enabling Strict ANSI/ISO Mode and Relaxed ANSI/ISO Mode (--strict_ansi and --relaxed_ansi
Options) .......................................................................................................... 177
6.14.3 Enabling Embedded C++ Mode (--embedded_cpp Option) .............................................. 178
6.15 GNU Language Extensions ............................................................................................. 178
6.15.1 Extensions ...................................................................................................... 178
6.15.2 Function Attributes ............................................................................................. 179
6.15.3 Variable Attributes ............................................................................................. 180
6.15.4 Type Attributes ................................................................................................. 180
6.15.5 Built-In Functions .............................................................................................. 181

SPRU187U – July 2012 Contents 5

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

7 Run-Time Environment ..................................................................................................... 182

7.1 Memory Model ............................................................................................................ 183
7.1.1 Sections .......................................................................................................... 183
7.1.2 C/C++ System Stack ........................................................................................... 184
7.1.3 Dynamic Memory Allocation ................................................................................... 185
7.1.4 Initialization of Variables in COFF ABI ....................................................................... 185
7.1.5 Data Memory Models ........................................................................................... 185
7.1.6 Trampoline Generation for Function Calls ................................................................... 186
7.1.7 Position Independent Data ..................................................................................... 187
7.2 Object Representation ................................................................................................... 188
7.2.1 Data Type Storage .............................................................................................. 188
7.2.2 Bit Fields ......................................................................................................... 194
7.2.3 Character String Constants .................................................................................... 195
7.3 Register Conventions .................................................................................................... 196
7.4 Function Structure and Calling Conventions ......................................................................... 197
7.4.1 How a Function Makes a Call ................................................................................. 197
7.4.2 How a Called Function Responds ............................................................................ 198
7.4.3 Accessing Arguments and Local Variables .................................................................. 199
7.5 Interfacing C and C++ With Assembly Language ................................................................... 200
7.5.1 Using Assembly Language Modules With C/C++ Code ................................................... 200
7.5.2 Accessing Assembly Language Variables From C/C++ ................................................... 202
7.5.3 Sharing C/C++ Header Files With Assembly Source ...................................................... 203
7.5.4 Using Inline Assembly Language ............................................................................. 204
7.5.5 Using Intrinsics to Access Assembly Language Statements .............................................. 204
7.5.6 The __x128_t Container Type ................................................................................. 223
7.5.7 The __float2_t Container Type ................................................................................ 224
7.5.8 Using Intrinsics for Interrupt Control and Atomic Sections ................................................ 225
7.5.9 Using Unaligned Data and 64-Bit Values .................................................................... 225
7.5.10 Using MUST_ITERATE and _nassert to Enable SIMD and Expand Compiler Knowledge of
Loops ............................................................................................................. 226
7.5.11 Methods to Align Data ......................................................................................... 227
7.5.12 SAT Bit Side Effects ........................................................................................... 229
7.5.13 IRP and AMR Conventions ................................................................................... 230
7.5.14 Floating Point Control Register Side Effects ............................................................... 230
7.6 Interrupt Handling ........................................................................................................ 230
7.6.1 Saving the SGIE Bit ............................................................................................ 230
7.6.2 Saving Registers During Interrupts ........................................................................... 230
7.6.3 Using C/C++ Interrupt Routines ............................................................................... 231
7.6.4 Using Assembly Language Interrupt Routines .............................................................. 232
7.7 Run-Time-Support Arithmetic Routines ............................................................................... 232
7.8 System Initialization ...................................................................................................... 234
7.8.1 COFF ABI Automatic Initialization of Variables ............................................................. 234
7.8.2 EABI Automatic Initialization of Variables ................................................................... 236
7.8.3 Initialization Tables ............................................................................................. 241
7.9 Support for Multi-Threaded Applications .............................................................................. 245
7.9.1 Compiling with OpenMP ....................................................................................... 245
7.9.2 Multi-Threading Runtime Support ............................................................................. 245
8 Using Run-Time-Support Functions and Building Libraries .................................................. 247
8.1 C and C++ Run-Time Support Libraries .............................................................................. 248
8.1.1 Linking Code With the Object Library ........................................................................ 248
8.1.2 Header Files ..................................................................................................... 249
8.1.3 Modifying a Library Function .................................................................................. 249
8.1.4 Minimal Support for Internationalization ...................................................................... 249

6 Contents SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

8.1.5 Allowable Number of Open Files .............................................................................. 249

8.1.6 Library Naming Conventions .................................................................................. 250
8.2 The C I/O Functions ..................................................................................................... 250
8.2.1 High-Level I/O Functions ....................................................................................... 251
8.2.2 Overview of Low-Level I/O Implementation ................................................................. 252
8.2.3 Device-Driver Level I/O Functions ............................................................................ 255
8.2.4 Adding a User-Defined Device Driver for C I/O ............................................................. 259
8.2.5 The device Prefix ................................................................................................ 260
8.3 Handling Reentrancy (_register_lock() and _register_unlock() Functions) ....................................... 262
8.4 C6700 FastMath Library ................................................................................................ 263
8.5 Library-Build Process .................................................................................................... 264
8.5.1 Required Non-Texas Instruments Software ................................................................. 264
8.5.2 Using the Library-Build Process ............................................................................... 264
8.5.3 Extending mklib ................................................................................................. 267
9 C++ Name Demangler ....................................................................................................... 268
9.1 Invoking the C++ Name Demangler ................................................................................... 269
9.2 C++ Name Demangler Options ........................................................................................ 269
9.3 Sample Usage of the C++ Name Demangler ........................................................................ 270
A Glossary ......................................................................................................................... 272

SPRU187U – July 2012 Contents 7

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

List of Figures
1-1. TMS320C6000 Software Development Flow .......................................................................... 15
3-1. Software-Pipelined Loop.................................................................................................. 60
4-1. 4-Bank Interleaved Memory ............................................................................................ 127
4-2. 4-Bank Interleaved Memory With Two Memory Spaces ............................................................ 127
7-1. Char and Short Data Storage Format ................................................................................. 189
7-2. 32-Bit Data Storage Format ............................................................................................ 190
7-3. Single-Precision Floating-Point Char Data Storage Format ........................................................ 190
7-4. 40-Bit Data Storage Format Signed __int40_t or 40-bit long....................................................... 191
7-5. Unsigned 40-bit __int40_t or long ..................................................................................... 191
7-6. 64-Bit Data Storage Format Signed 64-bit long ...................................................................... 192
7-7. Unsigned 64-bit long ..................................................................................................... 192
7-8. Double-Precision Floating-Point Data Storage Format ............................................................. 193
7-9. Bit-Field Packing in Big-Endian and Little-Endian Formats ........................................................ 194
7-10. Register Argument Conventions ....................................................................................... 198
7-11. Autoinitialization at Run Time .......................................................................................... 235
7-12. Initialization at Load Time ............................................................................................... 236
7-13. Autoinitialization at Run Time in EABI Mode ......................................................................... 237
7-14. Initialization at Load Time in EABI Mode ............................................................................. 241
7-15. Constructor Table for EABI Mode ..................................................................................... 241
7-16. Format of Initialization Records in the .cinit Section ................................................................ 242
7-17. Format of Initialization Records in the .pinit Section ................................................................ 244

8 List of Figures SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

List of Tables
2-1. Processor Options ........................................................................................................ 20
2-2. Optimization Options ..................................................................................................... 20
2-3. Debug Options ............................................................................................................. 20
2-4. Include Options ........................................................................................................... 21
2-5. Control Options ........................................................................................................... 21
2-6. Advanced Debug Options ................................................................................................ 21
2-7. Language Options ......................................................................................................... 21
2-8. Parser Preprocessing Options ........................................................................................... 22
2-9. Predefined Symbols Options ............................................................................................ 22
2-10. Diagnostics Options ....................................................................................................... 22
2-11. Run-Time Model Options ................................................................................................. 23
2-12. Advanced Optimization Options ........................................................................................ 24
2-13. Entry/Exit Hook Options .................................................................................................. 24
2-14. Feedback Options ........................................................................................................ 24
2-15. Library Function Assumptions Options ................................................................................. 24
2-16. Assembler Options ........................................................................................................ 25
2-17. File Type Specifier Options .............................................................................................. 25
2-18. Directory Specifier Options ............................................................................................... 25
2-19. Default File Extensions Options ......................................................................................... 26
2-20. Dynamic Linking Support Compiler Options ........................................................................... 26
2-21. Command Files Options .................................................................................................. 26
2-22. MISRA-C:2004 Options ................................................................................................... 26
2-23. Performance Advisor Options ............................................................................................ 26
2-24. Linker Basic Options ...................................................................................................... 27
2-25. File Search Path Options ................................................................................................. 27
2-26. Command File Preprocessing Options ................................................................................. 27
2-27. Diagnostic Options ........................................................................................................ 27
2-28. Linker Output Options ..................................................................................................... 27
2-29. Symbol Management Options ........................................................................................... 29
2-30. Run-Time Environment Options ......................................................................................... 29
2-31. Link-Time Optimization Options ......................................................................................... 29
2-32. Miscellaneous Options .................................................................................................... 29
2-33. Dynamic Linking Support Options ....................................................................................... 30
2-34. Compiler Options For Dynamic Linking................................................................................. 38
2-35. Linker Options For Dynamic Linking .................................................................................... 38
2-36. Compiler Backwards-Compatibility Options Summary ............................................................... 39
2-37. Predefined C6000 Macro Names ....................................................................................... 42
2-38. Raw Listing File Identifiers ............................................................................................... 49
2-39. Raw Listing File Diagnostic Identifiers .................................................................................. 49
3-1. Options That You Can Use With --opt_level=3 ........................................................................ 70
3-2. Selecting a File-Level Optimization Option ............................................................................ 70
3-3. Selecting a Level for the --gen_opt_info Option ....................................................................... 70
3-4. Selecting a Level for the --call_assumptions Option .................................................................. 71
3-5. Special Considerations When Using the --call_assumptions Option ............................................... 72
4-1. Options That Affect the Assembly Optimizer ......................................................................... 107
4-2. Assembly Optimizer Directives Summary ............................................................................ 113
5-1. Initialized Sections Created by the Compiler for COFFABI ........................................................ 143

SPRU187U – July 2012 List of Tables 9

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com

5-2. Initialized Sections Created by the Compiler for EABI .............................................................. 144
5-3. Uninitialized Sections Created by the Compiler for Both ABIs ..................................................... 144
6-1. TMS320C6000 C/C++ COFF ABI Data Types ....................................................................... 149
6-2. TMS320C6000 C/C++ EABI Data Types ............................................................................. 149
6-3. Valid Control Registers .................................................................................................. 150
6-4. GCC Language Extensions ............................................................................................. 178
7-1. Data Representation in Registers and Memory ..................................................................... 188
7-2. Register Usage .......................................................................................................... 196
7-3. C6000 C/C++ Intrinsics Support by Device .......................................................................... 205
7-4. TMS320C6000 C/C++ Compiler Intrinsics ............................................................................ 212
7-5. TMS320C6400, C6400+, C6740, and C6600 C/C++ Compiler Intrinsics ........................................ 214
7-6. TMS320C6400+, C6740, and C6600 C/C++ Compiler Intrinsics .................................................. 216
7-7. TMS320C6700, C6700+, C6740, and C6600 C/C++ Compiler Intrinsics ........................................ 217
7-8. TMS320C6600 C/C++ Compiler Intrinsics ............................................................................ 218
7-9. Vector-in-Scalar Support C/C++ Compiler v7.2 Intrinsics .......................................................... 224
7-10. Summary of Run-Time-Support Arithmetic Functions .............................................................. 232
8-1. The mklib Program Options ............................................................................................ 266

10 List of Tables SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Preface
SPRU187U – July 2012

Read This First

About This Manual

The TMS320C6000 Optimizing Compiler User's Guide explains how to use these compiler tools:
• Compiler
• Assembly optimizer
• Library build utility
• C++ name demangler
The compiler accepts C and C++ code conforming to the International Organization for Standardization
(ISO) standards for these languages. The compiler supports the 1989 version of the C language and the
1998 version of the C++ language.
This user's guide discusses the characteristics of the C/C++ compiler. It assumes that you already know
how to write C/C++ programs. The C Programming Language (second edition), by Brian W. Kernighan
and Dennis M. Ritchie, describes C based on the ISO C standard. You can use the Kernighan and Ritchie
(hereafter referred to as K&R) book as a supplement to this manual. References to K&R C (as opposed to
ISO C) in this manual refer to the C language as defined in the first edition of Kernighan and Ritchie's The
C Programming Language.

Notational Conventions
This document uses the following conventions:
• Program listings, program examples, and interactive displays are shown in a special typeface.
Interactive displays use a bold version of the special typeface to distinguish commands that you enter
from items that the system displays (such as prompts, command output, error messages, etc.).
Here is a sample of C code:
#include <stdio.h>
main()
{ printf("hello, cruel world\n");
}
• In syntax descriptions, the instruction, command, or directive is in a bold typeface and parameters are
in an italic typeface. Portions of a syntax that are in bold should be entered as shown; portions of a
syntax that are in italics describe the type of information that should be entered.
• Square brackets ( [ and ] ) identify an optional parameter. If you use an optional parameter, you specify
the information within the brackets. Unless the square brackets are in the bold typeface, do not enter
the brackets themselves. The following is an example of a command that has an optional parameter:

cl6x [options] [filenames] [--run_linker [link_options] [object files]]

• Braces ( { and } ) indicate that you must choose one of the parameters within the braces; you do not
enter the braces themselves. This is an example of a command with braces that are not included in the
actual syntax but indicate that you must specify either the --rom_model or --ram_model option:

cl6x --run_linker {--rom_model | --ram_model} filenames [--output_file= name.out]

--library= libraryname

SPRU187U – July 2012 Read This First 11

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Related Documentation www.ti.com

• In assembler syntax statements, column 1 is reserved for the first character of a label or symbol. If the
label or symbol is optional, it is usually not shown. If it is a required parameter, it is shown starting
against the left margin of the box, as in the example below. No instruction, command, directive, or
parameter, other than a symbol or label, can begin in column 1.

symbol .usect "section name", size in bytes[, alignment]

• Some directives can have a varying number of parameters. For example, the .byte directive. This
syntax is shown as [, ..., parameter].
• The TMS320C6200™ core is referred to as C6200. The TMS320C6400 core is referred to as C6400.
The TMS320C6700 core is referred to as C6700. TMS320C6000 and C6000 can refer to any of
C6200, C6400, C6400+, C6700, C6700+, C6740, or C6600.

Related Documentation
You can use the following books to supplement this user's guide:
ANSI X3.159-1989, Programming Language - C (Alternate version of the 1989 C Standard), American
National Standards Institute
ISO/IEC 9899:1989, International Standard - Programming Languages - C (The 1989 C Standard),
International Organization for Standardization
ISO/IEC 9899:1999, International Standard - Programming Languages - C (The C Standard),
International Organization for Standardization
ISO/IEC 14882-1998, International Standard - Programming Languages - C++ (The C++ Standard),
International Organization for Standardization
The C Programming Language (second edition), by Brian W. Kernighan and Dennis M. Ritchie,
published by Prentice-Hall, Englewood Cliffs, New Jersey, 1988
The Annotated C++ Reference Manual, Margaret A. Ellis and Bjarne Stroustrup, published by Addison-
Wesley Publishing Company, Reading, Massachusetts, 1990
C: A Reference Manual (fourth edition), by Samuel P. Harbison, and Guy L. Steele Jr., published by
Prentice Hall, Englewood Cliffs, New Jersey
Programming Embedded Systems in C and C++, by Michael Barr, Andy Oram (Editor), published by
O'Reilly & Associates; ISBN: 1565923545, February 1999
Programming in C, Steve G. Kochan, Hayden Book Company
The C++ Programming Language (second edition), Bjarne Stroustrup, published by Addison-Wesley
Publishing Company, Reading, Massachusetts, 1990
Tool Interface Standards (TIS) DWARF Debugging Information Format Specification Version 2.0,
TIS Committee, 1995
DWARF Debugging Information Format Version 3, DWARF Debugging Information Format Workgroup,
Free Standards Group, 2005 (http://dwarfstd.org)

12 Read This First SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Related Documentation From Texas Instruments

TMS320C6200, TMS320C6000 are trademarks of Texas Instruments.

All other trademarks are the property of their respective owners.

SPRU187U – July 2012 Read This First 13

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 1
SPRU187U – July 2012

Introduction to the Software Development Tools

The TMS320C6000™ is supported by a set of software development tools, which includes an optimizing
C/C++ compiler, an assembly optimizer, an assembler, a linker, and assorted utilities.
This chapter provides an overview of these tools and introduces the features of the optimizing C/C++
compiler. The assembly optimizer is discussed in Chapter 4. The assembler and linker are discussed in
detail in the TMS320C6000 Assembly Language Tools User's Guide.

Topic ........................................................................................................................... Page

1.1 Software Development Tools Overview ................................................................ 15

1.2 C/C++ Compiler Overview ................................................................................... 16

14 Introduction to the Software Development Tools SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Software Development Tools Overview

1.1 Software Development Tools Overview

Figure 1-1 illustrates the software development flow. The shaded portion of the figure highlights the most
common path of software development for C language programs. The other portions are peripheral
functions that enhance the development process.

Figure 1-1. TMS320C6000 Software Development Flow

C/C++
source
files

Macro
source C/C++ Linear
files compiler assembly

Assembler Assembly
Archiver
source optimizer

Macro Assembly
library Assembler optimized
file

Debugging
Library-build tools
Object
Archiver utility
files

Run-time-
Library of support
object library
files Link step

Executable
object file

Hex-conversion
utility

EPROM Cross-reference Object file C6000

Absolute lister
programmer lister utilities

The following list describes the tools that are shown in Figure 1-1:
• The assembly optimizer allows you to write linear assembly code without being concerned with the
pipeline structure or with assigning registers. It accepts assembly code that has not been register-
allocated and is unscheduled. The assembly optimizer assigns registers and uses loop optimization to
turn linear assembly into highly parallel assembly that takes advantage of software pipelining. See
Chapter 4.
• The compiler accepts C/C++ source code and produces C6000 assembly language source code. See
Chapter 2.

SPRU187U – July 2012 Introduction to the Software Development Tools 15

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
C/C++ Compiler Overview www.ti.com

• The assembler translates assembly language source files into machine language relocatable object
files. The TMS320C6000 Assembly Language Tools User's Guide explains how to use the assembler.
• The linker combines relocatable object files into a single absolute executable object file. As it creates
the executable file, it performs relocation and resolves external references. The linker accepts
relocatable object files and object libraries as input. See Chapter 5. The TMS320C6000 Assembly
Language Tools User's Guide provides a complete description of the linker.
• The archiver allows you to collect a group of files into a single archive file, called a library.
Additionally, the archiver allows you to modify a library by deleting, replacing, extracting, or adding
members. One of the most useful applications of the archiver is building a library of object files. The
TMS320C6000 Assembly Language Tools User's Guide explains how to use the archiver.
• The run-time-support libraries contain the standard ISO C and C++ library functions, compiler-utility
functions, floating-point arithmetic functions, and C I/O functions that are supported by the compiler.
See Chapter 8.
You can use the library-build utility to build your own customized run-time-support library. See
Section 8.5. Source code for the standard run-time-support library functions for C and C++ are
provided in the self-contained rtssrc.zip file.
• The hex conversion utility converts an object file into other object formats. You can download the
converted file to an EPROM programmer. The TMS320C6000 Assembly Language Tools User's Guide
explains how to use the hex conversion utility and describes all supported formats.
• The absolute lister accepts linked object files as input and creates .abs files as output. You can
assemble these .abs files to produce a listing that contains absolute, rather than relative, addresses.
Without the absolute lister, producing such a listing would be tedious and would require many manual
operations. The TMS320C6000 Assembly Language Tools User's Guide explains how to use the
absolute lister.
• The cross-reference lister uses object files to produce a cross-reference listing showing symbols,
their definitions, and their references in the linked source files. The TMS320C6000 Assembly
Language Tools User's Guide explains how to use the cross-reference utility.
• The C++ name demangler is a debugging aid that converts names mangled by the compiler back to
their original names as declared in the C++ source code. As shown in Figure 1-1, you can use the C++
name demangler on the assembly file that is output by the compiler; you can also use this utility on the
assembler listing file and the linker map file. See Chapter 9.
• The disassembler decodes object files to show the assembly instructions that they represent. The
TMS320C6000 Assembly Language Tools User's Guide explains how to use the disassembler.
• The main product of this development process is an executable object file that can be executed in a
TMS320C6000 device. You can use one of several debugging tools to refine and correct your code.
Available products include:
– An instruction-level and clock-accurate software simulator
– An XDS emulator

1.2 C/C++ Compiler Overview

The following subsections describe the key features of the compiler.

1.2.1 ANSI/ISO Standard

The C and C++ language features in the compiler are implemented in conformance with these ISO
standards:
• ISO-standard C
The C/C++ compiler conforms to the C Standard ISO/IEC 9889:1990. The ISO standard supercedes
and is the same as the ANSI C standard. There is also a 1999 version of the ISO standard, but the TI
compiler conforms to the 1990 standard, not the 1999 standard. The language is also described in the
second edition of Kernighan and Ritchie's The C Programming Language (K&R).
• ISO-standard C++
The C/C++ compiler conforms to the C++ Standard ISO/IEC 14882:1998. The language is also
described in Ellis and Stroustrup's The Annotated C++ Reference Manual (ARM), but this is not the

16 Introduction to the Software Development Tools SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com C/C++ Compiler Overview

standard. The compiler also supports embedded C++. For a description of unsupported C++ features,
see Section 6.2.
• ISO-standard run-time support
The compiler tools come with an extensive run-time library. All library functions conform to the ISO
C/C++ library standard. The library includes functions for standard input and output, string
manipulation, dynamic memory allocation, data conversion, timekeeping, trigonometry, and exponential
and hyperbolic functions. Functions for signal handling are not included, because these are target-
system specific. For more information, see Chapter 8.

1.2.2 Output Files

These types of output files are created by the compiler:
• COFF object files
Common object file format (COFF) allows you to define your system's memory map at link time. This
maximizes performance by enabling you to link C/C++ code and data objects into specific memory
areas. COFF also supports source-level debugging.
• ELF object files
Executable and linking format (ELF) enables supporting modern language features like early template
instantiation and exporting inline functions.

1.2.3 Compiler Interface

These features enable interfacing with the compiler:
• Compiler program
The compiler tools include a compiler program (cl6x) that you use to compile, optimize, assemble, and
link programs in a single step. For more information, see Section 2.1
• Flexible assembly language interface
The compiler has straightforward calling conventions, so you can write assembly and C functions that
call each other. For more information, see Chapter 7.

1.2.4 Utilities
These features are compiler utilities:
• Library-build utility
The library-build utility lets you custom-build object libraries from source for any combination of run-
time models. For more information, see Section 8.5.
• C++ name demangler
The C++ name demangler (dem6x) is a debugging aid that translates each mangled name it detects in
compiler-generated assembly code, disassembly output, or compiler diagnostic messages to its
original name found in the C++ source code. For more information, see Chapter 9.
• Hex conversion utility
For stand-alone embedded applications, the compiler has the ability to place all code and initialization
data into ROM, allowing C/C++ code to run from reset. The COFFor ELF files output by the compiler
can be converted to EPROM programmer data files by using the hex conversion utility, as described in
the TMS320C6000 Assembly Language Tools User's Guide.

SPRU187U – July 2012 Introduction to the Software Development Tools 17

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 2
SPRU187U – July 2012

Using the C/C++ Compiler

The compiler translates your source program into machine language object code that the TMS320C6000
can execute. Source code must be compiled, assembled, and linked to create an executable object file. All
of these steps are executed at once by using the compiler.

Topic ........................................................................................................................... Page

2.1 About the Compiler ........................................................................................... 19

2.2 Invoking the C/C++ Compiler .............................................................................. 19
2.3 Changing the Compiler's Behavior With Options ................................................... 20
2.4 Controlling the Compiler Through Environment Variables ...................................... 39
2.5 Precompiled Header Support .............................................................................. 41
2.6 Controlling the Preprocessor .............................................................................. 42
2.7 Understanding Diagnostic Messages ................................................................... 45
2.8 Other Messages ................................................................................................ 48
2.9 Generating Cross-Reference Listing Information (--gen_acp_xref Option) ................ 48
2.10 Generating a Raw Listing File (--gen_acp_raw Option) ........................................... 48
2.11 Using Inline Function Expansion ......................................................................... 50
2.12 Interrupt Flexibility Options (--interrupt_threshold Option) ..................................... 53
2.13 Linking C6400 Code With C6200/C6700/Older C6400 Object Code ........................... 54
2.14 Using Interlist ................................................................................................... 54
2.15 Generating and Using Performance Advice .......................................................... 55
2.16 Controlling Application Binary Interface ............................................................... 55
2.17 Enabling Entry Hook and Exit Hook Functions ..................................................... 57

18 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com About the Compiler

2.1 About the Compiler

The compiler lets you compile, assemble, and optionally link in one step. The compiler performs the
following steps on one or more source modules:
• The compiler accepts C/C++ source code and assembly code, and produces object code.
You can compile C, C++, and assembly files in a single command. The compiler uses the filename
extensions to distinguish between different file types. See Section 2.3.9 for more information.
• The linker combines object files to create a static executable or dynamic object file. The linker is
optional, so you can compile and assemble many modules independently and link them later. See
Chapter 5 for information about linking the files.

Invoking the Linker

NOTE: By default, the compiler does not invoke the linker. You can invoke the linker by using the --
run_linker compiler option.

For a complete description of the assembler and the linker, see the TMS320C6000 Assembly Language
Tools User's Guide.

2.2 Invoking the C/C++ Compiler

To invoke the compiler, enter:

cl6x [options] [filenames] [--run_linker [link_options] object files]]

cl6x Command that runs the compiler and the assembler.

options Options that affect the way the compiler processes input files. The options are
listed in Table 2-5 through Table 2-32.
filenames One or more C/C++ source files, assembly language source files, linear
assembly files, or object files.
--run_linker Option that invokes the linker. The --run_linker option's short form is -z. See
Chapter 5 for more information.
link_options Options that control the linking process.
object files Name of the additional object files for the linking process.

The arguments to the compiler are of three types:

• Compiler options
• Link options
• Filenames
The --run_linker option indicates linking is to be performed. If the --run_linker option is used, any compiler
options must precede the --run_linker option, and all link options must follow the --run_linker option.
Source code filenames must be placed before the --run_linker option. Additional object file filenames can
be placed after the --run_linker option.
For example, if you want to compile two files named symtab.c and file.c, assemble a third file named
seek.asm, and link to create an executable program called myprogram.out, you will enter:
cl6x symtab.c file.c seek.asm --run_linker --library=lnk.cmd
--library=rts6200.lib --output_file=myprogram.out

SPRU187U – July 2012 Using the C/C++ Compiler 19

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

2.3 Changing the Compiler's Behavior With Options

Options control the operation of the compiler. This section provides a description of option conventions
and an option summary table. It also provides detailed descriptions of the most frequently used options,
including options used for type-checking and assembling.
For a help screen summary of the options, enter cl6x with no parameters on the command line.
The following apply to the compiler options:
• Options are preceded by one or two hyphens.
• Options are case sensitive.
• Options are either single letters or sequences of characters.
• Individual options cannot be combined.
• An option with a required parameter should be specified with an equal sign before the parameter to
clearly associate the parameter with the option. For example, the option to undefine a constant can be
expressed as --undefine=name. Although not recommended, you can separate the option and the
parameter with or without a space, as in --undefine name or -undefinename.
• An option with an optional parameter should be specified with an equal sign before the parameter to
clearly associate the parameter with the option. For example, the option to specify the maximum
amount of optimization can be expressed as -O=3. Although not recommended, you can specify the
parameter directly after the option, as in -O3. No space is allowed between the option and the optional
parameter, so -O 3 is not accepted.
• Files and options except the --run_linker option can occur in any order. The --run_linker option must
follow all other compile options and precede any link options.
You can define default options for the compiler by using the C6X_C_OPTION environment variable. For a
detailed description of the environment variable, see Section 2.4.1.
Table 2-5 through Table 2-32 summarize all options (including link options). Use the references in the
tables for more complete descriptions of the options.

Table 2-1. Processor Options

Option Alias Effect Section
--silicon_version=id -mv Selects target version. Defaults to 6200. Section 2.3.4
--big_endian -me Produces object code in big-endian format Section 2.13
--abi={coffabi|eabi} Selects application binary interface. Default is coffabi. Section 2.16

Table 2-2. Optimization Options (1)

Option Alias Effect Section
--opt_level=0 -O0 Optimizes register usage Section 3.1
--opt_level=1 -O1 Uses -O0 optimizations and optimizes locally Section 3.1
--opt_level=2 -O2 or -O Uses -O1 optimizations and optimizes globally (default) Section 3.1
--opt_level=3 -O3 Uses -O2 optimizations and optimizes the file Section 3.1
Section 3.6
--opt_for_space=n -ms Controls code size on four levels (0, 1, 2, and 3) Section 3.5

(1)
Note: Machine-specific options (see Table 2-11) can also affect optimization.

Table 2-3. Debug Options

Option Alias Effect Section
--symdebug:dwarf -g Enables symbolic debugging Section 2.3.5
Section 3.14.3.1
--symdebug:coff Enables symbolic debugging using the alternate STABS debugging Section 2.3.5
format. STABS format is not supported for C6400+ or C674x, or Section 3.14.3.1
when using ELF.
--symdebug:none Disables all symbolic debugging Section 2.3.5

20 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

Table 2-3. Debug Options (continued)

Option Alias Effect Section
--symdebug:profile_coff Enables profiling using the alternate STABS debugging format. Section 2.3.5
STABS format is not supported for C6400+ or C674x, or when using
ELF.
--symdebug:skeletal Enables minimal symbolic debugging that does not hinder Section 2.3.5
optimizations (default behavior)
--disable_push_pop Disables push-pop optimization --
--optimize_with_debug -mn Reenables optimizations disabled with --symdebug:dwarf Section 3.14.3.1
--symdebug:dwarf_version=2|3 Specifies the DWARF format version Section 2.3.5

Table 2-4. Include Options

Option Alias Effect Section
--include_path=directory -I Defines #include search path Section 2.6.2.1
--preinclude=filename Includes filename at the beginning of compilation Section 2.3.2

Table 2-5. Control Options

Option Alias Effect Section
--compile_only -c Disables linking (negates --run_linker) Section 5.1.3
--help -h Prints (on the standard output device) a description of the options Section 2.3.1
understood by the compiler.
--run_linker -z Enables linking Section 2.3.1
--skip_assembler -n Compiles or assembly optimizes only Section 2.3.1

Table 2-6. Advanced Debug Options

Option Alias Effect Section
--machine_regs Displays reg operands as machine registers in assembly code Section 2.3.11
--symdebug:keep_all_types Keep unreferenced type information (default for ELF with debug) Section 2.3.5

Table 2-7. Language Options

Option Alias Effect Section
--cpp_default -fg Processes all source files with a C extension as C++ source files. Section 2.3.7
--create_pch=filename Creates a precompiled header file with the name specified Section 2.5
--embedded_cpp -pe Enables embedded C++ mode Section 6.14.3
--exceptions Enables C++ exception handling Section 6.6
--extern_c_can_throw Allow extern C functions to propagate exceptions (EABI only) --
--gcc Enables support for GCC extensions Section 6.15
--gen_acp_raw -pl Generates a raw listing file Section 2.10
--gen_acp_xref -px Generates a cross-reference listing file Section 2.9
--keep_unneeded_statics Keeps unreferenced static variables. Section 2.3.2
--kr_compatible -pk Allows K&R compatibility Section 6.14.1
--multibyte_chars -pc Enables support for multibyte character sequences in comments, --
string literals and character constants.
--no_inlining -pi Disables definition-controlled inlining (but --opt_level=3 (or -O3) Section 2.11
optimizations still perform automatic inlining)
--no_intrinsics -pn Disables intrinsic functions. No predefinition of compiler-supplied --
intrinsic functions.
--pch Creates or uses precompiled header files Section 2.5

SPRU187U – July 2012 Using the C/C++ Compiler 21

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

Table 2-7. Language Options (continued)

Option Alias Effect Section
--pch_dir=directory Specifies the path where the precompiled header file resides Section 2.5.2
--pch_verbose Displays a message for each precompiled header file that is Section 2.5.3
considered but not used
--program_level_compile -pm Combines source files to perform program-level optimization Section 3.7
--relaxed_ansi -pr Enables relaxed mode; ignores strict ISO violations Section 6.14.2
--rtti -rtti Enables run time type information (RTTI) –-
--static_template_instantiation Instantiate all template entities with internal linkage –-
--strict_ansi -ps Enables strict ISO mode (for C/C++, not K&R C) Section 6.14.2
--use_pch=filename Specifies the precompiled header file to use for this compilation Section 2.5.2

Table 2-8. Parser Preprocessing Options

Option Alias Effect Section
--preproc_dependency[=filename] -ppd Performs preprocessing only, but instead of writing preprocessed Section 2.6.7
output, writes a list of dependency lines suitable for input to a
standard make utility
--preproc_includes[=filename] -ppi Performs preprocessing only, but instead of writing preprocessed Section 2.6.8
output, writes a list of files included with the #include directive
--preproc_macros[=filename] -ppm Performs preprocessing only. Writes list of predefined and user- Section 2.6.9
defined macros to a file with the same name as the input but with a
.pp extension.
--preproc_only -ppo Performs preprocessing only. Writes preprocessed output to a file Section 2.6.3
with the same name as the input but with a .pp extension.
--preproc_with_comment -ppc Performs preprocessing only. Writes preprocessed output, keeping Section 2.6.5
the comments, to a file with the same name as the input but with a
.pp extension.
--preproc_with_compile -ppa Continues compilation after preprocessing Section 2.6.4
--preproc_with_line -ppl Performs preprocessing only. Writes preprocessed output with line- Section 2.6.6
control information (#line directives) to a file with the same name as
the input but with a .pp extension.

Table 2-9. Predefined Symbols Options

Option Alias Effect Section
--define=name[=def] -D Predefines name Section 2.3.1
--undefine=name -U Undefines name Section 2.3.1

Table 2-10. Diagnostics Options

Option Alias Effect Section
--compiler_revision Prints out the compiler release revision and exits --
--diag_error=num -pdse Categorizes the diagnostic identified by num as an error Section 2.7.1
--diag_remark=num -pdsr Categorizes the diagnostic identified by num as a remark Section 2.7.1
--diag_suppress=num -pds Suppresses the diagnostic identified by num Section 2.7.1
--diag_warning=num -pdsw Categorizes the diagnostic identified by num as a warning Section 2.7.1
--display_error_number -pden Displays a diagnostic's identifiers along with its text Section 2.7.1
--emit_warnings_as_errors -pdew Treat warnings as errors Section 2.7.1
--issue_remarks -pdr Issues remarks (nonserious warnings) Section 2.7.1
--no_warnings -pdw Suppresses warning diagnostics (errors are still issued) Section 2.7.1
--quiet -q Suppresses progress messages (quiet) --
--set_error_limit=num -pdel Sets the error limit to num. The compiler abandons compiling after Section 2.7.1
this number of errors. (The default is 100.)

22 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

Table 2-10. Diagnostics Options (continued)

Option Alias Effect Section
--super_quiet -qq Super quiet mode --
--tool_version -version Displays version number for each tool --
--verbose Display banner and function progress information --
--verbose_diagnostics -pdv Provides verbose diagnostics that display the original source with Section 2.7.1
line-wrap
--write_diagnostics_file -pdf Generates a diagnostics information file. Compiler only option. Section 2.7.1

Table 2-11. Run-Time Model Options

Option Alias Effect Section
--debug_software_pipeline -mw Produce verbose software pipelining report Section 3.2.2
--disable_software_pipelining -mu Turns off software pipelining Section 3.2.1
--dprel Specifies that all non-const data is addressed using DP-relative Section 7.1.5.2
addressing
f--fp_not_associative -mc Prevents reordering of associative floating-point operations Section 3.11
--fp_reassoc={on|off} Enables or disables the reassociation of floating-point arithmetic Section 2.3.3
--gen_func_subsections={on|off} -mo Puts each function in a separate subsection in the object file Section 5.2.1
--interrupt_threshold[=num] -mi Specifies an interrupt threshold value Section 2.12
--mem_model:const= Allows const objects to be made far independently of the -- Section 7.1.5.3
{far_aggregates|far|data} mem_model:data option
--mem_model:data= Determines data access model Section 7.1.5.1
{far_aggregates|near|far}
--no_bad_aliases -mt Allows certain assumptions about aliasing and loops Section 3.10.2
Section 4.6.2
--no_compress Prevents compression on C6400+, C6740, and C6600
--no_reload_errors Turns off all reload-related loop buffer error messages for C6400+, -
C6740, and C6600
--profile:breakpt Enables breakpoint-based profiling Section 2.3.5
Section 3.14.3.2
--profile:power Enables power profiling Section 2.3.5
Section 3.14.3.2
--sat_reassoc={on|off} Enables or disables the reassociation of saturating arithmetic. Section 2.3.2
Default is --sat_reassoc=off.
--small_enum --small-enum Uses the smallest possible size for the enumeration type Section 2.3.3
--speculate_loads=n -mh Specifies speculative load byte count threshold. Allows Section 3.2.3.1
speculative execution of loads with bounded address ranges.
--speculate_unknown_loads Allows speculative execution of loads with unbounded addresses Section 2.3.3
--target_compatiblity_6200 -mb Enables C62xx compatibility with C6400 code Section 2.13
--use_const_for_alias_analysis -ox Uses const to disambiguate pointers Section 2.3.3
--wchar_t={32|16} Sets the size of the C/C++ type wchar_t. Default is 16 bits. Section 2.3.3

SPRU187U – July 2012 Using the C/C++ Compiler 23

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com
(1)
Table 2-12. Advanced Optimization Options
Option Alias Effect Section
--auto_inline=[size] -oi Sets automatic inlining size (--opt_level=3 only). If size is not Section 3.13
specified, the default is 1.
--call_assumptions=0 -op0 Specifies that the module contains functions and variables that are Section 3.7.1
called or modified from outside the source code provided to the
compiler
--call_assumptions=1 -op1 Specifies that the module contains variables modified from outside Section 3.7.1
the source code provided to the compiler but does not use functions
called from outside the source code
--call_assumptions=2 -op2 Specifies that the module contains no functions or variables that are Section 3.7.1
called or modified from outside the source code provided to the
compiler (default)
--call_assumptions=3 -op3 Specifies that the module contains functions that are called from Section 3.7.1
outside the source code provided to the compiler but does not use
variables modified from outside the source code
--gen_opt_info=0 -on0 Disables the optimization information file Section 3.6.2
--gen_opt_info=1 -on1 Produces an optimization information file Section 3.6.2
--gen_opt_info=2 -on2 Produces a verbose optimization information file Section 3.6.2
--openmp --omp Enables support for OpenMP 3.0 Section 7.9.1
--opt_for_speed[=n] -mf Controls speed over space (0-5 range) (Default is 4.) Section 3.14.4
--optimizer_interlist -os Interlists optimizer comments with assembly statements Section 3.14.2
--remove_hooks_when_inlining Removes entry/exit hooks for auto-inlined functions Section 2.17
--single_inline Inlines functions that are only called once --
--aliased_variables -ma Assumes called functions create hidden aliases (rare) Section 3.10.1

(1)
Note: Machine-specific options (see Table 2-11) can also affect optimization.

Table 2-13. Entry/Exit Hook Options

Option Alias Effect Section
--entry_hook[=name] Enables entry hooks Section 2.17
--entry_parm={none|name| Specifies the parameters to the function to the --entry_hook option Section 2.17
address}
--exit_hook[=name] Enables exit hooks Section 2.17
--exit_parm={none|name|address} Specifies the parameters to the function to the --exit_hook option Section 2.17

Table 2-14. Feedback Options

Option Alias Effect Section
--analyze={codecov|callgraph} Generate analysis info from profile data Section 3.9.4.2
--analyze_only Only generate analysis Section 3.9.4.2
--gen_profile_info Generates instrumentation code to collect profile information Section 3.8.1.3
--use_profile_info=file1[, file2,...] Specifies the profile information file(s) Section 3.8.1.3

Table 2-15. Library Function Assumptions Options

Option Alias Effect Section
--printf_support={nofloat|full| Enables support for smaller, limited versions of the printf and sprintf Section 2.3.2
minimal} run-time-support functions.
--std_lib_func_defined -ol1 or -oL1 Informs the optimizer that your file declares a standard library Section 3.6.1
function
--std_lib_func_not_defined -ol2 or -oL2 Informs the optimizer that your file does not declare or alter library Section 3.6.1
functions. Overrides the -ol0 and -ol1 options (default).
--std_lib_func_redefined -ol0 or -oL0 Informs the optimizer that your file alters a standard library function Section 3.6.1

24 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

Table 2-16. Assembler Options

Option Alias Effect Section
--keep_asm -k Keeps the assembly language (.asm) file Section 2.3.11
--asm_listing -al Generates an assembly listing file Section 2.3.11
--c_src_interlist -ss Interlists C source and assembly statements Section 2.14
Section 3.14.2
--src_interlist -s Interlists optimizer comments (if available) and assembly source Section 2.3.1
statements; otherwise interlists C and assembly source statements
--absolute_listing -aa Enables absolute listing Section 2.3.11
--asm_define=name[=def] -ad Sets the name symbol Section 2.3.11
--asm_dependency -apd Performs preprocessing; lists only assembly dependencies Section 2.3.11
--asm_includes -api Performs preprocessing; lists only included #include files Section 2.3.11
--asm_undefine=name -au Undefines the predefined constant name Section 2.3.11
--copy_file=filename -ahc Copies the specified file for the assembly module Section 2.3.11
--cross_reference -ax Generates the cross-reference file Section 2.3.11
--include_file=filename -ahi Includes the specified file for the assembly module Section 2.3.11
--no_const_clink Stops generation of .clink directives for const global arrays. Section 2.3.2
--output_all_syms -as Puts labels in the symbol table Section 2.3.11
--strip_coff_underscore Aids in transitioning hand-coded assembly from COFF to EABI
--syms_ignore_case -ac Makes case insignificant in assembly source files Section 2.3.11

Table 2-17. File Type Specifier Options

Option Alias Effect Section
--ap_file=filename -fl Identifies filename as a linear assembly source file regardless of its Section 2.3.7
extension. By default, the compiler and assembly optimizer treat .sa
files as linear assembly source files.
--asm_file=filename -fa Identifies filename as an assembly source file regardless of its Section 2.3.7
extension. By default, the compiler and assembler treat .asm files as
assembly source files.
--c_file=filename -fc Identifies filename as a C source file regardless of its extension. By Section 2.3.7
default, the compiler treats .c files as C source files.
--cpp_file=filename -fp Identifies filename as a C++ file, regardless of its extension. By Section 2.3.7
default, the compiler treats .C, .cpp, .cc and .cxx files as a C++ files.
--obj_file=filename -fo Identifies filename as an object code file regardless of its extension. Section 2.3.7
By default, the compiler and linker treat .obj files as object code files.

Table 2-18. Directory Specifier Options

Option Alias Effect Section
--abs_directory=directory -fb Specifies an absolute listing file directory. By default, the compiler Section 2.3.10
uses the .obj directory.
--asm_directory=directory -fs Specifies an assembly file directory. By default, the compiler uses Section 2.3.10
the current directory.
--list_directory=directory -ff Specifies an assembly listing file and cross-reference listing file Section 2.3.10
directory By default, the compiler uses the .obj directory.
--obj_directory=directory -fr Specifies an object file directory. By default, the compiler uses the Section 2.3.10
current directory.
--output_file=filename -fe Specifies a compilation output file name; can override -- Section 2.3.10
obj_directory.
--pp_directory=dir Specifies a preprocessor file directory. By default, the compiler uses Section 2.3.10
the current directory.
--temp_directory=directory -ft Specifies a temporary file directory. By default, the compiler uses the Section 2.3.10
current directory.

SPRU187U – July 2012 Using the C/C++ Compiler 25

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

Table 2-19. Default File Extensions Options

Option Alias Effect Section
--ap_extension=[.]extension -el Sets a default extension for linear assembly source files. Section 2.3.9
--asm_extension=[.]extension -ea Sets a default extension for assembly source files Section 2.3.9
--c_extension=[.]extension -ec Sets a default extension for C source files Section 2.3.9
--cpp_extension=[.]extension -ep Sets a default extension for C++ source files Section 2.3.9
--listing_extension=[.]extension -es Sets a default extension for listing files Section 2.3.9
--obj_extension=[.]extension -eo Sets a default extension for object files Section 2.3.9

Table 2-20. Dynamic Linking Support Compiler Options (1)

Option Alias Description
--dsbt Generates addressing via Dynamic Segment Base Table
--export_all_cpp_vtbl Exports C++ virtual tables by default
--import_helper_functions Treats compiler helper functions as imported references
--import_undef[={off|on}] Imports all undefined symbols. Default is on.
--inline_plt[={off|on}] Inlines the import function call stub. Default is on.
--linux Generates code for Linux
--pic[={near|far}] Generates position independent addressing for a shared object. Default is near.
--visibility={hidden|fhidden|default| Specifies visibility of global symbols
protected}

(1)
See Section 2.3.12 for more information.

Table 2-21. Command Files Options

Option Alias Effect Section
--cmd_file=filename -@ Interprets contents of a file as an extension to the command line. Section 2.3.1
Multiple -@ instances can be used.

Table 2-22. MISRA-C:2004 Options

Option Alias Effect Section
--check_misra[={all|required| Enables checking of the specified MISRA-C:2004 rules. Default is Section 2.3.2
advisory|none|rulespec}] all.
--misra_advisory={error|warning| Sets the diagnostic severity for advisory MISRA-C:2004 rules Section 2.3.2
remark|suppress}
--misra_required={error|warning| Sets the diagnostic severity for required MISRA-C:2004 rules Section 2.3.2
remark|suppress}

Table 2-23. Performance Advisor Options

Option Alias Effect Section
--advice:performance[={all|none}] Generates compiler optimization advice. Default is all. Section 2.15
--advice:performance_file={stdout| Specifies that advice be written to stdout, stderr, or a file. Section 2.15
stderr|user_specified_filename}
--advice:performance_dir= Specifies that advice file be created in the named directory. Section 2.15
{user_specified_directory_name}

The following tables list the linker options. See the TMS320C6000 Assembly Language Tools User's
Guide for details on these options.

26 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

Table 2-24. Linker Basic Options

Option Alias Description
--output_file=file -o Names the executable output file. The default filename is a.out.
--map_file=file -m Produces a map or listing of the input and output sections, including holes, and places
the listing in filename
--stack_size=size [-]-stack Sets C system stack size to size bytes and defines a global symbol that specifies the
stack size. Default = 1K bytes
--heap_size=size [-]-heap Sets heap size (for the dynamic memory allocation in C) to size bytes and defines a
global symbol that specifies the heap size. Default = 1K bytes

Table 2-25. File Search Path Options

Option Alias Description
--library=file -l Names an archive library or link command file as linker input
--search_path=pathname -I Alters library-search algorithms to look in a directory named with pathname before
looking in the default location. This option must appear before the --library option.
--priority -priority Satisfies unresolved references by the first library that contains a definition for that
symbol
--reread_libs -x Forces rereading of libraries, which resolves back references
--disable_auto_rts Disables the automatic selection of a run-time-support library

Table 2-26. Command File Preprocessing Options

Option Alias Description
--define=name=value Predefines name as a preprocessor macro.
--undefine=name Removes the preprocessor macro name.
--disable_pp Disables preprocessing for command files

Table 2-27. Diagnostic Options

Option Alias Description
--diag_error=num Categorizes the diagnostic identified by num as an error
--diag_remark=num Categorizes the diagnostic identified by num as a remark
--diag_suppress=num Suppresses the diagnostic identified by num
--diag_warning=num Categorizes the diagnostic identified by num as a warning
--display_error_number Displays a diagnostic's identifiers along with its text
--emit_warnings_as_errors -pdew Treat warnings as errors
--issue_remarks Issues remarks (nonserious warnings)
--no_demangle Disables demangling of symbol names in diagnostics
--no_warnings Suppresses warning diagnostics (errors are still issued)
--set_error_limit=count Sets the error limit to count. The linker abandons linking after this number of errors. (The
default is 100.)
--verbose_diagnostics Provides verbose diagnostics that display the original source with line-wrap
--warn_sections -w Displays a message when an undefined output section is created

Table 2-28. Linker Output Options

Option Alias Description
--absolute_exe -a Produces an absolute, executable object file. This is the default; if neither --absolute_exe
nor --relocatable is specified, the linker acts as if --absolute_exe were specified.
--mapfile_contents=attribute Controls the information that appears in the map file.
--relocatable -r Produces a nonexecutable, relocatable output object file
--rom Creates a ROM object

SPRU187U – July 2012 Using the C/C++ Compiler 27

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

Table 2-28. Linker Output Options (continued)

Option Alias Description
--run_abs -abs Produces an absolute listing file
--xml_link_info=file Generates a well-formed XML file containing detailed information about the result of a
link

28 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

Table 2-29. Symbol Management Options

Option Alias Description
--entry_point=symbol -e Defines a global symbol that specifies the primary entry point for the executable object
file
--globalize=pattern Changes the symbol linkage to global for symbols that match pattern
--hide=pattern Hides symbols that match the specified pattern
--localize=pattern Make the symbols that match the specified pattern local
--make_global=symbol -g Makes symbol global (overrides -h)
--make_static -h Makes all global symbols static
--no_sym_merge -b Disables merge of symbolic debugging information in COFF object files
--no_symtable -s Strips symbol table information and line number entries from the executable object file
--retain={symbol| Specifies a symbol or section to be retained by the linker
section specification}
--scan_libraries -scanlibs Scans all libraries for duplicate symbol definitions
--symbol_map=refname=defname Specifies a symbol mapping; references to the refname symbol are replaced with
references to the defname symbol
--undef_sym=symbol -u Adds symbol to the symbol table as an unresolved symbol
--unhide=pattern Excludes symbols that match the specified pattern from being hidden

Table 2-30. Run-Time Environment Options

Option Alias Description
--arg_size=size --args Reserve size bytes for the argc/argv memory area
--fill_value=value -f Sets default fill value for holes within output sections
--ram_model -cr Initializes variables at load time
--rom_model -c Autoinitializes variables at run time
--trampolines[=off|on] Generates far call trampolines. Default is on.

Table 2-31. Link-Time Optimization Options

Option Description
--cinit_compression[=compression_kind] Specifies the type of compression to apply to the c auto initialization data. Default is rle.
--compress_dwarf[=off|on] Aggressively reduces the size of DWARF information from input object files. Default is
on.
--copy_compression[=compression_kind] Compresses data copied by linker copy tables. Default is rle.
--unused_section_elimination[=off|on] Eliminates sections that are not needed in the executable module. Default is on.

Table 2-32. Miscellaneous Options

Option Alias Description
--disable_clink -j Disables conditional linking of COFF object files
--linker_help [-]-help Displays information about syntax and available options
--minimize_trampolines[=off| Places sections to minimize number of far trampolines required. Default is postorder.
postorder]
--preferred_order=function Prioritizes placement of functions
--strict_compatibility[=off|on] Performs more conservative and rigorous compatibility checking of input object files.
Default is on.
--trampoline_min_spacing=size When trampoline reservations are spaced more closely than the specified limit, tries to
make them adjacent
--zero_init=[off|on] Controls preinitialization of uninitialized variables. Default is on.

SPRU187U – July 2012 Using the C/C++ Compiler 29

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

Table 2-33. Dynamic Linking Support Options

Option Description
--bind_now Disables Linux default lazy binding mechanisms
--dsbt_index=index Specifies the Data Segment Base Table (DSBT) index of this component
--dsbt_size=size Specifies the size of the DSBT in words
--dynamic[=exe|lib]] Generates dynamic executable or a dynamic library. Default is .exe.
--export=symbol Specifies symbol exported by ELF object
--fini=symbol Specifies symbol name of termination code
--forced_static_binding[=off|on] Forces all import references to bind during static linking; defaults to on
--import=symbol Specifies symbol imported by ELF object
--init=symbol Specifies symbol name of termination code
--rpath=dir Adds directory to beginning of library search path
--runpath=dir Adds directory to end of library search path
--soname=soname Specifies ELF shared object file name

2.3.1 Frequently Used Options

Following are detailed descriptions of options that you will probably use frequently:

--c_src_interlist Invokes the interlist feature, which interweaves original C/C++ source
with compiler-generated assembly language. The interlisted C
statements may appear to be out of sequence. You can use the interlist
feature with the optimizer by combining the --optimizer_interlist and --
c_src_interlist options. See Section 3.14.2. The --c_src_interlist option
can have a negative performance and/or code size impact.
--cmd_file=filename Appends the contents of a file to the option set. You can use this option
to avoid limitations on command line length or C style comments
imposed by the host operating system. Use a # or ; at the beginning of a
line in the command file to include comments. You can also include
comments by delimiting them with /* and */. To specify options, surround
hyphens with quotation marks. For example, "--"quiet.
You can use the --cmd_file option multiple times to specify multiple files.
For instance, the following indicates that file3 should be compiled as
source and file1 and file2 are --cmd_file files:
cl6x --cmd_file=file1 --cmd_file=file2 file3
--compile_only Suppresses the linker and overrides the --run_linker option, which
specifies linking. The --compile_only option's short form is -c. Use this
option when you have --run_linker specified in the C6X_C_OPTION
environment variable and you do not want to link. See Section 5.1.3.
--define=name[=def] Predefines the constant name for the preprocessor. This is equivalent to
inserting #define name def at the top of each C source file. If the
optional[=def] is omitted, the name is set to 1. The --define option's short
form is -D.
If you want to define a quoted string and keep the quotation marks, do
one of the following:
• For Windows, use --define=name="\"string def\"". For example, --
define=car="\"sedan\""
• For UNIX, use --define=name='"string def"'. For example, --
define=car='"sedan"'
• For Code Composer Studio, enter the definition in a file and include
that file with the --cmd_file option.

30 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

--help Displays the syntax for invoking the compiler and lists available options.
If the --help option is followed by another option or phrase, detailed
information about the option or phrase is displayed. For example, to see
information about debugging options use --help debug.
--include_path=directory Adds directory to the list of directories that the compiler searches for
#include files. The --include_path option's short form is -I. You can use
this option several times to define several directories; be sure to
separate the --include_path options with spaces. If you do not specify a
directory name, the preprocessor ignores the --include_path option. See
Section 2.6.2.1.
--keep_asm Retains the assembly language output from the compiler or assembly
optimizer. Normally, the compiler deletes the output assembly language
file after assembly is complete. The --keep_asm option's short form is -k.
--quiet Suppresses banners and progress information from all the tools. Only
source filenames and error messages are output. The --quiet option's
short form is -q.
--run_linker Runs the linker on the specified object files. The --run_linker option and
its parameters follow all other options on the command line. All
arguments that follow --run_linker are passed to the linker. The --
run_linker option's short form is -z. See Section 5.1.
--skip_assembler Compiles only. The specified source files are compiled but not
assembled or linked. The --skip_assembler option's short form is -n. This
option overrides --run_linker. The output is assembly language output
from the compiler.
--src_interlist Invokes the interlist feature, which interweaves optimizer comments or
C/C++ source with assembly source. If the optimizer is invoked (--
opt_level=n option), optimizer comments are interlisted with the
assembly language output of the compiler, which may rearrange code
significantly. If the optimizer is not invoked, C/C++ source statements are
interlisted with the assembly language output of the compiler, which
allows you to inspect the code generated for each C/C++ statement. The
--src_interlist option implies the --keep_asm option. The --src_interlist
option's short form is -s.
--tool_version Prints the version number for each tool in the compiler. No compiling
occurs.
--undefine=name Undefines the predefined constant name. This option overrides any --
define options for the specified constant. The --undefine option's short
form is -U.
--verbose Displays progress information and toolset version while compiling.
Resets the --quiet option.

2.3.2 Miscellaneous Useful Options

Following are detailed descriptions of miscellaneous options:

--check_misra={all|required| Displays the specified amount or type of MISRA-C documentation.

advisory|none|rulespec} The rulespec parameter is a comma-separated list of specifiers. See
Section 6.3 for details.
--fp_reassoc={on|off} Enables or disables the reassociation of floating-point arithmetic. If --
strict_ansi is set, --fp_reassoc=off is set since reassociation of
floating-point arithmetic is an ANSI violation.

SPRU187U – July 2012 Using the C/C++ Compiler 31

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

--keep_unneeded_statics Does not delete unreferenced static variables. The parser by default
remarks about and then removes any unreferenced static variables.
The --keep_unneeded_statics option keeps the parser from deleting
unreferenced static variables and any static functions that are
referenced by these variable definitions. Unreferenced static functions
will still be removed.
--no_const_clink Tells the compiler to not generate .clink directives for const global
arrays. By default, these arrays are placed in a .const subsection and
conditionally linked.
--misra_advisory={error| Sets the diagnostic severity for advisory MISRA-C:2004 rules.
warning|remark|suppress}
--misra_required={error| Sets the diagnostic severity for required MISRA-C:2004 rules.
warning|remark|suppress}
--preinclude=filename Includes the source code of filename at the beginning of the
compilation. This can be used to establish standard macro definitions.
The filename is searched for in the directories on the include search
list. The files are processed in the order in which they were specified.
--printf_support={full| Enables support for smaller, limited versions of the printf and sprintf
nofloat|minimal} run-time-support functions. The valid values are:
• full: Supports all format specifiers. This is the default.
• nofloat: Excludes support for printing and scanning floating-point
values. Supports all format specifiers except %f, %F, %g, %G, %e,
and %E.
• minimal: Supports the printing and scanning of integer, char, or
string values without width or precision flags. Specifically, only
the %%, %d, %o, %c, %s, and %x format specifiers are supported
There is no run-time error checking to detect if a format specifier is
used for which support is not included. The --printf_support option
precedes the --run_linker option, and must be used when performing
the final link.
--sat_reassoc={on|off} Enables or disables the reassociation of saturating arithmetic.

2.3.3 Run-Time Model Options

These options are specific to the TMS302C6000 toolset. See the referenced sections for more
information. TMS320C6000-specific assembler options are listed in Section 2.3.11.

--abi={eabi|coffabi} Specifies application binary interface (ABI). Default support is for

COFF ABI. See Section 2.16. Also, refer to The C6000 Embedded
Application Binary Interface Application Report (SPRAB89).
All code in an EABI application must be built for EABI. Make sure all
your libraries are available in EABI mode before migrating your
existing COFF ABI systems to C6000 EABI. See
http://tiexpressdsp.com/index.php/EABI_Support_in_C6000_Compiler
for full details.
--big_endian Produces code in big-endian format. By default, little-endian code is
produced.
--advice:performance Generates compile-time optimization Advice. See Section 2.15 .
--debug_software_pipeline Produces verbose software pipelining report. See Section 3.2.2.
--disable_software_pipelining Turns off software pipelining. See Section 3.2.1.
--fp_not_associative Compiler does not reorder floating-point operations. See
Section 3.11.

32 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

--interrupt_threshold=n Specifies an interrupt threshold value n that sets the maximum cycles
the compiler can disable interrupts. See Section 2.12.
--mem_model:const=type Allows const objects to be made far independently of the --
mem_model:data option. The type can be data, far, or
far_aggregates. See Section 7.1.5.3
--mem_model:data=type Specifies data access model as type far, far_aggregates, or near.
Default is far_aggregates. See Section 7.1.5.1.
--silicon_version=num Selects the target CPU version. See Section 2.3.4.
--small_enum By default, the C6000 compiler uses 32 bits for every enum. When
you use the --small_enum option, the smallest possible byte size for
the enumeration type is used. For example, enum example_enum
{first = -128, second = 0, third = 127} uses only one byte instead of 32
bits when the --small_enum option is used. Similarly, enum
a_short_enum {bottom = -32768, middle = 0, top = 32767} fits into
two bytes instead of four. Do not link object files compiled with the --
small_enum option with object files that have been compiled without
it. If you use the --small_enum option, you must use it with all of your
C/C++ files; otherwise, you will encounter errors that cannot be
detected until run time.
--speculate_loads=n Specifies speculative load byte count threshold. Allows speculative
execution of loads with bounded addresses. See Section 3.2.3.1.
--speculate_unknown_loads Allows speculative execution of loads with unbounded addresses.
--target_compatibility_6200 Compiles C6400 code that is compatible with array alignment
restrictions of version 4.0 tools or C6200/C6700 object code. This
option is deprecated. See Section 2.13
--use_const_for_alias_analysis Uses const to disambiguate pointers.
--wchar_t={32|16} Sets the size (in bits) of the C/C++ type wchar_t. The --abi=eabi
option is required when -wchar_t=32 is used. By default the compiler
generates 16-bit wchar_t. In COFF ABI mode, a warning is generated
and --wchar_t=32 is ignored. 16-bit wchar_t objects are not
compatible with the 32-bit wchar_t objects; an error is generated if
they are combined. When the --linux option is specified, it implies --
wchar_t=32 since Linux uses 32-bit extended characters.

2.3.4 Selecting Target CPU Version (--silicon_version Option)

Select the target CPU version using the last four digits of the TMS320C6000 part number. This selection
controls the use of target-specific instructions and alignment, such as --silicon_version=6701 or --
silicon_version=6412. Alternatively, you can also specify the family of the part, for example, --
silicon_version=6400 or --silicon_version=6700. If this option is not used, the compiler generates code for
the C6200 parts. If the --silicon_version option is not specified, the code generated runs on all C6000
parts; however, the compiler does not take advantage of target-specific instructions or alignment. This
option has the alias -mv. Common target CPU version options include:
• -mv6200
• -mv6700
• -mv6700+
• -mv6400+
• -mv6740
• -mv6600

SPRU187U – July 2012 Using the C/C++ Compiler 33

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

2.3.5 Symbolic Debugging and Profiling Options

The following options are used to select symbolic debugging or profiling:

--profile:breakpt Disables optimizations that would cause incorrect behavior when using a
breakpoint-based profiler.
--profile:power Enables power profiling support by inserting NOPs into the frame code.
These NOPs can then be instrumented by the power profiling tooling to
track the power usage of functions. If the power profiling tool is not used,
this option increases the cycle count of each function because of the
NOPs. The --profile:power option also disables optimizations that cannot
be handled by the power-profiler.
--symdebug:coff Enables symbolic debugging using the alternate STABS debugging
format. This may be necessary to allow debugging with older debuggers
or custom tools, which do not read the DWARF format. STABS format is
not supported for C6400+ or ELF.
--symdebug:dwarf Generates directives that are used by the C/C++ source-level debugger
and enables assembly source debugging in the assembler. The --
symdebug:dwarf option's short form is -g. The --symdebug:dwarf option
disables many code generator optimizations, because they disrupt the
debugger. You can use the --symdebug:dwarf option with the --opt_level
(aliased as -O) option to maximize the amount of optimization that is
compatible with debugging (see Section 3.14.3.1).
For more information on the DWARF debug format, see The DWARF
Debugging Standard.
--symdebug:keep_all_types Effects the ability to view unused types in the debugger that are from a
COFF executable. Use this option to view the details of a type that is
defined but not used to define any symbols. Such unused types are not
included in the debug information by default for COFF. However, in EABI
mode, all types are included in the debug information and this option has
no effect.
--symdebug:dwarf_ Specifies the DWARF debugging format version (2 or 3) to be generated
version={2|3} when --symdebug:dwarf or --symdebug:skeletal is specified. By default,
the compiler generates DWARF version 3 debug information. For more
information on TI extensions to the DWARF language, see The Impact of
DWARF on TI Object Files (SPRAAB5).
--symdebug:none Disables all symbolic debugging output. This option is not recommended;
it prevents debugging and most performance analysis capabilities.
--symdebug:profile_coff Adds the necessary debug directives to the object file which are needed
by the profiler to allow function level profiling with minimal impact on
optimization (when used). Using --symdebug:coff may hinder some
optimizations to ensure that debug ability is maintained, while this option
will not hinder optimization. STABS format is not supported for C6400+
or ELF.
You can set breakpoints and profile on function-level boundaries in Code
Composer Studio, but you cannot single-step through code as with full
debug ability.
--symdebug:skeletal Generates as much symbolic debugging information as possible without
hindering optimization. Generally, this consists of global-scope
information only. This option reflects the default behavior of the compiler.
See Section 2.3.13 for a list of deprecated symbolic debugging options.

34 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

2.3.6 Specifying Filenames

The input files that you specify on the command line can be C source files, C++ source files, assembly
source files, linear assembly files, or object files. The compiler uses filename extensions to determine the
file type.

Extension File Type

.asm, .abs, or .s* (extension begins with s) Assembly source
.c C source
.C Depends on operating system
.cpp, .cxx, .cc C++ source
.obj .o* .dll .so Object
.sa Linear assembly

NOTE: Case Sensitivity in Filename Extensions

Case sensitivity in filename extensions is determined by your operating system. If your
operating system is not case sensitive, a file with a .C extension is interpreted as a C file. If
your operating system is case sensitive, a file with a .C extension is interpreted as a C++ file.

For information about how you can alter the way that the compiler interprets individual filenames, see
Section 2.3.7. For information about how you can alter the way that the compiler interprets and names the
extensions of assembly source and object files, see Section 2.3.10.
You can use wildcard characters to compile or assemble multiple files. Wildcard specifications vary by
system; use the appropriate form listed in your operating system manual. For example, to compile all of
the files in a directory with the extension .cpp, enter the following:
cl6x *.cpp

NOTE: No Default Extension for Source Files is Assumed

If you list a filename called example on the command line, the compiler assumes that the
entire filename is example not example.c. No default extensions are added onto files that do
not contain an extension.

2.3.7 Changing How the Compiler Interprets Filenames

You can use options to change how the compiler interprets your filenames. If the extensions that you use
are different from those recognized by the compiler, you can use the filename options to specify the type
of file. You can insert an optional space between the option and the filename. Select the appropriate
option for the type of file you want to specify:

--ap_file=filename for a linear assembly source file

--asm_file=filename for an assembly language source file
--c_file=filename for a C source file
--cpp_file=filename for a C++ source file
--obj_file=filename for an object file

For example, if you have a C source file called file.s and an assembly language source file called assy,
use the --asm_file and --c_file options to force the correct interpretation:
cl6x --c_file=file.s --asm_file=assy

You cannot use the filename options with wildcard specifications.

SPRU187U – July 2012 Using the C/C++ Compiler 35

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

2.3.8 Changing How the Compiler Processes C Files

The --cpp_default option causes the compiler to process C files as C++ files. By default, the compiler
treats files with a .c extension as C files. See Section 2.3.9 for more information about filename extension
conventions.

2.3.9 Changing How the Compiler Interprets and Names Extensions

You can use options to change how the compiler program interprets filename extensions and names the
extensions of the files that it creates. The filename extension options must precede the filenames they
apply to on the command line. You can use wildcard specifications with these options. An extension can
be up to nine characters in length. Select the appropriate option for the type of extension you want to
specify:

--ap_extension=new extension for a linear assembly source file

--asm_extension=new extension for an assembly language file
--c_extension=new extension for a C source file
--cpp_extension=new extension for a C++ source file
--listing_extension=new extension sets default extension for listing files
--obj_extension=new extension for an object file

The following example assembles the file fit.rrr and creates an object file named fit.o:
cl6x --asm_extension=.rrr --obj_extension=.o fit.rrr

The period (.) in the extension is optional. You can also write the example above as:
cl6x --asm_extension=rrr --obj_extension=o fit.rrr

2.3.10 Specifying Directories

By default, the compiler program places the object, assembly, and temporary files that it creates into the
current directory. If you want the compiler program to place these files in different directories, use the
following options:

--abs_directory=directory Specifies the destination directory for absolute listing files. The default is
to use the same directory as the object file directory. For example:
cl6x --abs_directory=d:\abso_list
--asm_directory=directory Specifies a directory for assembly files. For example:
cl6x --asm_directory=d:\assembly
--list_directory=directory Specifies the destination directory for assembly listing files and cross-
reference listing files. The default is to use the same directory as the
object file directory. For example:
cl6x --list_directory=d:\listing
--obj_directory=directory Specifies a directory for object files. For example:
cl6x --obj_directory=d:\object
--output_file=filename Specifies a compilation output file name; can override --obj_directory . For
example:
cl6x --output_file=transfer
--pp_directory=directory Specifies a preprocessor file directory for object files (default is .). For
example:
cl6x --pp_directory=d:\preproc
--temp_directory=directory Specifies a directory for temporary intermediate files. For example:
cl6x --temp_directory=d:\temp

36 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the Compiler's Behavior With Options

2.3.11 Assembler Options

Following are assembler options that you can use with the compiler. For more information, see the
TMS320C6000 Assembly Language Tools User's Guide.

--absolute_listing Generates a listing with absolute addresses rather than section-relative

offsets.
--asm_define=name[=def] Predefines the constant name for the assembler; produces a .set directive
for a constant or a .arg directive for a string. If the optional [=def] is
omitted, the name is set to 1. If you want to define a quoted string and
keep the quotation marks, do one of the following:
• For Windows, use --asm_define=name="\"string def\"". For example: -
-asm_define=car="\"sedan\""
• For UNIX, use --asm_define=name='"string def"'. For example: --
asm_define=car='"sedan"'
• For Code Composer Studio, enter the definition in a file and include
that file with the --cmd_file option.
--asm_dependency Performs preprocessing for assembly files, but instead of writing
preprocessed output, writes a list of dependency lines suitable for input to
a standard make utility. The list is written to a file with the same name as
the source file but with a .ppa extension.
--asm_includes Performs preprocessing for assembly files, but instead of writing
preprocessed output, writes a list of files included with the #include
directive. The list is written to a file with the same name as the source file
but with a .ppa extension.
--asm_listing Produces an assembly listing file.
--asm_undefine=name Undefines the predefined constant name. This option overrides any --
asm_define options for the specified name.
--copy_file=filename Copies the specified file for the assembly module; acts like a .copy
directive. The file is inserted before source file statements. The copied file
appears in the assembly listing files.
--cross_reference Produces a symbolic cross-reference in the listing file.
--include_file=filename Includes the specified file for the assembly module; acts like a .include
directive. The file is included before source file statements. The included
file does not appear in the assembly listing files.
--machine_regs Displays reg operands as machine registers in the assembly file for
debugging purposes.
--no_compress Prevents compression in the assembler. For C6400+, C6740, and C6600,
compression is the changing of 32-bit instructions to 16-bit instructions,
where possible/profitable.
--no_reload_errors Turns off all reload-related loop buffer error messages in assembly code
for C6400+, C6740, and C6600.
--output_all_syms Puts labels in the symbol table. Label definitions are written to the COFF
symbol table for use with symbolic debugging.
--strip_coff_underscore Aids in transitioning hand-coded assembly from COFF to EABI.
--syms_ignore_case Makes letter case insignificant in the assembly language source files. For
example, --syms_ignore_case makes the symbols ABC and abc
equivalent. If you do not use this option, case is significant (this is the
default).

SPRU187U – July 2012 Using the C/C++ Compiler 37

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the Compiler's Behavior With Options www.ti.com

2.3.12 Dynamic Linking

The C6000 Code Generation Tools (CGT) (beginning with v7.3) support dynamic linking provided you
build with EABI. For details on dynamic linking with the C6000 CGT, see the TMS320C6000 Assembly
Language Tools User's Guide and http://processors.wiki.ti.com/index.php/C6000_Dynamic_Linking.
If you are not already familiar with the limitations of EABI support in the C6000 compiler, see
http://processors.wiki.ti.com/index.php/EABI_Support_in_C6000_Compiler.
For more information about support for C6000 Linux ABI in the C6000 Code Generation Tools, see
http://processors.wiki.ti.com/index.php/C6000_Linux_Support
Table 2-34 and Table 2-35 provide a brief summary of the compiler and linker options that are related to
support for the Dynamic Linking Model in the C6000 CGT.

Table 2-34. Compiler Options For Dynamic Linking

Option Description
--abi=eabi Specifies that EABI run-time model is to be used.
--dsbt Generates addressing via Dynamic Segment Base Table.
--export_all_cpp_vtbl Exports C++ virtual tables by default.
Specifies that all global symbol references that are not defined in a module are imported. Default
--import_undef[=off|on]
is on.
Specifies that all compiler generated calls to run-time-support functions are treated as calls to
--import_helper_functions
imported functions.
--inline_plt[=off|on] Inlines the import function call stub. Default is on.
--linux Generates C6000 Linux ABI compliant code.
--pic Generates position independent code suitable for a dynamic shared object.
--visibility=={hidden|fhidden|
Specifies a default visibility to be assumed for global symbols.
default|protected}
-wchar_t Generates 32-bit wchar_t type when --abi=eabi is specified.

Table 2-35. Linker Options For Dynamic Linking

Option Description
Requests a specific Data Segment Base Table (DSBT) index to be associated with the current
--dsbt_index=int output file. If the DSBT model is being used, and you do not request a specific DSBT index for the
output file, then a DSBT index is assigned to the module at load time.
Specifies the size of the Data Segment Base Table (DSBT) for the current output file, in words. If
--dsbt_size=int the DSBT model is being used, this option can be used to override the default DSBT size (8
words).
--dynamic[=exe] Specifies that the result of a link will be a lightweight dynamic executable.
--dynamic=lib Specifies that the result of a link will be a dynamic library.
--export=symbol Specifies that symbol is exported by the ELF object that is generated for this link.
--fini=symbol Specifies the symbol name of the termination code for the output file currently being linked.
--import=symbol Specifies that symbol is imported by the ELF object that is generated for this link.
--init=symbol Specifies the symbol name of the initialization code for the output file currently being linked.
--rpath=dir Adds a directory to the beginning of the dynamic library search path.
--runpath=dir Adds a directory to the end of the dynamic library search path.
--shared Generates an ELF dynamic shared object (DSO)
Specifies shared object name to be used to identify this ELF object to the any downstream ELF
--soname=string
object consumers.
--sysv Generates SysV ELF dynamic object module.

38 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Controlling the Compiler Through Environment Variables

2.3.13 Deprecated Options

Several compiler options have been deprecated. The compiler continues to accept these options, but they
are not recommended for use. Future releases of the tools will not support these options. Table 2-36 lists
the deprecated options and the options that have replaced them.

Table 2-36. Compiler Backwards-Compatibility Options Summary

Old Option Effect New Option
-gp Allows function-level profiling of optimized code --symdebug:dwarf or -g
-gt Enables symbolic debugging using the alternate STABS debugging format --symdebug:coff
-gw Enables symbolic debugging using the DWARF debugging format --symdebug:dwarf or -g

Additionally, the --symdebug:profile_coff option has been added to enable function-level profiling of
optimized code with symbolic debugging using the STABS debugging format (the --symdebug:coff or -gt
option).
Since C6400+, C6740, and C6600 produce only DWARF debug information, the -gp, -gt/--symdebug:coff,
and --symdebug:profile_coff options are not supported for C6400+, C6740, and C6600.

2.4 Controlling the Compiler Through Environment Variables

An environment variable is a system symbol that you define and assign a string to. Setting environment
variables is useful when you want to run the compiler repeatedly without re-entering options, input
filenames, or pathnames.

NOTE: C_OPTION and C_DIR

The C_OPTION and C_DIR environment variables are deprecated. Use the device-specific
environment variables instead.

2.4.1 Setting Default Compiler Options (C6X_C_OPTION)

You might find it useful to set the compiler, assembler, and linker default options using the
C6X_C_OPTION environment variable. If you do this, the compiler uses the default options and/or input
filenames that you name C6X_C_OPTION every time you run the compiler.
Setting the default options with these environment variables is useful when you want to run the compiler
repeatedly with the same set of options and/or input files. After the compiler reads the command line and
the input filenames, it looks for the C6X_C_OPTION environment variable and processes it.
The table below shows how to set the C6X_C_OPTION environment variable. Select the command for
your operating system:

Operating System Enter

UNIX (Bourne shell) C6X_C_OPTION=" option1 [option2 . . .]"; export C6X_C_OPTION
Windows set C6X_C_OPTION= option1 [;option2 . . .]

Environment variable options are specified in the same way and have the same meaning as they do on
the command line. For example, if you want to always run quietly (the --quiet option), enable C/C++
source interlisting (the --src_interlist option), and link (the --run_linker option) for Windows, set up the
C6X_C_OPTION environment variable as follows:
set C6X_C_OPTION=--quiet --src_interlist --run_linker

SPRU187U – July 2012 Using the C/C++ Compiler 39

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling the Compiler Through Environment Variables www.ti.com

In the following examples, each time you run the compiler, it runs the linker. Any options following --
run_linker on the command line or in C6X_C_OPTION are passed to the linker. Thus, you can use the
C6X_C_OPTION environment variable to specify default compiler and linker options and then specify
additional compiler and linker options on the command line. If you have set --run_linker in the environment
variable and want to compile only, use the compiler --compile_only option. These additional examples
assume C6X_C_OPTION is set as shown above:
cl6x *c ; compiles and links
cl6x --compile_only *.c ; only compiles
cl6x *.c --run_linker lnk.cmd ; compiles and links using a command file
cl6x --compile_only *.c --run_linker lnk.cmd
; only compiles (--compile_only overrides --run_linker)

For details on compiler options, see Section 2.3. For details on linker options, see the Linker Description
chapter in the TMS320C6000 Assembly Language Tools User's Guide.

2.4.2 Naming an Alternate Directory (C6X_C_DIR)

The linker uses the C6X_C_DIR environment variable to name alternate directories that contain object
libraries. The command syntaxes for assigning the environment variable are:

Operating System Enter

UNIX (Bourne shell) C6X_C_DIR=" pathname1 ; pathname2 ;..."; export C6X_C_DIR
Windows set C6X_C_DIR= pathname1 ; pathname2 ;...

The pathnames are directories that contain input files. The pathnames must follow these constraints:
• Pathnames must be separated with a semicolon.
• Spaces or tabs at the beginning or end of a path are ignored. For example, the space before and after
the semicolon in the following is ignored:
set C6X_C_DIR=c:\path\one\to\tools ; c:\path\two\to\tools
• Spaces and tabs are allowed within paths to accommodate Windows directories that contain spaces.
For example, the pathnames in the following are valid:
set C6X_C_DIR=c:\first path\to\tools;d:\second path\to\tools

The environment variable remains set until you reboot the system or reset the variable by entering:

Operating System Enter

UNIX (Bourne shell) unset C6X_C_DIR
Windows set C6X_C_DIR=

40 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Precompiled Header Support

2.5 Precompiled Header Support

Precompiled header files may reduce the compile time for applications whose source files share a
common set of headers, or a single file which has a large set of header files. Using precompiled headers,
some recompilation is avoided thus saving compilation time.
There are two ways to use precompiled header files. One is the automatic precompiled header file
processing and the other is called the manual precompiled header file processing.

2.5.1 Automatic Precompiled Header

The option to turn on automatic precompiled header processing is: --pch. Under this option, the compile
step takes a snapshot of all the code prior to the header stop point, and dump it out to a file with suffix
.pch. This snapshot does not have to be recompiled in the future compilations of this file or compilations of
files with the same header files.
The stop point typically is the first token in the primary source file that does not belong to a preprocessing
directive. For example, in the following the stopping point is before int i:
#include "x.h"
#include "y.h"
int i;

Carefully organizing the include directives across multiple files so that their header files maximize common
usage can increase the compile time savings when using precompiled headers.
A precompiled header file is produced only if the header stop point and the code prior to it meet certain
requirements.

2.5.2 Manual Precompiled Header

You can manually control the creation and use of precompiled headers by using several command line
options. You specify a precompiled header file with a specific filename as follows:
--create_pch=filename
The --use_pch=filename option specifies that the indicated precompiled header file should be used for this
compilation. If this precompiled header file is invalid, if its prefix does not match the prefix for the current
primary source file for example, a warning is issued and the header file is not used.
If --create_pch=filename or --use_pch=filename is used with --pch_dir, the indicated filename, which can
be a path name, is tacked on to the directory name, unless the filename is an absolute path name.
The --create_pch, --use_pch, and --pch options cannot be used together. If more than one of these
options is specified, only the last one is applied. In manual mode, the header stop points are determined
in the same way as in automatic mode. The precompiled header file applicability is determined in the
same manner.

2.5.3 Additional Precompiled Header Options

The --pch_verbose option displays a message for each precompiled header file that is considered but not
used. The --pch_dir=pathname option specifies the path where the precompiled header file resides.

SPRU187U – July 2012 Using the C/C++ Compiler 41

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling the Preprocessor www.ti.com

2.6 Controlling the Preprocessor

This section describes specific features that control the preprocessor, which is part of the parser. A
general description of C preprocessing is in section A12 of K&R. The C/C++ compiler includes standard
C/C++ preprocessing functions, which are built into the first pass of the compiler. The preprocessor
handles:
• Macro definitions and expansions
• #include files
• Conditional compilation
• Various preprocessor directives, specified in the source file as lines beginning with the # character
The preprocessor produces self-explanatory error messages. The line number and the filename where the
error occurred are printed along with a diagnostic message.

2.6.1 Predefined Macro Names

The compiler maintains and recognizes the predefined macro names listed in Table 2-37.

Table 2-37. Predefined C6000 Macro Names

Macro Name Description
(1)
_ _DATE_ _ Expands to the compilation date in the form mmm dd yyyy
_ _FILE_ _ (1) Expands to the current source filename
_ _LINE_ _ (1) Expands to the current line number
_ _STDC_ _ (1) Defined to indicate that compiler conforms to ISO C Standard. See Section 6.1 for
exceptions to ISO C conformance.
_ _STDC_VERSION_ _ C standard macro
_ _TI_32BIT_LONG_ _ Defined to 1 if the EABI ABI is enabled (see Section 2.16); otherwise, it is undefined.
_ _TI_40BIT_LONG_ _ Defined to 1 if _ _TI_32BIT_LONG_ _ is not defined; otherwise, it is undefined.
_ _TI_COMPILER_VERSION_ _ Defined to a 7-9 digit integer, depending on if X has 1, 2, or 3 digits. The number does not
contain a decimal. For example, version 3.2.1 is represented as 3002001. The leading
zeros are dropped to prevent the number being interpreted as an octal.
_ _TI_EABI_ _ Defined to 1 if the EABI is enabled (see Section 2.16); otherwise, it is undefined.
_ _TI_GNU_ATTRIBUTE_SUPPORT_ Defined if GCC extensions are enabled (the --gcc option is used); otherwise, it is
_ undefined.
_ _TI_STRICT_ANSI_MODE__ Defined if strict ANSI/ISO mode is enabled (the --strict_ansi option is used); otherwise, it
is undefined.
_ _TI_WCHAR_T_BITS_ _
_ _TIME_ _ (1) Expands to the compilation time in the form "hh:mm:ss"
_ _WCHAR_T_TYPE_ _
_BIG_ENDIAN Defined if big-endian mode is selected (the --big_endian option is used); otherwise, it is
undefined.
_INLINE Expands to 1 if optimization is used (--opt_level or -O option); undefined otherwise.
Regardless of any optimization, always undefined when --no_inlining is used.
_LITTLE_ENDIAN Defined if little-endian mode is selected (the --big_endian option is not used); otherwise, it
is undefined.
_TMS320C6X Always defined
_TMS320C6200 Defined if target is C6200
_TMS320C6400 Defined if target is C6400, C6400+, C6740, or C6600
_TMS320C6400_PLUS Defined if target is C6400+, C6740, or C6600
_TMS320C6600 Defined if target is C6600
_TMS320C6700 Defined if target is C6700, C6700+, C6740, or C6600
_TMS320C6700_PLUS Defined if target is C6700+, C6740, or C6600
_TMS320C6740 Defined if target is C6740 or C6600
_ _TMS320C6X_ _ Always defined for use as alternate name for _TMS320C6x
(1)
Specified by the ISO standard

42 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Controlling the Preprocessor

You can use the names listed in Table 2-37 in the same manner as any other defined name. For example,
printf ( "%s %s" , __TIME__ , __DATE__);

translates to a line such as:

printf ("%s %s" , "13:58:17", "Jan 14 1997");

2.6.2 The Search Path for #include Files

The #include preprocessor directive tells the compiler to read source statements from another file. When
specifying the file, you can enclose the filename in double quotes or in angle brackets. The filename can
be a complete pathname, partial path information, or a filename with no path information.
• If you enclose the filename in double quotes (" "), the compiler searches for the file in the following
directories in this order:
1. The directory of the file that contains the #include directive and in the directories of any files that
contain that file.
2. Directories named with the --include_path option.
3. Directories set with the C6X_C_DIR environment variable.
• If you enclose the filename in angle brackets (< >), the compiler searches for the file in the following
directories in this order:
1. Directories named with the --include_path option.
2. Directories set with the C6X_C_DIR environment variable.
See Section 2.6.2.1 for information on using the --include_path option. See Section 2.4.2 for more
information on input file directories.

2.6.2.1 Changing the #include File Search Path (--include_path Option)

The --include_path option names an alternate directory that contains #include files. The --include_path
option's short form is -I. The format of the --include_path option is:
--include_path=directory1 [--include_path= directory2 ...]
There is no limit to the number of --include_path options per invocation of the compiler; each --
include_path option names one directory. In C source, you can use the #include directive without
specifying any directory information for the file; instead, you can specify the directory information with the -
-include_path option. For example, assume that a file called source.c is in the current directory. The file
source.c contains the following directive statement:
#include "alt.h"

Assume that the complete pathname for alt.h is:

UNIX /tools/files/alt.h
Windows c:\tools\files\alt.h
The table below shows how to invoke the compiler. Select the command for your operating system:

Operating System Enter

UNIX cl6x --include_path=/tools/files source.c
Windows cl6x --include_path=c:\tools\files source.c

SPRU187U – July 2012 Using the C/C++ Compiler 43

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling the Preprocessor www.ti.com

NOTE: Specifying Path Information in Angle Brackets

If you specify the path information in angle brackets, the compiler applies that information
relative to the path information specified with --include_path options and the C6X_C_DIR
environment variable.
For example, if you set up C6X_C_DIR with the following command:
C6X_C_DIR "/usr/include;/usr/ucb"; export C6X_C_DIR

or invoke the compiler with the following command:

cl6x --include_path=/usr/include file.c

and file.c contains this line:

#include <sys/proc.h>

the result is that the included file is in the following path:

/usr/include/sys/proc.h

2.6.3 Generating a Preprocessed Listing File (--preproc_only Option)

The --preproc_only option allows you to generate a preprocessed version of your source file with an
extension of .pp. The compiler's preprocessing functions perform the following operations on the source
file:
• Each source line ending in a backslash (\) is joined with the following line.
• Trigraph sequences are expanded.
• Comments are removed.
• #include files are copied into the file.
• Macro definitions are processed.
• All macros are expanded.
• All other preprocessing directives, including #line directives and conditional compilation, are expanded.

2.6.4 Continuing Compilation After Preprocessing (--preproc_with_compile Option)

If you are preprocessing, the preprocessor performs preprocessing only; it does not compile your source
code. To override this feature and continue to compile after your source code is preprocessed, use the --
preproc_with_compile option along with the other preprocessing options. For example, use --
preproc_with_compile with --preproc_only to perform preprocessing, write preprocessed output to a file
with a .pp extension, and compile your source code.

2.6.5 Generating a Preprocessed Listing File With Comments (--preproc_with_comment

Option)
The --preproc_with_comment option performs all of the preprocessing functions except removing
comments and generates a preprocessed version of your source file with a .pp extension. Use the --
preproc_with_comment option instead of the --preproc_only option if you want to keep the comments.

2.6.6 Generating a Preprocessed Listing File With Line-Control Information (--

preproc_with_line Option)
By default, the preprocessed output file contains no preprocessor directives. To include the #line
directives, use the --preproc_with_line option. The --preproc_with_line option performs preprocessing only
and writes preprocessed output with line-control information (#line directives) to a file named as the
source file but with a .pp extension.

44 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Understanding Diagnostic Messages

2.6.7 Generating Preprocessed Output for a Make Utility (--preproc_dependency Option)

The --preproc_dependency option performs preprocessing only, but instead of writing preprocessed
output, writes a list of dependency lines suitable for input to a standard make utility. If you do not supply
an optional filename, the list is written to a file with the same name as the source file but with a .pp
extension.

2.6.8 Generating a List of Files Included With the #include Directive (--preproc_includes
Option)
The --preproc_includes option performs preprocessing only, but instead of writing preprocessed output,
writes a list of files included with the #include directive. If you do not supply an optional filename, the list is
written to a file with the same name as the source file but with a .pp extension.

2.6.9 Generating a List of Macros in a File (--preproc_macros Option)

The --preproc_macros option generates a list of all predefined and user-defined macros. If you do not
supply an optional filename, the list is written to a file with the same name as the source file but with a .pp
extension. Predefined macros are listed first and indicated by the comment /* Predefined */. User-defined
macros are listed next and indicated by the source filename.

2.7 Understanding Diagnostic Messages

One of the compiler's primary functions is to report diagnostics for the source program. The new linker
also reports diagnostics. When the compiler or linker detects a suspect condition, it displays a message in
the following format:
"file.c", line n : diagnostic severity : diagnostic message

"file.c" The name of the file involved

line n : The line number where the diagnostic applies
diagnostic severity The diagnostic message severity (severity category descriptions follow)
diagnostic message The text that describes the problem

Diagnostic messages have an associated severity, as follows:

• A fatal error indicates a problem so severe that the compilation cannot continue. Examples of such
problems include command-line errors, internal errors, and missing include files. If multiple source files
are being compiled, any source files after the current one will not be compiled.
• An error indicates a violation of the syntax or semantic rules of the C/C++ language. Compilation
continues, but object code is not generated.
• A warning indicates something that is valid but questionable. Compilation continues and object code is
generated (if no errors are detected).
• A remark is less serious than a warning. It indicates something that is valid and probably intended, but
may need to be checked. Compilation continues and object code is generated (if no errors are
detected). By default, remarks are not issued. Use the --issue_remarks compiler option to enable
remarks.
Diagnostics are written to standard error with a form like the following example:
"test.c", line 5: error: a break statement may only be used within a loop or switch
break;
^

By default, the source line is omitted. Use the --verbose_diagnostics compiler option to enable the display
of the source line and the error position. The above example makes use of this option.
The message identifies the file and line involved in the diagnostic, and the source line itself (with the
position indicated by the ^ character) follows the message. If several diagnostics apply to one source line,
each diagnostic has the form shown; the text of the source line is displayed several times, with an
appropriate position indicated each time.

SPRU187U – July 2012 Using the C/C++ Compiler 45

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Understanding Diagnostic Messages www.ti.com

Long messages are wrapped to additional lines, when necessary.

You can use the --display_error_number command-line option to request that the diagnostic's numeric
identifier be included in the diagnostic message. When displayed, the diagnostic identifier also indicates
whether the diagnostic can have its severity overridden on the command line. If the severity can be
overridden, the diagnostic identifier includes the suffix -D (for discretionary); otherwise, no suffix is
present. For example:
"Test_name.c", line 7: error #64-D: declaration does not declare anything
struct {};
^
"Test_name.c", line 9: error #77: this declaration has no storage class or type specifier
xxxxx;
^

Because an error is determined to be discretionary based on the error severity associated with a specific
context, an error can be discretionary in some cases and not in others. All warnings and remarks are
discretionary.
For some messages, a list of entities (functions, local variables, source files, etc.) is useful; the entities are
listed following the initial error message:
"test.c", line 4: error: more than one instance of overloaded function "f"
matches the argument list:
function "f(int)"
function "f(float)"
argument types are: (double)
f(1.5);
^

In some cases, additional context information is provided. Specifically, the context information is useful
when the front end issues a diagnostic while doing a template instantiation or while generating a
constructor, destructor, or assignment operator function. For example:
"test.c", line 7: error: "A::A()" is inaccessible
B x;
^
detected during implicit generation of "B::B()" at line 7

Without the context information, it is difficult to determine to what the error refers.

2.7.1 Controlling Diagnostics

The C/C++ compiler provides diagnostic options to control compiler- and linker-generated diagnostics. The
diagnostic options must be specified before the --run_linker option.

--diag_error=num Categorizes the diagnostic identified by num as an error. To determine the

numeric identifier of a diagnostic message, use the --display_error_number
option first in a separate compile. Then use --diag_error=num to recategorize
the diagnostic as an error. You can only alter the severity of discretionary
diagnostics.
--diag_remark=num Categorizes the diagnostic identified by num as a remark. To determine the
numeric identifier of a diagnostic message, use the --display_error_number
option first in a separate compile. Then use --diag_remark=num to
recategorize the diagnostic as a remark. You can only alter the severity of
discretionary diagnostics.
--diag_suppress=num Suppresses the diagnostic identified by num. To determine the numeric
identifier of a diagnostic message, use the --display_error_number option first
in a separate compile. Then use --diag_suppress=num to suppress the
diagnostic. You can only suppress discretionary diagnostics.
--diag_warning=num Categorizes the diagnostic identified by num as a warning. To determine the
numeric identifier of a diagnostic message, use the --display_error_number
option first in a separate compile. Then use --diag_warning=num to
recategorize the diagnostic as a warning. You can only alter the severity of
discretionary diagnostics.

46 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Understanding Diagnostic Messages

--display_error_number Displays a diagnostic's numeric identifier along with its text. Use this option in
determining which arguments you need to supply to the diagnostic
suppression options (--diag_suppress, --diag_error, --diag_remark, and --
diag_warning). This option also indicates whether a diagnostic is discretionary.
A discretionary diagnostic is one whose severity can be overridden. A
discretionary diagnostic includes the suffix -D; otherwise, no suffix is present.
See Section 2.7.
--emit_warnings_as_ Treats all warnings as errors. This option cannot be used with the --
errors no_warnings option. The --diag_remark option takes precedence over this
option. This option takes precedence over the --diag_warning option.
--issue_remarks Issues remarks (nonserious warnings), which are suppressed by default.
--no_warnings Suppresses warning diagnostics (errors are still issued).
--set_error_limit=num Sets the error limit to num, which can be any decimal value. The compiler
abandons compiling after this number of errors. (The default is 100.)
--verbose_diagnostics Provides verbose diagnostics that display the original source with line-wrap
and indicate the position of the error in the source line
--write_diagnostics_file Produces a diagnostics information file with the same source file name with an
.err extension. (The --write_diagnostics_file option is not supported by the
linker.)

2.7.2 How You Can Use Diagnostic Suppression Options

The following example demonstrates how you can control diagnostic messages issued by the compiler.
You control the linker diagnostic messages in a similar manner.
int one();
int I;
int main()
{
switch (I){
case 1;
return one ();
break;
default:
return 0;
break;
}
}

If you invoke the compiler with the --quiet option, this is the result:
"err.c", line 9: warning: statement is unreachable
"err.c", line 12: warning: statement is unreachable

Because it is standard programming practice to include break statements at the end of each case arm to
avoid the fall-through condition, these warnings can be ignored. Using the --display_error_number option,
you can find out the diagnostic identifier for these warnings. Here is the result:
[err.c]
"err.c", line 9: warning #111-D: statement is unreachable
"err.c", line 12: warning #111-D: statement is unreachable

Next, you can use the diagnostic identifier of 111 as the argument to the --diag_remark option to treat this
warning as a remark. This compilation now produces no diagnostic messages (because remarks are
disabled by default).
Although this type of control is useful, it can also be extremely dangerous. The compiler often emits
messages that indicate a less than obvious problem. Be careful to analyze all diagnostics emitted before
using the suppression options.

SPRU187U – July 2012 Using the C/C++ Compiler 47

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Other Messages www.ti.com

2.8 Other Messages

Other error messages that are unrelated to the source, such as incorrect command-line syntax or inability
to find specified files, are usually fatal. They are identified by the symbol >> preceding the message.

2.9 Generating Cross-Reference Listing Information (--gen_acp_xref Option)

The --gen_acp_xref option generates a cross-reference listing file that contains reference information for
each identifier in the source file. (The --gen_acp_xref option is separate from --cross_reference, which is
an assembler rather than a compiler option.) The cross-reference listing file has the same name as the
source file with a .crl extension.
The information in the cross-reference listing file is displayed in the following format:
sym-id name X filename line number column number

sym-id An integer uniquely assigned to each identifier

name The identifier name
X One of the following values:
D Definition
d Declaration (not a definition)
M Modification
A Address taken
U Used
C Changed (used and modified in a single operation)
R Any other kind of reference
E Error; reference is indeterminate
filename The source file
line number The line number in the source file
column number The column number in the source file

2.10 Generating a Raw Listing File (--gen_acp_raw Option)

The --gen_acp_raw option generates a raw listing file that can help you understand how the compiler is
preprocessing your source file. Whereas the preprocessed listing file (generated with the --preproc_only, --
preproc_with_comment, --preproc_with_line, and --preproc_dependency preprocessor options) shows a
preprocessed version of your source file, a raw listing file provides a comparison between the original
source line and the preprocessed output. The raw listing file has the same name as the corresponding
source file with an .rl extension.
The raw listing file contains the following information:
• Each original source line
• Transitions into and out of include files
• Diagnostics
• Preprocessed source line if nontrivial processing was performed (comment removal is considered
trivial; other preprocessing is nontrivial)
Each source line in the raw listing file begins with one of the identifiers listed in Table 2-38.

48 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Generating a Raw Listing File (--gen_acp_raw Option)

Table 2-38. Raw Listing File Identifiers

Identifier Definition
N Normal line of source
X Expanded line of source. It appears immediately following the normal line of source
if nontrivial preprocessing occurs.
S Skipped source line (false #if clause)
L Change in source position, given in the following format:
L line number filename key
Where line number is the line number in the source file. The key is present only
when the change is due to entry/exit of an include file. Possible values of key are:
1 = entry into an include file
2 = exit from an include file

The --gen_acp_raw option also includes diagnostic identifiers as defined in Table 2-39.

Table 2-39. Raw Listing File Diagnostic Identifiers

Diagnostic Identifier Definition
E Error
F Fatal
R Remark
W Warning

Diagnostic raw listing information is displayed in the following format:

S filename line number column number diagnostic

S One of the identifiers in Table 2-39 that indicates the severity of the diagnostic
filename The source file
line number The line number in the source file
column number The column number in the source file
diagnostic The message text for the diagnostic
Diagnostics after the end of file are indicated as the last line of the file with a column number of 0. When
diagnostic message text requires more than one line, each subsequent line contains the same file, line,
and column information but uses a lowercase version of the diagnostic identifier. For more information
about diagnostic messages, see Section 2.7.

SPRU187U – July 2012 Using the C/C++ Compiler 49

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Inline Function Expansion www.ti.com

2.11 Using Inline Function Expansion

When an inline function is called, the C/C++ source code for the function is inserted at the point of the call.
This is known as inline function expansion. Inline function expansion is advantageous in short functions for
the following reasons:
There are several types of inline function expansion:
• Inlining with intrinsic operators (intrinsics are always inlined)
• Automatic inlining
• Definition-controlled inlining with the unguarded inline keyword
• Definition-controlled inlining with the guarded inline keyword

NOTE: Function Inlining Can Greatly Increase Code Size

Expanding functions inline increases code size, especially inlining a function that is called in
a number of places. Function inlining is optimal for functions that are called only from a small
number of places and for small functions. If your code size seems too large, see Section 3.5.

2.11.1 Inlining Intrinsic Operators

There are many intrinsic operators for the C6000. All of them are automatically inlined by the compiler.
The inlining happens automatically whether or not you use the optimizer.
For details about intrinsics, and a list of the intrinsics, see Section 7.5.5.

2.11.2 Automatic Inlining

When optimizing with the --opt_level=3 or --opt_level=2 option (aliased as -O3 or -O2), the compiler
automatically inlines certain functions. For more information, see Section 3.13.

2.11.3 Unguarded Definition-Controlled Inlining

The inline keyword specifies that a function is expanded inline at the point at which it is called rather than
by using standard calling procedures. The compiler performs inline expansion of functions declared with
the inline keyword.
You must invoke the optimizer with any --opt_level option (--opt_level=0, --opt_level=1, --opt_level=2, or --
opt_level=3) to turn on definition-controlled inlining. Automatic inlining is also turned on when using --
opt_level=3.
The --no_inlining option turns off definition-controlled inlining. This option is useful when you need a
certain level of optimization but do not want definition-controlled inlining.
Example 2-1 shows usage of the inline keyword, where the function call is replaced by the code in the
called function.

Example 2-1. Using the Inline Keyword

inline float volume_sphere(float r)

{
return 4.0/3.0 * PI * r * r * r;
}
int foo(...)
{
...
volume = volume_sphere(radius);
...
}

50 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Inline Function Expansion

2.11.4 Guarded Inlining and the _INLINE Preprocessor Symbol

When declaring a function in a header file as static inline, you must follow additional procedures to avoid a
potential code size increase when inlining is turned off with --no_inlining or the optimizer is not run.
To prevent a static inline function in a header file from causing an increase in code size when inlining gets
turned off, use the following procedure. This allows external-linkage when inlining is turned off; thus, only
one function definition will exist throughout the object files.
• Prototype a static inline version of the function. Then, prototype an alternative, nonstatic, externally-
linked version of the function. Conditionally preprocess these two prototypes with the _INLINE
preprocessor symbol, as shown in Example 2-2.
• Create an identical version of the function definition in a .c or .cpp file, as shown in Example 2-3.
In the following examples there are two definitions of the strlen function. The first (Example 2-2), in the
header file, is an inline definition. This definition is enabled and the prototype is declared as static inline
only if _INLINE is true (_INLINE is automatically defined for you when the optimizer is used and --
no_inlining is not specified).
The second definition (see Example 2-3) for the library, ensures that the callable version of strlen exists
when inlining is disabled. Since this is not an inline function, the _INLINE preprocessor symbol is
undefined (#undef) before string.h is included to generate a noninline version of strlen's prototype.

Example 2-2. Header File string.h

/*****************************************************************************/
/* string.h vx.xx (Excerpted) */
/* Copyright (c) 1993-2011 Texas Instruments Incorporated */
/*****************************************************************************/
#ifdef _INLINE
#define _IDECL static inline
#else
#define _IDECL extern _CODE_ACCESS
#endif

_IDECL size_t strlen(const char *_string);

#ifdef _INLINE

/****************************************************************************/
/* strlen */
/****************************************************************************/
static inline size_t strlen(const char *string)
{
size_t n = (size_t)-1;
const char *s = string - 1;

do n++; while (*++s);

return n
}

#endif

SPRU187U – July 2012 Using the C/C++ Compiler 51

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Inline Function Expansion www.ti.com

Example 2-3. Library Definition File

/****************************************************************************/
/* strlen */
/****************************************************************************/
#undef _INLINE

#include <string.h>
{
_CODE_ACCESS size_t strlen(cont char * string)
size_t n = (size_t)-1;
const char *s = string - 1;

do n++; while (*++s);

return n;
}

2.11.5 Inlining Restrictions

There are several restrictions on what functions can be inlined for both automatic inlining and definition-
controlled inlining. Functions with local static variables or a variable number of arguments are not inlined,
with the exception of functions declared as static inline. In functions declared as static inline, expansion
occurs despite the presence of local static variables. In addition, a limit is placed on the depth of inlining
for recursive or nonleaf functions. Furthermore, inlining should be used for small functions or functions that
are called in a few places (though the compiler does not enforce this).
At a given call site, a function may be disqualified from inlining if it:
• Is not defined in the current compilation unit
• Never returns
• Is recursive
• Has a FUNC_CANNOT_INLINE pragma
• Has a variable length argument list
• Has a different number of arguments than the call site
• Has an argument whose type is incompatible with the corresponding call site argument
• Has a class, structe or union parameter
• Contains a volatile local variable or argument
• Is not declared inline and contains an asm() statement that is not a comment
• Is not declared inline and it is main()
• Is not declared inline and it is an interrupt function
• Is not declared inline and returns void but its return value is needed.
• Is not declared inline and will require too much stack space for local array or structure variables.

52 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Interrupt Flexibility Options (--interrupt_threshold Option)

2.12 Interrupt Flexibility Options (--interrupt_threshold Option)

On the C6000 architecture, interrupts cannot be taken in the delay slots of a branch. In some instances
the compiler can generate code that cannot be interrupted for a potentially large number of cycles. For a
given real-time system, there may be a hard limit on how long interrupts can be disabled.
The --interrupt_threshold=n option specifies an interrupt threshold value n. The threshold value specifies
the maximum number of cycles that the compiler can disable interrupts. If the n is omitted, the compiler
assumes that the code is never interrupted. In Code Composer Studio, to specify that the code is never
interrupted, select the Interrupt Threshold check box and leave the text box blank in the Build Options
dialog box on the Compiler tab, Advanced category.
If the --interrupt_threshold=n option is not specified, then interrupts are only explicitly disabled around
software pipelined loops. When using the --interrupt_threshold=n option, the compiler analyzes the loop
structure and loop counter to determine the maximum number of cycles it takes to execute a loop. If it can
determine that the maximum number of cycles is less than the threshold value, the compiler generates the
fastest/optimal version of the loop. If the loop is smaller than six cycles, interrupts are not able to occur
because the loop is always executing inside the delay slots of a branch. Otherwise, the compiler
generates a loop that can be interrupted (and still generate correct results—single assignment code),
which in most cases degrades the performance of the loop.
The --interrupt_threshold=n option does not comprehend the effects of the memory system. When
determining the maximum number of execution cycles for a loop, the compiler does not compute the
effects of using slow off-chip memory or memory bank conflicts. It is recommended that a conservative
threshold value is used to adjust for the effects of the memory system.
See Section 6.9.11 or the TMS320C6000 Programmer's Guide for more information.

RTS Library Files Are Not Built With the --interrupt_threshold Option
NOTE: The run-time-support library files provided with the compiler are not built with the interrupt
flexibility option. Refer to the readme file to see how the run-time-support library files were
built for your release. See Section 8.5 to build your own run-time-support library files with the
interrupt flexibility option.

Special Cases With the --interrupt_threshold Option

NOTE: The --interrupt_threshold=0 option generates the same code to disable interrupts around
software-pipelined loops as when the --interrupt_threshold option is not used.
The --interrupt_threshold option (the threshold value is omitted) means that no code is added
to disable interrupts around software pipelined loops, which means that the code cannot be
safely interrupted. Also, loop performance does not degrade because the compiler is not
trying to make the loop interruptible by ensuring that there is at least one cycle in the loop
kernel that is not in the delay slot of a branch instruction.

SPRU187U – July 2012 Using the C/C++ Compiler 53

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Linking C6400 Code With C6200/C6700/Older C6400 Object Code www.ti.com

2.13 Linking C6400 Code With C6200/C6700/Older C6400 Object Code

In order to facilitate certain packed-data optimizations, the alignment of top-level arrays for the C6400
family was changed from 4 bytes to 8 bytes. (For C6200 and C6700 code, the alignment for top-level
arrays is always 4 bytes.)
If you are linking C6400/C6400+/C6740/C6600 with C6200/C6700 code or older C6400 code, you may
need to take steps to ensure compatibility. The following lists the potential alignment conflicts and possible
solutions.
Potential alignment conflicts occur when:
• Linking new C6400/C6400+/C6740/C6600 code with any C6400 code already compiled with the 4.0
tools.
• Linking new C6400/C6400+/C6740/C6600 code with code already compiled with any version of the
tools for the C6200 or C6700 family.
Solutions (pick one):
• Recompile the entire application with the --silicon_version=6400 switch. This solution, if possible, is
recommended because it can lead to better performance.
• Compile the new code with the --target_compatibility_6200 option. When using this option, link with the
6200 RTS library to ensure proper compatibility. The --target_compatibility_6200 option changes the
alignment of top-level arrays to 4 bytes when the --silicon_version=6400 or --silicon_version=6400+
option is used.

This option is deprecated; it may be removed in the future.

The alignment of top-level arrays for the C6600 family is 16 bytes to facilitate compatibility with future
C6600 family devices. This change in alignment does not have any compatibility issues with the
C6400/C6400+/C6740 device code as the C6600 can safely accept top-level arrays aligned to an 8-byte
boundary.

2.14 Using Interlist

The compiler tools include a feature that interlists C/C++ source statements into the assembly language
output of the compiler. The interlist feature enables you to inspect the assembly code generated for each
C statement. The interlist behaves differently, depending on whether or not the optimizer is used, and
depending on which options you specify.
The easiest way to invoke the interlist feature is to use the --c_src_interlist option. To compile and run the
interlist on a program called function.c, enter:
cl6x --c_src_interlist function

The --c_src_interlist option prevents the compiler from deleting the interlisted assembly language output
file. The output assembly file, function.asm, is assembled normally.
When you invoke the interlist feature without the optimizer, the interlist runs as a separate pass between
the code generator and the assembler. It reads both the assembly and C/C++ source files, merges them,
and writes the C/C++ statements into the assembly file as comments.
Using the --c_src_interlist option can cause performance and/or code size degradation.
Example 2-4 shows a typical interlisted assembly file.
For more information about using the interlist feature with the optimizer, see Section 3.14.2.

54 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Generating and Using Performance Advice

Example 2‑4. An Interlisted Assembly Language File

_main:

STW .D2 B3,*SP--(12)

STW .D2 A10,*+SP(8)
;------------------------------------------------------------------------------
; 5 | printf("Hello, world\n");
;------------------------------------------------------------------------------
B .S1 _printf
NOP 2
MVKL .S1 SL1+0,A0
MVKH .S1 SL1+0,A0
|| MVKL .S2 RL0,B3
STW .D2 A0,*+SP(4)
|| MVKH .S2 RL0,B3
RL0: ; CALL OCCURS
;------------------------------------------------------------------------------
; 6 | return 0;
;------------------------------------------------------------------------------
ZERO .L1 A10
MV .L1 A10,A4
LDW .D2 *+SP(8),A10
LDW .D2 *++SP(12),B3
NOP 4
B .S2 B3
NOP 5
; BRANCH OCCURS

2.15 Generating and Using Performance Advice

The compiler can do better optimization in some cases, if the user aids the compiler by providing
additional information in the code. The compiler can prompt you to take certain actions to improve
performance, by emitting "Advice". To get this Advice, use the --advice:performance option:
cl6x --advice:performance -o3 filename.c

This Performance Advice is of 3 different types :

• Advice to use correct compiler options
• Advice to prevent software pipeline disqualification
• Advice to improve loop performance
For more details on Using Performance Advice to Optimize your Code, see Section 3.14

2.16 Controlling Application Binary Interface

An Application Binary Interface (ABI) defines the low level interface between object files, and between an
executable and its execution environment. An ABI allows ABI-compliant object code to link together,
regardless of its source, and allows the resulting executable to run on any system that supports that ABI
Object files conforming to different ABIs cannot be linked together. The linker detects this situation and
generates an error.
The C6000 compiler supports two ABIs. The ABI is chosen through the --abi option as follows:
• COFF ABI (--abi=coffabi)
The COFF ABI is the original ABI format. There is no COFF to ELF conversion possible; recompile or
reassemble assembly code.
• C6000 EABI (--abi=eabi)
Use this option to select the C6000 Embedded Application Binary Interface (EABI).
All code in an EABI application must be built for EABI. Make sure all your libraries are available in

SPRU187U – July 2012 Using the C/C++ Compiler 55

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling Application Binary Interface www.ti.com

EABI mode before migrating your existing COFF ABI systems to C6000 EABI. See
http://tiexpressdsp.com/index.php/EABI_Support_in_C6000_Compiler for full details.
For more details on the different ABIs, see Section 6.11.

56 Using the C/C++ Compiler SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Enabling Entry Hook and Exit Hook Functions

2.17 Enabling Entry Hook and Exit Hook Functions

An entry hook is a routine that is called upon entry to each function in the program. An exit hook is a
routine that is called upon exit of each function. Applications for hooks include debugging, trace, profiling,
and stack overflow checking.
Entry and exit hooks are enabled using the following options:

--entry_hook[=name] Enables entry hooks. If specified, the hook function is called name. Otherwise,
the default entry hook function name is __entry_hook.
--entry_parm{=name| Specify the parameters to the hook function. The name parameter specifies
address|none} that the name of the calling function is passed to the hook function as an
argument. In this case the signature for the hook function is: void hook(const
char *name);
The address parameter specifies that the address of the calling function is
passed to the hook function. In this case the signature for the hook function is:
void hook(void (*addr)());
The none parameter specifies that the hook is called with no parameters. This
is the default. In this case the signature for the hook function is: void
hook(void);
--exit_hook[=name] Enables exit hooks. If specified, the hook function is called name. Otherwise,
the default exit hook function name is __exit_hook.
--exit_parm{=name| Specify the parameters to the hook function. The name parameter specifies
address|none} that the name of the calling function is passed to the hook function as an
argument. In this case the signature for the hook function is: void hook(const
char *name);
The address parameter specifies that the address of the calling function is
passed to the hook function. In this case the signature for the hook function is:
void hook(void (*addr)());
The none parameter specifies that the hook is called with no parameters. This
is the default. In this case the signature for the hook function is: void
hook(void);

The presence of the hook options creates an implicit declaration of the hook function with the given
signature. If a declaration or definition of the hook function appears in the compilation unit compiled with
the options, it must agree with the signatures listed above.
In C++, the hooks are declared extern "C". Thus you can define them in C (or assembly) without being
concerned with name mangling.
Hooks can be declared inline, in which case the compiler tries to inline them using the same criteria as
other inline functions.
Entry hooks and exit hooks are independent. You can enable one but not the other, or both. The same
function can be used as both the entry and exit hook.
You must take care to avoid recursive calls to hook functions. The hook function should not call any
function which itself has hook calls inserted. To help prevent this, hooks are not generated for inline
functions, or for the hook functions themselves.
You can use the --remove_hooks_when_inlining option to remove entry/exit hooks for functions that are
auto-inlined by the optimizer.
See Section 6.9.23 for information about the NO_HOOKS pragma.

SPRU187U – July 2012 Using the C/C++ Compiler 57

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 3
SPRU187U – July 2012

Optimizing Your Code

The compiler tools can perform many optimizations to improve the execution speed and reduce the size of
C and C++ programs by simplifying loops, software pipelining, rearranging statements and expressions,
and allocating variables into registers.
This chapter describes how to invoke different levels of optimization and describes which optimizations are
performed at each level. This chapter also describes how you can use the Interlist feature when
performing optimization and how you can profile or debug optimized code.

Topic ........................................................................................................................... Page

3.1 Invoking Optimization ........................................................................................ 59

3.2 Optimizing Software Pipelining ........................................................................... 60
3.3 Redundant Loops .............................................................................................. 67
3.4 Utilizing the Loop Buffer Using SPLOOP on C6400+, C6740, and C6600 .................. 68
3.5 Reducing Code Size (--opt_for_space (or -ms) Option) .......................................... 68
3.6 Performing File-Level Optimization (--opt_level=3 option) ...................................... 70
3.7 Performing Program-Level Optimization (--program_level_compile and --
opt_level=3 options) .......................................................................................... 71
3.8 Using Feedback Directed Optimization ................................................................ 73
3.9 Using Profile Information to Get Better Program Cache Layout and Analyze Code
Coverage .......................................................................................................... 77
3.10 Indicating Whether Certain Aliasing Techniques Are Used ..................................... 87
3.11 Prevent Reordering of Associative Floating-Point Operations ................................ 88
3.12 Use Caution With asm Statements in Optimized Code ........................................... 89
3.13 Automatic Inline Expansion (--auto_inline Option) ................................................. 89
3.14 Using Performance Advice to Optimize your Code ................................................ 90

58 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Invoking Optimization

3.1 Invoking Optimization

The C/C++ compiler is able to perform various optimizations. High-level optimizations are performed in the
optimizer and low-level, target-specific optimizations occur in the code generator. Use high-level
optimization levels, such as --opt_level=2 and --opt_level=3, to achieve optimal code.
The easiest way to invoke optimization is to use the compiler program, specifying the --opt_level=n option
on the compiler command line. You can use -On to alias the --opt_level option. The n denotes the level of
optimization (0, 1, 2, and 3), which controls the type and degree of optimization.
• --opt_level=0 or -O0
– Performs control-flow-graph simplification
– Allocates variables to registers
– Performs loop rotation
– Eliminates unused code
– Simplifies expressions and statements
– Expands calls to functions declared inline
• --opt_level=1 or -O1
Performs all --opt_level=0 (-O0) optimizations, plus:
– Performs local copy/constant propagation
– Removes unused assignments
– Eliminates local common expressions
• --opt_level=2 or -O2
Performs all --opt_level=1 (-O1) optimizations, plus:
– Performs software pipelining (see Section 3.2)
– Performs loop optimizations
– Eliminates global common subexpressions
– Eliminates global unused assignments
– Converts array references in loops to incremented pointer form
– Performs loop unrolling
The optimizer uses --opt_level=2 (-O2) as the default if you use --opt_level (-O) without an optimization
level.
• --opt_level=3 or -O3
Performs all --opt_level=2 (-O2) optimizations, plus:
– Removes all functions that are never called
– Simplifies functions with return values that are never used
– Inlines calls to small functions
– Reorders function declarations; the called functions attributes are known when the caller is
optimized
– Propagates arguments into function bodies when all calls pass the same value in the same
argument position
– Identifies file-level variable characteristics
If you use --opt_level=3 (-O3), see Section 3.6 and Section 3.7 for more information.
The levels of optimizations described above are performed by the stand-alone optimization pass. The
code generator performs several additional optimizations, particularly processor-specific optimizations. It
does so regardless of whether you invoke the optimizer. These optimizations are always enabled,
although they are more effective when the optimizer is used.

SPRU187U – July 2012 Optimizing Your Code 59

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Optimizing Software Pipelining www.ti.com

Do Not Lower the Optimization Level to Control Code Size

NOTE: To reduce code size, do not lower the level of optimization. Instead, use the --opt_for_space
option to control the code size/performance tradeoff. Higher optimization levels (--opt_level
or -O) combined with high --opt_for_space levels result in the smallest code size. For more
information, see Section 3.5.

The --opt_level= n (-O n ) Option Applies to the Assembly Optimizer

NOTE: The --opt_level=n (-On) option should also be used with the assembly optimizer. Although
the assembly optimizer does not perform all the optimizations described here, key
optimizations such as software pipelining and loop unrolling require the --opt_level (-O)
option.

3.2 Optimizing Software Pipelining

Software pipelining schedules instructions from a loop so that multiple iterations of the loop execute in
parallel. At optimization levels --opt_level=2 (or -O2) and --opt_level=3 (or -O3), the compiler usually
attempts to software pipeline your loops. The --opt_for_space option also affects the compiler's decision to
attempt to software pipeline loops. In general, code size and performance are better when you use the --
opt_level=2 or --opt_level=3 options. (See Section 3.1.)
Figure 3-1 illustrates a software-pipelined loop. The stages of the loop are represented by A, B, C, D, and
E. In this figure, a maximum of five iterations of the loop can execute at one time. The shaded area
represents the loop kernel. In the loop kernel, all five stages execute in parallel. The area above the kernel
is known as the pipelined loop prolog, and the area below the kernel is known as the pipelined loop epilog.

Figure 3-1. Software-Pipelined Loop

A1
B1 A2
Pipelined-loop prolog
C1 B2 A3
D1 C2 B3 A4
E1 D2 C3 B4 A5 Kernel
E2 D3 C4 B5
E3 D4 C5
Pipelined-loop epilog
E4 D5
E5

If you enter comments on instructions in your linear assembly input file, the compiler moves the comments
to the output file along with additional information. It attaches a 2-tuple <x, y> to the comments to specify
the iteration and cycle of the loop an instruction is on in the software pipeline. The zero-based number x
represents the iteration the instruction is on during the first execution of the loop kernel. The zero-based
number y represents the cycle that the instruction is scheduled on within a single iteration of the loop.
For more information about software pipelining, see the TMS320C6000 Programmer's Guide.

60 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Optimizing Software Pipelining

3.2.1 Turn Off Software Pipelining (--disable_software_pipelining Option)

At optimization levels --opt_level=2 (or -O2) and -O3, the compiler attempts to software pipeline your
loops. You might not want your loops to be software pipelined for debugging reasons. Software-pipelined
loops are sometimes difficult to debug because the code is not presented serially. The --
disable_software_pipelining option affects both compiled C/C++ code and assembly optimized code.

Software Pipelining May Increase Code Size

NOTE: Software pipelining without the use of SPLOOP can lead to significant increases in code
size. To control code size for loops that get software pipelined, it is preferable to use the --
opt_for_space option rather than the --disable_software_pipelining option. The --
opt_for_space option is capable of disabling non-SPLOOP software pipelining if necessary to
achieve code size savings, but it does not affect the SPLOOP capability of C64x+ and C674x
devices. SPLOOP does not significantly increase code size, but can greatly speed up a loop.
Using --disable_software_pipelining disables all software pipelining including SPLOOP.

3.2.2 Software Pipelining Information

The compiler embeds software pipelined loop information in the .asm file. This information is used to
optimize C/C++ code or linear assembly code.
The software pipelining information appears as a comment in the .asm file before a loop and for the
assembly optimizer the information is displayed as the tool is running. Example 3-1 illustrates the
information that is generated for each loop.
The --debug_software_pipeline option adds additional information displaying the register usage at each
cycle of the loop kernel and displays the instruction ordering of a single iteration of the software pipelined
loop.

More Details on Software Pipelining Information

NOTE: Refer to the TMS320C6000 Programmer’s Guide for details on the information and
messages that can appear in the Software Pipelining Information comment block before each
loop.

SPRU187U – July 2012 Optimizing Your Code 61

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Optimizing Software Pipelining www.ti.com

Example 3-1. Software Pipelining Information

*----------------------------------------------------------------------------*

;* SOFTWARE PIPELINE INFORMATION

;*
;* Known Minimum Trip Count : 2
;* Known Maximum Trip Count : 2
;* Known Max Trip Count Factor : 2
;* Loop Carried Dependency Bound(^) : 4
;* Unpartitioned Resource Bound : 4
;* Partitioned Resource Bound(*) : 5
;* Resource Partition:
;* A-side B-side
;* .L units 2 3
;* .S units 4 4
;* .D units 1 0
;* .M units 0 0
;* .X cross paths 1 3
;* .T address paths 1 0
;* Long read paths 0 0
;* Long write paths 0 0
;* Logical ops (.LS) 0 1 (.L or .S unit)
;* Addition ops (.LSD) 6 3 (.L or .S or .D unit)
;* Bound(.L .S .LS) 3 4
;* Bound(.L .S .D .LS .LSD) 5* 4
;*

;* Searching for software pipeline schedule at ...

;* ii = 5 Register is live too long
;* ii = 6 Did not find schedule
;* ii = 7 Schedule found with 3 iterations in parallel
;* done
;*
;* Epilog not entirely removed
;* Collapsed epilog stages : 1
;*
;* Prolog not removed
;* Collapsed prolog stages : 0
;*
;* Minimum required memory pad : 2 bytes
;*
;* Minimum safe trip count : 2
;*
;*----------------------------------------------------------------------------*

The terms defined below appear in the software pipelining information. For more information on each
term, see the TMS320C6000 Programmer's Guide.
• Loop unroll factor. The number of times the loop was unrolled specifically to increase performance
based on the resource bound constraint in a software pipelined loop.
• Known minimum trip count. The minimum number of times the loop will be executed.
• Known maximum trip count. The maximum number of times the loop will be executed.
• Known max trip count factor. Factor that would always evenly divide the loops trip count. This
information can be used to possibly unroll the loop.
• Loop label. The label you specified for the loop in the linear assembly input file. This field is not
present for C/C++ code.
• Loop carried dependency bound. The distance of the largest loop carry path. A loop carry path
occurs when one iteration of a loop writes a value that must be read in a future iteration. Instructions
that are part of the loop carry bound are marked with the ^ symbol.

62 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Optimizing Software Pipelining

• Initiation interval (ii). The number of cycles between the initiation of successive iterations of the loop.
The smaller the initiation interval, the fewer cycles it takes to execute a loop.
• Resource bound. The most used resource constrains the minimum initiation interval. If four
instructions require a .D unit, they require at least two cycles to execute (4 instructions/2 parallel .D
units).
• Unpartitioned resource bound. The best possible resource bound values before the instructions in
the loop are partitioned to a particular side.
• Partitioned resource bound (*). The resource bound values after the instructions are partitioned.
• Resource partition. This table summarizes how the instructions have been partitioned. This
information can be used to help assign functional units when writing linear assembly. Each table entry
has values for the A-side and B-side registers. An asterisk is used to mark those entries that determine
the resource bound value. The table entries represent the following terms:
– .L units is the total number of instructions that require .L units.
– .S units is the total number of instructions that require .S units.
– .D units is the total number of instructions that require .D units.
– .M units is the total number of instructions that require .M units.
– .X cross paths is the total number of .X cross paths.
– .T address paths is the total number of address paths.
– Long read path is the total number of long read port paths.
– Long write path is the total number of long write port paths.
– Logical ops (.LS) is the total number of instructions that can use either the .L or .S unit.
– Addition ops (.LSD) is the total number of instructions that can use either the .L or .S or .D unit
• Bound(.L .S .LS). The resource bound value as determined by the number of instructions that use the
.L and .S units. It is calculated with the following formula:
Bound(.L .S .LS ) = ceil((.L + .S + .LS) / 2)
• Bound(.L .S .D .LS .LSD). The resource bound value as determined by the number of instructions that
use the .D, .L, and .S units. It is calculated with the following formula:
Bound(.L .S .D .LS .SLED) = ceil((.L + .S + .D + .LS + .LSD) / 3)
• Minimum required memory pad. The number of bytes that are read if speculative execution is
enabled. See Section 3.2.3 for more information.

3.2.2.1 Loop Disqualified for Software Pipelining Messages

The following messages appear if the loop is completely disqualified for software pipelining:
• Bad loop structure. This error is very rare and can stem from the following:
– An asm statement inserted in the C code inner loop
– Parallel instructions being used as input to the Linear Assembly Optimizer
– Complex control flow such as GOTO statements and breaks
• Loop contains a call. Sometimes the compiler may not be able to inline a function call that is in a
loop. Because the compiler could not inline the function call, the loop could not be software pipelined.
• Too many instructions. There are too many instructions in the loop to software pipeline.
• Software pipelining disabled. Software pipelining has been disabled by a command-line option, such
as when using the --disable_software_pipelining option, not using the --opt_level=2 (or -O2) or --
opt_level=3 (or -O3) option, or using the --opt_for_space=2 or --opt_for_space=3 option.
• Uninitialized trip counter. The trip counter may not have been set to an initial value.
• Suppressed to prevent code expansion. Software pipelining may be suppressed because of the --
opt_for_space=1 option. When the --opt_for_space=1 option is used, software pipelining is disabled in
less promising cases to reduce code size. To enable pipelining, use --opt_for_space=0 or omit the --
opt_for_space option altogether.

SPRU187U – July 2012 Optimizing Your Code 63

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Optimizing Software Pipelining www.ti.com

• Loop carried dependency bound too large. If the loop has complex loop control, try --
speculate_loads according to the recommendations in Section 3.2.3.2.
• Cannot identify trip counter. The loop trip counter could not be identified or was used incorrectly in
the loop body.

3.2.2.2 Pipeline Failure Messages

The following messages can appear when the compiler or assembly optimizer is processing a software
pipeline and it fails:
• Address increment is too large. An address register's offset must be adjusted because the offset is
out of range of the C6000's offset addressing mode. You must minimize address register offsets.
• Cannot allocate machine registers. A software pipeline schedule was found, but it cannot allocate
machine registers for the schedule. Simplification of the loop may help.
The register usage for the schedule found at the given ii is displayed. This information can be used
when writing linear assembly to balance register pressure on both sides of the register file. For
example:
ii = 11 Cannot allocate machine registers
Regs Live Always : 3/0 (A/B-side)
Max Regs Live : 20/14
Max Condo Regs Live : 2/1
– Regs Live Always. The number of values that must be assigned a register for the duration of the
whole loop body. This means that these values must always be allocated registers for any given
schedule found for the loop.
– Max Regs Live. Maximum number of values live at any given cycle in the loop that must be
allocated to a register. This indicates the maximum number of registers required by the schedule
found.
– Max Cond Regs Live. Maximum number of registers live at any given cycle in the loop kernel that
must be allocated to a condition register.
• Cycle count too high. Never profitable. With the schedule that the compiler found for the loop, it is
more efficient to use a non-software-pipelined version.
• Did not find schedule. The compiler was unable to find a schedule for the software pipeline at the
given ii (iteration interval). You should simplify the loop and/or eliminate loop carried dependencies.
• Iterations in parallel > minimum or maximum trip count. A software pipeline schedule was found,
but the schedule has more iterations in parallel than the minimum or maximum loop trip count. You
must enable redundant loops or communicate the trip information.
• Speculative threshold exceeded. It would be necessary to speculatively load beyond the threshold
currently specified by the --speculate_loads option. You must increase the --speculate_loads threshold
as recommended in the software-pipeline feedback located in the assembly file.
• Register is live too long. A register must have a value that exists (is live) for more than ii cycles. You
may insert MV instructions to split register lifetimes that are too long.
If the assembly optimizer is being used, the .sa file line numbers of the instructions that define and use
the registers that are live too long are listed after this failure message. For example:
ii = 9 Register is live too long
|10| -> |17|
This means that the instruction that defines the register value is on line 10 and the instruction that uses
the register value is on line 17 in the .sa file.
• Too many predicates live on one side. The C6000 has predicate, or conditional, registers available
for use with conditional instructions. There are five predicate registers on the C6200 and C6700, and
six predicate registers on the C6400, C6400+, and C6700+. There are two or three on the A side and
three on the B side. Sometimes the particular partition and schedule combination requires more than
these available registers.
• Schedule found with N iterations in parallel. (This is not a failure message.) A software pipeline
schedule was found with N iterations executing in parallel.

64 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Optimizing Software Pipelining

• Too many reads of one register. The same register can be read a maximum of four times per cycle
with the C6200 or C6700 core. The C6400 core can read the same register any number of times per
cycle.
• Trip variable used in loop - Cannot adjust trip count. The loop trip counter has a use in the loop
other than as a loop trip counter.

3.2.2.3 Register Usage Table Generated by the --debug_software_pipeline Option

The --debug_software_pipeline option places additional software pipeline feedback in the generated
assembly file. This information includes a single scheduled iteration view of the software pipelined loop.
If software pipelining succeeds for a given loop, and the --debug_software_pipeline option was used
during the compilation process, a register usage table is added to the software pipelining information
comment block in the generated assembly code.
The numbers on each row represent the cycle number within the loop kernel.
Each column represents one register on the TMS320C6000. The registers are labeled in the first three
rows of the register usage table and should be read columnwise.
An * in a table entry indicates that the register indicated by the column header is live on the kernel
execute packet indicated by the cycle number labeling each row.
An example of the register usage table follows:
;* Searching for software pipeline schedule at
;* ii = 15 Schedule found with 2 iterations in parallel
;*
;* Register Usage Table:
;* +---------------------------------+
;* |AAAAAAAAAAAAAAAA|BBBBBBBBBBBBBBBB|
;* |0000000000111111|0000000000111111|
;* |0123456789012345|0123456789012345|
;* |----------------+----------------|
;* 0: |*** **** |*** ****** |
;* 1: |**** **** |*** ****** |
;* 2: |**** **** |*** ****** |
;* 3: | ** ***** |*** ****** |
;* 4: | ** ***** |*** ****** |
;* 5: | ** ***** |*** ****** |
;* 6: | ** ***** |********** |
;* 7: |*** ***** |** ******* |
;* 8: |**** ***** |*********** |
;* 9: |********** |** ******** |
;* 10: |*********** |** ********* |
;* 11: |*********** |** ********* |
;* 12: |********** |************ |
;* 13: |**** ***** |** ******* * |
;* 14: |*** ***** |*** ****** * |
;* +---------------------------------+
;*

This example shows that on cycle 0 (first execute packet) of the loop kernel, registers A0, A1, A2, A6, A7,
A8, A9, B0, B1, B2, B4, B5, B6, B7, B8, and B9 are all live during this cycle.

3.2.3 Collapsing Prologs and Epilogs for Improved Performance and Code Size
When a loop is software pipelined, a prolog and epilog are generally required. The prolog is used to pipe
up the loop and epilog is used to pipe down the loop.
In general, a loop must execute a minimum number of iterations before the software-pipelined version can
be safely executed. If the minimum known trip count is too small, either a redundant loop is added or
software pipelining is disabled. Collapsing the prolog and epilog of a loop can reduce the minimum trip
count necessary to safely execute the pipelined loop.

SPRU187U – July 2012 Optimizing Your Code 65

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Optimizing Software Pipelining www.ti.com

Collapsing can also substantially reduce code size. Some of this code size growth is due to the redundant
loop. The remainder is due to the prolog and epilog.
The prolog and epilog of a software-pipelined loop consists of up to p-1 stages of length ii, where p is the
number of iterations that are executed in parallel during the steady state and ii is the cycle time for the
pipelined loop body. During prolog and epilog collapsing the compiler tries to collapse as many stages as
possible. However, over-collapsing can have a negative performance impact. Thus, by default, the
compiler attempts to collapse as many stages as possible without sacrificing performance. When the --
opt_for_space=0 or --opt_for_space=1 options are invoked, the compiler increasingly favors code size
over performance.

3.2.3.1 Speculative Execution

When prologs and epilogs are collapsed, instructions might be speculatively executed, thereby causing
loads to addresses beyond either end of the range explicitly read within the loop. By default, the compiler
cannot speculate loads because this could cause an illegal memory location to be read. Sometimes, the
compiler can predicate these loads to prevent over execution. However, this can increase register
pressure and might decrease the total amount collapsing which can be performed.
When the --speculate_loads=n option is used, the speculative threshold is increased from the default of 0
to n. When the threshold is n, the compiler can allow a load to be speculatively executed as the memory
location it reads will be no more than n bytes before or after some location explicitly read within the loop. If
the n is omitted, the compiler assumes the speculative threshold is unlimited. To specify this in Code
Composer Studio, select the Speculate Threshold check box and leave the text box blank in the Build
Options dialog box on the Compiler tab, Advanced category.
Collapsing can usually reduce the minimum safe trip count. If the minimum known trip count is less than
the minimum safe trip count, a redundant loop is required. Otherwise, pipelining must be suppressed. Both
these values can be found in the comment block preceding a software pipelined loop.
;* Known Minimum Trip Count : 1
....
;* Minimum safe trip count : 7

If the minimum safe trip count is greater than the minimum known trip count, use of --speculate_loads is
highly recommended, not only for code size, but for performance.
When using --speculate_loads, you must ensure that potentially speculated loads will not cause illegal
reads. This can be done by padding the data sections and/or stack, as needed, by the required memory
pad in both directions. The required memory pad for a given software-pipelined loop is also provided in the
comment block for that loop.
;* Minimum required memory pad : 8 bytes

3.2.3.2 Selecting the Best Threshold Value

When a loop is software pipelined, the comment block preceding the loop provides the following
information:
• Required memory pad for this loop
• The minimum value of n needed to achieve this software pipeline schedule and level of collapsing
• Suggestion for a larger value of n to use which might allow additional collapsing
This information shows up in the comment block as follows:
;* Minimum required memory pad : 5 bytes
;* Minimum threshold value : --speculate_loads=7
;*
;* For further improvement on this loop, try option --speculate_loads=14

For safety, the example loop requires that array data referenced within this loop be preceded and followed
by a pad of at least 5 bytes. This pad can consist of other program data. The pad will not be modified. In
many cases, the threshold value (namely, the minimum value of the argument to --speculate_loads that is
needed to achieve a particular schedule and level of collapsing) is the same as the pad. However, when it
is not, the comment block will also include the minimum threshold value. In the case of this loop, the
threshold value must be at least 7 to achieve this level of collapsing.

66 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Redundant Loops

Beginning with v7.4.0 of the C6000 Code Generation Tools, the compiler and linker can provide automatic
load speculation via the auto argument to the --speculate_loads option (i.e. --speculate_loads=auto or -
mh=auto). Use of the auto argument makes it easier to use and benefit from speculative load
optimizations. This option can generate speculative loads of up to 256 bytes beyond memory that the
compiler can prove to be allocated.
In addition, the compiler communicates information to the linker to help automatically ensure the required
pre- and post-padding:
• If the symbol of the speculatively loaded buffer is known during compile time, the linker will ensure the
object pointed to by the symbol has the required padding to let the speculative load access legal
memory.
• If the symbol information is not known during compile time, the linker will ensure that the placement of
data sections will allow legal accessing beyond the boundaries of the data sections. The linker does
this by simply padding the start and end of the memory range(s) where the data sections are placed.
However, you can also specify the speculative loads threshold explicitly via the --speculate_loads=n
option, where n is at least the minimum required pad (as explained earlier), but you also need to consider
whether a larger threshold value would facilitate additional collapsing. This information is also provided, if
applicable. For example, in the above comment block, a threshold value of 14 might facilitate further
collapsing. If you choose the auto argument to --speculate_loads, the compiler will consider the larger
threshold value automatically.

3.3 Redundant Loops

Every loop iterates some number of times before the loop terminates. The number of iterations is called
the trip count. The variable used to count each iteration is the trip counter. When the trip counter reaches
a limit equal to the trip count, the loop terminates. The C6000 tools use the trip count to determine
whether or not a loop can be pipelined. The structure of a software pipelined loop requires the execution
of a minimum number of loop iterations (a minimum trip count) in order to fill or prime the pipeline.
The minimum trip count for a software pipelined loop is determined by the number of iterations executing
in parallel. In Figure 3-1, the minimum trip count is five. In the following example A, B, and C are
instructions in a software pipeline, so the minimum trip count for this single-cycle software pipelined loop is
three.

A
B A
C B A ←Three iterations in parallel = minimum trip count
C B
C

When the C6000 tools cannot determine the trip count for a loop, then by default two loops and control
logic are generated. The first loop is not pipelined, and it executes if the run-time trip count is less than the
loop's minimum trip count. The second loop is the software pipelined loop, and it executes when the run-
time trip count is greater than or equal to the minimum trip count. At any given time, one of the loops is a
redundant loop. For example:
foo(N) /* N is the trip count */
{
for (I=0; I < N; I++) /* I is the trip counter */
}

After finding a software pipeline for the loop, the compiler transforms foo() as below, assuming the
minimum trip count for the loop is 3. Two versions of the loop would be generated and the following
comparison would be used to determine which version should be executed:
foo(N)
{
if (N < 3)
{
for (I=0; I < N; I++) /* Unpipelined version */
}

SPRU187U – July 2012 Optimizing Your Code 67

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Utilizing the Loop Buffer Using SPLOOP on C6400+, C6740, and C6600 www.ti.com

else
}
for (I=0; I < N; I++) /* Pipelined version */
}
}
foo(50); /* Execute software pipelined loop */
foo(2); /* Execute loop (unpipelined)*/

You may be able to help the compiler avoid producing redundant loops with the use of --
program_level_compile --opt_level=3 (see Section 3.7) or the use of the MUST_ITERATE pragma (see
Section 6.9.20).

Turning Off Redundant Loops

NOTE: Specifying any --opt_for_space option turns off redundant loops.

3.4 Utilizing the Loop Buffer Using SPLOOP on C6400+, C6740, and C6600
The C6400+, C6740, and C6600 ISAs have a loop buffer which improves performance and reduces code
size for software pipelined loops. The loop buffer provides the following benefits:
• Code size. A single iteration of the loop is stored in program memory.
• Interrupt latency. Loops executing out of the loop buffer are interruptible.
• Improves performance for loops with unknown trip counts and eliminates redundant loops.
• Reduces or eliminates the need for speculated loads.
• Reduces power usage.
You can tell that the compiler is using the loop buffer when you find SPLOOP(D/W) at the beginning of a
software pipelined loop followed by an SPKERNEL at the end. Refer to the TMS320C6400/C6400+ CPU
and Instruction Set Reference Guide for information on SPLOOP.
When the --opt_for_space option is not used, the compiler will not use the loop buffer if it can find a faster
software pipelined loop without it. When using the --opt_for_space option, the compiler will use the loop
buffer when it can.
The compiler does not generate code for the loop buffer (SPLOOP/D/W) when any of the following occur:
• ii (initiation interval) > 14 cycles
• Dynamic length (of a single iteration) > 48 cycles
• The optimizer completely unrolls the loop
• Code contains elements that disqualify normal software pipelining (call in loop, complex control code in
loop, etc.). See the TMS320C6000 Programmer's Guide for more information.

3.5 Reducing Code Size (--opt_for_space (or -ms) Option)

When using the --opt_level=n option (or -On), you are telling the compiler to optimize your code. The
higher the value of n, the more effort the compiler invests in optimizing your code. However, you might still
need to tell the compiler what your optimization priorities are. By default, when --opt_level=2 or -
opt_level=3 is specified, the compiler optimizes primarily for performance. (Under lower optimization
levels, the priorities are compilation time and debugging ease.) You can adjust the priorities between
performance and code size by using the code size flag --opt_for_space=n. The --opt_for_space=0, --
opt_for_space=1, --opt_for_space=2 and --opt_for_space=3 options increasingly favor code size over
performance.
When you specify --silicon_version=6400+ in conjunction with the --opt_for_space option, the code will be
tailored for compression. That is, more instructions are tailored so they will more likely be converted from
32-bit to 16-bit instructions when assembled.

68 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Reducing Code Size (--opt_for_space (or -ms) Option)

It is recommended that a code size flag not be used with the most performance-critical code. Using --
opt_for_space=0 or --opt_for_space=1 is recommended for all but the most performance-critical code.
Using --opt_for_space=2 or --opt_for_space=3 is recommended for seldom-executed code. Either --
opt_for_space=2 or --opt_for_space=3 should be used if you need minimum code size. It is generally
recommended that the code size flags be combined with --opt_level=2 or --opt_level=3.

Disabling Code-Size Optimizations or Reducing the Optimization Level

NOTE: If you reduce optimization and/or do not use code size flags, you are disabling code-size
optimizations and sacrificing performance.

The --opt_for_space Option is Equivalent to --opt_for_space=0

NOTE: If you use --opt_for_space with no code size level number specified, the option level defaults
to --opt_for_space=0.

SPRU187U – July 2012 Optimizing Your Code 69

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Performing File-Level Optimization (--opt_level=3 option) www.ti.com

3.6 Performing File-Level Optimization (--opt_level=3 option)

The --opt_level=3 option (aliased as the -O3 option) instructs the compiler to perform file-level
optimization. You can use the --opt_level=3 option alone to perform general file-level optimization, or you
can combine it with other options to perform more specific optimizations. The options listed in Table 3-1
work with --opt_level=3 to perform the indicated optimization:

Table 3-1. Options That You Can Use With --opt_level=3

If You ... Use this Option See
Have files that redeclare standard library functions --std_lib_func_defined Section 3.6.1
--std_lib_func_redefined
Want to create an optimization information file --gen_opt_level=n Section 3.6.2
Want to compile multiple source files --program_level_compile Section 3.7

Do Not Lower the Optimization Level to Control Code Size

NOTE: When trying to reduce code size, do not lower the level of optimization, as you might see an
increase in code size. Instead, use the --opt_for_space option to control the code.

3.6.1 Controlling File-Level Optimization (--std_lib_func_def Options)

When you invoke the compiler with the --opt_level=3 option, some of the optimizations use known
properties of the standard library functions. If your file redeclares any of these standard library functions,
these optimizations become ineffective. Use Table 3-2 to select the appropriate file-level optimization
option.

Table 3-2. Selecting a File-Level Optimization Option

If Your Source File... Use this Option
Declares a function with the same name as a standard library function --std_lib_func_redefined
Contains but does not alter functions declared in the standard library --std_lib_func_defined
Does not alter standard library functions, but you used the --std_lib_func_redefined or -- --std_lib_func_not_defined
std_lib_func_defined option in a command file or an environment variable. The --
std_lib_func_not_defined option restores the default behavior of the optimizer.

3.6.2 Creating an Optimization Information File (--gen_opt_info Option)

When you invoke the compiler with the --opt_level=3 option, you can use the --gen_opt_info option to
create an optimization information file that you can read. The number following the option denotes the
level (0, 1, or 2). The resulting file has an .nfo extension. Use Table 3-3 to select the appropriate level to
append to the option.

Table 3-3. Selecting a Level for the --gen_opt_info Option

If you... Use this option
Do not want to produce an information file, but you used the --gen_opt_level=1 or --gen_opt_level=2 --gen_opt_info=0
option in a command file or an environment variable. The --gen_opt_level=0 option restores the
default behavior of the optimizer.
Want to produce an optimization information file --gen_opt_info=1
Want to produce a verbose optimization information file --gen_opt_info=2

70 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Performing Program-Level Optimization (--program_level_compile and --opt_level=3 options)

3.7 Performing Program-Level Optimization (--program_level_compile and --opt_level=3

options)
You can specify program-level optimization by using the --program_level_compile option with the --
opt_level=3 option (aliased as -O3). With program-level optimization, all of your source files are compiled
into one intermediate file called a module. The module moves to the optimization and code generation
passes of the compiler. Because the compiler can see the entire program, it performs several
optimizations that are rarely applied during file-level optimization:
• If a particular argument in a function always has the same value, the compiler replaces the argument
with the value and passes the value instead of the argument.
• If a return value of a function is never used, the compiler deletes the return code in the function.
• If a function is not called directly or indirectly by main(), the compiler removes the function.
To see which program-level optimizations the compiler is applying, use the --gen_opt_level=2 option to
generate an information file. See Section 3.6.2 for more information.
In Code Composer Studio, when the --program_level_compile option is used, C and C++ files that have
the same options are compiled together. However, if any file has a file-specific option that is not selected
as a project-wide option, that file is compiled separately. For example, if every C and C++ file in your
project has a different set of file-specific options, each is compiled separately, even though program-level
optimization has been specified. To compile all C and C++ files together, make sure the files do not have
file-specific options. Be aware that compiling C and C++ files together may not be safe if previously you
used a file-specific option.

Compiling Files With the --program_level_compile and --keep_asm Options

NOTE: If you compile all files with the --program_level_compile and --keep_asm options, the
compiler produces only one .asm file, not one for each corresponding source file.

3.7.1 Controlling Program-Level Optimization (--call_assumptions Option)

You can control program-level optimization, which you invoke with --program_level_compile --opt_level=3,
by using the --call_assumptions option. Specifically, the --call_assumptions option indicates if functions in
other modules can call a module's external functions or modify a module's external variables. The number
following --call_assumptions indicates the level you set for the module that you are allowing to be called or
modified. The --opt_level=3 option combines this information with its own file-level analysis to decide
whether to treat this module's external function and variable declarations as if they had been declared
static. Use Table 3-4 to select the appropriate level to append to the --call_assumptions option.

Table 3-4. Selecting a Level for the --call_assumptions Option

If Your Module … Use this Option
Has functions that are called from other modules and global variables that are modified in other --call_assumptions=0
modules
Does not have functions that are called by other modules but has global variables that are modified in --call_assumptions=1
other modules
Does not have functions that are called by other modules or global variables that are modified in other --call_assumptions=2
modules
Has functions that are called from other modules but does not have global variables that are modified --call_assumptions=3
in other modules

In certain circumstances, the compiler reverts to a different --call_assumptions level from the one you
specified, or it might disable program-level optimization altogether. Table 3-5 lists the combinations of --
call_assumptions levels and conditions that cause the compiler to revert to other --call_assumptions
levels.

SPRU187U – July 2012 Optimizing Your Code 71

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Performing Program-Level Optimization (--program_level_compile and --opt_level=3 options) www.ti.com

Table 3-5. Special Considerations When Using the --call_assumptions Option

Then the --call_assumptions
If Your Option is... Under these Conditions... Level...
Not specified The --opt_level=3 optimization level was specified Defaults to --call_assumptions=2
Not specified The compiler sees calls to outside functions under the -- Reverts to --call_assumptions=0
opt_level=3 optimization level
Not specified Main is not defined Reverts to --call_assumptions=0
--call_assumptions=1 or No function has main defined as an entry point and functions are Reverts to --call_assumptions=0
--call_assumptions=2 not identified by the FUNC_EXT_CALLED pragma
--call_assumptions=1 or No interrupt function is defined Reverts to --call_assumptions=0
--call_assumptions=2
--call_assumptions=1 or Functions are identified by the FUNC_EXT_CALLED pragma Remains --call_assumptions=1
--call_assumptions=2 or --call_assumptions=2
--call_assumptions=3 Any condition Remains --call_assumptions=3

In some situations when you use --program_level_compile and --opt_level=3, you must use a --
call_assumptions option or the FUNC_EXT_CALLED pragma. See Section 3.7.2 for information about
these situations.

3.7.2 Optimization Considerations When Mixing C/C++ and Assembly

If you have any assembly functions in your program, you need to exercise caution when using the --
program_level_compile option. The compiler recognizes only the C/C++ source code and not any
assembly code that might be present. Because the compiler does not recognize the assembly code calls
and variable modifications to C/C++ functions, the --program_level_compile option optimizes out those
C/C++ functions. To keep these functions, place the FUNC_EXT_CALLED pragma (see Section 6.9.10)
before any declaration or reference to a function that you want to keep.
Another approach you can take when you use assembly functions in your program is to use the --
call_assumptions=n option with the --program_level_compile and --opt_level=3 options (see
Section 3.7.1).
In general, you achieve the best results through judicious use of the FUNC_EXT_CALLED pragma in
combination with --program_level_compile --opt_level=3 and --call_assumptions=1 or --
call_assumptions=2.
If any of the following situations apply to your application, use the suggested solution:
Situation — Your application consists of C/C++ source code that calls assembly functions. Those
assembly functions do not call any C/C++ functions or modify any C/C++ variables.
Solution — Compile with --program_level_compile --opt_level=3 --call_assumptions=2 to tell the compiler
that outside functions do not call C/C++ functions or modify C/C++ variables. See Section 3.7.1 for
information about the --call_assumptions=2 option.
If you compile with the --program_level_compile --opt_level=3 options only, the compiler reverts
from the default optimization level (--call_assumptions=2) to --call_assumptions=0. The compiler
uses --call_assumptions=0, because it presumes that the calls to the assembly language functions
that have a definition in C/C++ may call other C/C++ functions or modify C/C++ variables.
Situation — Your application consists of C/C++ source code that calls assembly functions. The assembly
language functions do not call C/C++ functions, but they modify C/C++ variables.
Solution — Try both of these solutions and choose the one that works best with your code:
• Compile with --program_level_compile --opt_level=3 --call_assumptions=1.
• Add the volatile keyword to those variables that may be modified by the assembly functions and
compile with --program_level_compile --opt_level=3 --call_assumptions=2.
See Section 3.7.1 for information about the --call_assumptions=n option.

72 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Feedback Directed Optimization

Situation — Your application consists of C/C++ source code and assembly source code. The assembly
functions are interrupt service routines that call C/C++ functions; the C/C++ functions that the
assembly functions call are never called from C/C++. These C/C++ functions act like main: they
function as entry points into C/C++.
Solution — Add the volatile keyword to the C/C++ variables that may be modified by the interrupts. Then,
you can optimize your code in one of these ways:
• You achieve the best optimization by applying the FUNC_EXT_CALLED pragma to all of the
entry-point functions called from the assembly language interrupts, and then compiling with --
program_level_compile --opt_level=3 --call_assumptions=2. Be sure that you use the pragma
with all of the entry-point functions. If you do not, the compiler might remove the entry-point
functions that are not preceded by the FUNC_EXT_CALLED pragma.
• Compile with --program_level_compile --opt_level=3 --call_assumptions=3. Because you do not
use the FUNC_EXT_CALLED pragma, you must use the --call_assumptions=3 option, which is
less aggressive than the --call_assumptions=2 option, and your optimization may not be as
effective.
Keep in mind that if you use --program_level_compile --opt_level=3 without additional options, the
compiler removes the C functions that the assembly functions call. Use the FUNC_EXT_CALLED
pragma to keep these functions.

3.8 Using Feedback Directed Optimization

Feedback directed optimization provides a method for finding frequently executed paths in an application
using compiler-based instrumentation. This information is fed back to the compiler and is used to perform
optimizations. This information is also used to provide you with information about application behavior.

3.8.1 Feedback Directed Optimization

Feedback directed optimization uses run-time feedback to identify and optimize frequently executed
program paths. Feedback directed optimization is a two-phase process.

3.8.1.1 Phase 1: Collect Program Profile Information

In this phase the compiler is invoked with the option --gen_profile_info, which instructs the compiler to add
instrumentation code to collect profile information. The compiler inserts a minimal amount of
instrumentation code to determine control flow frequencies. Memory is allocated to store counter
information.
The instrumented application program is executed on the target using representative input data sets. The
input data sets should correlate closely with the way the program is expected to be used in the end
product environment. When the program completes, a run-time-support function writes the collected
information into a profile data file called a PDAT file. Multiple executions of the program using different
input data sets can be performed and in such cases, the run-time-support function appends the collected
information into the PDAT file. The resulting PDAT file is post-processed using a tool called the Profile
Data Decoder or pdd6x. The pdd6x tool consolidates multiple data sets and formats the data into a
feedback file (PRF file, see Section 3.8.2) for consumption by phase 2 of feedback directed optimization.

3.8.1.2 Phase 2: Use Application Profile Information for Optimization

In this phase, the compiler is invoked with the --use_profile_info=file.prf option, which reads the specified
PRF file generated in phase 1. In phase 2, optimization decisions are made using the data generated
during phase 1. The profile feedback file is used to guide program optimization. The compiler optimizes
frequently executed program paths more aggressively.
The compiler uses data in the profile feedback file to guide certain optimizations of frequently executed
program paths.

SPRU187U – July 2012 Optimizing Your Code 73

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Feedback Directed Optimization www.ti.com

3.8.1.3 Generating and Using Profile Information

There are two options that control feedback directed optimization:

--gen_profile_info tells the compiler to add instrumentation code to collect profile information. When
the program executes the run-time-support exit() function, the profile data is
written to a PDAT file. If the environment variable TI_PROFDATA on the host is
set, the data is written into the specified file name. Otherwise, it uses the default
filename: pprofout.pdat. The full pathname of the PDAT file (including the directory
name) can be specified using the TI_PROFDATA host environment variable.
By default, the RTS profile data output routine uses the C I/O mechanism to write
data to the PDAT file. You can install a device handler for the PPHNDL device that
enables you to re-direct the profile data to a custom device driver routine.
Feedback directed optimization requires you to turn on at least skeletal debug
information when using the --gen_profile_info option. This enables the compiler to
output debug information that allows pdd6x to correlate compiled functions and
their associated profile data.
--use_profile_info specifies the profile information file(s) to use for performing phase 2 of feedback
directed optimization. More than one profile information file can be specified on the
command line; the compiler uses all input data from multiple information files. The
syntax for the option is:
--use_profile_info==file1[, file2, ..., filen]
If no filename is specified, the compiler looks for a file named pprofout.prf in the
directory where the compiler in invoked.

3.8.1.4 Example Use of Feedback Directed Optimization

These steps illustrate the creation and use of feedback directed optimization.
1. Generate profile information. (Skeletal debug is on by default.)
cl6x -mv6400+ --opt_level=2 --gen_profile_info foo.c --run_linker --output_file=foo.out
--library=lnk.cmd --library=rts64plus.lib
2. Execute the application.
The execution of the application creates a PDAT file named pprofout.pdat in the current (host)
directory. The application can be run on a simulator or on target hardware connected to a host
machine.
3. Process the profile data.
After running the application with multiple data-sets, run pdd6x on the PDAT files to create a profile
information (PRF) file to be used with --use_profile_info.
ppd6x -e foo.out -o pprofout.prf pprofout.pdat
4. Re-compile using the profile feedback file. Skeletal debug is not required.
cl6x -mv6400+ --opt_level=2 --use_profile_info=pprofout.prf foo.c --run_linker
--output_file=foo.out --library=lnk.cmd --library=rts64plus.lib

3.8.1.5 The .ppdata Section

The profile information collected in phase 1 is stored in the .ppdata section, which must be allocated into
target memory. The .ppdata section contains profiler counters for all functions compiled with --
gen_profile_info. The default lnk.cmd file in code generation tools version 6.1 and later has directives to
place the .ppdata section in data memory. If the link command file has no section directive for allocating
.ppdata section, the link step places the .ppdata section in a writable memory range.
The .ppdata section must be allocated memory in multiples of 32 bytes. Please refer to the linker
command file in the distribution for example usage.

74 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Feedback Directed Optimization

3.8.1.6 Feedback Directed Optimization and Code Size Tune

Feedback directed optimization is different from the Code Size Tune feature in Code Composer Studio
(CCS). The code size tune feature uses CCS profiling to select specific compilation options for each
function in order to minimize code size while still maintaining a specific performance point. Code size tune
is coarse-grained, since it is selecting an option set for the whole function. Feedback directed optimization
selects different optimization goals along specific regions within a function.

3.8.1.7 Instrumented Program Execution Overhead

During profile collection, the execution time of the application may increase. The amount of increase
depends on the size of the application and the number of files in the application compiled for profiling.
The profiling counters increase the code and data size of the application. Consider using the --
opt_for_space (-ms) code size options when using profiling to mitigate the code size increase. This has no
effect on the accuracy of the profile data being collected. Since profiling only counts execution frequency
and not cycle counts, code size optimization flags do not affect profiler measurements.

3.8.1.8 Invalid Profile Data

When recompiling with --use_profile_info, the profile information is invalid in the following cases:
• The source file name changed between the generation of profile information (gen-profile) and the use
of the profile information (use-profile).
• The source code was modified since gen-profile. In this case, profile information is invalid for the
modified functions.
• Certain compiler options used with gen-profile are different from those with used with use-profile. In
particular, options that affect parser behavior could invalidate profile data during use-profile. In general,
using different optimization options during use-profile should not affect the validity of profile data.

3.8.2 Profile Data Decoder

The code generation tools include a new tool called the profile data decoder or pdd6x, which is used for
post processing profile data (PDAT) files. The pdd6x tool generates a profile feedback (PRF) file. See
Section 3.8.1 for a discussion on where pdd6x fits in the profiling flow. The pdd6x tool is invoked with this
syntax:

pdd6x -e exec.out -o application.prf filename .pdat

-a Computes the average of the data values in the data sets instead of
accumulating data values
-e exec.out Specifies exec.out is the name of the application executable.
-o application.prf Specifies application.prf is the formatted profile feedback file that is used as the
argument to --use_profile_info during recompilation. If no output file is specified,
the default output filename is pprofout.prf.
filename .pdat Is the name of the profile data file generated by the run-time-support function.
This is the default name and it can be overridden by using the host environment
variable TI_PROFDATA.

SPRU187U – July 2012 Optimizing Your Code 75

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Feedback Directed Optimization www.ti.com

The run-time-support function and pdd6x append to their respective output files and do not overwrite
them. This enables collection of data sets from multiple runs of the application.

Profile Data Decoder Requirements

NOTE: Your application must be compiled with at least skeletal (dwarf) debug support to enable
feedback directed optimization. When compiling for feedback directed optimization, the
pdd6x tool relies on basic debug information about each function in generating the formatted
.prf file.
The pprofout.pdat file generated by the run-time support is a raw data file of a fixed format
understood only by pdd6x. You should not modify this file in any way.

3.8.3 Feedback Directed Optimization API

There are two user interfaces to the profiler mechanism. You can start and stop profiling in your
application by using the following run-time-support calls.
• TI_start_pprof_collection()
This interface informs the run-time support that you wish to start profiling collection from this point on
and causes the run-time support to clear all profiling counters in the application (that is, discard old
counter values).
• TI_stop_pprof_collection()
This interface directs the run-time support to stop profiling collection and output profiling data into the
output file (into the default file or one specified by the TI_PROFDATA host environment variable). The
run-time support also disables any further output of profile data into the output file during exit(), unless
you call TI_start_pprof_collection() again.

3.8.4 Feedback Directed Optimization Summary

Options

--gen_profile_info Adds instrumentation to the compiled code. Execution of the code results in
profile data being emitted to a PDAT file.
--use_profile_info=file.prf Uses profile information for optimization and/or generating code coverage
information.
--analyze=codecov Generates a code coverage information file and continues with profile-based
compilation. Must be used with --use_profile_info.
--analyze_only Generates only a code coverage information file. Must be used with --
use_profile_info. You must specify both --analyze=codecov and --
analyze_only to do code coverage analysis of the instrumented application.

Host Environment Variables

TI_PROFDATA Writes profile data into the specified file

TI_COVDIR Creates code coverage files in the specified directory
TI_COVDATA Writes code coverage data into the specified file

API

TI_start_pprof_collection() Clears the profile counters to file

TI_stop_pprof_collection() Writes out all profile counters to file
PPHDNL Device driver handle for low-level C I/O based driver for writing out profile
data from a target program.

76 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage

Files Created

*.pdat Profile data file, which is created by executing an instrumented program and
used as input to the profile data decoder
*.prf Profiling feedback file, which is created by the profile data decoder and
used as input to the re-compilation step

3.9 Using Profile Information to Get Better Program Cache Layout and Analyze Code
Coverage
There are two different types of analysis information you can get from the path profiler: code coverage
information and call graph information.
The program cache layout tool helps you to develop better program instruction cache efficiency into your
applications. Program cache layout is the process of controlling the relative placement of code sections
into memory to minimize the occurrence of conflict misses in the program instruction cache.

3.9.1 Background and Motivation

Effective utilization of the program instruction cache is an important part of getting the best performance
from a C6000. The dedicated program instruction cache (L1P) provides fast instruction fetches, but a
cache miss can be very costly. Some applications (e.g. h264) can spend 30%+ of the processor's time
stalling due to L1P cache misses. A cache miss occurs when a fetch fails to read an instruction from L1P
and the process is required to access the instruction from the next level of memory. A request to L2 or
external memory has a much higher latency than an access from L1P.
Careful placement of code sections can greatly reduce the number of cache misses. The C6000 L1P is
especially sensitive to code placement because it is direct-mapped.
Many L1P cache misses are conflict misses. Conflict misses occur when the cache has recently evicted a
block of code that is now needed again. In a program instruction cache this often occurs when two
frequently executed blocks of code (usually from different functions) interleave their execution and are
mapped to the same cache line.
For example, suppose there is a call to function B from inside a loop in function A. Suppose also that the
code for function A's loop is mapped to the same cache line as a block of code from function B that is
executed every time that B is called. Each time B is called from within this loop, the loop code in function
A is evicted from the cache by the code in B that is mapped to the same cache line. Even worse, when B
returns to A, the loop code in A evicts the code from function B that is mapped to the same cache line.
Every iteration through the loop will cause two program instruction cache conflict misses. If the loop is
heavily traversed, then the number of processor cycles lost to program instruction cache stalls can
become quite large.
Many program instruction cache conflict misses can be avoided with more intelligent placement of
functions that are active at the same time. Program instruction cache efficiency can be significantly
improved using code placement strategies that utilize dynamic profile information that is gathered during
the run of an instrumented application.
The program cache layout tool (clt6x) takes dynamic profile information in the form of a weighted call
graph and creates a preferred function order command file that can be used as input to the linker to guide
the placement of function subsections.
You can use the program cache layout tool to help improve your program locality and reduce the number
of L1P cache conflict misses that occur during the run of your application, thereby improving your
application's performance.

SPRU187U – July 2012 Optimizing Your Code 77

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage www.ti.com

3.9.2 Code Coverage

The information collected during feedback directed optimization can be used for generating code coverage
reports. As with feedback directed optimization, the program must be compiled with the --gen_profile_info
option.
Code coverage conveys the execution count of each line of source code in the file being compiled, using
data collected during profiling.

3.9.2.1 Phase1: Collect Program Profile Information

3.9.2.2 Phase 2: Generate Code Coverage Reports

In this phase, the compiler is invoked with the --use_profile_info=file.prf option, which indicates that the
compiler should read the specified PRF file generated in phase 1. The application must also be compiled
with either the --codecov or --onlycodecov option; the compiler generates a code-coverage info file. The --
codecov option directs the compiler to continue compilation after generating code-coverage information,
while the --onlycodecov option stops the compiler after generating code-coverage data. For example:
cl6x --opt_level=2 --use_profile_info=pprofout.prf --onlycodecov foo.c

You can specify two environment variables to control the destination of the code-coverage information file.
• The TI_COVDIR environment variable specifies the directory where the code-coverage file should be
generated. The default is the directory where the compiler is invoked.
• The TI_COVDATA environment variable specifies the name of the code-coverage data file generated
by the compiler. the default is filename.csv where filename is the base-name of the file being compiled.
For example, if foo.c is being compiled, the default code-coverage data file name is foo.csv.
If the code-coverage data file already exists, the compiler appends the new dataset at the end of the file.
Code-coverage data is a comma-separated list of data items that can be conveniently handled by data-
processing tools and scripting languages. The following is the format of code-coverage data:
"filename-with-full-path","funcname",line#,column#,exec-frequency,"comments"

"filename-with-full-path" Full pathname of the file corresponding to the entry

"funcname" Name of the function
line# Line number of the source line corresponding to frequency data
column# Column number of the source line
exec-frequency Execution frequency of the line
"comments" Intermediate-level representation of the source-code generated by the parser

The full filename, function name, and comments appear within quotation marks ("). For example:
"/some_dir/zlib/c64p/deflate.c","_deflateInit2_",216,5,1,"( strm->zalloc )"
Other tools, such as a spreadsheet program, can be used to format and view the code coverage data.

78 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage

3.9.3 What Performance Improvements Can You Expect to See?

If your application does not suffer from inefficient usage of the L1P cache, then the program cache layout
capability will not have any effect on the performance of your application. Before applying the program
cache layout tooling to your application, analyze the L1P cache performance in your application.

3.9.3.1 Evaluating L1P Cache Performance

Evaluating the L1P cache usage efficiency of your application will not only help you determine whether or
not your application might benefit from using program cache layout, but it also gives you a rough estimate
as to how much performance improvement you can reasonably expect from applying program cache
layout.
There are several resources available to help you evaluate L1P cache usage in your application. One way
of doing this is to use the Function Profiling capability in Code Composer Studio (CCS). This capability is
available in the C6400+ Megamodule Cycle Accurate Simulator target configuration under CCS. You can
find further information about using the CCS Function Profiling capabilities at
http://tiexpressdsp.com/index.php/Profiler. You can find more information about how to use this capability
in conjunction with the program cache layout tool at
http://tiexpressdsp.com/index.php/Program_Cache_Layout.
The number of CPU stall cycles that occur due to L1P cache misses gives you a reasonable upper bound
estimate of the number of CPU cycles that you may be able to recover with the use of the program cache
layout tooling in your application. Please be aware that the performance impact due to program cache
layout will tend to vary for the different data sets that are run through your application.

3.9.4 Program Cache Layout Related Features and Capabilities

Version 7.0 of the C6000 code generation tools introduce some features and capabilities that can be used
in conjunction with the program cache layout tool, clt6x. The following is a summary:

3.9.4.1 Path Profiler

The C6000 tools include a path profiling utility, pprof6x, that is run from the compiler, cl6x. The pprof6x
utility is invoked by the compiler when the --gen_profile or the --use_profile command is used from the
compiler command line:

cl6x --gen_profile ... file.c

cl6x --use_profile ... file.c

For further information about profile-based optimization and a more detailed description of the profiling
infrastructure within the C6000, see Section 3.8.

3.9.4.2 Analysis Options

The path profiling utility, pprof6x, appends code coverage or weighted call graph analysis information to
existing CSV (comma separated values) files that contain the same type of analysis information.
The utility checks to make sure that an existing CSV file contains analysis information that is consistent
with the type of analysis information it is being asked to generate (whether it be code coverage or
weighted call graph analysis). Attempts to mix code coverage and weighted call graph analysis information
in the same output CSV file will be detected and pprof6x will emit a fatal error and abort.

--analyze=callgraph Instructs the compiler to generate weighted call graph analysis information.
--analyze=codecov Instructs the compiler to generate code coverage analysis information. This
option replaces the previous --codecov option.
--analyze_only Halts compilation after generation of analysis information is completed.

SPRU187U – July 2012 Optimizing Your Code 79

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage www.ti.com

3.9.4.3 Environment Variables

To assist with the management of output CSV analysis files, pprof6x supports two new environment
variables:

TI_WCGDATA Allows you to specify a single output CSV file for all weighted call graph analysis
information. New information is appended to the CSV file identified by this
environment variable, if the file already exists.
TI_ANALYSIS_DIR Specifies the directory in which the output analysis file will be generated. The
same environment variable can be used for both code coverage information and
weighted call graph information (all analysis files generated by pprof6x will be
written to the directory specified by the TI_ANALYSIS_DIR environment variable).

TI_COVDIR Environment Variable

NOTE: The existing TI_COVDIR environment variable is still supported when generating code
coverage analysis, but is overridden in the presence of a defined TI_ANALYSIS_DIR
environment variable.

3.9.4.4 Program Cache Layout Tool, clt6x

The program cache layout tool creates a preferred function order command file from input weighted call
graph (WCG) information. The syntax is:

clt6x CSV files with WCG info -o forder.cmd

3.9.4.5 Linker
The compiler prioritizes the placement of a function relative to others based on the order in which --
preferred_order options are encountered during the linker invocation. The syntax is:

--preferred_order=function specification

3.9.4.6 Linker Command File Operator unordered()

The new linker command file keyword unordered relaxes placement constraints placed on an output
section whose specification includes an explicit list of which input sections are contained in the output
section. The syntax is:

unordered()

3.9.5 Program Instruction Cache Layout Development Flow

Once you have determined that your application is experiencing some inefficiencies in its usage of the
program instruction cache, you may decide to include the program cache layout tooling in your
development to attempt to recover some of the CPU cycles that are being lost to stalls due to program
instruction cache conflict misses.

3.9.5.1 Gather Dynamic Profile Information

The program cache layout tool, clt6x, relies on the availability of dynamic profile information in the form of
a weighted call graph in order to produce a preferred function order command file that can be used to
guide function placement at link-time when your application is re-built.

80 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage

There are several ways in which this dynamic profile information can be collected. For example, if you are
running your application on hardware, you may have the capability to collect a PC discontinuity trace. The
discontinuity trace can then be post-processed to construct weighted call graph input information for the
clt6x.
The method for collecting dynamic profile information that is presented here relies on the path profiling
capabilities in the C6000 code generation tools. Here is how it works:
1. Build an instrumented application using the --gen_profile_info option.
Using --gen_profile_info instructs the compiler to embed counters into the code along the execution
paths of each function.
To compile only use:

cl6x options --gen_profile_info src_file(s)

The compile and link use:

cl6x options --gen_profile_info src_file(s) -run_linker --library lnk.cmd

2. Run an instrumented application to generate a .pdat file.

When the application runs, the counters embedded into the application by --gen_profile_info keep track
of how many times a particular execution path through a function is traversed. The data collected in
these counters is written out to a profile data file named pprofout.pdat.
The profile data file is automatically generated. For example, if you are using the C64+ simulator under
CCS, you can load and run your instrumented program, and you will see that a new pprofout.pdat file
is created in your working directory (where the instrumented application is loaded from).

3. Decode the profile data file.

Once you have a profile data file, the file is decoded by the profile data decoder tool, pdd6x, as follows:

pdd6x -e=instrumented app out file -o=pprofout.prf pprofout.pdat

Using pdd6x produces a .prf file which is then fed into the re-compile of the application that uses the
profile information to generate weighted call graph input data.

4. Use decoded profile information to generate weighted call graph input.

The compiler now supports a new option, --analyze, which is used to tell the compiler to generate
weighted call graph or code coverage analysis information. Its syntax are as follows:

--analyze=callgraph Instructs the compiler to generate weighted call graph information.

--analyze=codecov Instructs the compiler to generate code coverage information. This option
replaces the previous --codecov option.
The compiler also supports a new --analyze_only option which instructs the compiler to halt
compilation after the generation of analysis information has been completed. This option replaces the
previous --onlycodecov option.
To make use of the dynamic profile information that you gathered, re-compile the source code for your
application using the --analyze=callgraph option in combination with the --use_profile_info option:

cl6x options -mo --analyze=callgraph --use_profile_info=pprofout.prf src_file(s)

The use of -mo instructs the compiler to generate code for each function into its own subsection. This
option provides the linker with the means to directly control the placement of the code for a given
function.

SPRU187U – July 2012 Optimizing Your Code 81

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage www.ti.com

The compiler generates a CSV file containing weighted call graph information for each source file that
is specified on the command line. If such a CSV file already exists, then new call graph analysis
information will be appended to the existing CSV file. These CSV files are then input to the cache
layout tool (clt6x) to produce a preferred function order command file for your application.
For more details on the content of the CSV files (containing weighted call graph information) generated
by the compiler, see Section 3.9.6.

3.9.5.2 Generate Preferred Function Order from Dynamic Profile Information

At this point, the compiler has generated a CSV file for each C/C++ source file specified on the command
line of the re-compile of the application. Each CSV file contains weighted call graph information about all
of the call sites in each function defined in the C/C++ source file.
The program cache layout tool, clt6x, collects all of the weighted call graph information in these CSV files
into a single, merged weighted call graph. The weighted call graph is processed to produce a preferred
function order command file that is fed into the linker to guide the placement of the functions defined in
your application source files. This is the syntax for clt6x:

clt6x *.csv -o forder.cmd

The output of clt6x is a text file containing a sequence of --preferred_order=function specification options.
By default, the name of the output file is forder.cmd, but you can specify your own file name with the -o
option. The order in which functions appear in this file is their preferred function order as determined by
the clt6x.
In general, the proximity of one function to another in the preferred function order list is a reflection of how
often the two functions call each other. If two functions are very close to each other in the list, then the
linker interprets this as a suggestion that the two functions should be placed very near to one another.
Functions that are placed close together are less likely to create a cache conflict miss at run time when
both functions are active at the same time. The overall effect should be an improvement in program
instruction cache efficiency and performance.

3.9.5.3 Utilize Preferred Function Order in Re-Build of Application

Finally, the preferred function order command file that is produced by the clt6x is fed into the linker during
the re-build of the application, as follows:

cl6x options --run_linker *.obj forder.cmd -llnk.cmd

The preferred function order command file, forder.cmd, contains a list of --preferred_order=function
specification options. The linker prioritizes the placement of functions relative to each other in the order
that the --preferred_order options are encountered during the linker invocation.
Each --preferred_order option contains a function specification. A function specification can describe
simply the name of the function for a global function, or it can provide the path name and source file name
where the function is defined. A function specification that contains path and file name information is used
to distinguish one static function from another that has the same function name.
The --preferred_order options are interpreted by the linker as suggestions to guide the placement of
functions relative to each other. They are not explicit placement instructions. If an object file or input
section is explicitly mentioned in a linker command file SECTIONS directive, then the placement
instruction specified in the linker command file takes precedence over any suggestion from a --
preferred_order option that is associated with a function that is defined in that object file or input section.
This precedence can be relaxed by applying the unordered() operator to an output specification as
described in Section 3.9.7.

82 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage

3.9.6 Comma-Separated Values (CSV) Files with Weighted Call Graph (WCG) Information
The format of the CSV files generated by the compiler under the --analyze=callgraph --use_profile_info
option combination is as follows:

"caller","callee","weight" [CR][LF]
caller spec,callee spec,call frequency [CR][LF]
caller spec,callee spec,call frequency [CR][LF]
caller spec,callee spec,call frequency [CR][LF]
...

Keep the following points in mind:

• Line 1 of the CSV file is the header line. It specifies the meaning of each field in each line of the
remainder of the CSV file. In the case of CSV files that contain weighted call graph information, each
line will have a caller function specification, followed by a callee function specification, followed by an
unsigned integer that provides the number of times a call was executed during run time.
• There may be instances where the caller and callee function specifications are identical on multiple
lines in the CSV file. This will happen when a caller function has multiple call sites to the callee
function. In the merged weighted call graph that is created by the clt6x, the weights of each line that
has the same caller and callee function specifications will be added together.
• The CSV file that is generated by the compiler using the path profiling instrumentation will not include
information about indirect function calls or calls to runtime support helper functions (like _remi or _divi).
However, you may be able to gather information about such calls with another method (like the PC
discontinuity trace mentioned earlier).
• The format of these CSV files is in compliance with the RFC-4180 specification of Comma-Separated
Values (CSV) files. For more details on this specification, please see http://tools.ietf.org/html/rfc4180.

3.9.7 Linker Command File Operator - unordered()

A new unordered() operator is now available for use in a linker command file. The effect of this operator is
to relax the placement constraints placed on an output section specification in which the content of the
output section is explicitly stated.
Consider an example output section specification:
SECTIONS
{
.text:
{
file.obj(.text:func_a)
file.obj(.text:func_b)
file.obj(.text:func_c)
file.obj(.text:func_d)
file.obj(.text:func_e)
file.obj(.text:func_f)
file.obj(.text:func_g)
file.obj(.text:func_h)

*(.text)
} > PMEM
...
}

In this SECTIONS directive, the specification of .text explicitly dictates the order in which functions are laid
out in the output section. Thus by default, the linker will layout func_a through func_h in exactly the order
that they are specified, regardless of any other placement priority criteria (such as a preferred function
order list that is enumerated by --preferred_order options).
The unordered() operator can be used to relax this constraint on the placement of the functions in the
'.text' output section so that placement can be guided by other placement priority criteria.
The unordered() operator can be applied to an output section as in Example 3-2.

SPRU187U – July 2012 Optimizing Your Code 83

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage www.ti.com

Example 3-2. Output Section for unordered() Operator

SECTIONS
{
.text: unordered()
{
file.obj(.text:func_a)
file.obj(.text:func_b)
file.obj(.text:func_c)
file.obj(.text:func_d)
file.obj(.text:func_e)
file.obj(.text:func_f)
file.obj(.text:func_g)
file.obj(.text:func_h)

*(.text)
} > PMEM
...
}

So that, given this list of --preferred_order options:

• --preferred_order="func_g"
• --preferred_order="func_b"
• --preferred_order="func_d"
• --preferred_order="func_a"
• --preferred_order="func_c"
• --preferred_order="func_f"
• --preferred_order="func_h"
• --preferred_order="func_e"
The placement of the functions in the .text output section is guided by this preferred function order list.
This placement will be reflected in a linker generated map file, as follows:

Example 3-3. Generated Linker Map File for Example 3-2

SECTION ALLOCATION MAP

output attributes/
section page origin length input sections
-------- ---- ---------- ---------- ----------------
.text 0 00000020 00000120
00000020 00000020 file.obj (.text:func_g:func_g)
00000040 00000020 file.obj (.text:func_b:func_b)
00000060 00000020 file.obj (.text:func_d:func_d)
00000080 00000020 file.obj (.text:func_a:func_a)
000000a0 00000020 file.obj (.text:func_c:func_c)
000000c0 00000020 file.obj (.text:func_f:func_f)
000000e0 00000020 file.obj (.text:func_h:func_h)
00000100 00000020 file.obj (.text:func_e:func_e)

3.9.7.1 About Dot (.) Expressions in the Presence of unordered()

Another aspect of the unordered() operator that should be taken into consideration is that even though the
operator causes the linker to relax constraints imposed by the explicit specification of an output section's
contents, the unordered() operator will still respect the position of a dot (.) expression within such a
specification.

84 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage

Consider the output section specification in Example 3-4.

Example 3-4. Respecting Position of a . Expression

SECTIONS
{
.text: unordered()
{
file.obj(.text:func_a)
file.obj(.text:func_b)
file.obj(.text:func_c)
file.obj(.text:func_d)

. += 0x100;

file.obj(.text:func_e)
file.obj(.text:func_f)
file.obj(.text:func_g)
file.obj(.text:func_h)

*(.text)

} > PMEM
...
}

In Example 3-4, a dot (.) expression, ". += 0x100;", separates the explicit specification of two groups of
functions in the output section. In this case, the linker will honor the specified position of the dot (.)
expression with respect to the functions on either side of the expression. That is, the unordered() operator
will allow the preferred function order list to guide the placement of func_a through func_d relative to each
other, but none of those functions will be placed after the hole that is created by the dot (.) expression.
Likewise, the unordered() operator allows the preferred function order list to influence the placement of
func_e through func_h relative to each other, but none of those functions will be placed before the hole
that is created by the dot (.) expression.

3.9.7.2 GROUPs and UNIONs

The unordered() operator can only be applied to an output section. This includes members of a GROUP or
UNION directive.

Example 3-5. Applying unordered() to GROUPs

SECTIONS
{
GROUP
{
.grp1:
{
file.obj(.grp1:func_a)
file.obj(.grp1:func_b)
file.obj(.grp1:func_c)
file.obj(.grp1:func_d)
} unordered()

.grp2:
{
file.obj(.grp2:func_e)
file.obj(.grp2:func_f)
file.obj(.grp2:func_g)
file.obj(.grp2:func_h)
}

SPRU187U – July 2012 Optimizing Your Code 85

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Profile Information to Get Better Program Cache Layout and Analyze Code Coverage www.ti.com

Example 3-5. Applying unordered() to GROUPs (continued)

.text: { *(.text) }

} > PMEM
...
}

The SECTIONS directive in Example 3-5 applies the unordered() operator to the first member of the
GROUP. The .grp1 output section layout can then be influenced by other placement priority criteria (like
the preferred function order list), whereas the .grp2 output section will be laid out as explicitly specified.
The unordered() operator cannot be applied to an entire GROUP or UNION. Attempts to do so will result
in a linker command file syntax error and the link will be aborted.

3.9.8 Things To Be Aware Of

There are some behavioral characteristics and limitations of the program cache layout development flow
that you should bear in mind:
• Generation of Path Profiling Data File (.pdat)
When running an application that has been instrumented to collect path-profiling data (using --
gen_profile_info compiler option during build), the application will use functions in the run-time-support
library to write out information to the path profiling data file (pprofout.pdat in above tutorial). If there is a
path profiling data file already in existence when the application starts to run, then any new path
profiling data generated will be appended to the existing file.
To prevent combining path profiling data from separate runs of an application, you need to either
rename the path profiling data file from the previous run of the application or remove it before running
the application again.
• Indirect Calls Not Recognized by Path Profiling Mechanisms
When using available path profiling mechanisms to collect weighted call graph information from the
path profiling data, pprof6x does not recognize indirect calls. An indirect call site will not be
represented in the CSV output file that is generated by pprof6x.
You can work around this limitation by introducing your own information about indirect call sites into the
relevant CSV file(s). If you take this approach, please be sure to follow the format of the callgraph
analysis CSV file ("caller", "callee","call frequency").
If you are able to get weighted call graph information from a PC trace into a callgraph analysis CSV,
this limitation will no longer apply (as the PC trace can always identify the callee of an indirect call).
• Multiple --preferred_order Options Associated With Single Function
There may be cases in which you might want to input more than one preferred function order
command file to the linker during the link of an application. For example, you may have developed or
received a separate preferred function order command file for one or more of the object libraries that
are used by your application.
In such cases, it is possible that one function may be specified in multiple preferred function order
command files. If this happens, the linker will honor only the first instance of the --preferred_order
option in which the function is specified.

86 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Indicating Whether Certain Aliasing Techniques Are Used

3.10 Indicating Whether Certain Aliasing Techniques Are Used

Aliasing occurs when you can access a single object in more than one way, such as when two pointers
point to the same object or when a pointer points to a named object. Aliasing can disrupt optimization,
because any indirect reference can refer to another object. The compiler analyzes the code to determine
where aliasing can and cannot occur, then optimizes as much as possible while preserving the
correctness of the program. The compiler behaves conservatively.
The following sections describe some aliasing techniques that may be used in your code. These
techniques are valid according to the ISO C standard and are accepted by the C6000 compiler; however,
they prevent the optimizer from fully optimizing your code.

3.10.1 Use the --aliased_variables Option When Certain Aliases are Used
The compiler, when invoked with optimization, assumes that any variable whose address is passed as an
argument to a function is not subsequently modified by an alias set up in the called function. Examples
include:
• Returning the address from a function
• Assigning the address to a global variable
If you use aliases like this in your code, you must use the --aliased_variables option when you are
optimizing your code. For example, if your code is similar to this, use the --aliased_variables option:
int *glob_ptr;

g()
{
int x = 1;
int *p = f(&x);

*p = 5; /* p aliases x */
*glob_ptr = 10; /* glob_ptr aliases x */

h(x);
}

int f(int arg)

{
glob_ptr = arg;
return arg;
}

3.10.2 Use the --no_bad_aliases Option to Indicate That These Techniques Are Not Used
The --no_bad_aliases option informs the compiler that it can make certain assumptions about how aliases
are used in your code. These assumptions allow the compiler to improve optimization. The --
no_bad_aliases option also specifies that loop-invariant counter increments and decrements are non-zero.
Loop invariant means the value of an expression does not change within the loop.
• The --no_bad_aliases option indicates that your code does not use the aliasing technique described in
Section 3.10.1. If your code uses that technique, do not use the --no_bad_aliases option. You must
compile with the --aliased_variables option.
Do not use the --aliased_variables option with the --no_bad_aliases option. If you do, the --
no_bad_aliases option overrides the --aliased_variables option.
• The --no_bad_aliases option indicates that a pointer to a character type does not alias (point to) an
object of another type. That is, the special exception to the general aliasing rule for these types given
in section 3.3 of the ISO specification is ignored. If you have code similar to the following example, do
not use the --no_bad_aliases option:

SPRU187U – July 2012 Optimizing Your Code 87

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Prevent Reordering of Associative Floating-Point Operations www.ti.com

{
long l;
char *p = (char *) &l;

p[2] = 5;
}
• The --no_bad_aliases option indicates that indirect references on two pointers, P and Q, are not
aliases if P and Q are distinct parameters of the same function activated by the same call at run time.
If you have code similar to the following example, do not use the --no_bad_aliases option:
g(int j)
{

int a[20];

f(&a, &a) /* Bad */

f(&a+42, &a+j) /* Also Bad */
}

f(int ptr1, int ptr2)

{
...
}
• The --no_bad_aliases option indicates that each subscript expression in an array reference A[E1]..[En]
evaluates to a nonnegative value that is less than the corresponding declared array bound. Do not use
--no_bad_aliases if you have code similar to the following example:
static int ary[20][20];

int g()
{
return f(5, -4); /* -4 is a negative index */
return f(0, 96); /* 96 exceeds 20 as an index */
return f(4, 16); /* This one is OK */
}

int f(int I, int j)

{
return ary[i][j];
}
In this example, ary[5][-4], ary[0][96], and ary[4][16] access the same memory location. Only the
reference ary[4][16] is acceptable with the --no_bad_aliases option because both of its indices are
within the bounds (0..19).
• The --no_bad_aliases option indicates that loop-invariant counter increments and decrements of loop
counters are non-zero. Loop invariant means a value of an expression does not change within the
loop.
If your code does not contain any of the aliasing techniques described above, you should use the --
no_bad_aliases option to improve the optimization of your code. However, you must use discretion with
the --no_bad_aliases option; unexpected results may occur if these aliasing techniques appear in your
code and the --no_bad_aliases option is used.

3.10.3 Using the --no_bad_aliases Option With the Assembly Optimizer

The --no_bad_aliases option allows the assembly optimizer to assume there are no memory aliases in
your linear assembly; i.e., no memory references ever depend on each other. However, the assembly
optimizer still recognizes any memory dependencies you point out with the .mdep directive. For more
information about the .mdep directive, see and Section 4.6.4 .

3.11 Prevent Reordering of Associative Floating-Point Operations

The compiler freely reorders associative floating-point operations. If you do not wish to have the compiler
reorder associative floating point operations, use the --fp_not_associative option. Specifying the --
fp_not_associative option may decrease performance.
88 Optimizing Your Code SPRU187U – July 2012
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Use Caution With asm Statements in Optimized Code

3.12 Use Caution With asm Statements in Optimized Code

You must be extremely careful when using asm (inline assembly) statements in optimized code. The
compiler rearranges code segments, uses registers freely, and can completely remove variables or
expressions. Although the compiler never optimizes out an asm statement (except when it is
unreachable), the surrounding environment where the assembly code is inserted can differ significantly
from the original C/C++ source code.
It is usually safe to use asm statements to manipulate hardware controls such as interrupt masks, but asm
statements that attempt to interface with the C/C++ environment or access C/C++ variables can have
unexpected results. After compilation, check the assembly output to make sure your asm statements are
correct and maintain the integrity of the program.

3.13 Automatic Inline Expansion (--auto_inline Option)

When optimizing with the --opt_level=3 option or --opt_level=2 option (aliased as -O3 and -O2,
respectively), the compiler automatically inlines small functions. A command-line option, --
auto_inline=size, specifies the size threshold for automatic inlining. This option controls only the inlining of
functions that are not explicitly declared as inline.
When the --auto_inline option is not used, the compiler sets the size limit based on the optimization level
and the optimization goal (performance versus code size). If the -auto_inline size parameter is set to 0,
automatic inline expansion is disabled. If the --auto_inline size parameter is set to a non-zero integer, the
compiler automatically inlines any function smaller than size. (This is a change from previous releases,
which inlined functions for which the product of the function size and the number of calls to it was less
than size. The new scheme is simpler, but will usually lead to more inlining for a given value of size.)
The compiler measures the size of a function in arbitrary units; however the optimizer information file
(created with the --gen_opt_info=1 or --gen_opt_info=2 option) reports the size of each function in the
same units that the --auto_inline option uses. When --auto_inline is used, the compiler does not attempt to
prevent inlining that causes excessive growth in compile time or size; use with care.
When --auto_inline option is not used, the decision to inline a function at a particular call-site is based on
an algorithm that attempts to optimize benefit and cost. The compiler inlines eligible functions at call-sites
until a limit on size or compilation time is reached.
When deciding what to inline, the compiler collects all eligible call-sites in the module being compiled and
sorts them by the estimated benefit over cost. Functions declared static inline are ordered first, then leaf
functions, then all others eligible. Functions that are too big are not included.
Inlining behavior varies, depending on which compile-time options are specified:
• The code size limit is smaller when compiling for code size rather than performance. The --auto_inline
option overrides this size limit.
• At --opt_level=3, the compiler auto-inlines aggressively if compiling for performance.
• At --opt_level=2, the compiler only automatically inlines small functions.

Some Functions Cannot Be Inlined

NOTE: For a call-site to be considered for inlining, it must be legal to inline the function and inlining
must not be disabled in some way. See the inlining restrictions in Section 2.11.5.

Optimization Level 3 or 2 and Inlining

NOTE: In order to turn on automatic inlining, you must use the --opt_level=3 option or --opt_level=2
option. At --opt_level=2, only small functions are auto-inlined. If you desire the --opt_level=3
or 2 optimizations, but not automatic inlining, use --auto_inline=0 with the --opt_level=3 or 2
option.

SPRU187U – July 2012 Optimizing Your Code 89

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

Inlining and Code Size

NOTE: Expanding functions inline increases code size, especially inlining a function that is called in
a number of places. Function inlining is optimal for functions that are called only from a small
number of places and for small functions. To prevent increases in code size because of
inlining, use the --auto_inline=0 and --no_inlining options. These options, used together,
cause the compiler to inline intrinsics only.

3.14 Using Performance Advice to Optimize your Code

You can use compiler generated Performance Advice to optimize your code. To get Performance Advice,
compile with with the following options:

--advice:performance Instructs the compiler to emit advice to stdout (default).

--advice:performance_file Instructs the compiler to emit advice into a file.
--advice:performance_dir Instructs the compiler to emit advice into a file in a specific directory.

If an Advice file is requested, but there is no advice, the advice file will not be created;
rather the compiler prints a message to stdout :
"filename.c": advice #27004: No Performance Advice is generated.

Example 3‑6. Output Advice to stdout (default)

cl6x -mv6400 -o2 -k --advice:performance func.c

"func.c", line 10: advice #30006: Loop at line 8 cannot be scheduled efficiently
as it contains a function call ("_init"). Try making "_init" an inline
function.
"func.c", line 12: advice #30000: Loop at line 8 cannot be scheduled efficiently
as it contains a function call ("_calculate"). Try to inline call or
consider rewriting loop.

Note that Advice to prevent Software Pipeline Disqualification (such as that presented above) will also be
printed in the .asm file. So, func.asm will contain :
;*----------------------------------------------------------------------------*
;* SOFTWARE PIPELINE INFORMATION
;* Disqualified loop: Loop contains a call
;* Loop at line 8 cannot be scheduled efficiently as it contains a
;* function call ("_init"). Try making "_init" an inline function.
;* Disqualified loop: Loop contains non-pipelinable instructions
;* Disqualified loop: Loop contains a call
;* Loop at line 8 cannot be scheduled efficiently as it contains a
;* function call ("_calculate"). Try to inline call or consider
;* rewriting loop.
;* Disqualified loop: Loop contains non-pipelinable instructions
;*----------------------------------------------------------------------------*

Example 3‑7. Output Advice to a file filename.advice

cl6x -mv6400 --advice:performance --advice:performance_file=filename.advice func.c

;******************************************************************************
;* TMS320C6x C/C++ Codegen Unix v7.5.0P12047 (a0322878 - Feb 16 2012) *
;* Date/Time created: Thu Feb 16 10:26:02 2012 *
;* *
;* *
;* This file contains Performance Advice generated by the compiler under the *

90 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

Example 3‑7. Output Advice to a file filename.advice (continued)

;* --advice:performance option. Additional information on the specific piece *
;* of advice can be found at: *
;* *
;* http://processors.wiki.ti.com/index.php/Compiler/diagnostic_messages/C6000 *
;* *
;* or by searching for `Advice #advice_id`. *
;* *
;* Warning: This file is auto generated by the compiler and can be *
;* overwritten during the next compile. *
;* *
;******************************************************************************
;* User Options: --silicon_version=6400

"func.c": advice #27000: Detecting compilation without optimization. Use

optimization option -o2 or higher.

Example 3‑8. Output advice to a specific file "myfile.adv" in directory "mydir"

You can do this by using the --advice:performance_file and --advice:performance_dir option :

cl6x -mv6400 -o2 -k --advice:performance_file=myfile.adv --advice:performance_dir=mydir basicloop.c

Or by specifying the full path name to the --advice:performance_file option :

cl6x -mv6400 -o2 -k --advice:performance_file=mydir/myfile.adv basicloop.c

If --advice_dir option and full pathname are specified together, --advice:performance_dir option is ignored, and
the advice is generated in the full pathname advice file. Also, note that directory "mydir" must already exist for
an advice file to be created in there.

3.14.1 Types of Advice

3.14.1.1 Advice #27000

advice #27000: Detecting compilation without optimization. Use optimization option -o2 or higher.

Your compilation is being done without any optimization options (-o0 and above). This prevents the
compiler from using its most powerful optimization techniques, since the -o (--opt_level) options are the
foundtions for most other optimizations. You could get substantially better performance using -o2 (or
above) optimization. For C6000, optimization option -o2 is required for the software pipelining loop
optimization, which is crucial to getting good performance.

The C/C++ compiler is able to perform various optimizations, but you need to specify optimization options
on the command line so that these optimizations are performed. The easiest way to invoke optimization is
to specify the --opt_level=n option on the compiler command line. You can use -On to alias the --opt_level
option. The n denotes the level of optimization (0, 1, 2, and 3), which controls the type and degree of
optimization. See "Invoking Optimization" in Section 3.1 for more information on Optimization Options.

3.14.1.2 Advice #27001: Increase Optimization Level

advice #27001: Detecting compilation with low optimization level.
Use optimization option -o2 or higher.

Your compilation uses low-level optimization options (-o1 and below), which prevents the compiler from
using its most powerful optimization techniques.

SPRU187U – July 2012 Optimizing Your Code 91

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

The C/C++ compiler is able to perform various optimizations, but you can control the level of these
optimizations. High-level optimizations are performed in the optimizer and low-level, target-specific
optimizations occur in the code generator. You must use high-level optimizations to achieve optimal code.
You can invoke optimization by specifying the --opt_level=n option on the compiler command line.
See "Invoking Optimization" in Section 3.1 for more information on Optimization Options. Also see
information for Advice #27000 in Section 3.14.1.1.

3.14.1.3 Advice #27002: Do not turn off software pipelining

advice #27002: Detecting compilation with "-mu" which turns off
software pipelining. To optimize, turn off this option.

Your compilation is being done using -mu, which turns off software-pipelining. Software-pipelining is a key
optimization for achieving good performance. This Advice is issued to alert you to NOT use compiler
option -mu. -mu is a good option for debugging, but it is recommended that this option not be used for
production code because of the negative performance implications.
In general, to achieve maximal performance, avoid using the following in production code :
• -g: Support full symbolic debug, and is a good option for debugging. This inhibits code reordering
across source line boundaries and limits optimizations around function boundaries. This results in less
parallelism, more nops and generally less efficient schedules. Using this option can cause a 30-50%
performance degradation for control code, generally somewhat less but still significant degradation for
performance critical code.
• -ss: Interlist source code into assembly file. As with -g, this option can negatively impact performance.
• -mu: Turns off software-pipelining, which is a key optimization for achieving good performance. This is
a good option for debugging, but is not recommended for use in production code due to negative
performance implications.

3.14.1.4 Advice #27003: Avoid compiling with debug options

advice #27003: Detecting compilation with debug option "-g",
which hinders optimization. To optimize, remove -g or
compile with --optimize_with_debug_option.

The compiler detects that your compilation is being done using -g, which limits software-pipelining.
Software-pipelining is a key optimization for achieving good performance. This Advice is issued to alert
you to NOT use compiler option -g. -g is a good option for debugging, but it is recommended that this
option not be used for production code because of the negative performance implications. Also see Advice
#27002 in Section 3.14.1.3.

3.14.1.5 Advice #27004: No Performance Advice generated

advice #27004: No Performance Advice is generated.

The compiler detects that your compilation is being done using --advice:performance option, but the
compiler has no Advice to report. This Advice is issued to alert you to the fact that no Advice is being
emitted, and an Advice file will not be created (if one was requested).

3.14.1.6 Advice #30000: Prevent Loop Disqualification due to call

advice #30000: Loop at line 10 cannot be scheduled efficiently,
as it contains a function call ("function_name").
Try to inline call or consider rewriting loop.

The compiler attempts to perform the software pipeline loop optimization at optimization levels --
opt_level=2 (or -O2) and -O3. If there is a call in the loop, the compiler will attempt to completely inline the
called function, but sometimes this is not possible. If the compiler cannot inline the called function,
software pipelining cannot be performed. This can severely reduce the performance of the loop.
In the testcase below, the call to the function "func2" prevents software pipelining. Inlining function "func2"
or rewriting the loop to avoid a function call can avoid pipeline disqualification. If the loop pipelines
successfully you may see performance improvement.
void func1(int *p, int *q, int n)

92 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code
{
unsigned int i;

for (i = 0; i < n; i++)

{
int t = func2(i);

p[i] = q[i] + t;
}
}

3.14.1.7 Advice #30001: Prevent Loop Disqualification due to rts-call

advice #30001: Loop at line 18 cannot be scheduled efficiently, as it
contains conversion from "type-a" to "type-b".
Try to use "suggested" type.

The compiler can insert calls to special functions in the run-time support library (RTS) to support
operations that are not natively supported by the ISA. For instance, while C6000 floating-point ISAs
support instructions to convert between floating-point and signed integer values, they don't support
conversion between floating-point and unsigned integer values. If you use unsigned variables in floating
point expressions, the compiler will generate a call to an RTS routine to carry out this function. Such a call
will disable software pipelining.
You can change the unsigned variables in your code to signed variables and prevent this from happening.
The compiler will then be able to use the native hardware instead of adding the special function call, so
you may get better performance.

3.14.1.8 Advice #30002: Prevent Loop Disqualification due to asm statement

advice #30002: Loop at line 8 cannot be scheduled efficiently, as it
contains an asm() statement. Try to replace the asm()
statement with C or intrinsic statement.

An asm statement inserted in a C code loop will disqualify the loop for software pipelining. Software-
pipelining is a key optimization for achieving good performance. You may see reduced performance
without software pipelining.
Replace the asm() statement with native C, or an intrinsic function call to prevent this from happening.

3.14.1.9 Advice #30003: Prevent Loop Disqualification due to complex condition

advice #30003: Loop at line 8 cannot be scheduled efficiently, as it
contains complex conditional expression. Try to simplify
condition.

Your code contains a complex conditional expression, possibly a large "if" clause, within a loop, which is
preventing optimization. The C6000 compiler will optimize small “if” statements (“if” statements with “if”
and “else” blocks that are short or empty). The compiler will not optimize large "if" statements, and such
large if statements within the loop body will disqualify the loop for software pipelining. Software-pipelining
is a key optimization; you may see reduced performance without it.
In the examples below, Example 1 will pipeline, but Example 2 won't :

Example 1:
for (i=0; i < N; i++)
{
if (!flag)
{
//statements
}
else
{
x[i] = y[i];
}
}

SPRU187U – July 2012 Optimizing Your Code 93

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

Example 2:
for (i = 0; i < n; i++)
{
if (!flag)
{
//statements
}
else
{
if (flag == 1) x[i] = y[i];
}
}

Example 1 will have significantly better performance than Example 2 becaues it pipelines successfully. But
Example 2 can be pipelined if the code is modified to eliminate the nested "if" :
for (i = 0; i < n; i++)
{
if (!flag)
{
//statements
}
else
{
p = (flag == 1);
x[i] = !p * x[i] + p * y[i] ;
}
}

3.14.1.10 Advice #30004: Prevent Loop Disqualification due to switch statement

advice #30004: Loop at line 257 cannot be scheduled efficiently, as it
contains a switch statement. Try to rewrite loop.

There is a switch statement within the loop. A switch statement in a loop will disqualify the loop for
software pipelining. Software-pipelining is a key optimization; you may see reduced performance without
it.
Try and rewrite the loop without a switch statement.

3.14.1.11 Advice #30005: Prevent Loop Disqualification due to arithmetic operation

advice #30005: Loop at line 5 cannot be scheduled efficiently, as it
contains a "division" operation. Rewrite using simpler
operations if possible.

The compiler can insert calls to special functions in the run-time support library (RTS) to support
operations that are not natively supported by the ISA. For example, the compiler calls __c6xabi_divi()
(_divi() in COFF) function to perform 32-bit integer divide operation. Such functions are called compiler
helper functions, and result in a function call withing the loop body. In the example below, the compiler will
accomplish the division operation by calling the compiler helper function "_divi" :
void func(float *p, float n)
{
int i;

for (i = 1; i < 1000; i++)

{
p[i] /= n;
}
}

However if we modify this loop, like below, the loop pipelines :

void func_adjusted(float *p, float n)
{
int i;

94 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

float inv = 1/n;

for (i = 1; i < 1000; i++)

{
p[i] *= inv;
}
}

3.14.1.12 Advice #30006: Prevent Loop Disqualification due to call(2)

advice #30006: Loop at line 22 cannot be scheduled efficiently, as it
contains a function call ("function_name"). Try making "function_name" an
inline function.

For improved performance, at optimization levels --opt_level=2 (-O2) and --opt_level=3 (-O3), the compiler
attempts to software pipeline your loops. Sometimes the compiler may not be able to inline a function call
that is in a loop. Because the compiler could not inline the function call, the loop could not be software
pipelined, and the loop could not be efficiently scheduled.
For example, in the testcase below, call to function "func2" prevents software pipelining:
void func1(int *p, int *q, int n)
{
unsigned int i;

for (i = 0; i < n; i++)

{
int t = func2(i);

; other operations
}
}
int function func2() { . . . }

However if function func2 is inlined, it saves the overhead of a function call. The compiler is free to
optimize the function in context with surrounding code. Automatic inlining is controlled by the "inline"
keyword; use it to allow inlining of specific functions :
inline int function func2() { . . . }

Also see #Advice 30000 in Section 3.14.1.6.

3.14.1.13 Advice #30007: Prevent Loop Disqualification due to rts-call(2)

advice #30007: Attempting to use floating-point operation "__mpyd" on
fixed-point device, at line 5 (there may be other instances
of this). Such calls reduce loop performance; use fixed point
operation if possible.

The compiler inserts calls to special functions in the run-time support library (RTS) to support operations
that are not natively supported by the instruction set architecture (ISA). For example, C6000 fixed point
ISAs do not support floating-point instructions and the compiler will generate a call to an RTS routine to
carry out the floating point operation. In the testcase below, the floating-point multiplication is unavailable
for a fixed-point device such as C6200:
void func(float *p, float *q, int n)
{
unsigned int i;

for (i = 1; i < n; i++)

{
p[i] = (q[i] * 12.4) / p[i - 1];
}
}

SPRU187U – July 2012 Optimizing Your Code 95

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

If compiled for C6200 (compiler option -mv6200) the compiler will use an RTS call to carry out the
operation. Such a call will disable software pipelining. You can rewrite the operation, or use a fixed point
operation to prevent this.
Also see Advice #30001 in Section 3.14.1.7.

3.14.1.14 Advice #30008: Improve Loop; Qualify with restrict

advice #30008: Consider adding the restrict qualifier to the definition
of inp1, inp2 if they don't access the same memory location.

3.14.1.15 Advice #30009: Improve Loop; Add MUST_ITERATE pragma

advice #30009: If you know that this loop will always execute at a
multiple of <2> and at least <2> times, try adding
"#pragma MUST_ITERATE(2, ,2)" just before the loop.

The C6000 architecture is partitioned into two nearly symmetric halves. The resource breakdown
displayed in the software pipelining information in the asm file, is computed after the compiler has
partitioned instructions to either the A-side or the B-side. If the resources are imbalanced (i.e.; some
resources on one side are used more than resources on the other) software pipelinging is resource-
bound, and the loop cannot be efficiently scheduled. If the compiler has information about the trip-count
for the loop, it can unroll the loop to balance resource usage, and get better pipelining. You can give loop
trip-count information to the compiler using the "MUST_ITERATE" pragma.
To see more information on using the MUST_ITERATE pragma, refer to Section 6.9.20

3.14.1.16 Advice #30010: Improve Loop; Add MUST_ITERATE pragma(2)

advice #30010: If you know that this loop will always execute atlest
<2> times, try adding "#pragma MUST_ITERATE(2)" just before the loop.

See Advice #30009 at Section 3.14.1.15.

3.14.1.17 Advice #30011: Improve Loop; Add _nasssert()

advice #30011: Consider adding assertions to indicate n-byte alignment
of variables input1, input2, output if they are actually n-byte
aligned: _nassert((int)(input1) % 8 == 0).

Most loops have memory access instructions. The compiler attempts to use wider load instructions, and
aligned memory accesses instead of non-aligned memory accesses to reduce/balance out resources used
for the memory access instructions. One of the ways to let the compiler know that it is safe to use "wider"
loads is to use the keyword "_nassert".
To find out more on using the _nassert keyword, see Section 7.5.10.

3.14.2 Using the Interlist Feature With Optimization

You control the output of the interlist feature when compiling with optimization (the --opt_level=n or -On
option) with the --optimizer_interlist and --c_src_interlist options.
• The --optimizer_interlist option interlists compiler comments with assembly source statements.
• The --c_src_interlist and --optimizer_interlist options together interlist the compiler comments and the
original C/C++ source with the assembly code.

96 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

When you use the --optimizer_interlist option with optimization, the interlist feature does not run as a
separate pass. Instead, the compiler inserts comments into the code, indicating how the compiler has
rearranged and optimized the code. These comments appear in the assembly language file as comments
starting with ;**. The C/C++ source code is not interlisted, unless you use the --c_src_interlist option also.
The interlist feature can affect optimized code because it might prevent some optimization from crossing
C/C++ statement boundaries. Optimization makes normal source interlisting impractical, because the
compiler extensively rearranges your program. Therefore, when you use the --optimizer_interlist option,
the compiler writes reconstructed C/C++ statements.
Example 3-9 shows a function that has been compiled with optimization (--opt_level=2) and the --
optimizer_interlist option. The assembly file contains compiler comments interlisted with assembly code.

Impact on Performance and Code Size

NOTE: The --c_src_interlist option can have a negative effect on performance and code size.

When you use the --c_src_interlist and --optimizer_interlist options with optimization, the compiler inserts
its comments and the interlist feature runs before the assembler, merging the original C/C++ source into
the assembly file.
Example 3-10 shows the function from Example 3-9 compiled with the optimization (--opt_level=2) and the
--c_src_interlist and --optimizer_interlist options. The assembly file contains compiler comments and C
source interlisted with assembly code.

SPRU187U – July 2012 Optimizing Your Code 97

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

Example 3-9. The Function From Example 2-4 Compiled With the -O2 and --optimizer_interlist Options

_main:
;** 5 ----------------------- printf("Hello, world\n");
;** 6 ----------------------- return 0;
STW .D2 B3,*SP--(12)
.line 3
B .S1 _printf
NOP 2
MVKL .S1 SL1+0,A0
MVKH .S1 SL1+0,A0
|| MVKL .S2 RL0,B3
STW .D2 A0,*+SP(4)
|| MVKH .S2 RL0,B3
RL0: ; CALL OCCURS
.line 4
ZERO .L1 A4
.line 5
LDW .D2 *++SP(12),B3
NOP 4
B .S2 B3
NOP 5
; BRANCH OCCURS

Example 3-10. The Function From Example 2-4 Compiled with the --opt_level=2, --optimizer_interlist, and
--c_src_interlist Options

_main:
;** 5 ----------------------- printf("Hello, world\n");
;** 6 ----------------------- return 0;
STW .D2 B3,*SP--(12)
;------------------------------------------------------------------------------
; 5 | printf("Hello, world\n");
;------------------------------------------------------------------------------
B .S1 _printf
NOP 2
MVKL .S1 SL1+0,A0
MVKH .S1 SL1+0,A0
|| MVKL .S2 RL0,B3
STW .D2 A0,*+SP(4)
|| MVKH .S2 RL0,B3
RL0: ; CALL OCCURS
;------------------------------------------------------------------------------
; 6 | return 0;
;------------------------------------------------------------------------------
ZERO .L1 A4
LDW .D2 *++SP(12),B3
NOP 4
B .S2 B3
NOP 5
; BRANCH OCCURS

98 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

3.14.3 Debugging and Profiling Optimized Code

Debugging fully optimized code is not recommended, because the compiler's extensive rearrangement of
code and the many-to-many allocation of variables to registers often make it difficult to correlate source
code with object code. Profiling code that has been built with the --symdebug:dwarf (aliased as -g) option
or the --symdebug:coff option (STABS debug) is not recommended either, because these options can
significantly degrade performance. To remedy these problems, you can use the options described in the
following sections to optimize your code in such a way that you can still debug or profile the code.

3.14.3.1 Debugging Optimized Code (--symdebug:dwarf, --symdebug:coff, and --opt_level Options)

To debug optimized code, use the --opt_level (aliased as -O) option in conjunction with one of the
symbolic debugging options (--symdebug:dwarf or --symdebug:coff). The symbolic debugging options
generate directives that are used by the C/C++ source-level debugger, but they disable many compiler
optimizations. When you use the --opt_level option (which invokes optimization) with the --
symdebug:dwarf or --symdebug:coff option, you turn on the maximum amount of optimization that is
compatible with debugging.
If you want to use symbolic debugging and still generate fully optimized code, use the --
optimize_with_debug option. This option reenables the optimizations disabled by --symdebug:dwarf or --
symdebug:coff. However, if you use the --optimize_with_debug option, portions of the debugger's
functionality will be unreliable.
If you are having trouble debugging loops in your code, you can use the --disable_software_pipelining
option to turn off software pipelining. See Section 3.2.1 for more information.

Symbolic Debugging Options Affect Performance and Code Size

NOTE: Using the --symdebug:dwarf or --symdebug:coff option can cause a significant performance
and code size degradation of your code. Use these options for debugging only. Using --
symdebug:dwarf or --symdebug:coff when profiling is not recommended.

C6400+, C6740, and C6600 Support Only DWARF Debugging

NOTE: Since C6400+, C6740, and C6600 produce only DWARF debug information, the --
symdebug:coff option is not supported when compiling with -mv6400+, -mv6740, or -mv6600.

3.14.3.2 Profiling Optimized Code

To profile optimized code, use optimization (--opt_level=0 through --opt_level=3) without any debug option.
By default, the compiler generates a minimal amount of debug information without affecting optimizations,
code size, or performance.
If you have a breakpoint-based profiler, use the --profile:breakpt option with the --opt_level option. The --
profile:breakpt option disables optimizations that would cause incorrect behavior when using a breakpoint-
based profiler.
If you have a power profiler, use the --profile:power option with the --opt_level option. The --profile:power
option produces instrument code for the power profiler.
If you need to profile code at a finer grain that the function level in Code Composer Studio, you can use
the --symdebug:dwarf or --symdebug:coff option, although this is not recommended. You might see a
significant performance degradation because the compiler cannot use all optimizations with --
symdebug:dwarf or --symdebug:coff. It is recommended that outside of Code Composer Studio, you use
the clock( ) function.

Profile Points
NOTE: In Code Composer Studio, when symbolic debugging is not used, profile points can only be
set at the beginning and end of functions.

SPRU187U – July 2012 Optimizing Your Code 99

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

3.14.4 Controlling Code Size Versus Speed

The latest mechanism for controlling the goal of optimizations in the compiler is represented by the --
opt_for_speed=num option. The num denotes the level of optimization (0-5), which controls the type and
degree of code size or code speed optimization:
• --opt_for_speed=0
Enables optimizations geared towards improving the code size with a high risk of worsening or
impacting performance.
• --opt_for_speed=1
Enables optimizations geared towards improving the code size with a medium risk of worsening or
impacting performance.
• --opt_for_speed=2
Enables optimizations geared towards improving the code size with a low risk of worsening or
impacting performance.
• --opt_for_speed=3
Enables optimizations geared towards improving the code performance/speed with a low risk of
worsening or impacting code size.
• --opt_for_speed=4
Enables optimizations geared towards improving the code performance/speed with a medium risk of
worsening or impacting code size.
• --opt_for_speed=5
Enables optimizations geared towards improving the code performance/speed with a high risk of
worsening or impacting code size.
If you specify the option without a parameter, the default setting is --opt_for_speed=4.

The initial mechanism for controlling code space, the --opt_for_space option, has the following
equivalences with the --opt_for_speed option:

--opt_for_space --opt_for_speed
none =4
=0 =3
=1 =2
=2 =1
=3 =0

100 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

3.14.5 What Kind of Optimization Is Being Performed?

The TMS320C6000 C/C++ compiler uses a variety of optimization techniques to improve the execution
speed of your C/C++ programs and to reduce their size.
Following are some of the optimizations performed by the compiler:

Optimization See
Cost-based register allocation Section 3.14.5.1
Alias disambiguation Section 3.14.5.1
Branch optimizations and control-flow simplification Section 3.14.5.3
Data flow optimizations Section 3.14.5.4
• Copy propagation
• Common subexpression elimination
• Redundant assignment elimination
Expression simplification Section 3.14.5.5
Inline expansion of functions Section 3.14.5.6
Function Symbol Aliasing Section 3.14.5.7
Induction variable optimizations and strength reduction Section 3.14.5.8
Loop-invariant code motion Section 3.14.5.9
Loop rotation Section 3.14.5.10
Instruction scheduling Section 3.14.5.11

C6000-Specific Optimization See

3.14.5.1 Cost-Based Register Allocation

The compiler, when optimization is enabled, allocates registers to user variables and compiler temporary
values according to their type, use, and frequency. Variables used within loops are weighted to have
priority over others, and those variables whose uses do not overlap can be allocated to the same register.
Induction variable elimination and loop test replacement allow the compiler to recognize the loop as a
simple counting loop and software pipeline, unroll, or eliminate the loop. Strength reduction turns the array
references into efficient pointer references with autoincrements.

3.14.5.2 Alias Disambiguation

C and C++ programs generally use many pointer variables. Frequently, compilers are unable to determine
whether or not two or more I values (lowercase L: symbols, pointer references, or structure references)
refer to the same memory location. This aliasing of memory locations often prevents the compiler from
retaining values in registers because it cannot be sure that the register and memory continue to hold the
same values over time.
Alias disambiguation is a technique that determines when two pointer expressions cannot point to the
same location, allowing the compiler to freely optimize such expressions.

SPRU187U – July 2012 Optimizing Your Code 101

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using Performance Advice to Optimize your Code www.ti.com

3.14.5.3 Branch Optimizations and Control-Flow Simplification

The compiler analyzes the branching behavior of a program and rearranges the linear sequences of
operations (basic blocks) to remove branches or redundant conditions. Unreachable code is deleted,
branches to branches are bypassed, and conditional branches over unconditional branches are simplified
to a single conditional branch.
When the value of a condition is determined at compile time (through copy propagation or other data flow
analysis), the compiler can delete a conditional branch. Switch case lists are analyzed in the same way as
conditional branches and are sometimes eliminated entirely. Some simple control flow constructs are
reduced to conditional instructions, totally eliminating the need for branches.

3.14.5.4 Data Flow Optimizations

Collectively, the following data flow optimizations replace expressions with less costly ones, detect and
remove unnecessary assignments, and avoid operations that produce values that are already computed.
The compiler with optimization enabled performs these data flow optimizations both locally (within basic
blocks) and globally (across entire functions).
• Copy propagation. Following an assignment to a variable, the compiler replaces references to the
variable with its value. The value can be another variable, a constant, or a common subexpression.
This can result in increased opportunities for constant folding, common subexpression elimination, or
even total elimination of the variable.
• Common subexpression elimination. When two or more expressions produce the same value, the
compiler computes the value once, saves it, and reuses it.
• Redundant assignment elimination. Often, copy propagation and common subexpression elimination
optimizations result in unnecessary assignments to variables (variables with no subsequent reference
before another assignment or before the end of the function). The compiler removes these dead
assignments.

3.14.5.5 Expression Simplification

For optimal evaluation, the compiler simplifies expressions into equivalent forms, requiring fewer
instructions or registers. Operations between constants are folded into single constants. For example, a =
(b + 4) - (c + 1) becomes a = b - c + 3.

3.14.5.6 Inline Expansion of Functions

The compiler replaces calls to small functions with inline code, saving the overhead associated with a
function call as well as providing increased opportunities to apply other optimizations.

3.14.5.7 Function Symbol Aliasing

The compiler recognizes a function whose definition contains only a call to another function. If the two
functions have the same signature (same return value and same number of parameters with the same
type, in the same order), then the compiler can make the calling function an alias of the called function.
For example, consider the following:
int bbb(int arg1, char *arg2);

int aaa(int n, char *str)

{
return bbb(n, str);
}

For this example, the compiler makes aaa an alias of bbb, so that at link time all calls to function aaa
should be redirected to bbb. If the linker can successfully redirect all references to aaa, then the body of
function aaa can be removed and the symbol aaa is defined at the same address as bbb.

102 Optimizing Your Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Using Performance Advice to Optimize your Code

3.14.5.8 Induction Variables and Strength Reduction

Induction variables are variables whose value within a loop is directly related to the number of executions
of the loop. Array indices and control variables for loops are often induction variables.
Strength reduction is the process of replacing inefficient expressions involving induction variables with
more efficient expressions. For example, code that indexes into a sequence of array elements is replaced
with code that increments a pointer through the array.
Induction variable analysis and strength reduction together often remove all references to your loop-
control variable, allowing its elimination.

3.14.5.9 Loop-Invariant Code Motion

This optimization identifies expressions within loops that always compute to the same value. The
computation is moved in front of the loop, and each occurrence of the expression in the loop is replaced
by a reference to the precomputed value.

3.14.5.10 Loop Rotation

The compiler evaluates loop conditionals at the bottom of loops, saving an extra branch out of the loop. In
many cases, the initial entry conditional check and the branch are optimized out.

3.14.5.11 Instruction Scheduling

The compiler performs instruction scheduling, which is the rearranging of machine instructions in such a
way that improves performance while maintaining the semantics of the original order. Instruction
scheduling is used to improve instruction parallelism and hide pipeline latencies. It can also be used to
reduce code size.

3.14.5.12 Register Variables

The compiler helps maximize the use of registers for storing local variables, parameters, and temporary
values. Accessing variables stored in registers is more efficient than accessing variables in memory.
Register variables are particularly effective for pointers.

3.14.5.13 Register Tracking/Targeting

The compiler tracks the contents of registers to avoid reloading values if they are used again soon.
Variables, constants, and structure references such as (a.b) are tracked through straight-line code.
Register targeting also computes expressions directly into specific registers when required, as in the case
of assigning to register variables or returning values from functions.

3.14.5.14 Software Pipelining

Software pipelining is a technique use to schedule from a loop so that multiple iterations of a loop execute
in parallel. See Section 3.2 for more information.

SPRU187U – July 2012 Optimizing Your Code 103

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 4
SPRU187U – July 2012

Using the Assembly Optimizer

The assembly optimizer allows you to write assembly code without being concerned with the pipeline
structure of the C6000 or assigning registers. It accepts linear assembly code, which is assembly code
that may have had register-allocation performed and is unscheduled. The assembly optimizer assigns
registers and uses loop optimizations to turn linear assembly into highly parallel assembly.

Topic ........................................................................................................................... Page

4.1 Code Development Flow to Increase Performance ............................................... 105

4.2 About the Assembly Optimizer .......................................................................... 106
4.3 What You Need to Know to Write Linear Assembly .............................................. 107
4.4 Assembly Optimizer Directives .......................................................................... 113
4.5 Avoiding Memory Bank Conflicts With the Assembly Optimizer ............................ 127
4.6 Memory Alias Disambiguation ........................................................................... 133

104 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Code Development Flow to Increase Performance

4.1 Code Development Flow to Increase Performance

You can achieve the best performance from your C6000 code if you follow this flow when you are writing
and debugging your code:

Phase 1: Write C/C++ code

Develop C/C++ code
Compile

Profile

Efficient Yes
Complete
enough?

No
Refine C/C++ code
Phase 2:
Refine C/C++ Compile
code
Profile

Efficient Yes
Complete
enough?

No
Yes
More C/C++
optimizations?

No
Write/refine linear assembly
Phase 3:
Write linear Assembly optimize
assembly
Profile

No
Efficient
enough?

Yes
Complete

There are three phases of code development for the C6000:

• Phase 1: write in C
You can develop your C/C++ code for phase 1 without any knowledge of the C6000. Use a simulator
after compiling with the --opt_level=3 option without any --debug option to identify any inefficient areas
in your C/C++ code. See Section 3.14.3 for more information about debugging and profiling optimized
code. To improve the performance of your code, proceed to phase 2.

SPRU187U – July 2012 Using the Assembly Optimizer 105

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
About the Assembly Optimizer www.ti.com

• Phase 2: refine your C/C++ code

In phase 2, use the intrinsics and compiler options that are described in this book to improve your
C/C++ code. Use a simulator to check the performance of your altered code. Refer to the
TMS320C6000 Programmer's Guide for hints on refining C/C++ code. If your code is still not as
efficient as you would like it to be, proceed to phase 3.
• Phase 3: write linear assembly
In this phase, you extract the time-critical areas from your C/C++ code and rewrite the code in linear
assembly. You can use the assembly optimizer to optimize this code. When you are writing your first
pass of linear assembly, you should not be concerned with the pipeline structure or with assigning
registers. Later, when you are refining your linear assembly code, you might want to add more details
to your code, such as partitioning registers.
Improving performance in this stage takes more time than in phase 2, so try to refine your code as
much as possible before using phase 3. Then, you should have smaller sections of code to work on in
this phase.

4.2 About the Assembly Optimizer

If you are not satisfied with the performance of your C/C++ code after you have used all of the C/C++
optimizations that are available, you can use the assembly optimizer to make it easier to write assembly
code for the C6000.
The assembly optimizer performs several tasks including the following:
• Optionally, partitions instructions and/or registers
• Schedules instructions to maximize performance using the instruction-level parallelism of the C6000
• Ensures that the instructions conform to the C6000 latency requirements
• Optionally, allocates registers for your source code
Like the C/C++ compiler, the assembly optimizer performs software pipelining. Software pipelining is a
technique used to schedule instructions from a loop so that multiple iterations of the loop execute in
parallel. The code generation tools attempt to software pipeline your code with inputs from you and with
information that it gathers from your program. For more information, see Section 3.2.
To invoke the assembly optimizer, use the compiler program (cl6x). The assembly optimizer is
automatically invoked by the compiler program if one of your input files has a .sa extension. You can
specify C/C++ source files along with your linear assembly files. For more information about the compiler
program, see Chapter 2.

106 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com What You Need to Know to Write Linear Assembly

4.3 What You Need to Know to Write Linear Assembly

By using the C6000 profiling tools, you can identify the time-critical sections of your code that need to be
rewritten as linear assembly. The source code that you write for the assembly optimizer is similar to
assembly source code. However, linear assembly code does not need to be partitioned, scheduled, or
register allocated. The intention is for you to let the assembly optimizer determine this information for you.
When you are writing linear assembly code, you need to know about these items:
• Assembly optimizer directives
Your linear assembly file can be a combination of linear assembly code segments and regular
assembly source. Use the assembly optimizer directives to differentiate the assembly optimizer code
from the regular assembly code and to provide the assembly optimizer with additional information
about your code. The assembly optimizer directives are described in Section 4.4.
• Options that affect what the assembly optimizer does
The compiler options in Table 4-1 affect the behavior of the assembly optimizer.

Table 4-1. Options That Affect the Assembly Optimizer

Option Effect See
--ap_extension Changes the default extension for assembly optimizer source files Section 2.3.9
--ap_file Changes how assembly optimizer source files are identified Section 2.3.7
--disable_software_pipelining Turns off software pipelining Section 3.2.1
--debug_software_pipeline Generates verbose software pipelining information Section 3.2.2
--interrupt_threshold=n Specifies an interrupt threshold value Section 2.12
--keep_asm Keeps the assembly language (.asm) file Section 2.3.1
--no_bad_aliases Presumes no memory aliasing Section 3.10.3
--opt_for_space=n Controls code size on four levels (n=0, 1, 2, or 3) Section 3.5
--opt_level=n Increases level of optimization (n=0, 1, 2, or 3) Section 3.1
--quiet Suppresses progress messages Section 2.3.1
--silicon_version=n Select target version Section 2.3.4
--skip_assembler Compiles or assembly optimizes only (does not assemble) Section 2.3.1
--speculate_loads=n Allows speculative execution of loads with bounded address ranges Section 3.2.3

• TMS320C6000 instructions
When you are writing your linear assembly, your code does not need to indicate the following:
– Pipeline latency
– Register usage
– Which unit is being used
As with other code generation tools, you might need to modify your linear assembly code until you are
satisfied with its performance. When you do this, you will probably want to add more detail to your
linear assembly. For example, you might want to partition or assign some registers.

Do Not Use Scheduled Assembly Code as Source

NOTE: The assembly optimizer assumes that the instructions in the input file are placed in the
logical order in which you would like them to occur (that is, linear assembly code). Parallel
instructions are illegal.
If the compiler cannot make your instructions linear (non-parallel), it produces an error
message. The compiler assumes instructions occur in the order the instructions appear in
the file. Scheduled code is illegal (even non-parallel scheduled code). Scheduled code may
not be detected by the compiler but the resulting output may not be what you intended.

• Linear assembly source statement syntax

The linear assembly source programs consist of source statements that can contain assembly
optimizer directives, assembly language instructions, and comments. See Section 4.3.1 for more
information on the elements of a source statement.

SPRU187U – July 2012 Using the Assembly Optimizer 107

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
What You Need to Know to Write Linear Assembly www.ti.com

• Specifying registers or register sides

Registers can be assigned explicitly to user symbols. Alternatively, symbols can be assigned to the A-
side or B-side leaving the compiler to do the actual register allocation. See Section 4.3.2 for
information on specifying registers.
• Specifying the functional unit
The functional unit specifier is optional in linear assembly code. Data path information is respected;
unit information is ignored.
• Source comments
The assembly optimizer attaches the comments on instructions from the input linear assembly to the
output file. It attaches the 2-tuple <x, y> to the comments to specify which iteration and cycle of the
loop an instruction is on in the software pipeline. The zero-based number x represents the iteration the
instruction is on during the first execution of the kernel. The zero-based number y represents the cycle
the instruction is scheduled on within a single iteration of the loop. See Section 4.3.4, for an illustration
of the use of source comments and the resulting assembly optimizer output.

4.3.1 Linear Assembly Source Statement Format

A source statement can contain five ordered fields (label, mnemonic, unit specifier, operand list, and
comment). The general syntax for source statements is as follows:

label[:] Labels are optional for all assembly language instructions and for most (but not all)
assembly optimizer directives. When used, a label must begin in column 1 of a source
statement. A label can be followed by a colon.
[ register ] Square brackets ([ ]) enclose conditional instructions. The machine-instruction
mnemonic is executed based on the value of the register within the brackets; valid
register names are A0 for C6400, C6400+, C6740, and C6600 only; A1, A2, B0, B1,
B2, or symbolic.
mnemonic The mnemonic is a machine-instruction (such as ADDK, MVKH, B) or assembly
optimizer directive (such as .proc, .trip)
unit specifier The optional unit specifier enables you to specify the functional unit operand. Only the
specified unit side is used; other specifications are ignored. The preferred method is
specifying register sides.
operand list The operand list is not required for all instructions or directives. The operands can be
symbols, constants, or expressions and must be separated by commas.
comment Comments are optional. Comments that begin in column 1 must begin with a
semicolon or an asterisk; comments that begin in any other column must begin with a
semicolon.

The C6000 assembly optimizer reads up to 200 characters per line. Any characters beyond 200 are
truncated. Keep the operational part of your source statements (that is, everything other than comments)
less than 200 characters in length for correct assembly. Your comments can extend beyond the character
limit, but the truncated portion is not included in the .asm file.
Follow these guidelines in writing linear assembly code:
• All statements must begin with a label, a blank, an asterisk, or a semicolon.
• Labels are optional; if used, they must begin in column 1.
• One or more blanks must separate each field. Tab characters are interpreted as blanks. You must
separate the operand list from the preceding field with a blank.
• Comments are optional. Comments that begin in column 1 can begin with an asterisk or a semicolon (*
or ;) but comments that begin in any other column must begin with a semicolon.
• If you set up a conditional instruction, the register must be surrounded by square brackets.
• A mnemonic cannot begin in column 1 or it is interpreted as a label.
Refer to the TMS320C6000 Assembly Language Tools User's Guide for information on the syntax of
C6000 instructions, including conditional instructions, labels, and operands.

108 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com What You Need to Know to Write Linear Assembly

4.3.2 Register Specification for Linear Assembly

There are only two cross paths in the C6000. This limits the C6000 to one source read from each data
path's opposite register file per cycle. The compiler must select a side for each instruction; this is called
partitioning.
It is recommended that you do not initially partition the linear assembly source code by hand. This allows
the compiler more freedom to partition and optimize your code. If the compiler does not find an optimal
partition in a software pipelined loop, then you can partition enough instructions by hand to force optimal
partitioning by partitioning registers.
The assembly optimizer chooses a register for you such that its use agrees with the functional units
chosen for the instructions that operate on the value.
Registers can be directly partitioned through two directives. The .rega directive is used to constrain a
symbolic name to A-side registers. The .regb directive is used to constrain a symbolic name to B-side
registers. See the .rega/.regb topic for further details on these directives. The .reg directive allows you to
use descriptive names for values that are stored in registers. See the .reg topic for further details and
examples of the .reg directive.
Example 4-1 is a hand-coded linear assembly program that computes a dot product; compare to
Example 4-2, which illustrates C code.

Example 4-1. Linear Assembly Code for Computing a Dot Product

_dotp: .cproc a_0, b_0

.rega a_4, tmp0, sum0, prod1, prod2

.regb b_4, tmp1, sum1, prod3, prod4
.reg cnt, sum
.reg val0, val1

ADD 4, a_0, a_4

ADD 4, b_0, b_4
MVK 100, cnt
ZERO sum0
ZERO sum1

loop: .trip 25
LDW *a_0++[2], val0 ; load a[0-1]
LDW *b_0++[2], val1 ; load b[0-1]
MPY val0, val1, prod1 ; a[0] * b[0]
MPYH val0, val1, prod2 ; a[1] * b[1]
ADD prod1, prod2, tmp0 ; sum0 += (a[0]*b[0]) +
ADD tmp0, sum0, sum0 ; (a[1]*b[1])

LDW *a_4++[2], val0 ; load a[2-3]

LDW *b_4++[2], val1 ; load b[2-3]
MPY val0, val1, prod3 ; a[2] * b[2]
MPYH val0, val1, prod4 ; a[3] * b[3]
ADD prod3, prod4, tmp1 ; sum1 =+ (a[2]*b[2]) +
ADD tmp1, sum1, sum1 ; (a[3]*b[3])

[cnt] SUB cnt, 4, cnt ; cnt -= 4

[cnt] B loop ; if (cnt!=0) goto loop

ADD sum0, sum1, sum ; compute final result

.return sum
.endproc

Example 4-2 is refined C code for computing a dot product.

SPRU187U – July 2012 Using the Assembly Optimizer 109

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
What You Need to Know to Write Linear Assembly www.ti.com

Example 4-2. C Code for Computing a Dot Product

int dotp(short a[], shortb[])

{
int sum0 = 0;
int sum1 = 0;

int sum, I;

for (I = 0; I < 100/4; I +=4)

{
sum0 += a[i] * b[i];
sum0 += a[i+1] * b[i+1];

sum1 += a[i+2] * b[i+2];

sum1 += a[i+3] * [b[i+3];
}
return
}

The old method of partitioning registers indirectly by partitioning instructions can still be used. Side and
functional unit specifiers can still be used on instructions. However, functional unit specifiers (.L/.S/.D/.M)
are ignored. Side specifiers are translated into partitioning constraints on the corresponding symbolic
names, if any. For example:
MV .1 x, y ; translated to .REGA y
LDW .D2T2 *u, v:w ; translated to .REGB u, v, w

In the linear assembler, you can also specify register pairs using the .cproc and/or .reg directive as in
Example 4-3:

Example 4-3. Specifying a Register Pair

.global foopair
foopair: .cproc q1:q0,s0
.reg r1:r0
ADD q1:q0, s0, r1:r0
.return r1:r0
.endproc

In Example 4-3, the expression "q1:q0" means that the first argument into the linear assembly function is a
register pair. By the C calling conventions, the pair "q1:q0" symbols are mapped to register pair "a5:a4".
When a register pair syntax is used as the argument to a .reg directive (as shown), it means that the two
register symbols are constrained to be an aligned register pair when the compiler processes the linear
assembler source and allocates actual registers that the register pair symbols map to "r1:r0" as shown.
The 7.2. Beta compiler supports a register quad syntax (C6600 only), in order to specify 128-bit operands
of 128-bit capable instructions in linear assembly and assembly source code. Example 4-4 illustrates how
you can specify register quads:

Example 4-4. Specifying a Register Quad (C6600 Only)

.global fooquad
fooquad: .cproc q3:q2:q1:q0, s3:s2:s1:s0
.reg r3:r2:r1:r0
QMPY32 s3:s2:s1:s0, q3:q2:q1:q0, r3:r2:r1:r0
.return r3:r2:r1:r0
.endproc

110 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com What You Need to Know to Write Linear Assembly

In Example 4-4, the expression "q3:q2:q1:q0" means that the first argument into the linear assembly
function is a register quad. By the C calling conventions, the quad "q3:q2:q1:q0" symbols are mapped to
register quad "a7:a6:a5:a4". When a register quad syntax is used as the argument to a .reg directive (as
shown), it means that the four register symbols are constrained to be an aligned register quad when the
compiler processes the linear assembler source and allocates actual registers that the register quad
symbols map to "r3:r2:r1:r0" as shown.

4.3.3 Functional Unit Specification for Linear Assembly

Specifying functional units has been deprecated by the ability to partition registers directly. (See
Section 4.3.2 for details.) While you can use the unit specifier field in linear assembly, only the register
side information is used by the compiler.
You specify a functional unit by following the assembler instruction with a period (.) and a functional unit
specifier. One instruction can be assigned to each functional unit in a single instruction cycle. There are
eight functional units, two of each functional type, and two address paths. The two of each functional type
are differentiated by the data path each uses, A or B.

.D1 and .D2 Data/addition/subtraction operations

.L1 and .L2 Arithmetic logic unit (ALU)/compares/long data arithmetic
.M1 and .M2 Multiply operations
.S1 and .S2 Shift/ALU/branch/field operations
.T1 and .T2 Address paths

There are several ways to enter the unit specifier filed in linear assembly. Of these, only the specific
register side information is recognized and used:
• You can specify the particular functional unit (for example, .D1).
• You can specify the .D1 or .D2 functional unit followed by T1 or T2 to specify that the nonmemory
operand is on a specific register side. T1 specifies side A and T2 specifies side B. For example:
LDW .D1T2 *A3[A4], B3
LDW .D1T2 *src, dst
• You can specify only the data path (for example, .1), and the assembly optimizer assigns the functional
type (for example, .L1).
For more information on functional units refer to the TMS320C6000 CPU and Instruction Set Reference
Guide.

4.3.4 Using Linear Assembly Source Comments

Your comments in linear assembly can begin in any column and extend to the end of the source line. A
comment can contain any ASCII character, including blanks. Your comments are printed in the linear
assembly source listing, but they do not affect the linear assembly.
A source statement that contains only a comment is valid. If it begins in column 1, it can start with a
semicolon ( ; ) or an asterisk ( * ). Comments that begin anywhere else on the line must begin with a
semicolon. The asterisk identifies a comment only if it appears in column 1.
The assembly optimizer schedules instructions; that is, it rearranges instructions. Stand-alone comments
are moved to the top of a block of instructions. Comments at the end of an instruction statement remain in
place with the instruction.
Example 4-5 shows code for a function called Lmac that contains comments.

SPRU187U – July 2012 Using the Assembly Optimizer 111

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
What You Need to Know to Write Linear Assembly www.ti.com

Example 4-5. Lmac Function Code Showing Comments

Lmac: .cproc A4,B4

.reg t0,t1,p,i,sh:sl

MVK 100,i
ZERO sh
ZERO sl

loop: .trip 100

LDH *a4++, t0 ; t0 = a[i]

LDH *b4++, t1 ; t1 = b[i]
MPY t0,t1,p ; prod = t0 * t1
ADD p,sh:sl,sh:sl ; sum += prod
[I] ADD -1,i,i ; --I
[I] B loop ; if (I) goto loop

.return sh:sl

.endproc

4.3.5 Assembly File Retains Your Symbolic Register Names

In the output assembly file, register operands contain your symbolic name. This aids you in debugging
your linear assembly files and in gluing snippets of linear assembly output into assembly files.
A .map directive (see the .map topic) at the beginning of an assembly function associates the symbolic
name with the actual register. In other words, the symbolic name becomes an alias for the actual register.
The .map directive can be used in assembly and linear assembly code.
When the compiler splits a user symbol into two symbols and each is mapped to distinct machine register,
a suffix is appended to instances of the symbolic name to generate unique names so that each unique
name is associated with one machine register.
For example, if the compiler associated the symbolic name y with A5 in some instructions and B6 in some
others, the output assembly code might look like:
.MAP y/A5
.MAP y'/B6
...
ADD .S2X y, 4, y' ; Equivalent to add A5, 4, B6

To disable this format with symbolic names and display assembly instructions with actual registers
instead, compile with the --machine_regs option.

112 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Assembly Optimizer Directives

4.4 Assembly Optimizer Directives

Assembly optimizer directives supply data for and control the assembly optimization process. The
assembly optimizer optimizes linear assembly code that is contained within procedures; that is, code
within the .proc and .endproc directives or within the .cproc and .endproc directives. If you do not use
.cproc/.proc directives in your linear assembly file, your code will not be optimized by the assembly
optimizer. This section describes these directives and others that you can use with the assembly
optimizer.
Table 4-2 summarizes the assembly optimizer directives. It provides the syntax for each directive, a
description of each directive, and any restrictions that you should keep in mind. See the specific directive
topic for more detail.
In Table 4-2 and the detailed directive topics, the following terms for parameters are used:
argument— Symbolic variable name or machine register
memref— Symbol used for a memory reference (not a register)
register— Machine (hardware) register
symbol— Symbolic user name or symbolic register name
variable— Symbolic variable name or machine register

Table 4-2. Assembly Optimizer Directives Summary

Syntax Description Restrictions
.call [ret_reg =] func_name (argument1 , Calls a function Valid only within procedures
argument2 , ...)
.circ symbol1 / register1 [, symbol2 / Declares circular addressing Must manually insert setup/teardown code
register2 ] for circular addressing. Valid only within
procedures
label .cproc [argument1 [, argument2 , …]] Start a C/C++ callable procedure Must use with .endproc
.endproc End a C/C++ callable procedure Must use with .cproc
.endproc [variable1 [, variable2,…]] End a procedure Must use with .proc
.map symbol1 / register1 [, symbol2 / Assigns a symbol to a register Must use an actual machine register
register2 ]
.mdep [memref1 [, memref2 ]] Indicates a memory dependence Valid only within procedures
.mptr {variable|memref}, base [+ offset] Avoid memory bank conflicts Valid only within procedures
[, stride]
.no_mdep No memory aliases in the function Valid only within procedures
.pref symbol / register1 [/register2 /...] Assigns a symbol to a register in a set Must use actual machine registers
label .proc [variable1 [, variable2 , …]] Start a procedure Must use with .endproc
.reg symbol1 [, symbol2 ,…] Declare variables Valid only within procedures
.rega symbol1 [, symbol2 ,…] Partition symbol to A-side register Valid only within procedures
.regb symbol1 [, symbol2 ,…] Partition symbol to B-side register Valid only within procedures
.reserve [register1 [, register2 ,…]] Prevents the compiler from allocating a Valid only within procedures
register
.return [argument] Return a value to a procedure Valid only within .cproc procedures
label .trip min Specify trip count value Valid only within procedures
.volatile memref1 [, memref2 ,…] Designate memory reference volatile Use --interrupt_threshold=1 if reference
may be modified during an interrupt

SPRU187U – July 2012 Using the Assembly Optimizer 113

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.call — Calls a Function www.ti.com

.call Calls a Function

Syntax .call [ret_reg=] func_name ([argument1, argument2,...])

Description Use the .call directive to call a function. Optionally, you can specify a register that is
assigned the result of the call. The register can be a symbolic or machine register. The
.call directive adheres to the same register and function calling conventions as the
C/C++ compiler. For information, see Section 7.3 and Section 7.4. There is no support
for alternative register or function calling conventions.
You cannot call a function that has a variable number of arguments, such as printf. No
error checking is performed to ensure the correct number and/or type of arguments is
passed. You cannot pass or return structures through the .call directive.
Following is a description of the .call directive parameters:

ret_reg (Optional) Symbolic/machine register that is assigned the result of the

call. If not specified, the assembly optimizer presumes the call
overwrites the registers A5 and A4 with a result.
func_name The name of the function to call, or the name of the symbolic/
machine register for indirect calls. A register pair is not allowed. The
label of the called function must be defined in the file. If the code for
the function is not in the file, the label must be defined with the .global
or .ref directive (refer to the TMS320C6000 Assembly Language
Tools User's Guide for details). If you are calling a C/C++ function,
you must use the appropriate linkname of that function. See
Section 6.12 for more information.
arguments (Optional) Symbolic/machine registers passed as an argument. The
arguments are passed in this order and cannot be a constant,
memory reference, or other expression.

By default, the compiler generates near calls and the linker utilizes trampolines if the
near call will not reach its destination. To force a far call, you must explicitly load the
address of the function into a register, and then issue an indirect call. For example:
MVK func,reg
MVKH func,reg
.call reg(op1) ; forcing a far call

If you want to use * for indirection, you must abide by C/C++ syntax rules, and use the
following alternate syntax:
.call [ret_reg =] (* ireg)([arg1, arg2,...])
For example:
.call (*driver)(op1, op2) ; indirect call

.reg driver
.call driver(op1, op2) ; also an indirect call

Here are other valid examples that use the .call syntax.
.call fir(x, h, y) ; void function

.call minimal( ) ; no arguments

.call sum = vecsum(a, b) ; returns an int

.call hi:lo = _atol(string) ; returns a long

Since you can use machine register names anywhere you can use symbolic registers, it
may appear you can change the function calling convention. For example:
.call A6 = compute()

114 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .circ — Declare Circular Registers

It appears that the result is returned in A6 instead of A4. This is incorrect. Using machine
registers does not override the calling convention. After returning from the compute
function with the returned result in A4, a MV instruction transfers the result to A6.

Example Here is a complete .call example:

.global _main
.global _puts, _rand, _ltoa
.sect ".const"
string1: .string "The random value returned is ", 0
string2: .string " ", 10, 0 ; '10' == newline
.bss charbuf, 20
.text
_main: .cproc
.reg random_value, bufptr, ran_val_hi:ran_val_lo
.call random_value = _rand() ; get a random value

MVKL string1, bufptr ; load address of string1

MVKH string1, bufptr
.call _puts(bufptr) ; print out string1
MV random_value, ran_val_lo
SHR ran_val_lo, 31, ran_val_hi ; sign extend random value
.call _ltoa(ran_val_hi:ran_val_lo, bufptr) ; convert it to a string
.call _puts(bufptr) ; print out the random value
MVKL string2, bufptr ; load address of string2
MVKH string2, bufptr
.call _puts(bufptr) ; print out a newline
.endproc

.circ Declare Circular Registers

Syntax .circ symbol1 /register1 [, symbol2 lregister2 , ...]

Description The .circ directive assigns a symbolic register name to a machine register and declares
the symbolic register as available for circular addressing. The compiler then assigns the
variable to the register and ensures that all code transformations are safe in this
situation. You must insert setup/teardown code for circular addressing.

symbol A valid symbol name to be assigned to the register. The variable is up

to 128 characters long and must begin with a letter. Remaining
characters of the variable can be a combination of alphanumeric
characters, the underscore (_), and the dollar sign ($).
register Name of the actual register to be assigned a variable.

The compiler assumes that it is safe to speculate any load using an explicitly declared
circular addressing variable as the address pointer and may exploit this assumption to
perform optimizations.
When a symbol is declared with the .circ directive, it is not necessary to declare that
symbol with the .reg directive.
The .circ directive is equivalent to using .map with a circular declaration.

Example Here the symbolic name Ri is assigned to actual machine register Mi and Ri is declared
as potentially being used for circular addressing.
.CIRC R1/M1, R2/M2 ...

SPRU187U – July 2012 Using the Assembly Optimizer 115

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.cproc/.endproc — Define a C Callable Procedure www.ti.com

.cproc/.endproc Define a C Callable Procedure

Syntax label .cproc [argument1 [, argument2 , …]]

.endproc

Description Use the .cproc/.endproc directive pair to delimit a section of your code that you want
the assembly optimizer to optimize and treat as a C/C++ callable function. This section is
called a procedure. The .cproc directive is similar to the .proc directive in that you use
.cproc at the beginning of a section and .endproc at the end of a section. In this way, you
can set off sections of your assembly code that you want to be optimized, like functions.
The directives must be used in pairs; do not use .cproc without the corresponding
.endproc. Specify a label with the .cproc directive. You can have multiple procedures in a
linear assembly file.
The .cproc directive differs from the .proc directive in that the compiler treats the .cproc
region as a C/C++ callable function. The assembly optimizer performs some operations
automatically in a .cproc region in order to make the function conform to the C/C++
calling conventions and to C/C++ register usage conventions.
These operations include the following:
• When you use save-on-entry registers (A10 to A15 and B10 to B15), the assembly
optimizer saves the registers on the stack and restores their original values at the
end of the procedure.
• If the compiler cannot allocate machine registers to symbolic register names specified
with the .reg directive (see the .reg topic) it uses local temporary stack variables. With
.cproc, the compiler manages the stack pointer and ensures that space is allocated
on the stack for these variables.
For more information, see Section 7.3 and Section 7.4.
Use the optional argument to represent function parameters. The argument entries are
very similar to parameters declared in a C/C++ function. The arguments to the .cproc
directive can be of the following types:
• Machine-register names. If you specify a machine-register name, its position in the
argument list must correspond to the argument passing conventions for C (see
Section 7.4). For example, the C/C++ compiler passes the first argument to a
function in register A4. This means that the first argument in a .cproc directive must
be A4 or a symbolic name. Up to ten arguments can be used with the .cproc
directive.
• Variable names.If you specify a variable name, then the assembly optimizer ensures
that either the variable name is allocated to the appropriate argument passing
register or the argument passing register is copied to the register allocated for the
variable name. For example, the first argument in a C/C++ call is passed in register
A4, so if you specify the following .cproc directive:
frame .cproc arg1
The assembly optimizer either allocates arg1 to A4, or arg1 is allocated to a different
register (such as B7) and an MV A4, B7 is automatically generated.
• Register pairs. A register pair is specified as arghi:arglo and represents a 40-bit
argument or a 64-bit type double argument.
For example, the .cproc defined as follows:
_fcn: .cproc arg1, arg2hi:arg2lo, arg3, B6, arg5, B9:B8
...
.return res
...
.endproc
corresponds to a C function declared as:
int fcn(int arg1, long arg2, int arg3, int arg4, int arg5, long arg6);

116 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .cproc/.endproc — Define a C Callable Procedure

In this example, the fourth argument of .cproc is register B6. This is allowed since the
fourth argument in the C/C++ calling conventions is passed in B6. The sixth
argument of .cproc is the actual register pair B9:B8. This is allowed since the sixth
argument in the C/C++ calling conventions is passed in B8 or B9:B8 for longs.
• Register quads (C6600 only). A register quad is specified as r3:r2:r1:r0 and
represents a 128-bit type, __x128_t. See Example 4-4.
If you are calling a procedure from C++ source, you must use the appropriate linkname
for the procedure label. Otherwise, you can force C naming conventions by using the
extern C declaration. See Section 6.12 and Section 7.5 for more information.
When .endproc is used with a .cproc directive, it cannot have arguments. The live out set
for a .cproc region is determined by any .return directives that appear in the .cproc
region. (A value is live out if it has been defined before or within the procedure and is
used as an output from the procedure.) Returning a value from a .cproc region is
handled by the .return directive. The return branch is automatically generated in a .cproc
region. See the .return topic for more information.
Only code within procedures is optimized. The assembly optimizer copies any code that
is outside of procedures to the output file and does not modify it. See Section 4.4.1 for a
list of instruction types that cannot appear in a .cproc region.

Example Here is an example in which .cproc and .endproc are used:

_if_then: .cproc a, cword, mask, theta

.reg cond, if, ai, sum, cntr

MVK 32,cntr ; cntr = 32

ZERO sum ; sum = 0

LOOP:
AND cword,mask,cond ; cond = codeword & mask
[cond] MVK 1,cond ; !(!(cond))
CMPEQ theta,cond,if ; (theta == !(!(cond)))
LDH *a++,ai ; a[i]
[if] ADD sum,ai,sum ; sum += a[i]
[!if] SUB sum,ai,sum ; sum -= a[i]
SHL mask,1,mask ; mask = mask << 1
[cntr] ADD -1,cntr,cntr ; decrement counter
[cntr] B LOOP ; for LOOP

.return sum

.endproc

SPRU187U – July 2012 Using the Assembly Optimizer 117

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.map — Assign a Variable to a Register www.ti.com

.map Assign a Variable to a Register

Syntax .map symbol1 / register1 [, symbol2 / register2 , ...]

Description The .map directive assigns symbol names to machine registers. Symbols are stored in
the substitution symbol table. The association between symbolic names and actual
registers is wiped out at the beginning and end of each linear assembly function. The
.map directive can be used in assembly and linear assembly files.

variable A valid symbol name to be assigned to the register. The substitution

symbol is up to 128 characters long and must begin with a letter.
Remaining characters of the variable can be a combination of
alphanumeric characters, the underscore (_), and the dollar sign ($).
register Name of the actual register to be assigned a variable.

When a symbol is declared with the .map directive, it is not necessary to declare that
symbol with the .reg directive.

Example Here the .map directive is used to assign x to register A6 and y to register B7. The
symbols are used with a move statement.
.map x/A6, y/B7
MV x, y ; equivalent to MV A6, B7

.mdep Indicates a Memory Dependence

Syntax .mdep memref1 , memref2

Description The .mdep directive identifies a specific memory dependence.

Following is a description of the .mdep directive parameters:

memref The symbol parameter is the name of the memory reference.

The symbol used to name a memory reference has the same syntax restrictions as any
assembly symbol. (For more information about symbols, refer to the TMS320C6000
Assembly Language Tools User's Guide.) It is in the same space as the symbolic
registers. You cannot use the same name for a symbolic register and annotating a
memory reference.
The .mdep directive tells the assembly optimizer that there is a dependence between
two memory references.
The .mdep directive is valid only within procedures; that is, within occurrences of the
.proc and .endproc directive pair or the .cproc and .endproc directive pair.

Example Here is an example in which .mdep is used to indicate a dependence between two
memory references.
.mdep ld1, st1

LDW *p1++{ld1}, inp1 ;memory reference "ld1"

;other code ...
STW outp2, *p2++{st1} ;memory reference "st1"

118 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .mptr — Avoid Memory Bank Conflicts

.mptr Avoid Memory Bank Conflicts

Syntax .mptr {variable | memref}, base [+ offset] [, stride]

Description The .mptr directive associates a register with the information that allows the assembly
optimizer to determine automatically whether two memory operations have a memory
bank conflict. If the assembly optimizer determines that two memory operations have a
memory bank conflict, then it does not schedule them in parallel.
A memory bank conflict occurs when two accesses to a single memory bank in a given
cycle result in a memory stall that halts all pipeline operation for one cycle while the
second value is read from memory. For more information on memory bank conflicts,
including how to use the .mptr directive to prevent them, see Section 4.5.
Following are descriptions of the .mptr directive parameters:

variable|memref The name of the register symbol or memory reference used to identify
a load or store involved in a dependence.
base A symbolic address that associates related memory accesses
offset The offset in bytes from the starting base symbol. The offset is an
optional parameter and defaults to 0.
stride The register loop increment in bytes. The stride is an optional
parameter and defaults to 0.

The .mptr directive tells the assembly optimizer that when the symbol or memref is used
as a memory pointer in an LD(B/BU)(H/HU)(W) or ST(B/H/W) instruction, it is initialized
to point to base + offset and is incremented by stride each time through the loop.
The .mptr directive is valid within procedures only; that is, within occurrences of the .proc
and .endproc directive pair or the .cproc and .endproc directive pair.
The symbolic addresses used for base symbol names are in a name space separate
from all other labels. This means that a symbolic register or assembly label can have the
same name as a memory bank base name. For example:
.mptr Darray,Darray

Example Here is an example in which .mptr is used to avoid memory bank conflicts.
_blkcp: .cproc I

.reg ptr1, ptr2, tmp1, tmp2

MVK 0x0, ptr1 ; ptr1 = address 0

MVK 0x8, ptr2 ; ptr2 = address 8

loop: .trip 50

.mptr ptr1, a+0, 4

.mptr foo, a+8, 4

; potential conflict
LDW *ptr1++, tmp1 ; load *0, bank 0
STW tmp1, *ptr2++{foo} ; store *8, bank 0

[I] ADD -1,i,i ; I--

[I] B loop ; if (!0) goto loop

.endproc

SPRU187U – July 2012 Using the Assembly Optimizer 119

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.no_mdep — No Memory Aliases in the Function www.ti.com

.no_mdep No Memory Aliases in the Function

Syntax .no_mdep

Description The .no_mdep directive tells the assembly optimizer that no memory dependencies
occur within that function, with the exception of any dependencies pointed to with the
.mdep directive.

Example Here is an example in which .no_mdep is used.

fn: .cproc dst, src, cnt
.no_mdep ;no memory aliasing in this function
...
.endproc

.pref Assign a Variable to a Register in a Set

Syntax .pref symbol / register1 [/register2...]

Description The .pref directive communicates a preference to assign a variable to one of a list of
registers. The preference is used only in the .cproc or .proc region the .pref directive is
declared in and is valid only until the end of the region.

symbol A valid symbol name to be assigned to the register. The substitution

symbol is up to 128 characters long and must begin with a letter.
Remaining characters of the symbol can be a combination of
alphanumeric characters, the underscore (_), and the dollar sign ($).
register List of actual registers to be assigned a variable.

There is no guarantee that the symbol will be assigned to any register in the specified
group. The compiler may ignore the preference.
When a symbol is declared with the .pref directive, it is not necessary to declare that
variable with the .reg directive.

Example Here x is given a preference to be assigned to either A6 or B7. However, It would be

correct for the compiler to assign x to B3 (for example) instead.
.PREF x/A6/B7 ; Preference to assign x to either A6 or B7

.proc/.endproc Define a Procedure

Syntax label .proc [variable1 [, variable2 , …]]

.endproc [register1 [, register2 , …]]

Description Use the .proc/.endproc directive pair to delimit a section of your code that you want the
assembly optimizer to optimize. This section is called a procedure. Use .proc at the
beginning of the section and .endproc at the end of the section. In this way, you can set
off sections of unscheduled assembly instructions that you want optimized by the
compiler. The directives must be used in pairs; do not use .proc without the
corresponding .endproc. Specify a label with the .proc directive. You can have multiple
procedures in a linear assembly file.
Use the optional variable parameter in the .proc directive to indicate which registers are
live in, and use the optional register parameter of the .endproc directive to indicate which
registers are live out for each procedure. The variable can be an actual register or a
symbolic name. For example:
.PROC x, A5, y, B7
...
.ENDPROC y

120 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .reg — Declare Symbolic Registers

A value is live in if it has been defined before the procedure and is used as an input to
the procedure. A value is live out if it has been defined before or within the procedure
and is used as an output from the procedure. If you do not specify any registers with the
.endproc directive, it is assumed that no registers are live out.
Only code within procedures is optimized. The assembly optimizer copies any code that
is outside of procedures to the output file and does not modify it.
See Section 4.4.1 for a list of instruction types that cannot appear in a .proc region.

Example Here is a block move example in which .proc and .endproc are used:
move .proc A4, B4, B0
.no_mdep

loop:
LDW *B4++, A1
MV A1, B1
STW B1, *A4++
ADD -4, B0, B0
[B0] B loop
.endproc

.reg Declare Symbolic Registers

Syntax .reg symbol1 [, symbol2 , …]

Description The .reg directive allows you to use descriptive names for values that are stored in
registers. The assembly optimizer chooses a register for you such that its use agrees
with the functional units chosen for the instructions that operate on the value.
The .reg directive is valid within procedures only; that is, within occurrences of the .proc
and .endproc directive pair or the .cproc and .endproc directive pair.
Declaring register pairs (or register quads for C6600) explicitly is optional. Doing so is
only necessary if the registers should be allocated as a pair, but they are not used that
way. It is a best practice to declare register pairs and register quads with the pair/quad
syntax. Here is an example of declaring a register pair:
.reg A7:A6

Example 1 This example uses the same code as the block move example shown for .proc/.endproc
but the .reg directive is used:
move .cproc dst, src, cnt

.reg tmp1, tmp2

loop:
LDW *src++, tmp1
MV tmp1, tmp2
STW tmp2, *dst++
ADD -4, cnt, cnt
[cnt] B loop

Notice how this example differs from the .proc example: symbolic registers declared with
.reg are allocated as machine registers.

Example 2 The code in the following example is invalid, because a variable defined by the .reg
directive cannot be used outside of the defined procedure:
move .proc A4
.reg tmp
LDW *A4++, top
MV top, B5
.endproc
MV top, B6 ; WRONG: top is invalid outside of the procedure

SPRU187U – July 2012 Using the Assembly Optimizer 121

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.rega/.regb — Partition Registers Directly www.ti.com

.rega/.regb Partition Registers Directly

Syntax .rega symbol1 [, symbol2 , …]

.regb symbol1 [, symbol2 , …]

Description Registers can be directly partitioned through two directives. The .rega directive is used
to constrain a symbol name to A-side registers. The .regb directive is used to constrain
a symbol name to B-side registers. For example:
.REGA y
.REGB u, v, w
MV x, y
LDW *u, v:w

The .rega and .regb directives are valid within procedures only; that is, within
occurrences of the .proc and .endproc directive pair or the .cproc and .endproc directive
pair.
When a symbol is declared with the .rega or .regb directive, it is not necessary to declare
that symbol with the .reg directive.
The old method of partitioning registers indirectly by partitioning instructions can still be
used. Side and functional unit specifiers can still be used on instructions. However,
functional unit specifiers (.L/.S/.D/.M) and crosspath information are ignored. Side
specifiers are translated into partitioning constraints on the corresponding symbol
names, if any. For example:
MV .1X z, y ; translated to .REGA y
LDW .D2T2 *u, v:w ; translated to .REGB u, v, w

.reserve Reserve a Register

Syntax .reserve [register1 [, register2 , …]]

Description The .reserve directive prevents the assembly optimizer from using the specified register
in a .proc or .cproc region.
If a .reserved register is explicitly assigned in a .proc or .cproc region, then the assembly
optimizer can also use that register. For example, the variable tmp1 can be allocated to
register A7, even though it is in the .reserve list, since A7 was explicitly defined in the
ADD instruction:
.cproc
.reserve a7
.reg tmp1
....
ADD a6, b4, a7
....
.endproc

Reserving Registers A4 and A5

NOTE: When inside of a .cproc region that contains a .call statement, A4 and A5
cannot be specified in a .reserve statement. The calling convention
mandates that A4 and A5 are used as the return registers for a .call
statement.

122 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .return — Return a Value to a C callable Procedure

Example 1 The .reserve in this example guarantees that the assembly optimizer does not use A10
to A13 or B10 to B13 for the variables tmp1 to tmp5:
test .proc a4, b4
.reg tmp1, tmp2, tmp3, tmp4, tmp5
.reserve a10, a11, a12, a13, b10, b11, b12, b13
.....
.endproc a4

Example 2 The assembly optimizer may generate less efficient code if the available register pool is
overly restricted. In addition, it is possible that the available register pool is constrained
such that allocation is not possible and an error message is generated. For example, the
following code generates an error since all of the conditional registers have been
reserved, but a conditional register is required for the variable tmp:
.cproc ...
.reserve a1,a2,b0,b1,b2
.reg tmp
....
[tmp] ....
....
.endproc

.return Return a Value to a C callable Procedure

Syntax .return [argument]

Description The .return directive function is equivalent to the return statement in C/C++ code. It
places the optional argument in the appropriate register for a return value as per the
C/C++ calling conventions (see Section 7.4).
The optional argument can have the following meanings:
• Zero arguments implies a .cproc region that has no return value, similar to a void
function in C/C++ code.
• An argument implies a .cproc region that has a 32-bit return value, similar to an int
function in C/C++ code.
• A register pair of the format hi:lo implies a .cproc region that has a 40-bit long, a 64-
bit long long, or a 64-bit type double return value; similar to a long/long long/double
function in C/C++ code.
Arguments to the .return directive can be either symbolic register names or machine-
register names.
All return statements in a .cproc region must be consistent in the type of the return value.
It is not legal to mix a .return arg with a .return hi:lo in the same .cproc region.
The .return directive is unconditional. To perform a conditional .return, simply use a
conditional branch around a .return. The assembly optimizer removes the branch and
generates the appropriate conditional code. For example, to return if condition cc is true,
code the return as:
[!cc] B around
.return
around:

Example This example uses a symbolic register, tmp, and a machine-register, A5, as .return
arguments:
.cproc ...
.reg tmp
...
.return tmp = legal symbolic name
...
.return a5 = legal actual name

SPRU187U – July 2012 Using the Assembly Optimizer 123

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.trip — Specify Trip Count Values www.ti.com

.trip Specify Trip Count Values

Syntax label .trip minimum value [, maximum value[, factor]]

Description The .trip directive specifies the value of the trip count. The trip count indicates how
many times a loop iterates. The .trip directive is valid within procedures only. Following
are descriptions of the .trip directive parameters:

label The label represents the beginning of the loop. This is a required
parameter.
minimum value The minimum number of times that the loop can iterate. This is a
required parameter. The default is 1.
maximum value The maximum number of times that the loop can iterate. The
maximum value is an optional parameter.
factor The factor used, along with minimum value and maximum value, to
determine the number of times that the loop can iterate. In the
following example, the loop executes some multiple of 8, between 8
and 48, times:
loop: .trip 8, 48, 8

A factor of 2 states that your loop always executes an even number

of times allowing the compiler to unroll once; this can result in a
performance increase.
The factor is optional when the maximum value is specified.

If the assembly optimizer cannot ensure that the trip count is large enough to pipeline a
loop for maximum performance, a pipelined version and an unpipelined version of the
same loop are generated. This makes one of the loops a redundant loop. The pipelined
or the unpipelined loop is executed based on a comparison between the trip count and
the number of iterations of the loop that can execute in parallel. If the trip count is
greater or equal to the number of parallel iterations, the pipelined loop is executed;
otherwise, the unpipelined loop is executed. For more information about redundant
loops, see Section 3.3.
You are not required to specify a .trip directive with every loop; however, you should use
.trip if you know that a loop iterates some number of times. This generally means that
redundant loops are not generated (unless the minimum value is really small) saving
code size and execution time.
If you know that a loop always executes the same number of times whenever it is called,
define maximum value (where maximum value equals minimum value) as well. The
compiler may now be able to unroll your loop thereby increasing performance.
When you are compiling with the interrupt flexibility option (--interrupt_threshold=n),
using a .trip maximum value allows the compiler to determine the maximum number of
cycles that the loop can execute. Then, the compiler compares that value to the
threshold value given by the --interrupt_threshold option. See Section 2.12 for more
information.

124 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com .volatile — Declare Memory References as Volatile

Example The .trip directive states that the loop will execute 16, 24, 32, 40 or 48 times when the
w_vecsum routine is called.
w_vecsum: .cproc ptr_a, ptr_b, ptr_c, weight, cnt
.reg ai, bi, prod, scaled_prod, ci
.no_mdep

loop: .trip 16, 48, 8

ldh *ptr_a++, ai
ldh *ptr_b++, bi
mpy weight, ai, prod
shr prod, 15, scaled_prod
add scaled_prod, bi, ci
sth ci, *ptr_c++
[cnt] sub cnt, 1, cnt
[cnt] b loop
.endproc

.volatile Declare Memory References as Volatile

Syntax .volatile memref1 [, memref2 , …]

Description The .volatile directive allows you to designate memory references as volatile. Volatile
loads and stores are not deleted. Volatile loads and stores are not reordered with
respect to other volatile loads and stores.
If the .volatile directive references a memory location that may be modified during an
interrupt, compile with the --interrupt_threshold=1 option to ensure all code referencing
the volatile memory location can be interrupted.

Example The st and ld memory references are designated as volatile.

.volatile st, ld

STW W, *X{st} ; volatile store

STW U, *V
LDW *Y{ld}, Z ; volatile load

SPRU187U – July 2012 Using the Assembly Optimizer 125

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
.volatile — Declare Memory References as Volatile www.ti.com

4.4.1 Instructions That Are Not Allowed in Procedures

These types of instructions are not allowed in .cproc or .proc topic regions:
• The stack pointer (register B15) can be read, but it cannot be written to. Instructions that write to B15
are not allowed in a .proc or .cproc region. Stack space can be allocated by the assembly optimizer in
a .proc or .cproc region for storage of temporary values. To allocate this storage area, the stack pointer
is decremented on entry to the region and incremented on exit from the region. Since the stack pointer
can change value on entry to the region, the assembly optimizer does not allow code that changes the
stack pointer register.
• Indirect branches are not allowed in a .proc or .cproc region so that the .proc or .cproc region exit
protocols cannot be bypassed. Here is an example of an indirect branch:
B B4 <= illegal
• Direct branches to labels not defined in the .proc or .cproc region are not allowed so that the .proc or
.cproc region exit protocols cannot be bypassed. Here is an example of a direct branch outside of a
.proc region:
.proc
...
B outside = illegal
.endproc
outside:
• Direct branches to the label associated with a .proc directive are not allowed. If you require a branch
back to the start of the linear assembly function, then use the .call directive. Here is an example of a
direct branch to the label of a .proc directive:
_func: .proc
...
B _func <= illegal
...
.endproc
• An .if/.endif loop must be entirely inside or outside of a proc or .cproc region. It is not allowed to have
part of an .if/.endif loop inside of a .proc or .cproc region and the other part of the .if/.endif loop outside
of the .proc or .cproc region. Here are two examples of legal .if/.endif loops. The first loop is outside a
.cproc region, the second loop is inside a .proc region:
.if
.cproc
...
.endproc
.endif

.proc
.if
...
.endif
.endproc
Here are two examples of .if/.endif loops that are partly inside and partly outside of a .cproc or .proc
region:
.if
.cproc
.endif
.endproc

.proc
.if
...
.else
.endproc
.endif
• The following assembly instructions cannot be used from linear assembly:
– EFI
– SPLOOP, SPLOOPD and SPLOOPW and all other loop-buffer related instructions

126 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Avoiding Memory Bank Conflicts With the Assembly Optimizer

– C6700+ instructions
– ADDKSP and DP-relative addressing

4.5 Avoiding Memory Bank Conflicts With the Assembly Optimizer

The internal memory of the C6000 family varies from device to device. See the appropriate device data
sheet to determine the memory spaces in your particular device. This section discusses how to write code
to avoid memory bank conflicts.
Most C6000 devices use an interleaved memory bank scheme, as shown in Figure 4-1. Each number in
the diagram represents a byte address. A load byte (LDB) instruction from address 0 loads byte 0 in
bank 0. A load halfword (LDH) from address 0 loads the halfword value in bytes 0 and 1, which are also in
bank 0. A load word (LDW) from address 0 loads bytes 0 through 3 in banks 0 and 1.
Because each bank is single-ported memory, only one access to each bank is allowed per cycle. Two
accesses to a single bank in a given cycle result in a memory stall that halts all pipeline operation for one
cycle while the second value is read from memory. Two memory operations per cycle are allowed without
any stall, as long as they do not access the same bank.

Figure 4-1. 4-Bank Interleaved Memory

0 1 2 3 4 5 6 7
8 9 10 11 12 13 14 15

8N 8N + 1 8N + 2 8N + 3 8N + 4 8N + 5 8N + 6 8N + 7
Bank 0 Bank 1 Bank 2 Bank 3

For devices that have more than one memory space (Figure 4-2), an access to bank 0 in one memory
space does not interfere with an access to bank 0 in another memory space, and no pipeline stall occurs.

Figure 4-2. 4-Bank Interleaved Memory With Two Memory Spaces

Memory
0 1 2 3 4 5 6 7
space 0
8 9 10 11 12 13 14 15

8N 8N + 1 8N + 2 8N + 3 8N + 4 8N + 5 8N + 6 8N + 7

Bank 0 Bank 1 Bank 2 Bank 3

Memory 8M 8M + 1 8M + 2 8M + 3 8M + 4 8M + 5 8M + 6 8M + 7
space 1

Bank 0 Bank 1 Bank 2 Bank 3

SPRU187U – July 2012 Using the Assembly Optimizer 127

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Avoiding Memory Bank Conflicts With the Assembly Optimizer www.ti.com

4.5.1 Preventing Memory Bank Conflicts

The assembly optimizer uses the assumptions that memory operations do not have bank conflicts. If it
determines that two memory operations have a bank conflict on any loop iteration it does not schedule the
operations in parallel. The assembly optimizer checks for memory bank conflicts only for those loops that
it is trying to software pipeline.
The information required for memory bank analysis indicates a base, an offset, a stride, a width, and an
iteration delta. The width is implicitly determined by the type of memory access (byte, halfword, word, or
double word for the C6400 and C6700). The iteration delta is determined by the assembly optimizer as it
constructs the schedule for the software pipeline. The base, offset, and stride are supplied by the load and
store instructions and/or by the .mptr directive.
An LD(B/BU)(H/HU)(W) or ST(B/H/W) operation in linear assembly can have memory bank information
associated with it implicitly, by using the .mptr directive. The .mptr directive associates a register with the
information that allows the assembly optimizer to determine automatically whether two memory operations
have a bank conflict. If the assembly optimizer determines that two memory operations have a memory
bank conflict, then it does not schedule them in parallel within a software pipelined loop. The syntax is:

.mptr variable , base + offset , stride

For example:
.mptr a_0,a+0,16
.mptr a_4,a+4,16
LDW *a_0++[4], val1 ; base=a, offset=0, stride=16
LDW *a_4++[4], val2 ; base=a, offset=4, stride=16
.mptr dptr,D+0,8
LDH *dptr++, d0 ; base=D, offset=0, stride=8
LDH *dptr++, d1 ; base=D, offset=2, stride=8
LDH *dptr++, d2 ; base=D, offset=4, stride=8
LDH *dptr++, d3 ; base=D, offset=6, stride=8

In this example, the offset for dptr is updated after every memory access. The offset is updated only when
the pointer is modified by a constant. This occurs for the pre/post increment/decrement addressing modes.
See the .mptr topic for more information.
Example 4-6 shows loads and stores extracted from a loop that is being software pipelined.

Example 4-6. Load and Store Instructions That Specify Memory Bank Information

.mptr Ain,IN,-16
.mptr Bin,IN-4,-16

.mptr Aco,COEF,16
.mptr Bco,COEF+4,16

.mptr Aout,optr+0,4
.mptr Bout,optr+2,4

LDW *Ain--[2],Ain12 ; IN(k-I) & IN(k-I+1)

LDW *Bin--[2],Bin23 ; IN(k-I-2) & IN(k-I-1)
LDW *Ain--[2],Ain34 ; IN(k-I-4) & IN(k-I-3)
LDW *Bin--[2],Bin56 ; IN(k-I-6) & IN(k-I-5)

LDW *Bco++[2],Bco12 ; COEF(I) & COEF(I+1)

LDW *Aco++[2],Aco23 ; COEF(I+2) & COEF(I+3)
LDW *Bco++[2],Bin34 ; COEF(I+4) & COEF(I+5)
LDW *Aco++[2],Ain56 ; COEF(I+6) & COEF(I+7)

STH Assum,Aout++[2] ; oPtr++ = (r >> 15)

STH Bssum,*Bout++[2] ; *oPtr++ = (I >> 15)

128 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Avoiding Memory Bank Conflicts With the Assembly Optimizer

4.5.2 A Dot Product Example That Avoids Memory Bank Conflicts

The C code in Example 4-7 implements a dot product function. The inner loop is unrolled once to take
advantage of the C6000's ability to operate on two 16-bit data items in a single 32-bit register. LDW
instructions are used to load two consecutive short values. The linear assembly instructions in Example 4-
8 implement the dotp loop kernel. Example 4-9 shows the loop kernel determined by the assembly
optimizer.
For this loop kernel, there are two restrictions associated with the arrays a[ ] and b[ ]:
• Because LDW is being used, the arrays must be aligned to start on word boundaries.
• To avoid a memory bank conflict, one array must start in bank 0 and the other array in bank 2. If they
start in the same bank, then a memory bank conflict occurs every cycle and the loop computes a result
every two cycles instead of every cycle, due to a memory bank stall. For example:
Bank conflict:
MVK 0, A0
|| MVK 8, B0
LDW *A0, A1
No bank conflict:
MVK 0, A0
|| MVK 4, B0
LDW *A0, A1
|| LDW *B0, B1

Example 4-7. C Code for Dot Product

int dot(short a[], short b[])

{
int sum0 = 0, sum1 = 0, sum, I;

for (I = 0; I < 100/2; I+= 2)

{
sum0 += a[i] * b[i];
sum1 += a[i + 1] * b[i + 1];
}
return sum0 + sum1;
}

SPRU187U – July 2012 Using the Assembly Optimizer 129

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Avoiding Memory Bank Conflicts With the Assembly Optimizer www.ti.com

Example 4-8. Linear Assembly for Dot Product

_dot: .cproc a, b
.reg sum0, sum1, I
.reg val1, val2, prod1, prod2

MVK 50,i ; I = 100/2

ZERO sum0 ; multiply result = 0
ZERO sum1 ; multiply result = 0

loop: .trip 50
LDW *a++,val1 ; load a[0-1] bank0
LDW *b++,val2 ; load b[0-1] bank2
MPY val1,val2,prod1 ; a[0] * b[0]
MPYH val1,val2,prod2 ; a[1] * b[1]
ADD prod1,sum0,sum0 ; sum0 += a[0] * b[0]
ADD prod2,sum1,sum1 ; sum1 += a[1] * b[1]

[I] ADD -1,i,i ; I--

[I] B loop ; if (!I) goto loop

ADD sum0,sum1,A4 ; compute final result

.return A4
.endproc

Example 4-9. Dot Product Software-Pipelined Kernel

L2: ; PIPED LOOP KERNEL

ADD .L2 B7,B4,B4 ; |14| <0,7> sum0 += a[0]*b[0]

|| ADD .L1 A5,A0,A0 ; |15| <0,7> sum1 += a[1]*b[1]
|| MPY .M2X B6,A4,B7 ; |12| <2,5> a[0] * b[0]
|| MPYH .M1X B6,A4,A5 ; |13| <2,5> a[1] * b[1]
|| [ B0] B .S1 L2 ; |18| <5,2> if (!I) goto loop
|| [ B0] ADD .S2 0xffffffff,B0,B0 ; |17| <6,1> I--
|| LDW .D2T2 *B5++,B6 ; |10| <7,0> load a[0-1] bank0
|| LDW .D1T1 *A3++,A4 ; |11| <7,0> load b[0-1] bank2
|| LDW *B0, B1

It is not always possible to control fully how arrays and other memory objects are aligned. This is
especially true when a pointer is passed into a function and that pointer may have different alignments
each time the function is called. A solution to this problem is to write a dot product routine that cannot
have memory hits. This would eliminate the need for the arrays to use different memory banks.
If the dot product loop kernel is unrolled once, then four LDW instructions execute in the loop kernel.
Assuming that nothing is known about the bank alignment of arrays a and b (except that they are word
aligned), the only safe assumptions that can be made about the array accesses are that a[0-1] cannot
conflict with a[2-3] and that b[0-1] cannot conflict with b[2-3]. Example 4-10 shows the unrolled loop
kernel.

130 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Avoiding Memory Bank Conflicts With the Assembly Optimizer

Example 4-10. Dot Product From Example 4-8 Unrolled to Prevent Memory Bank Conflicts

_dotp2: .cproc a_0, b_0

.reg a_4, b_4, sum0, sum1, I
.reg val1, val2, prod1, prod2

ADD 4,a_0,a_4
ADD 4,b_0,b_4
MVK 25,i ; I = 100/4
ZERO sum0 ; multiply result = 0
ZERO sum1 ; multiply result = 0

.mptr a_0,a+0,8
.mptr a_4,a+4,8
.mptr b_0,b+0,8
.mptr b_4,b+4,8

loop: .trip 25

LDW *a_0++[2],val1 ; load a[0-1] bankx

LDW *b_0++[2],val2 ; load b[0-1] banky
MPY val1,val2,prod1 ; a[0] * b[0]
MPYH val1,val2,prod2 ; a[1] * b[1]
ADD prod1,sum0,sum0 ; sum0 += a[0] * b[0]
ADD prod2,sum1,sum1 ; sum1 += a[1] * b[1]

LDW *a_4++[2],val1 ; load a[2-3] bankx+2

LDW *b_4++[2],val2 ; load b[2-3] banky+2
MPY val1,val2,prod1 ; a[2] * b[2]
MPYH val1,val2,prod2 ; a[3] * b[3]
ADD prod1,sum0,sum0 ; sum0 += a[2] * b[2]
ADD prod2,sum1,sum1 ; sum1 += a[3] * b[3]

[I] ADD -1,i,i ; I--

[I] B loop ; if (!0) goto loop

ADD sum0,sum1,A4 ; compute final result

.return A4
.endproc

SPRU187U – July 2012 Using the Assembly Optimizer 131

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Avoiding Memory Bank Conflicts With the Assembly Optimizer www.ti.com

The goal is to find a software pipeline in which the following instructions are in parallel:
LDW *a0++[2],val1 ; load a[0-1] bankx
|| LDW *a2++[2],val2 ; load a[2-3] bankx+2
LDW *b0++[2],val1 ; load b[0-1] banky
|| LDW *b2++[2],val2 ; load b[2-3] banky+2

Example 4-11. Unrolled Dot Product Kernel From Example 4-9

L2: ; PIPED LOOP KERNEL

[ B1] SUB .S2 B1,1,B1 ; <0,8>

|| ADD .L2 B9,B5,B9 ; |21| <0,8> ^ sum0 += a[0] * b[0]
|| ADD .L1 A6,A0,A0 ; |22| <0,8> ^ sum1 += a[1] * b[1]
|| MPY .M2X B8,A4,B9 ; |19| <1,6> a[0] * b[0]
|| MPYH .M1X B8,A4,A6 ; |20| <1,6> a[1] * b[1]
|| [ B0] B .S1 L2 ; |32| <2,4> if (!I) goto loop
|| [ B1] LDW .D1T1 *A3++(8),A4 ; |24| <3,2> load a[2-3] bankx+2
|| [ A1] LDW .D2T2 *B6++(8),B8 ; |17| <4,0> load a[0-1] bankx

[ A1] SUB .S1 A1,1,A1 ; <0,9>

|| ADD .L2 B5,B9,B5 ; |28| <0,9> ^ sum0 += a[2] * b[2]
|| ADD .L1 A6,A0,A0 ; |29| <0,9> ^ sum1 += a[3] * b[3]
|| MPY .M2X A4,B7,B5 ; |26| <1,7> a[2] * b[2]
|| MPYH .M1X A4,B7,A6 ; |27| <1,7> a[3] * b[3]
|| [ B0] ADD .S2 -1,B0,B0 ; |31| <3,3> I--
|| [ A1] LDW .D2T2 *B4++(8),B7 ; |25| <4,1> load b[2-3] banky+2
|| [ A1] LDW .D1T1 *A5++(8),A4 ; |18| <4,1> load b[0-1] banky

Without the .mptr directives in Example 4-10, the loads of a[0-1] and b[0-1] are scheduled in parallel, and
the loads of a[2-3] and b[2-3] might be scheduled in parallel. This results in a 50% chance that a memory
conflict will occur on every cycle. However, the loop kernel shown in Example 4-11 can never have a
memory bank conflict.
In Example 4-8, if .mptr directives had been used to specify that a and b point to different bases, then the
assembly optimizer would never find a schedule for a 1-cycle loop kernel, because there would always be
a memory bank conflict. However, it would find a schedule for a 2-cycle loop kernel.

4.5.3 Memory Bank Conflicts for Indexed Pointers

When determining memory bank conflicts for indexed memory accesses, it is sometimes necessary to
specify that a pair of memory accesses always conflict, or that they never conflict. This can be
accomplished by using the .mptr directive with a stride of 0.
A stride of 0 indicates that there is a constant relation between the memory accesses regardless of the
iteration delta. Essentially, only the base, offset, and width are used by the assembly optimizer to
determine a memory bank conflict. Recall that the stride is optional and defaults to 0.
In Example 4-12, the .mptr directive is used to specify which memory accesses conflict and which never
conflict.

Example 4-12. Using .mptr for Indexed Pointers

.mptr a,RS
.mptr b,RS
.mptr c,XY
.mptr d,XY+2
LDW *a++[i0a],A0 ; a and b always conflict with each other
LDW *b++[i0b],B0 ;
STH A1,*c++[i1a] ; c and d never conflict with each other
STH B2,*d++[i1b] ;

132 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Memory Alias Disambiguation

4.5.4 Memory Bank Conflict Algorithm

The assembly optimizer uses the following process to determine if two memory access instructions might
have a memory bank conflict:
1. If either access does not have memory bank information, then they do not conflict.
2. If both accesses do not have the same base, then they conflict.
3. The offset, stride, access width, and iteration delta are used to determine if a memory bank conflict will
occur. The assembly optimizer uses a straightforward analysis of the access patterns and determines if
they ever access the same relative bank. The stride and offset values are always expressed in bytes.
The iteration delta is the difference in the loop iterations of the memory references being scheduled in
the software pipeline. For example, given three instructions A, B, C and a software pipeline with a
single-cycle kernel, then A and C have an iteration delta of 2:
A
B A
C B A
C B
C

4.6 Memory Alias Disambiguation

Memory aliasing occurs when two instructions can access the same memory location. Such memory
references are called ambiguous. Memory alias disambiguation is the process of determining when such
ambiguity is not possible. When you cannot determine whether two memory references are ambiguous,
you presume they are ambiguous. This is the same as saying the two instructions have a memory
dependence between them.
Dependencies between instructions constrain the instruction schedule, including the software pipeline
schedule. In general, the fewer the Dependencies, the greater freedom you have in choosing a schedule
and the better the final schedule performs.

4.6.1 How the Assembly Optimizer Handles Memory References (Default)

The assembly optimizer assumes memory references are aliased, unless it can prove otherwise.
Because alias analysis is very limited in the assembly optimizer, this presumption is often overly
conservative. In such cases, the extra instruction Dependencies, due to the presumed memory aliases,
can cause the assembly optimizer to emit instruction schedules that have less parallelism and do not
perform well. To handle these cases, the assembly optimizer provides one option and two directives.

4.6.2 Using the --no_bad_aliases Option to Handle Memory References

In the assembly optimizer, the --no_bad_aliases option means no memory references ever depend on
each other. The --no_bad_aliases option does not mean the same thing to the C/C++ compiler. The
C/C++ compiler interprets the --no_bad_aliases switch to indicate several specific cases of memory
aliasing are guaranteed not to occur. For more information about using the --no_bad_aliases option, see
Section 3.10.2.

4.6.3 Using the .no_mdep Directive

You can specify the .no_mdep directive anywhere in a .(c)proc function. Whenever it is used, you
guarantee that no memory Dependencies occur within that function.

Memory Dependency Exception

NOTE: For both of these methods, --no_bad_aliases and .no_mdep, the assembly optimizer
recognizes any memory Dependencies you point out with the .mdep directive.

SPRU187U – July 2012 Using the Assembly Optimizer 133

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Memory Alias Disambiguation www.ti.com

4.6.4 Using the .mdep Directive to Identify Specific Memory Dependencies

You can use the .mdep directive to identify specific memory Dependencies by annotating each memory
reference with a name, and using those names with the .mdep directive to indicate the actual
dependence. Annotating a memory reference requires adding information right next to the memory
reference in the assembly stream. Include the following immediately after a memory reference:
{ memref }
The memref has the same syntax restrictions as any assembly symbol. (For more information about
symbols, refer to the TMS320C6000 Assembly Language Tools User's Guide.) It is in the same name
space as the symbolic registers. You cannot use the same name for a symbolic register and annotating a
memory reference.

Example 4-13. Annotating a Memory Reference

LDW *p1++ {ld1}, inp1 ;name memory reference "ld1"

;other code ...
STW outp2, *p2++ {st1} ;name memory reference "st1"

*<The directive to indicate...:

.mdep ld1, st1 <<bold>>

The directive to indicate a specific memory dependence in the previous example is as follows:
.mdep ld1, st1

This means that whenever ld1 accesses memory at location X, some later time in code execution, st1 may
also access location X. This is equivalent to adding a dependence between these two instructions. In
terms of the software pipeline, these two instructions must remain in the same order. The ld1 reference
must always occur before the st1 reference; the instructions cannot even be scheduled in parallel.
It is important to note the directional sense of the directive from ld1 to st1. The opposite, from st1 to ld1, is
not implied. In terms of the software pipeline, while every ld1 must occur before every st1, it is still legal to
schedule the ld1 from iteration n+1 before the st1 from iteration n.
Example 4-14 is a picture of the software pipeline with the instructions from two different iterations in
different columns. In the actual instruction sequence, instructions on the same horizontal line are in
parallel.

Example 4-14. Software Pipeline Using .mdep ld1, st1

iteration n iteration n+1

------------- -------------
LDW { ld1 }

... LDW { ld1 }

STW { st1 } ...

STW { st1 }

*<If that schedule...>

.mdep st1, ld1

If that schedule does not work because the iteration n st1 might write a value the iteration n+1 ld1 should
read, then you must note a dependence relationship from st1 to ld1.
.mdep st1, ld1

Both directives together force the software pipeline shown in Example 4-15.

134 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Memory Alias Disambiguation

Example 4-15. Software Pipeline Using .mdep st1, ld1 and .mdep ld1, st1

iteration n iteration n+1

------------- -------------
LDW { ld1 }

...

STW { st1 }

LDW { ld1 }

...

STW { st1 }

.mdep ld1, st1

.mdep st1, ld1

Indexed addressing, *+base[index], is a good example of an addressing mode where you typically do not
know anything about the relative sequence of the memory accesses, except they sometimes access the
same location. To correctly model this case, you need to note the dependence relation in both directions,
and you need to use both directives.
.mdep ld1, st1 .mdep st1, ld1

4.6.5 Memory Alias Examples

Following are memory alias examples that use the .mdep and .no_mdep directives.
• Example 1
The .mdep r1, r2 directive declares that LDW must be before STW. In this case, src and dst might
point to the same array.
fn: .cproc dst, src, cnt
.reg tmp
.no_mdep
.mdep r1, r2

LDW *src{r1}, tmp

STW cnt, *dst{r2}

.return tmp
.endproc
• Example 2
Here, .mdep r2, r1 indicates that STW must occur before LDW. Since STW is after LDW in the code,
the dependence relation is across loop iterations. The STW instruction writes a value that may be read
by the LDW instruction on the next iteration. In this case, a 6-cycle recurrence is created.
fn: .cproc dst, src, cnt
.reg tmp
.no_mdep
.mdep r2, r1

LOOP: .trip 100

LDW *src++{r1}, tmp
STW tmp, *dst++{r2}
[cnt] SUB cnt, 1, cnt
[cnt] B LOOP

.endproc

SPRU187U – July 2012 Using the Assembly Optimizer 135

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Memory Alias Disambiguation www.ti.com

Memory Dependence/Bank Conflict

NOTE: Do not confuse memory alias disambiguation with the handling of memory bank conflicts.
These may seem similar because they each deal with memory references and the effect of
those memory references on the instruction schedule. Alias disambiguation is a correctness
issue, bank conflicts are a performance issue. A memory dependence has a much broader
impact on the instruction schedule than a bank conflict. It is best to keep these two topics
separate.

Volatile References
NOTE: For volatile references, use .volatile rather than .mdep.

136 Using the Assembly Optimizer SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 5
SPRU187U – July 2012

Linking C/C++ Code

The C/C++ compiler and assembly language tools provide two methods for linking your programs:
• You can compile individual modules and link them together. This method is especially useful when you
have multiple source files.
• You can compile and link in one step. This method is useful when you have a single source module.
This chapter describes how to invoke the linker with each method. It also discusses special requirements
of linking C/C++ code, including the run-time-support libraries, specifying the type of initialization, and
allocating the program into memory. For a complete description of the linker, see the TMS320C6000
Assembly Language Tools User's Guide.

Topic ........................................................................................................................... Page

5.1 Invoking the Linker Through the Compiler (-z Option) .......................................... 138
5.2 Linker Code Optimizations ................................................................................ 140
5.3 Controlling the Linking Process ........................................................................ 141

SPRU187U – July 2012 Linking C/C++ Code 137

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Invoking the Linker Through the Compiler (-z Option) www.ti.com

5.1 Invoking the Linker Through the Compiler (-z Option)

This section explains how to invoke the linker after you have compiled and assembled your programs: as
a separate step or as part of the compile step.

5.1.1 Invoking the Linker Separately

This is the general syntax for linking C/C++ programs as a separate step:

cl6x --run_linker {--rom_model | --ram_model} filenames

[options] [--output_file= name.out] --library= library [lnk.cmd]

cl6x --run_linker The command that invokes the linker.

--rom_model | --ram_model Options that tell the linker to use special conventions defined by the
C/C++ environment. When you use cl6x --run_linker, you must use --
rom_model or --ram_model. The --rom_model option uses
automatic variable initialization at run time; the --ram_model option
uses variable initialization at load time.
filenames Names of object files, linker command files, or archive libraries. The
default extension for all input files is .obj; any other extension must be
explicitly specified. The linker can determine whether the input file is
an object or ASCII file that contains linker commands. The default
output filename is a.out, unless you use the --output_file option to
name the output file.
options Options affect how the linker handles your object files. Linker options
can only appear after the --run_linker option on the command line,
but otherwise may be in any order. (Options are discussed in detail in
the TMS320C6000 Assembly Language Tools User's Guide.)
--output_file= name.out Names the output file.
--library= library Identifies the appropriate archive library containing C/C++ run-time-
support and floating-point math functions, or linker command files. If
you are linking C/C++ code, you must use a run-time-support library.
You can use the libraries included with the compiler, or you can
create your own run-time-support library. If you have specified a run-
time-support library in a linker command file, you do not need this
parameter. The --library option's short form is -l.
lnk.cmd Contains options, filenames, directives, or commands for the linker.

When you specify a library as linker input, the linker includes and links only those library members that
resolve undefined references. The linker uses a default allocation algorithm to allocate your program into
memory. You can use the MEMORY and SECTIONS directives in the linker command file to customize
the allocation process. For information, see the TMS320C6000 Assembly Language Tools User's Guide.
You can link a C/C++ program consisting of object files prog1.obj, prog2.obj, and prog3.obj, with an
executable object file filename of prog.out with the command:
cl6x --run_linker --rom_model prog1 prog2 prog3 --output_file=prog.out
--library=rts6200.lib

138 Linking C/C++ Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Invoking the Linker Through the Compiler (-z Option)

5.1.2 Invoking the Linker as Part of the Compile Step

This is the general syntax for linking C/C++ programs as part of the compile step:

cl6xfilenames [options] --run_linker {--rom_model | --ram_model} filenames

[options] [--output_file= name.out] --library= library [lnk.cmd]

The --run_linker option divides the command line into the compiler options (the options before --
run_linker) and the linker options (the options following --run_linker). The --run_linker option must follow all
source files and compiler options on the command line.
All arguments that follow --run_linker on the command line are passed to the linker. These arguments can
be linker command files, additional object files, linker options, or libraries. These arguments are the same
as described in Section 5.1.1.
All arguments that precede --run_linker on the command line are compiler arguments. These arguments
can be C/C++ source files, assembly files, linear assembly files, or compiler options. These arguments are
described in Section 2.2.
You can compile and link a C/C++ program consisting of object files prog1.c, prog2.c, and prog3.c, with an
executable object file filename of prog.out with the command:
cl6x prog1.c prog2.c prog3.c --run_linker --rom_model --output_file=prog.out --library=rts6200.lib

NOTE: Order of Processing Arguments in the Linker

The order in which the linker processes arguments is important. The compiler passes
arguments to the linker in the following order:
1. Object filenames from the command line
2. Arguments following the --run_linker option on the command line
3. Arguments following the --run_linker option from the C6X_C_OPTION environment
variable

5.1.3 Disabling the Linker (--compile_only Compiler Option)

You can override the --run_linker option by using the --compile_only compiler option. The -run_linker
option's short form is -z and the --compile_only option's short form is -c.
The --compile_only option is especially helpful if you specify the --run_linker option in the C6X_C_OPTION
environment variable and want to selectively disable linking with the --compile_only option on the
command line.

SPRU187U – July 2012 Linking C/C++ Code 139

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Linker Code Optimizations www.ti.com

5.2 Linker Code Optimizations

5.2.1 Generating Function Subsections (--gen_func_subsections Compiler Option)

When the linker places code into an executable file, it allocates all the functions in a single source file as a
group. This means that if any function in a file needs to be linked into an executable, then all the functions
in the file are linked in. This can be undesirable if a file contains many functions and only a few are
required for an executable.
This situation may exist in libraries where a single file contains multiple functions, but the application only
needs a subset of those functions. An example is a library .obj file that contains a signed divide routine
and an unsigned divide routine. If the application requires only signed division, then only the signed divide
routine is required for linking. By default, both the signed and unsigned routines are linked in since they
exist in the same .obj file.
The --gen_func_subsections compiler option remedies this problem by placing each function in a file in its
own subsection. Thus, only the functions that are referenced in the application are linked into the final
executable. This can result in an overall code size reduction.
However, be aware that using the --gen_func_subsections compiler option can result in overall code size
growth if all or nearly all functions are being referenced. This is because any section containing code must
be aligned to a 32-byte boundary to support the C6000 branching mechanism. When the --
gen_func_subsections option is not used, all functions in a source file are usually placed in a common
section which is aligned. When --gen_func_subsections is used, each function defined in a source file is
placed in a unique section. Each of the unique sections requires alignment. If all the functions in the file
are required for linking, code size may increase due to the additional alignment padding for the individual
subsections.
Thus, the --gen_func_subsections compiler option is advantageous for use with libraries where normally
only a limited number of the functions in a file are used in any one executable.
The alternative to the --gen_func_subsections option is to place each function in its own file.

5.2.2 Conditional Linking

The conditional linking paradigm is different under COFF compared to ELF. In COFF, you must mark a
section with the .clink directive to make it eligible for removal during conditional linking. In ELF, all sections
are considered eligible for removal through conditional linking. Sections are not removed if they are
referenced or if they are marked with the .retain directive.
Under COFF, when you compile with the -gen_func_subsections option, in addition to placing each
function in a separate subsection, the compiler also annotates that subsection with the conditional linking
directive, .clink. This directive marks the section as a candidate to be removed if it is not referenced by
any other section in the program. The compiler does not place a .clink directive in a subsection for a trap
or interrupt function, as these may be needed by a program even though there is no symbolic reference to
them anywhere in the program.
Under COFF, if a section that has been marked for conditional linking is never referenced by any other
section in the program, that section is removed from the program. Under ELF, a section that is never
referenced by any other section in the program is removed from the program automatically, unless it is
marked with .retain. Conditional linking is disabled when performing a partial link or when relocation
information is kept with the output of the link. Conditional linking can also be disabled with the --
disable_clink linker option.

140 Linking C/C++ Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Controlling the Linking Process

5.3 Controlling the Linking Process

Regardless of the method you choose for invoking the linker, special requirements apply when linking
C/C++ programs. You must:
• Include the compiler's run-time-support library
• Specify the type of boot-time initialization
• Determine how you want to allocate your program into memory
This section discusses how these factors are controlled and provides an example of the standard default
linker command file.
For more information about how to operate the linker, see the linker description in the TMS320C6000
Assembly Language Tools User's Guide

5.3.1 Including the Run-Time-Support Library

You must link all C/C++ programs with a run-time-support library. The library contains standard C/C++
functions as well as functions used by the compiler to manage the C/C++ environment. The following
sections describe two methods for including the run-time-support library.

5.3.1.1 Automatic Run-Time-Support Library Selection

If the --rom_model or --ram_model option is specified during the linker and the entry point for the program
(normally c_int00) is not resolved by any specified object file or library, the linker attempts to automatically
include the best compatible run-time-support library for your program. The chosen run-time-support library
is linked in after any other libraries specified with the --library option on the command line. Alternatively,
you can force the linker to choose an appropriate run-time-support library by specifying “libc.a” as an
argument to the --library option, or when specifying the run-time-support library name explicitly in a linker
command file.
The automatic selection of a run-time-support library can be disabled with the --disable_auto_rts option.
If the --issue_remarks option is specified before the --run_linker option during the linker, a remark is
generated indicating which run-time support library was linked in. If a different run-time-support library is
desired, you must specify the name of the desired run-time-support library using the --library option and in
your linker command files when necessary.

Example 5-1. Using the --issue_remarks Option

cl6x --silicon_version=6400+ --issue_remarks main.c --run_linker --rom_model

remark: linking in "libc.a"

remark: linking in "rts64plus.lib" in place of "libc.a"

5.3.1.2 Manual Run-Time-Support Library Selection

You should use the --library linker option to specify which C6000 run-time-support library to use. The --
library option also tells the linker to look at the --search_path options and then the C6X_C_DIR
environment variable to find an archive path or object file. To use the --library linker option, type on the
command line:

cl6x --run_linker {--rom_model | --ram_model} filenames --library= libraryname

SPRU187U – July 2012 Linking C/C++ Code 141

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling the Linking Process www.ti.com

5.3.1.3 Library Order for Searching for Symbols

Generally, you should specify the run-time-support library as the last name on the command line because
the linker searches libraries for unresolved references in the order that files are specified on the command
line. If any object files follow a library, references from those object files to that library are not resolved.
You can use the --reread_libs option to force the linker to reread all libraries until references are resolved.
Whenever you specify a library as linker input, the linker includes and links only those library members
that resolve undefined references.
By default, if a library introduces an unresolved reference and multiple libraries have a definition for it, then
the definition from the same library that introduced the unresolved reference is used. Use the --priority
option if you want the linker to use the definition from the first library on the command line that contains
the definition.

5.3.2 Run-Time Initialization

C/C++ programs require initialization of the run-time environment before execution of the program itself
may begin. This initialization is performed by a bootstrap routine. This routine is responsible for creating
the stack, initializing global variables, and calling the function main. The bootstrap routine should be the
entry point for the program, and it typically should be the RESET interrupt handler. The bootstrap routine
is responsible for the following tasks:
1. Set up the stack by initializing SP
2. Set up the data page pointer DP (for architectures that have one)
3. Set configuration registers
4. Process the .cinit table to autoinitialize global variables (when using the --rom_model option)
5. Process the .pinit table to construct global C++ objects.
6. Call the function main with appropriate arguments
7. Call exit when main returns
When you compile a C program and use --rom_model or --ram_model, the linker looks for a bootstrap
routine named _c_int00. The run-time support library provides a sample _c_int00 in boot.obj, which
performs the required tasks. If you use the run-time support's bootstrap routine, you should set _c_int00
as the entry point.
A sample bootstrap routine is _c_int00, provided in boot.obj in the run-time support object libraries. The
entry point is usually set to the starting address of the bootstrap routine.

NOTE: The _c_int00 Symbol

If you use the --ram_model or --rom_model link option, _c_int00 is automatically defined as
the entry point for the program.

5.3.3 Global Object Constructors

Global C++ variables that have constructors and destructors require their constructors to be called during
program initialization and their destructors to be called during program termination. The C++ compiler
produces a table of constructors to be called at startup.
Constructors for global objects from a single module are invoked in the order declared in the source code,
but the relative order of objects from different object files is unspecified.
Global constructors are called after initialization of other global variables and before the function main is
called. Global destructors are invoked during the function exit, similar to functions registered through
atexit.
Section 7.8.1.3 discusses the format of the global constructor table for COFFABI mode and
Section 7.8.1.3 for EABI mode..

142 Linking C/C++ Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Controlling the Linking Process

5.3.4 Specifying the Type of Global Variable Initialization

The C/C++ compiler produces data tables for initializing global variables. Section 7.8.3 discusses the
format of these initialization tables for COFFABI. Section 7.8.2.4 discusses the format of these initialization
tables for EABI. The initialization tables are used in one of the following ways:
• Global variables are initialized at run time. Use the --rom_model linker option (see Section 7.8.1.1).
• Global variables are initialized at load time. Use the --ram_model linker option (see Section 7.8.1.2).
When you link a C/C++ program, you must use either the --rom_model or --ram_model option. These
options tell the linker to select initialization at run time or load time.
When you compile and link programs, the --rom_model option is the default. If used, the --rom_model
option must follow the --run_linker option (see Section 5.1). The following list outlines the linking
conventions for COFFABI used with --rom_model or --ram_model:
• The symbol _c_int00 is defined as the program entry point; it identifies the beginning of the C/C++ boot
routine in boot.obj. When you use --rom_model or --ram_model, _c_int00 is automatically referenced,
ensuring that boot.obj is automatically linked in from the run-time-support library.
• The initialization output section is padded with a termination record so that the loader (load-time
initialization) or the boot routine (run-time initialization) knows when to stop reading the initialization
tables.
• When initializing at load time (the --ram_model option), the following occur:
– The linker sets the initialization table symbol to -1. This indicates that the initialization tables are not
in memory, so no initialization is performed at run time.
– The STYP_COPY flag is set in the initialization table section header. STYP_COPY is the special
attribute that tells the loader to perform autoinitialization directly and not to load the initialization
table into memory. The linker does not allocate space in memory for the initialization table.
• When autoinitializing at run time (--rom_model option), the linker defines the initialization table symbol
as the starting address of the initialization table. The boot routine uses this symbol as the starting point
for autoinitialization.
For details on linking conventions for EABI used with --rom_model and --ram_model, see Section 7.8.2.3
and Section 7.8.2.5, respectively.

5.3.5 Specifying Where to Allocate Sections in Memory

The compiler produces relocatable blocks of code and data. These blocks, called sections, are allocated
in memory in a variety of ways to conform to a variety of system configurations.
The compiler creates two basic kinds of sections: initialized and uninitialized. Table 5-1 summarizes the
initialized sections created under the COFF ABI mode. Table 5-2 summarizes the initialized sections
created under the EABI mode. Table 5-3 summarizes the uninitialized sections. Be aware that the COFF
ABI .cinit and .pinit (.init_array in EABI) tables have different formats in EABI.

Table 5-1. Initialized Sections Created by the Compiler for COFFABI

Name Contents
.args Command argument for host-based loader; read-only (see the --arg_size option)
.cinit Tables for explicitly initialized global and static variables
.const Global and static const variables that are explicitly initialized and contain string
literals
.pinit Table of constructors to be called at startup
.ppdata Data tables for compiler-based profiling (see the --gen_profile_info option)
.ppinfo Correlation tables for compiler-based profiling ( see the --gen_profile_info option)
.switch Jump tables for large switch statements
.text Executable code and constants

SPRU187U – July 2012 Linking C/C++ Code 143

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Controlling the Linking Process www.ti.com

Table 5-2. Initialized Sections Created by the Compiler for EABI

Name Contents
.args Command argument for host-based loader; read-only (see the --arg_size option)
.binit Boot time copy tables (See the TMS320C6000 Assembly Language Tools User's
Guide for information on BINIT in linker command files.)
.cinit In EABI mode, the compiler does not generate a .cinit section. However, when the --
rom_mode linker option is specified, the linker creates this section, which contains
tables for explicitly initialized global and static variables.
.const Far, const global and static variables, and string constants
.c6xabi.exidx Index table for exception handling; read-only (see --exceptions option)
.c6xabi.extab Unwinded instructions for exception handling; read-only (see --exceptions option)
.fardata Far non-const global and static variables that are explicitly initialized
.init_array Table of constructors to be called at startup
.name.load Compressed image of section name; read-only (See the TMS320C6000 Assembly
Language Tools User's Guide for information on copy tables.)
.neardata Near non-const global and static variables that are explicitly initialized
.ppdata Data tables for compiler-based profiling (see the --gen_profile_info option)
.ppinfo Correlation tables for compiler-based profiling (see the --gen_profile_info option)
.rodata Global and static variables that have near and const qualifiers
.switch Jump tables for large switch statements
.text Executable code and constants

Table 5-3. Uninitialized Sections Created by the Compiler for Both ABIs
Name Contents
.bss Global and static variables
.far Global and static variables declared far
.stack Stack
.sysmem Memory for malloc functions (heap)

When you link your program, you must specify where to allocate the sections in memory. In general,
initialized sections are linked into ROM or RAM; uninitialized sections are linked into RAM. With the
exception of code sections, the initialized and uninitialized sections created by the compiler cannot be
allocated into internal program memory. See Section 7.1.1 for a complete description of how the compiler
uses these sections.
The linker provides MEMORY and SECTIONS directives for allocating sections. For more information
about allocating sections into memory, see the TMS320C6000 Assembly Language Tools User's Guide.

144 Linking C/C++ Code SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Controlling the Linking Process

5.3.6 A Sample Linker Command File

Example 5-2 shows a typical linker command file that links a C program. The command file in this
example is named lnk.cmd and lists several linker options:

--rom_model Tells the linker to use autoinitialization at run time.

--heap_size Tells the linker to set the C heap size at 0x2000 bytes.
--stack_size Tells the linker to set the stack size to 0x0100 bytes.
--library Tells the linker to use an archive library file, rts6200.lib, for input.

To link the program, use the following syntax:

cl6x --run_linker object_file(s) --output_file= outfile --map_file= mapfile lnk.cmd

The MEMORY and possibly the SECTIONS directives, might require modification to work with your
system. See the TMS320C6000 Assembly Language Tools User's Guide for more information on these
directives.

Example 5-2. Linker Command File

--rom_model
--heap_size=0x2000
--stack_size=0x0100
--library=rts6200.lib

MEMORY
{
VECS: o = 0x00000000 l = 0x000000400 /* reset & interrupt vectors */
PMEM: o = 0x00000400 l = 0x00000FC00 /* intended for initialization */
BMEM: o = 0x80000000 l = 0x000010000 /* .bss, .sysmem, .stack, .cinit */
}

SECTIONS
{
vectors > VECS
.text > PMEM
.data > BMEM
.stack > BMEM
.bss > BMEM
.sysmem > BMEM
.cinit > BMEM
.const > BMEM
.cio > BMEM
.far > BMEM
}

SPRU187U – July 2012 Linking C/C++ Code 145

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 6
SPRU187U – July 2012

TMS320C6000C/C++ Language Implementation

The C/C++ compiler supports the C/C++ language standard that was developed by a committee of the
American National Standards Institute (ANSI) and subsequently adopted by the International Standards
Organization (IS0).
The C++ language supported by the C6000 is defined by the ANSI/ISO/IEC 14882:1998 standard with
certain exceptions.

Topic ........................................................................................................................... Page

6.1 Characteristics of TMS320C6000 C .................................................................... 147

6.2 Characteristics of TMS320C6000 C++ ................................................................. 147
6.3 Using MISRA-C:2004 ........................................................................................ 148
6.4 Data Types ...................................................................................................... 149
6.5 Keywords ....................................................................................................... 150
6.6 C++ Exception Handling ................................................................................... 155
6.7 Register Variables and Parameters .................................................................... 155
6.8 The asm Statement .......................................................................................... 156
6.9 Pragma Directives ........................................................................................... 157
6.10 The _Pragma Operator ..................................................................................... 173
6.11 Application Binary Interface .............................................................................. 174
6.12 Object File Symbol Naming Conventions (Linknames) ......................................... 174
6.13 Initializing Static and Global Variables in COFF ABI Mode .................................... 175
6.14 Changing the ANSI/ISO C Language Mode .......................................................... 176
6.15 GNU Language Extensions ............................................................................... 178

146 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Characteristics of TMS320C6000 C

6.1 Characteristics of TMS320C6000 C

The compiler supports the C language as defined by ISO/IEC 9899:1990, which is equivalent to American
National Standard for Information Systems-Programming Language C X3.159-1989 standard, commonly
referred to as C89, published by the American National Standards Institute. The compiler can also accept
many of the language extensions found in the GNU C compiler (see Section 6.15). The compiler does not
support C99.
The ANSI/ISO standard identifies some features of the C language that are affected by characteristics of
the target processor, run-time environment, or host environment. For reasons of efficiency or practicality,
this set of features can differ among standard compilers.
Unsupported features of the C library are:
• The run-time library has minimal support for wide and multi-byte characters. The type wchar_t is
implemented as int. The wide character set is equivalent to the set of values of type char. The library
includes the header files <wchar.h> and <wctype.h>, but does not include all the functions specified in
the standard. So-called multi-byte characters are limited to single characters. There are no shift states.
The mapping between multi-byte characters and wide characters is simple equivalence; that is, each
wide character maps to and from exactly a single multi-byte character having the same value.
• The run-time library includes the header file <locale.h>, but with a minimal implementation. The only
supported locale is the C locale. That is, library behavior that is specified to vary by locale is hard-
coded to the behavior of the C locale, and attempting to install a different locale by way of a call to
setlocale() will return NULL.

6.2 Characteristics of TMS320C6000 C++

The C6000 compiler supports C++ as defined in the ANSI/ISO/IEC 14882:1998 standard, including these
features:
• Complete C++ standard library support, with exceptions noted below.
• Templates
• Exceptions, which are enabled with the --exceptions option; see Section 6.6.
• Run-time type information (RTTI), which can be enabled with the --rtti compiler option.
The exceptions to the standard are as follows:
• The compiler does not support embedded C++ run-time-support libraries.
• The library supports wide chars (wchar_t), in that template functions and classes that are defined for
char are also available for wchar_t. For example, wide char stream classes wios, wiostream,
wstreambuf and so on (corresponding to char classes ios, iostream, streambuf) are implemented.
However, there is no low-level file I/O for wide chars. Also, the C library interface to wide char support
(through the C++ headers <cwchar> and <cwctype>) is limited as described above in the C library.
• For COFF ABI only: If the definition of an inline function contains a static variable, and it appears in
multiple compilation units (usually because it’s a member function of a class defined in a header file),
the compiler generates multiple copies of the static variable rather than resolving them to a single
definition. The compiler emits a warning (#1369) in such cases.
• No support for bad_cast or bad_type_id is included in the typeinfo header.
• Two-phase name binding in templates, as described in [tesp.res] and [temp.dep] of the standard, is not
implemented.
• The export keyword for templates is not implemented.
• A typedef of a function type cannot include member function cv-qualifiers.
• A partial specialization of a class member template cannot be added outside of the class definition.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 147

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Using MISRA-C:2004 www.ti.com

6.3 Using MISRA-C:2004

You can alter your code to work with the MISRA-C:2004 rules. The following enable/disable the rules:
• The --check_misra option enables checking of the specified MISRA-C:2004 rules.
• The CHECK_MISRA pragma enables/disables MISRA-C:2004 rules at the source level. This pragma is
equivalent to using the --check_misra option. See Section 6.9.1.
• RESET_MISRA pragma resets the specified MISRA-C:2004 rules to the state they were before any
CHECK_MISRA pragmas were processed. See Section 6.9.25.
The syntax of the option and pragmas is:

--check_misra={all|required|advisory|none|rulespec}
#pragma CHECK_MISRA ("{all|required|advisory|none|rulespec}");
#pragma RESET_MISRA ("{all|required|advisory|rulespec}");

The rulespec parameter is a comma-separated list of these specifiers:

[-]X Enable (or disable) all rules in topic X.

[-]X-Z Enable (or disable) all rules in topics X through Z.
[-]X.A Enable (or disable) rule A in topic X.
[-]X.A-C Enable (or disable) rules A through C in topic X.

Example: --check_misra=1-5,-1.1,7.2-4
• Checks topics 1 through 5
• Disables rule 1.1 (all other rules from topic 1 remain enabled)
• Checks rules 2 through 4 in topic 7
Two options control the severity of certain MISRA-C:2004 rules:
• The --misra_required option sets the diagnostic severity for required MISRA-C:2004 rules.
• The --misra_advisory option sets the diagnostic severity for advisory MISRA-C:2004 rules.
The syntax for these options is:

--misra_advisory={error|warning|remark|suppress}
--misra_required={error|warning|remark|suppress}

148 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Data Types

6.4 Data Types

Table 6-1 lists the size, representation, and range of each scalar data type for the C6000 compiler for
COFF ABI. See Table 6-2 for the EABI data types. Many of the range values are available as standard
macros in the header file limits.h.

Table 6-1. TMS320C6000 C/C++ COFF ABI Data Types

Range
Type Size Representation Minimum Maximum
char, signed char 8 bits ASCII -128 127
unsigned char 8 bits ASCII 0 255
short 16 bits 2s complement -32 768 32 767
unsigned short 16 bits Binary 0 65 535
int, signed int 32 bits 2s complement -2 147 483 648 2 147 483 647
unsigned int 32 bits Binary 0 4 294 967 295
long, signed long 40 bits 2s complement -549 755 813 888 549 755 813 887
unsigned long 40 bits Binary 0 1 099 511 627 775
__int40_t 40 bits 2s complement -549 755 813 888 549 755 813 887
unsigned __int40_t 40 bits Binary 0 1 099 511 627 775
long long, signed long long 64 bits 2s complement -9 223 372 036 854 775 808 9 223 372 036 854 775 807
unsigned long long 64 bits Binary 0 18 446 744 073 709 551 615
enum 32 bits 2s complement -2 147 483 648 2 147 483 647
(1)
float 32 bits IEEE 32-bit 1.175 494e-38 3.40 282 346e+38
double 64 bits IEEE 64-bit 2.22 507 385e-308 (1) 1.79 769 313e+308
long double 64 bits IEEE 64-bit 2.22 507 385e-308 (1) 1.79 769 313e+308
pointers, references, pointer to 32 bits Binary 0 0xFFFFFFFF
data members
(1)
Figures are minimum precision.

Table 6-2. TMS320C6000 C/C++ EABI Data Types

Range
Type Size Representation Minimum Maximum
char, signed char 8 bits ASCII -128 127
unsigned char 8 bits ASCII 0 255
short 16 bits 2s complement -32 768 32 767
unsigned short 16 bits Binary 0 65 535
int, signed int 32 bits 2s complement -2 147 483 648 2 147 483 647
unsigned int 32 bits Binary 0 4 294 967 295
long, signed long 32 bits 2s complement -2 147 483 648 2 147 483 647
unsigned long 32 bits Binary 0 4 294 967 295
__int40_t 40 bits 2s complement -549 755 813 888 549 755 813 887
unsigned __int40_t 40 bits Binary 0 1 099 511 627 775
long long, signed long long 64 bits 2s complement -9 223 372 036 854 775 808 9 223 372 036 854 775 807
unsigned long long 64 bits Binary 0 18 446 744 073 709 551 615
enum 32 bits 2s complement -2 147 483 648 2 147 483 647
(1)
float 32 bits IEEE 32-bit 1.175 494e-38 3.40 282 346e+38
double 64 bits IEEE 64-bit 2.22 507 385e-308 (1) 1.79 769 313e+308
long double 64 bits IEEE 64-bit 2.22 507 385e-308 (1) 1.79 769 313e+308
pointers, references, pointer to 32 bits Binary 0 0xFFFFFFFF
data members
(1)
Figures are minimum precision.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 149

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Keywords www.ti.com

6.5 Keywords
The C6000 C/C++ compiler supports the standard const, register, restrict, and volatile keywords. In
addition, the C/C++ compiler extends the C/C++ language through the support of the cregister, interrupt,
near, and far keywords.

6.5.1 The const Keyword

The C/C++ compiler supports the ANSI/ISO standard keyword const. This keyword gives you greater
optimization and control over allocation of storage for certain data objects. You can apply the const
qualifier to the definition of any variable or array to ensure that its value is not altered.
If you define an object as far const, the .const section allocates storage for the object. The const data
storage allocation rule has two exceptions:
• If the keyword volatile is also specified in the definition of an object (for example, volatile const int x).
Volatile keywords are assumed to be allocated to RAM. (The program does not modify a const volatile
object, but something external to the program might.)
• If the object has automatic storage (allocated on the stack).
In both cases, the storage for the object is the same as if the const keyword were not used.
The placement of the const keyword within a definition is important. For example, the first statement below
defines a constant pointer p to a variable int. The second statement defines a variable pointer q to a
constant int:
int * const p = &x;
const int * q = &x;

Using the const keyword, you can define large constant tables and allocate them into system ROM. For
example, to allocate a ROM table, you could use the following definition:
far const int digits[] = {0,1,2,3,4,5,6,7,8,9};

6.5.2 The cregister Keyword

The compiler extends the C/C++ language by adding the cregister keyword to allow high level language
access to control registers.
When you use the cregister keyword on an object, the compiler compares the name of the object to a list
of standard control registers for the C6000 (see Table 6-3). If the name matches, the compiler generates
the code to reference the control register. If the name does not match, the compiler issues an error.

Table 6-3. Valid Control Registers

Register Description
AMR Addressing mode register
CSR Control status register
DESR (C6700+ only) dMAX event status register
DETR (C6700+ only) dMAX event trigger register
DNUM (C6400+ only) DSP core number register
ECR (C6400+ only) Exception clear register
EFR (C6400+ only) Exception flag register
FADCR (C6700 only) Floating-point adder configuration register
FAUCR (C6700 only) Floating-point auxiliary configuration register
FMCR (C6700 only) Floating-point multiplier configuration register
GFPGFR (C6400 only) Galois field polynomial generator function register
GPLYA (C6400+ only) GMPY A-side polynomial register
CPLYB (C6400+ only) GMPY B-side polynomial register
ICR Interrupt clear register
IER Interrupt enable register
IERR (C6400+ only) Internal exception report register

150 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Keywords

Table 6-3. Valid Control Registers (continued)

Register Description
IFR Interrupt flag register. (IFR is read only.)
ILC (C6400+ only) Inner loop count register
IRP Interrupt return pointer
ISR Interrupt set register
ISTP Interrupt service table pointer
ITSR (C6400+ only) Interrupt task state register
NRP Nonmaskable interrupt return pointer
NTSR (C6400+ only) NMI/exception task state register
REP (C6400+ only) Restricted entry point address register
RILC (C6400+ only) Reload inner loop count register
SSR (C6400+ only) Saturation status register
TSCH (C6400+ only) Time-stamp counter (high 32) register
TSCL (C6400+ only) Time-stamp counter (low 32) register
TSR (C6400+ only) Task state register

The cregister keyword can be used only in file scope. The cregister keyword is not allowed on any
declaration within the boundaries of a function. It can only be used on objects of type integer or pointer.
The cregister keyword is not allowed on objects of any floating-point type or on any structure or union
objects.
The cregister keyword does not imply that the object is volatile. If the control register being referenced is
volatile (that is, can be modified by some external control), then the object must be declared with the
volatile keyword also.
To use the control registers in Table 6-3, you must declare each register as follows. The c6x.h include file
defines all the control registers through this syntax:

extern cregister volatile unsigned int register ;

Once you have declared the register, you can use the register name directly. See the TMS320C62x DSP
CPU and Instruction Set Reference Guide, TMS320C64x/C64x+ DSP CPU and Instruction Set Reference
Guide, the TMS320C67x/C67x+ DSP CPU and Instruction Set Reference Guide, or TMS320C66x+ DSP
CPU and Instruction Set Reference Guide for detailed information on the control registers.
See Example 6-1 for an example that declares and uses control registers.

Example 6-1. Define and Use Control Registers

extern cregister volatile unsigned int AMR;

extern cregister volatile unsigned int CSR;
extern cregister volatile unsigned int IFR;
extern cregister volatile unsigned int ISR;
extern cregister volatile unsigned int ICR;
extern cregister volatile unsigned int IER;
extern cregister volatile unsigned int FADCR;
extern cregister volatile unsigned int FAUCR;
extern cregister volatile unsigned int FMCR;
main()
{
printf("AMR = %x\n", AMR);
}

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 151

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Keywords www.ti.com

6.5.3 The interrupt Keyword

The compiler extends the C/C++ language by adding the interrupt keyword, which specifies that a function
is treated as an interrupt function.
Functions that handle interrupts follow special register-saving rules and a special return sequence. The
implementation stresses safety. The interrupt routine does not assume that the C run-time conventions for
the various CPU register and status bits are in effect; instead, it re-establishes any values assumed by the
run-time environment. When C/C++ code is interrupted, the interrupt routine must preserve the contents of
all machine registers that are used by the routine or by any function called by the routine. When you use
the interrupt keyword with the definition of the function, the compiler generates register saves based on
the rules for interrupt functions and the special return sequence for interrupts.
You can only use the interrupt keyword with a function that is defined to return void and that has no
parameters. The body of the interrupt function can have local variables and is free to use the stack or
global variables. For example:
interrupt void int_handler()
{
unsigned int flags;
...
}

The name c_int00 is the C/C++ entry point. This name is reserved for the system reset interrupt. This
special interrupt routine initializes the system and calls the function main. Because it has no caller, c_int00
does not save any registers.
Use the alternate keyword, __interrupt, if you are writing code for strict ANSI/ISO mode (using the --
strict_ansi compiler option).

HWI Objects and the interrupt Keyword

NOTE: The interrupt keyword must not be used when BIOS HWI objects are used in conjunction
with C functions. The HWI_enter/HWI_exit macros and the HWI dispatcher contain this
functionality, and the use of the C modifier can cause negative results.

6.5.4 The near and far Keywords

The C6000 C/C++ compiler extends the C/C++ language with the near and far keywords to specify how
global and static variables are accessed and how functions are called.
Syntactically, the near and far keywords are treated as storage class modifiers. They can appear before,
after, or in between the storage class specifiers and types. With the exception of near and far, two storage
class modifiers cannot be used together in a single declaration. The following examples are legal
combinations of near and far with other storage class modifiers:
far static int x;
static near int x;
static int far x;
far int foo();
static far int foo();

6.5.4.1 near and far Data Objects

Global and static data objects can be accessed in the following two ways:

near keyword The compiler assumes that the data item can be accessed relative to the data page
pointer. For example:
LDW *+dp(_address),a0
far keyword The compiler cannot access the data item via the DP. This can be required if the
total amount of program data is larger than the offset allowed (32K) from the DP.
For example:
MVKL _address, a1
MVKH _address, a1
LDW *a1,a0

152 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Keywords

Once a variable has been defined to be far, all external references to this variable in other C files or
headers must also contain the far keyword. This is also true of the near keyword. However, you will get
compiler or linker errors when the far keyword is not used everywhere. Not using the near keyword
everywhere only leads to slower data access times.
If you use the DATA_SECTION pragma, the object is indicated as a far variable, and this cannot be
overridden. If you reference this object in another file, then you need to use extern far when declaring this
object in the other source file. This ensures access to the variable, since the variable might not be in the
.bss section. For details, see Section 6.9.6.

NOTE: Defining Global Variables in Assembly Code

If you also define a global variable in assembly code with the .usect directive (where the
variable is not assigned in the .bss section) or you allocate a variable into separate section
using a #pragma DATA_SECTION directive; and you want to reference that variable in C
code, you must declare the variable as extern far. This ensures the compiler does not try to
generate an illegal access of the variable by way of the data page pointer.

When data objects do not have the near or far keyword specified, the compiler will use far accesses to
aggregate data and near accesses to non-aggregate data. For more information on the data memory
model and ways to control accesses to data, see Section 7.1.5.1.

6.5.4.2 Near and far Function Calls

Function calls can be invoked in one of two ways:

near keyword The compiler assumes that destination of the call is within ± 1 M word of the caller.
Here the compiler uses the PC-relative branch instruction.
B _func
far keyword The compiler is told by you that the call is not within ± 1 M word.
MVKL _func, al
MVKH _func, al
B _func

By default, the compiler generates small-memory model code, which means that every function call is
handled as if it were declared near, unless it is actually declared far.
For more information on function calls, see Section 7.1.6.

6.5.5 The restrict Keyword

To help the compiler determine memory dependencies, you can qualify a pointer, reference, or array with
the restrict keyword. The restrict keyword is a type qualifier that can be applied to pointers, references,
and arrays. Its use represents a guarantee by you, the programmer, that within the scope of the pointer
declaration the object pointed to can be accessed only by that pointer. Any violation of this guarantee
renders the program undefined. This practice helps the compiler optimize certain sections of code
because aliasing information can be more easily determined.
In Example 6-2, the restrict keyword is used to tell the compiler that the function func1 is never called with
the pointers a and b pointing to objects that overlap in memory. You are promising that accesses through
a and b will never conflict; therefore, a write through one pointer cannot affect a read from any other
pointers. The precise semantics of the restrict keyword are described in the 1999 version of the ANSI/ISO
C Standard.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 153

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Keywords www.ti.com

Example 6-2. Use of the restrict Type Qualifier With Pointers

void func1(int * restrict a, int * restrict b)

{
/* func1's code here */
}

Example 6-3 illustrates using the restrict keyword when passing arrays to a function. Here, the arrays c
and d should not overlap, nor should c and d point to the same array.

Example 6-3. Use of the restrict Type Qualifier With Arrays

void func2(int c[restrict], int d[restrict])

{
int i;

for(i = 0; i < 64; i++)

{
c[i] += d[i];
d[i] += 1;
}
}

6.5.6 The volatile Keyword

The compiler eliminates redundant memory accesses whenever possible, using data flow analysis to
figure out when it is legal. However, some memory accesses may be special in some way that the
compiler cannot see, and in such cases you must use the volatile keyword to prevent the compiler from
optimizing away something important. The compiler does not optimize out any accesses to variables
declared volatile. The number and order of accesses of a volatile variable are exactly as they appear in
the C/C++ code, no more and no less.
There are different ways to understand how volatile works, but fundamentally it is a hint to the the
compiler that something it cannot understand is going on, and so the compiler should not try to be over-
clever.
Any variable which might be modified by something external to the obvious control flow of the program
(such as an interrupt service routine) must be declared volatile. This tells the compiler that an interrupt
function might modify the value at any time, so the compiler should not perform optimizations which will
change the number or order of accesses of that variable. This is the primary purpose of the volatile
keyword.
In the following example, the loop intends to wait for a location to be read as 0xFF:
unsigned int *ctrl;
while (*ctrl !=0xFF);

However, in this example, *ctrl is a loop-invariant expression, so the loop is optimized down to a single-
memory read. To get the desired result, define ctrl as:
volatile unsigned int *ctrl;

Here the *ctrl pointer is intended to reference a hardware location, such as an interrupt flag.
Volatile must also be used when accessing memory locations that represent memory-mapped peripheral
devices. Such memory locations might change value in ways that the compiler cannot predict. These
locations might change if accessed, or when some other memory location is accessed, or when some
signal occurs.
Volatile must also be used for local variables in a function which calls setjmp, if the value of the local
variables needs to remain valid if a longjmp occurs.

154 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com C++ Exception Handling

Example 6-4. Volatile for Local Variables With setjmp

#include <stdlib.h>
jmp_buf context;
void function()
{
volatile int x = 3;
switch(setjmp(context))
{
case 0: setup(); break;
default:
{
printf("x == %d\n", x); /* We can only reach here if longjmp
has occured; because x's lifetime
begins before the setjmp and lasts
through the longjmp, the C standard
requires x be declared "volatile" */
break;
}
}
}

The --interrupt_threshold=1 option should be used when compiling with volatiles.

6.6 C++ Exception Handling

The compiler supports all the C++ exception handling features as defined by the ANSI/ISO 14882 C++
Standard. More details are discussed in The C++ Programming Language, Third Edition by Bjarne
Stroustrup.
The compiler --exceptions option enables exception handling. The compiler’s default is no exception
handling support.
For exceptions to work correctly, all C++ files in the application must be compiled with the --exceptions
option, regardless of whether exceptions occur in a particular file. Mixing exception-enabled object files
and libraries with object files and libraries that do not have exceptions enabled can lead to undefined
behavior.
Exception handling requires support in the run-time-support library, which come in exception-enabled and
exception-disabled forms; you must link with the correct form. When using automatic library selection (the
default), the linker automatically selects the correct library Section 5.3.1.1. If you select the library
manually, you must use run-time-support libraries whose name contains _eh if you enable exceptions.
Using --exceptions causes the compiler to insert exception handling code. This code will increase the
code size of the program, particularly for COFF ABI. In addition, COFF ABI will increase the execution
time, even if an exception is never thrown. EABI will not increase code size as much, and has a minimal
execution time cost if exceptions are never thrown, but will slightly increase the data size for the
exception-handling tables.
See Section 8.1 for details on the run-time libraries.

6.7 Register Variables and Parameters

The C/C++ compiler treats register variables (variables defined with the register keyword) differently,
depending on whether you use the --opt_level (-O) option.
• Compiling with optimization
The compiler ignores any register definitions and allocates registers to variables and temporary values
by using an algorithm that makes the most efficient use of registers.
• Compiling without optimization
If you use the register keyword, you can suggest variables as candidates for allocation into registers.
The compiler uses the same set of registers for allocating temporary expression results as it uses for

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 155

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
The asm Statement www.ti.com

allocating register variables.

The compiler attempts to honor all register definitions. If the compiler runs out of appropriate registers, it
frees a register by moving its contents to memory. If you define too many objects as register variables,
you limit the number of registers the compiler has for temporary expression results. This limit causes
excessive movement of register contents to memory.
Any object with a scalar type (integral, floating point, or pointer) can be defined as a register variable. The
register designator is ignored for objects of other types, such as arrays.
The register storage class is meaningful for parameters as well as local variables. Normally, in a function,
some of the parameters are copied to a location on the stack where they are referenced during the
function body. The compiler copies a register parameter to a register instead of the stack, which speeds
access to the parameter within the function.
For more information about register conventions, see Section 7.3.

6.8 The asm Statement

The C/C++ compiler can embed assembly language instructions or directives directly into the assembly
language output of the compiler. This capability is an extension to the C/C++ language—the asm
statement. The asm (or __asm) statement provides access to hardware features that C/C++ cannot
provide. The asm statement is syntactically like a call to a function named asm, with one string constant
argument:

asm(" assembler text ");

The compiler copies the argument string directly into your output file. The assembler text must be
enclosed in double quotes. All the usual character string escape codes retain their definitions. For
example, you can insert a .byte directive that contains quotes as follows:
asm("STR: .byte \"abc\"");

The inserted code must be a legal assembly language statement. Like all assembly language statements,
the line of code inside the quotes must begin with a label, a blank, a tab, or a comment (asterisk or
semicolon). The compiler performs no checking on the string; if there is an error, the assembler detects it.
For more information about the assembly language statements, see the TMS320C6000 Assembly
Language Tools User's Guide.
The asm statements do not follow the syntactic restrictions of normal C/C++ statements. Each can appear
as a statement or a declaration, even outside of blocks. This is useful for inserting directives at the very
beginning of a compiled module.
Use the alternate statement __asm("assembler text") if you are writing code for strict ANSI/ISO C mode
(using the --strict_ansi option).

NOTE: Avoid Disrupting the C/C++ Environment With asm Statements

Be careful not to disrupt the C/C++ environment with asm statements. The compiler does not
check the inserted instructions. Inserting jumps and labels into C/C++ code can cause
unpredictable results in variables manipulated in or around the inserted code. Directives that
change sections or otherwise affect the assembly environment can also be troublesome.
Be especially careful when you use optimization with asm statements. Although the compiler
cannot remove asm statements, it can significantly rearrange the code order near them and
cause undesired results.

156 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9 Pragma Directives

Pragma directives tell the compiler how to treat a certain function, object, or section of code. The C6000
C/C++ compiler supports the following pragmas:
• CHECK_MISRA (See Section 6.9.1)
• CLINK (See Section 6.9.2)
• CODE_SECTION (See Section 6.9.3)
• DATA_ALIGN (See Section 6.9.4)
• DATA_MEM_BANK (See Section 6.9.5)
• DATA_SECTION (See Section 6.9.6)
• DIAG_SUPPRESS, DIAG_REMARK, DIAG_WARNING, DIAG_ERROR, and DIAG_DEFAULT (See
Section 6.9.7)
• FUNC_ALWAYS_INLINE (See Section 6.9.8)
• FUNC_CANNOT_INLINE (See Section 6.9.9)
• FUNC_EXT_CALLED (See Section 6.9.10)
• FUNC_INTERRUPT_THRESHOLD (See Section 6.9.11)
• FUNC_IS_PURE (See Section 6.9.12)
• FUNC_IS_SYSTEM (See Section 6.9.13)
• FUNC_NEVER_RETURNS (See Section 6.9.14)
• FUNC_NO_GLOBAL_ASG (See Section 6.9.15)
• FUNC_NO_IND_ASG (See Section 6.9.16)
• FUNCTION_OPTIONS (See Section 6.9.17)
• INTERRUPT (See Section 6.9.18)
• LOCATION (See Section 6.9.19)
• MUST_ITERATE (See Section 6.9.20)
• NMI_INTERRUPT (See Section 6.9.21)
• NOINIT (See Section 6.9.22)
• NO_HOOKS (See Section 6.9.23)
• PERSISTENT (See Section 6.9.22)
• PROB_ITERATE (See Section 6.9.24)
• RESET_MISRA (See Section 6.9.25)
• RETAIN (See Section 6.9.26)
• SET_CODE_SECTION (See Section 6.9.27)
• SET_DATA_SECTION (See Section 6.9.27)
• STRUCT_ALIGN (See Section 6.9.28)
• UNROLL (See Section 6.9.29)
Most of these pragmas apply to functions. Except for the DATA_MEM_BANK pragma, the arguments func
and symbol cannot be defined or declared inside the body of a function. You must specify the pragma
outside the body of a function; and the pragma specification must occur before any declaration, definition,
or reference to the func or symbol argument. If you do not follow these rules, the compiler issues a
warning and may ignore the pragma.
For the pragmas that apply to functions or symbols (except CLINK and RETAIN), the syntax for the
pragmas differs between C and C++. In C, you must supply the name of the object or function to which
you are applying the pragma as the first argument. In C++, the name is omitted; the pragma applies to the
declaration of the object or function that follows it.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 157

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.1 The CHECK_MISRA Pragma

The CHECK_MISRA pragma enables/disables MISRA-C:2004 rules at the source level. This pragma is
equivalent to using the --check_misra option.
The syntax of the pragma in C is:

#pragma CHECK_MISRA (" {all|required|advisory|none|rulespec} ");

The rulespec parameter is a comma-separated list of specifiers. See Section 6.3 for details.
The RESET_MISRA pragma can be used to reset any CHECK_MISRA pragmas; see Section 6.9.25.

6.9.2 The CLINK Pragma

The CLINK pragma can be applied to a code or data symbol. It causes a .clink directive to be generated
into the section that contains the definition of the symbol. The .clink directive indicates to the linker that the
section is eligible for removal during conditional linking. Therefore, if the section is not referenced by any
other section in the application that is being compiled and linked, it will not be included in the output file
result of the link.
The syntax of the pragma in C/C++ is:

#pragma CLINK (symbol )

The RETAIN pragma has the opposite effect of the CLINK pragma. See Section 6.9.26 for more details.

6.9.3 The CODE_SECTION Pragma

The CODE_SECTION pragma allocates space for the symbol in C, or the next symbol declared in C++, in
a section named section name.
The syntax of the pragma in C is:

#pragma CODE_SECTION (symbol , "section name ")

The syntax of the pragma in C++ is:

#pragma CODE_SECTION (" section name ")

The CODE_SECTION pragma is useful if you have code objects that you want to link into an area
separate from the .text section.
The following examples demonstrate the use of the CODE_SECTION pragma.

Example 6-5. Using the CODE_SECTION Pragma C Source File

#pragma CODE_SECTION(fn, "my_sect")

int fn(int x)
{
return x;
}

158 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

Example 6-6. Generated Assembly Code From Example 6-5

.sect "my_sect"
.global _fn

;******************************************************************************
;* FUNCTION NAME: _fn *
;* *
;* Regs Modified : SP *
;* Regs Used : A4,B3,SP *
;* Local Frame Size : 0 Args + 4 Auto + 0 Save = 4 byte *
;******************************************************************************
_fn:
;** --------------------------------------------------------------------------*
RET .S2 B3 ; |6|
SUB .D2 SP,8,SP ; |4|
STW .D2T1 A4,*+SP(4) ; |4|
ADD .S2 8,SP,SP ; |6|
NOP 2
; BRANCH OCCURS ; |6|

6.9.4 The DATA_ALIGN Pragma

The DATA_ALIGN pragma aligns the symbol in C, or the next symbol declared in C++, to an alignment
boundary. The alignment boundary is the maximum of the symbol's default alignment value or the value of
the constant in bytes. The constant must be a power of 2.
The syntax of the pragma in C is:

#pragma DATA_ALIGN ( symbol , constant );

The syntax of the pragma in C++ is:

#pragma DATA_ALIGN ( constant );

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 159

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.5 The DATA_MEM_BANK Pragma

The DATA_MEM_BANK pragma aligns a symbol or variable to a specified C6000 internal data memory
bank boundary. The constant specifies a specific memory bank to start your variables on. (See Figure 4-1
for a graphic representation of memory banks.) The value of constant depends on the C6000 device:

C6200 The C6200 devices contain four memory banks (0, 1, 2, and 3); constant can be 0 or 2.
C6400 The C6400 devices contain 8 memory banks; constant can be 0, 2, 4, or 6.
C6400+ The C6400+ devices contain 8 memory banks; constant can be 0, 2, 4, or 6.
C6600 The C6600 devices contain 8 memory banks; constant can be 0, 2, 4, or 6.
C6700 The C6700 devices contain 8 memory banks; constant can be 0, 2, 4, or 6.
C6740 The C6740 devices contain 8 memory banks; constant can be 0, 2, 4, or 6.

The syntax of the pragma in C is:

#pragma DATA_MEM_BANK ( symbol , constant );

The syntax of the pragma in C++ is:

#pragma DATA_MEM_BANK ( constant );

Both global and local variables can be aligned with the DATA_MEM_BANK pragma. The
DATA_MEM_BANK pragma must reside inside the function that contains the local variable being aligned.
The symbol can also be used as a parameter in the DATA_SECTION pragma.
When optimization is enabled, the tools may or may not use the stack to store the values of local
variables.
The DATA_MEM_BANK pragma allows you to align data on any data memory bank that can hold data of
the type size of the symbol. This is useful if you need to align data in a particular way to avoid memory
bank conflicts in your hand-coded assembly code versus padding with zeros and having to account for the
padding in your code.
This pragma increases the amount of space used in data memory by a small amount as padding is used
to align data onto the correct bank.
For C6200, the code in Example 6-7 guarantees that array x begins at an address ending in 4 or c (in
hexadecimal), and that array y begins at an address ending in 4 or c. The alignment for array y affects its
stack placement. Array z is placed in the .z_sect section, and begins at an address ending in 0 or 8.

Example 6-7. Using the DATA_MEM_BANK Pragma

#pragma DATA_MEM_BANK (x, 2);

short x[100];

#pragma DATA_MEM_BANK (z, 0);

#pragma DATA_SECTION (z, ".z_sect");
short z[100];

void main()
{
#pragma DATA_MEM_BANK (y, 2);
short y[100];
...
}

160 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.6 The DATA_SECTION Pragma

The DATA_SECTION pragma allocates space for the symbol in C, or the next symbol declared in C++, in
a section named section name.
The syntax of the pragma in C is:

#pragma DATA_SECTION ( symbol , " section name ");

The syntax of the pragma in C++ is:

#pragma DATA_SECTION (" section name ");

The DATA_SECTION pragma is useful if you have data objects that you want to link into an area separate
from the .bss section. If you allocate a global variable using a DATA_SECTION pragma and you want to
reference the variable in C code, you must declare the variable as extern far.
Example 6-8 through Example 6-10 demonstrate the use of the DATA_SECTION pragma.

Example 6-8. Using the DATA_SECTION Pragma C Source File

#pragma DATA_SECTION(bufferB, "my_sect")

char bufferA[512];
char bufferB[512];

Example 6-9. Using the DATA_SECTION Pragma C++ Source File

char bufferA[512];
#pragma DATA_SECTION("my_sect")
char bufferB[512];

Example 6-10. Using the DATA_SECTION Pragma Assembly Source File

.global _bufferA
.bss _bufferA,512,4
.global _bufferB

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 161

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.7 The Diagnostic Message Pragmas

The following pragmas can be used to control diagnostic messages in the same ways as the
corresponding command line options:

Pragma Option Description

DIAG_SUPPRESS num -pds=num[, num2, num3...] Suppress diagnostic num
DIAG_REMARK num -pdsr=num[, num2, num3...] Treat diagnostic num as a remark
DIAG_WARNING num -pdsw=num[, num2, num3...] Treat diagnostic num as a warning
DIAG_ERROR num -pdse=num[, num2, num3...] Treat diagnostic num as an error
DIAG_DEFAULT num n/a Use default severity of the diagnostic

The syntax of the pragmas in C is:

#pragma DIAG_XXX [=]num[, num2, num3...]

The diagnostic affected (num) is specified using either an error number or an error tag name. The equal
sign (=) is optional. Any diagnostic can be overridden to be an error, but only diagnostics with a severity of
discretionary error or below can have their severity reduced to a warning or below, or be suppressed. The
diag_default pragma is used to return the severity of a diagnostic to the one that was in effect before any
pragmas were issued (i.e., the normal severity of the message as modified by any command-line options).
The diagnostic identifier number is output along with the message when the -pden command line option is
specified.

6.9.8 The FUNC_ALWAYS_INLINE Pragma

The FUNC_ALWAYS_INLINE pragma instructs the compiler to always inline the named function. The
compiler only inlines the function if it is legal to inline the function and the compiler is invoked with any
level of optimization (--opt_level=0).
The pragma must appear before any declaration or reference to the function that you want to inline. In C,
the argument func is the name of the function that will be inlined. In C++, the pragma applies to the next
function declared.
The syntax of the pragma in C is:

#pragma FUNC_ALWAYS_INLINE ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_ALWAYS_INLINE;

Use Caution with the FUNC_ALWAYS_INLINE Pragma

NOTE: The FUNC_ALWAYS_INLINE pragma overrides the compiler's inlining decisions. Overuse of
the pragma could result in increased compilation times or memory usage, potentially enough
to consume all available memory and result in compilation tool failures.

162 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.9 The FUNC_CANNOT_INLINE Pragma

The FUNC_CANNOT_INLINE pragma instructs the compiler that the named function cannot be expanded
inline. Any function named with this pragma overrides any inlining you designate in any other way, such as
using the inline keyword. Automatic inlining is also overridden with this pragma; see Section 2.11.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function that cannot be inlined. In C++, the pragma applies to the
next function declared.
The syntax of the pragma in C is:

#pragma FUNC_CANNOT_INLINE ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_CANNOT_INLINE;

6.9.10 The FUNC_EXT_CALLED Pragma

When you use the --program_level_compile option, the compiler uses program-level optimization. When
you use this type of optimization, the compiler removes any function that is not called, directly or indirectly,
by main. You might have C/C++ functions that are called by hand-coded assembly instead of main.
The FUNC_EXT_CALLED pragma specifies to the optimizer to keep these C functions or any other
functions that these C/C++ functions call. These functions act as entry points into C/C++.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function that you do not want removed. In C++, the pragma applies
to the next function declared.
The syntax of the pragma in C is:

#pragma FUNC_EXT_CALLED ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_EXT_CALLED;

Except for _c_int00, which is the name reserved for the system reset interrupt for C/C++programs, the
name of the interrupt (the func argument) does not need to conform to a naming convention.
When you use program-level optimization, you may need to use the FUNC_EXT_CALLED pragma with
certain options. See Section 3.7.2.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 163

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.11 The FUNC_INTERRUPT_THRESHOLD Pragma

The compiler allows interrupts to be disabled around software pipelined loops for threshold cycles within
the function. This implements the --interrupt_threshold option for a single function (see Section 2.12). The
FUNC_INTERRUPT_THRESHOLD pragma always overrides the --interrupt_threshold=n command line
option. A threshold value less than 0 assumes that the function is never interrupted, which is equivalent to
an interrupt threshold of infinity.
The syntax of the pragma in C is:

#pragma FUNC_INTERRUPT_THRESHOLD ( func , threshold );

The syntax of the pragma in C++ is:

#pragma FUNC_INTERRUPT_THRESHOLD ( threshold );

The following examples demonstrate the use of different thresholds:

• The function foo( ) must be interruptible at least every 2,000 cycles:
#pragma FUNC_INTERRUPT_THRESHOLD (foo, 2000)
• The function foo( ) must always be interruptible.
#pragma FUNC_INTERRUPT_THRESHOLD (foo, 1)
• The function foo( ) is never interrupted.
#pragma FUNC_INTERRUPT_THRESHOLD (foo, -1)

6.9.12 The FUNC_IS_PURE Pragma

The FUNC_IS_PURE pragma specifies to the compiler that the named function has no side effects. This
allows the compiler to do the following:
• Delete the call to the function if the function's value is not needed
• Delete duplicate functions
The pragma must appear before any declaration or reference to the function. In C, the argument func is
the name of a function. In C++, the pragma applies to the next function declared.
The syntax of the pragma in C is:

#pragma FUNC_IS_PURE ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_IS_PURE;

164 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.13 The FUNC_IS_SYSTEM Pragma

The FUNC_IS_SYSTEM pragma specifies to the compiler that the named function has the behavior
defined by the ANSI/ISO standard for a function with that name.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function to treat as an ANSI/ISO standard function. In C++, the
pragma applies to the next function declared.
The syntax of the pragma in C is:

#pragma FUNC_IS_SYSTEM ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_IS_SYSTEM;

6.9.14 The FUNC_NEVER_RETURNS Pragma

The FUNC_NEVER_RETURNS pragma specifies to the compiler that the function never returns to its
caller.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function that does not return. In C++, the pragma applies to the next
function declared.
The syntax of the pragma in C is:

#pragma FUNC_NEVER_RETURNS ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_NEVER_RETURNS;

6.9.15 The FUNC_NO_GLOBAL_ASG Pragma

The FUNC_NO_GLOBAL_ASG pragma specifies to the compiler that the function makes no assignments
to named global variables and contains no asm statements.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function that makes no assignments. In C++, the pragma applies to
the next function declared.
The syntax of the pragma in C is:

#pragma FUNC_NO_GLOBAL_ASG ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_NO_GLOBAL_ASG;

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 165

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.16 The FUNC_NO_IND_ASG Pragma

The FUNC_NO_IND_ASG pragma specifies to the compiler that the function makes no assignments
through pointers and contains no asm statements.
The pragma must appear before any declaration or reference to the function that you want to keep. In C,
the argument func is the name of the function that makes no assignments. In C++, the pragma applies to
the next function declared.
The syntax of the pragma in C is:

#pragma FUNC_NO_IND_ASG ( func );

The syntax of the pragma in C++ is:

#pragma FUNC_NO_IND_ASG;

6.9.17 The FUNCTION_OPTIONS Pragma

The FUNCTION_OPTIONS pragma allows you to compile a specific function in a C or C++ file with
additional command-line compiler options. The affected function will be compiled as if the specified list of
options appeared on the command line after all other compiler options. In C, the pragma is applied to the
function specified. In C++, the pragma is applied to the next function.
The syntax of the pragma in C is:

#pragma FUNCTION_OPTIONS ( func, "additional options" );

The syntax of the pragma in C++ is:

#pragma FUNCTION_OPTIONS( "additional options" );

6.9.18 The INTERRUPT Pragma

The INTERRUPT pragma enables you to handle interrupts directly with C code. In C, the argument func is
the name of a function. In C++, the pragma applies to the next function declared.
The syntax of the pragma in C is:

#pragma INTERRUPT ( func );

The syntax of the pragma in C++ is:

#pragma INTERRUPT ;

The code for the function will return via the IRP (interrupt return pointer).
Except for _c_int00, which is the name reserved for the system reset interrupt for C programs, the name
of the interrupt (the func argument) does not need to conform to a naming convention.

HWI Objects and the INTERRUPT Pragma

NOTE: The INTERRUPT pragma must not be used when BIOS HWI objects are used in conjunction
with C functions. The HWI_enter/HWI_exit macros and the HWI dispatcher contain this
functionality, and the use of the C modifier can cause negative results.

166 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.19 The LOCATION Pragma

The compiler supports the ability to specify the run-time address of a variable at the source level. This can
be accomplished with the LOCATION pragma or attribute. Location support is only available in EABI.
The syntax of the pragma in C is:

#pragma LOCATION( x , address );

int x;

The syntax of the pragmas in C++ is:

#pragma LOCATION(address );
int x;

The syntax of the pragma for IAR is:

#pragma location=address
int x;

The syntax of the GCC attribute is:

int x attribute((location(address )));

The noinit pragma may be used in conjunction with the location pragma to map variables to special
memory locations; see Section 6.9.22.

6.9.20 The MUST_ITERATE Pragma

The MUST_ITERATE pragma specifies to the compiler certain properties of a loop. You guarantee that
these properties are always true. Through the use of the MUST_ITERATE pragma, you can guarantee
that a loop executes a specific number of times. Anytime the UNROLL pragma is applied to a loop,
MUST_ITERATE should be applied to the same loop. For loops the MUST_ITERATE pragma's third
argument, multiple, is the most important and should always be specified.
Furthermore, the MUST_ITERATE pragma should be applied to any other loops as often as possible. This
is because the information provided via the pragma (especially the minimum number of iterations) aids the
compiler in choosing the best loops and loop transformations (that is, software pipelining and nested loop
transformations). It also helps the compiler reduce code size.
No statements are allowed between the MUST_ITERATE pragma and the for, while, or do-while loop to
which it applies. However, other pragmas, such as UNROLL and PROB_ITERATE, can appear between
the MUST_ITERATE pragma and the loop.

6.9.20.1 The MUST_ITERATE Pragma Syntax

The syntax of the pragma for C and C++ is:

#pragma MUST_ITERATE ( min, max, multiple );

The arguments min and max are programmer-guaranteed minimum and maximum trip counts. The trip
count is the number of times a loop iterates. The trip count of the loop must be evenly divisible by multiple.
All arguments are optional. For example, if the trip count could be 5 or greater, you can specify the
argument list as follows:
#pragma MUST_ITERATE(5);

However, if the trip count could be any nonzero multiple of 5, the pragma would look like this:
#pragma MUST_ITERATE(5, , 5); /* Note the blank field for max */

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 167

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

It is sometimes necessary for you to provide min and multiple in order for the compiler to perform
unrolling. This is especially the case when the compiler cannot easily determine how many iterations the
loop will perform (that is, the loop has a complex exit condition).
When specifying a multiple via the MUST_ITERATE pragma, results of the program are undefined if the
trip count is not evenly divisible by multiple. Also, results of the program are undefined if the trip count is
less than the minimum or greater than the maximum specified.
If no min is specified, zero is used. If no max is specified, the largest possible number is used. If multiple
MUST_ITERATE pragmas are specified for the same loop, the smallest max and largest min are used.

6.9.20.2 Using MUST_ITERATE to Expand Compiler Knowledge of Loops

Through the use of the MUST_ITERATE pragma, you can guarantee that a loop executes a certain
number of times. The example below tells the compiler that the loop is guaranteed to run exactly 10 times:
#pragma MUST_ITERATE(10,10);

for(i = 0; i < trip_count; i++) { ...

In this example, the compiler attempts to generate a software pipelined loop even without the pragma.
However, if MUST_ITERATE is not specified for a loop such as this, the compiler generates code to
bypass the loop, to account for the possibility of 0 iterations. With the pragma specification, the compiler
knows that the loop iterates at least once and can eliminate the loop-bypassing code.
MUST_ITERATE can specify a range for the trip count as well as a factor of the trip count. For example:
pragma MUST_ITERATE(8, 48, 8);

for(i = 0; i < trip_count; i++) { ...

This example tells the compiler that the loop executes between 8 and 48 times and that the trip_count
variable is a multiple of 8 (8, 16, 24, 32, 40, 48). The multiple argument allows the compiler to unroll the
loop.
You should also consider using MUST_ITERATE for loops with complicated bounds. In the following
example:
for(i2 = ipos[2]; i2 < 40; i2 += 5) { ...

The compiler would have to generate a divide function call to determine, at run time, the exact number of
iterations performed. The compiler will not do this. In this case, using MUST_ITERATE to specify that the
loop always executes eight times allows the compiler to attempt to generate a software pipelined loop:
#pragma MUST_ITERATE(8, 8);

for(i2 = ipos[2]; i2 < 40; i2 += 5) { ...

6.9.21 The NMI_INTERRUPT Pragma

The NMI_INTERRUPT pragma enables you to handle non-maskable interrupts directly with C code. In C,
the argument func is the name of a function. In C++, the pragma applies to the next function declared.
The syntax of the pragma in C is:

#pragma NMI_INTERRUPT( func );

The syntax of the pragma in C++ is:

#pragma NMI_INTERRUPT;

The code generated for the function will return via the NRP versus the IRP as for a function declared with
the interrupt keyword or INTERRUPT pragma.
Except for _c_int00, which is the name reserved for the system reset interrupt for C programs, the name
of the interrupt (function) does not need to conform to a naming convention.

168 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.22 The NOINIT and PERSISTENT Pragmas

When using EABI, global and static variables will be zero-initialized. However, in applications using non-
volatile memory, it may be desirable to have variables that are not initialized. Noinit variables are global or
static variables that are not zero-initialized at startup or reset.
The NOINIT pragma may be used in conjunction with the location pragma to map variables to special
memory locations, like memory-mapped registers, without generating unwanted writes.
Noinit variables cannot have an initializer. However, statically-initialized variables may be declared
"persistent" to disable startup initialization. Persistent variables are given an initial value when the code is
loaded, but are never again initialized. Persistent and noinit variables behave identically with the exception
of whether or not they are initialized at load time.
The syntax of the pragmas in C is:

#pragma NOINIT (x );
int x;

#pragma PERSISTENT (x );
int x=10;

The syntax of the pragmas in C++ is:

#pragma NOINIT;
int x;

#pragma PERSISTENT;
int x=10;

The syntax of the GCC attributes is:

int x __attribute__((noinit));

int x =0 __attribute__((persistent));

6.9.23 The NO_HOOKS Pragma

The NO_HOOKS pragma prevents entry and exit hook calls from being generated for a function.
The syntax of the pragma in C is:

#pragma NO_HOOKS ( func );

The syntax of the pragma in C++ is:

#pragma NO_HOOKS;

See Section 2.17 for details on entry and exit hooks.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 169

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.24 The PROB_ITERATE Pragma

The PROB_ITERATE pragma specifies to the compiler certain properties of a loop. You assert that these
properties are true in the common case. The PROB_ITERATE pragma aids the compiler in choosing the
best loops and loop transformations (that is, software pipelining and nested loop transformations).
PROB_ITERATE is useful only when the MUST_ITERATE pragma is not used or the PROB_ITERATE
parameters are more constraining than the MUST_ITERATE parameters.
No statements are allowed between the PROB_ITERATE pragma and the for, while, or do-while loop to
which it applies. However, other pragmas, such as UNROLL and MUST_ITERATE, may appear between
the PROB_ITERATE pragma and the loop.
The syntax of the pragma for C and C++ is:

#pragma PROB_ITERATE( min , max )

Where min and max are the minimum and maximum trip counts of the loop in the common case. The trip
count is the number of times a loop iterates. Both arguments are optional.
For example, PROB_ITERATE could be applied to a loop that executes for eight iterations in the majority
of cases (but sometimes may execute more or less than eight iterations):
#pragma PROB_ITERATE(8, 8);

If only the minimum expected trip count is known (say it is 5), the pragma would look like this:
#pragma PROB_ITERATE(5);

If only the maximum expected trip count is known (say it is 10), the pragma would look like this:
#pragma PROB_ITERATE(, 10); /* Note the blank field for min */

6.9.25 The RESET_MISRA Pragma

The RESET_MISRA pragma resets the specified MISRA-C:2004 rules to the state they were before any
CHECK_MISRA pragmas (see Section 6.9.1) were processed. For instance, if a rule was enabled on the
command line but disabled in the source, the RESET_MISRA pragma resets it to enabled. This pragma
accepts the same format as the --check_misra option, except for the "none" keyword.
The syntax of the pragma in C is:

#pragma RESET_MISRA (" {all|required|advisory|rulespec} ")

The rulespec parameter is a comma-separated list of specifiers. See Section 6.3 for details.

6.9.26 The RETAIN Pragma

The RETAIN pragma can be applied to a code or data symbol. It causes a .retain directive to be
generated into the section that contains the definition of the symbol. The .retain directive indicates to the
linker that the section is ineligible for removal during conditional linking. Therefore, regardless whether or
not the section is referenced by another section in the application that is being compiled and linked, it will
be included in the output file result of the link.
The syntax of the pragma in C/C++ is:

#pragma RETAIN ( symbol )

The CLINK pragma has the opposite effect of the RETAIN pragma. See Section 6.9.2 for more details.

170 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Pragma Directives

6.9.27 The SET_CODE_SECTION and SET_DATA_SECTION Pragmas

These pragmas can be used to set the section for all declarations below the pragma.
The syntax of the pragmas in C/C++ is:

#pragma SET_CODE_SECTION ("section name")

#pragma SET_DATA_SECTION ("section name")

In Example 6-11 x and y are put in the section mydata. To reset the current section to the default used by
the compiler, a blank paramater should be passed to the pragma. An easy way to think of the pragma is
that it is like applying the CODE_SECTION or DATA_SECTION pragma to all symbols below it.

Example 6-11. Setting Section With SET_DATA_SECTION Pragma

#pragma SET_DATA_SECTION("mydata")
int x;
int y;
#pragma SET_DATA_SECTION()

The pragmas apply to both declarations and definitions. If applied to a declaration and not the definition,
the pragma that is active at the declaration is used to set the section for that symbol. Here is an example:

Example 6-12. Setting a Section With SET_CODE_SECTION Pragma

#pragma SET_CODE_SECTION("func1")
extern void func1();
#pragma SET_CODE_SECTION()
...
void func1() { ... }

In Example 6-12 func1 is placed in section func1. If conflicting sections are specified at the declaration
and definition, a diagnostic is issued.
The current CODE_SECTION and DATA_SECTION pragmas and GCC attributes can be used to override
the SET_CODE_SECTION and SET_DATA_SECTION pragmas. For example:

Example 6-13. Overriding SET_DATA_SECTION Setting

#pragma DATA_SECTION(x, "x_data")

#pragma SET_DATA_SECTION("mydata")
int x;
int y;
#pragma SET_DATA_SECTION()

In Example 6-13 x is placed in x_data and y is placed in mydata. No diagnostic is issued for this case.
The pragmas work for both C and C++. In C++, the pragmas are ignored for templates and for implictly
created objects, such as implicit constructors and virtual function tables.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 171

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Pragma Directives www.ti.com

6.9.28 The STRUCT_ALIGN Pragma

The STRUCT_ALIGN pragma is similar to DATA_ALIGN, but it can be applied to a structure, union type,
or typedef and is inherited by any symbol created from that type. The STRUCT_ALIGN pragma is
supported only in C.
The syntax of the pragma is:

#pragma STRUCT_ALIGN( type , constant expression )

This pragma guarantees that the alignment of the named type or the base type of the named typedef is at
least equal to that of the expression. (The alignment may be greater as required by the compiler.) The
alignment must be a power of 2. The type must be a type or a typedef name. If a type, it must be either a
structure tag or a union tag. If a typedef, its base type must be either a structure tag or a union tag.
Since ANSI/ISO C declares that a typedef is simply an alias for a type (i.e. a struct) this pragma can be
applied to the struct, the typedef of the struct, or any typedef derived from them, and affects all aliases of
the base type.
This example aligns any st_tag structure variables on a page boundary:
typedef struct st_tag
{
int a;
short b;
} st_typedef;

#pragma STRUCT_ALIGN (st_tag, 128);

#pragma STRUCT_ALIGN (st_typedef, 128);

Any use of STRUCT_ALIGN with a basic type (int, short, float) or a variable results in an error.

6.9.29 The UNROLL Pragma

The UNROLL pragma specifies to the compiler how many times a loop should be unrolled. The UNROLL
pragma is useful for helping the compiler utilize SIMD instructions on the C6400 family. It is also useful in
cases where better utilization of software pipeline resources are needed over a non-unrolled loop.
The optimizer must be invoked (use --opt_level=[1|2|3] or -O1, -O2, or -O3) in order for pragma-specified
loop unrolling to take place. The compiler has the option of ignoring this pragma.
No statements are allowed between the UNROLL pragma and the for, while, or do-while loop to which it
applies. However, other pragmas, such as MUST_ITERATE and PROB_ITERATE, can appear between
the UNROLL pragma and the loop.
The syntax of the pragma for C and C++ is:

#pragma UNROLL( n );

If possible, the compiler unrolls the loop so there are n copies of the original loop. The compiler only
unrolls if it can determine that unrolling by a factor of n is safe. In order to increase the chances the loop is
unrolled, the compiler needs to know certain properties:
• The loop iterates a multiple of n times. This information can be specified to the compiler via the
multiple argument in the MUST_ITERATE pragma.
• The smallest possible number of iterations of the loop
• The largest possible number of iterations of the loop
The compiler can sometimes obtain this information itself by analyzing the code. However, sometimes the
compiler can be overly conservative in its assumptions and therefore generates more code than is
necessary when unrolling. This can also lead to not unrolling at all.
Furthermore, if the mechanism that determines when the loop should exit is complex, the compiler may
not be able to determine these properties of the loop. In these cases, you must tell the compiler the
properties of the loop by using the MUST_ITERATE pragma.
172 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com The _Pragma Operator

Specifying #pragma UNROLL(1); asks that the loop not be unrolled. Automatic loop unrolling also is not
performed in this case.
If multiple UNROLL pragmas are specified for the same loop, it is undefined which pragma is used, if any.

6.10 The _Pragma Operator

The C6000 C/C++ compiler supports the C99 preprocessor _Pragma() operator. This preprocessor
operator is similar to #pragma directives. However, _Pragma can be used in preprocessing macros
(#defines).
The syntax of the operator is:

_Pragma (" string_literal ");

The argument string_literal is interpreted in the same way the tokens following a #pragma directive are
processed. The string_literal must be enclosed in quotes. A quotation mark that is part of the string_literal
must be preceded by a backward slash.
You can use the _Pragma operator to express #pragma directives in macros. For example, the
DATA_SECTION syntax:
#pragma DATA_SECTION( func ," section ");
Is represented by the _Pragma() operator syntax:
_Pragma ("DATA_SECTION( func ,\" section \")")
The following code illustrates using _Pragma to specify the DATA_SECTION pragma in a macro:
...

#define EMIT_PRAGMA(x) _Pragma(#x)

#define COLLECT_DATA(var) EMIT_PRAGMA(DATA_SECTION(var,"mysection"))

COLLECT_DATA(x)
int x;

...

The EMIT_PRAGMA macro is needed to properly expand the quotes that are required to surround the
section argument to the DATA_SECTION pragma.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 173

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Application Binary Interface www.ti.com

6.11 Application Binary Interface

Selecting one of the two ABIs supported by the C6000 compiler is discussed in Section 2.16.
An ABI should define how functions that are written separately, and compiled or assembled separately
can work together. This involves standardizing the data type representation, register conventions, and
function structure and calling conventions. It should define linkname generation from C symbol names. It
should define the object file format and the debug format. It should document how the system is initialized.
In the case of C++ it defines C++ name mangling and exception handling support.
An application must be only one of COFF ABI and EABI; these ABIs are not compatible.

6.11.1 COFF ABI

COFF ABI is the only ABI supported by older compilers. To generate object files compatible with older
COFF ABI object files, you must use COFF ABI (--abi=coffabi, the default). This option must also be used
when assembly hand-coded assembly source files intended to be used in a COFF ABI application.

6.11.2 EABI
EABI requires the ELF object file format which enables supporting modern language features like early
template instantiation and export inline functions support.
TI-specific information on EABI mode is described in Section 7.8.2.
To generate object files compatible with EABI, you must use C6000 compiler version 7.2 or greater; see
Section 2.16. The __TI_EABI__ predefined symbol is defined and set to 1 if compiling for EABI and is not
defined otherwise.

6.12 Object File Symbol Naming Conventions (Linknames)

Each externally visible identifier is assigned a unique symbol name to be used in the object file, a so-
called linkname. This name is assigned by the compiler according to an algorithm which depends on the
name, type, and source language of the symbol. This algorithm may add a prefix to the identifier (typically
an underscore), and it may mangle the name. This algorithm may mangle the name.
In COFF ABI, the linkname for all objects and functions is the same as the name in the C source with an
added underscore prefix. This prevents any C identifier from colliding with any identifier in the assembly
code namespace, such as an assembler keyword.
In EABI, no prefix is used. If a C identifier would collide with an assembler keyword, the compiler will
escape the identifier with double parallel bars, which instructs the assembler not to treat the identifier as a
keyword. You are responsible for making sure that C identifiers do not collide with user-defined assembly
code identifiers.
Name mangling encodes the types of the parameters of a function in the linkname for a function. Name
mangling only occurs for C++ functions which are not declared 'extern "C"'. Mangling allows function
overloading, operator overloading, and type-safe linking. Be aware that the return value of the function is
not encoded in the mangled name, as C++ functions cannot be overloaded based on the return value.
For COFF ABI, the mangling algorithm used closely follows that described in The Annotated Reference
Manual (ARM).
For example, the general form of a C++ linkname for a function named func is:
_func__F parmcodes
Where parmcodes is a sequence of letters that encodes the parameter types of func.
For this simple C++ source file:
int foo(int i){ } //global C++ function

This is the resulting assembly code:

_foo__Fi

174 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Initializing Static and Global Variables in COFF ABI Mode

The linkname of foo is _foo__Fi, indicating that foo is a function that takes a single argument of type int.
To aid inspection and debugging, a name demangling utility is provided that demangles names into those
found in the original C++ source. See Chapter 9 for more information.
For EABI, the mangling algorithm follows that described in the Itanium C++ ABI
(http://www.codesourcery.com/cxx-abi/abi.html).
int foo(int i) { } would be mangled "_Z3fooi"

6.13 Initializing Static and Global Variables in COFF ABI Mode

The ANSI/ISO C standard specifies that global (extern) and static variables without explicit initializations
must be initialized to 0 before the program begins running. This task is typically done when the program is
loaded. Because the loading process is heavily dependent on the specific environment of the target
application system, in COFF ABI mode the compiler itself makes no provision for initializing to 0 otherwise
uninitialized static storage class variables at run time. It is up to your application to fulfill this requirement.

Initialize Global Objects

NOTE: You should explicitly initialize all global objects which you expected the compiler would set to
zero by default.
In C6000 EABI mode the uninitialized variables are zero initialized automatically.

6.13.1 Initializing Static and Global Variables With the Linker

If your loader does not preinitialize variables, you can use the linker to preinitialize the variables to 0 in the
object file. For example, in the linker command file, use a fill value of 0 in the .bss section:
SECTIONS
{
...

.bss: {} = 0x00;
...
}

Because the linker writes a complete load image of the zeroed .bss section into the output COFF file, this
method can have the unwanted effect of significantly increasing the size of the output file (but not the
program).
If you burn your application into ROM, you should explicitly initialize variables that require initialization.
The preceding method initializes .bss to 0 only at load time, not at system reset or power up. To make
these variables 0 at run time, explicitly define them in your code.
For more information about linker command files and the SECTIONS directive, see the linker description
information in the TMS320C6000 Assembly Language Tools User's Guide.

6.13.2 Initializing Static and Global Variables With the const Type Qualifier
Static and global variables of type const without explicit initializations are similar to other static and global
variables because they might not be preinitialized to 0 (for the same reasons discussed in Section 6.13).
For example:
const int zero; /* may not be initialized to 0 */

However, the initialization of const global and static variables is different because these variables are
declared and initialized in a section called .const. For example:
const int zero = 0 /* guaranteed to be 0 */

This corresponds to an entry in the .const section:

.sect .const
_zero
.word 0

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 175

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Changing the ANSI/ISO C Language Mode www.ti.com

This feature is particularly useful for declaring a large table of constants, because neither time nor space
is wasted at system startup to initialize the table. Additionally, the linker can be used to place the .const
section in ROM.
You can use the DATA_SECTION pragma to put the variable in a section other than .const. For example,
the following C code:
#pragma DATA_SECTION (var, ".mysect");
const int zero=0;

is compiled into this assembly code:

.sect .mysect
_zero
.word 0

6.14 Changing the ANSI/ISO C Language Mode

The --kr_compatible, --relaxed_ansi, and --strict_ansi options let you specify how the C/C++ compiler
interprets your source code. You can compile your source code in the following modes:
• Normal ANSI/ISO mode
• K&R C mode
• Relaxed ANSI/ISO mode
• Strict ANSI/ISO mode
The default is normal ANSI/ISO mode. Under normal ANSI/ISO mode, most ANSI/ISO violations are
emitted as errors. Strict ANSI/ISO violations (those idioms and allowances commonly accepted by C/C++
compilers, although violations with a strict interpretation of ANSI/ISO), however, are emitted as warnings.
Language extensions, even those that conflict with ANSI/ISO C, are enabled.
K&R C mode does not apply to C++ code.

6.14.1 Compatibility With K&R C (--kr_compatible Option)

The ANSI/ISO C/C++ language is a superset of the de facto C standard defined in Kernighan and
Ritchie's The C Programming Language. Most programs written for other non-ANSI/ISO compilers
correctly compile and run without modification.
There are subtle changes, however, in the language that can affect existing code. Appendix C in The C
Programming Language (second edition, referred to in this manual as K&R) summarizes the differences
between ANSI/ISO C and the first edition's C standard (the first edition is referred to in this manual as
K&R C).
To simplify the process of compiling existing C programs with the ANSI/ISO C/C++ compiler, the compiler
has a K&R option (--kr_compatible) that modifies some semantic rules of the language for compatibility
with older code. In general, the --kr_compatible option relaxes requirements that are stricter for ANSI/ISO
C than for K&R C. The --kr_compatible option does not disable any new features of the language such as
function prototypes, enumerations, initializations, or preprocessor constructs. Instead, --kr_compatible
simply liberalizes the ANSI/ISO rules without revoking any of the features.
The specific differences between the ANSI/ISO version of C and the K&R version of C are as follows:
• The integral promotion rules have changed regarding promoting an unsigned type to a wider signed
type. Under K&R C, the result type was an unsigned version of the wider type; under ANSI/ISO, the
result type is a signed version of the wider type. This affects operations that perform differently when
applied to signed or unsigned operands; namely, comparisons, division (and mod), and right shift:
unsigned short u;
int i;
if (u < i) /* SIGNED comparison, unless --kr_compatible used */
• ANSI/ISO prohibits combining two pointers to different types in an operation. In most K&R compilers,
this situation produces only a warning. Such cases are still diagnosed when --kr_compatible is used,
but with less severity:
int *p;
char *q = p; /* error without --kr_compatible, warning with --kr_compatible */

176 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Changing the ANSI/ISO C Language Mode

• External declarations with no type or storage class (only an identifier) are illegal in ANSI/ISO but legal
in K&R:
a; /* illegal unless --kr_compatible used */
• ANSI/ISO interprets file scope definitions that have no initializers as tentative definitions. In a single
module, multiple definitions of this form are fused together into a single definition. Under K&R, each
definition is treated as a separate definition, resulting in multiple definitions of the same object and
usually an error. For example:
int a;
int a; /* illegal if --kr_compatible used, OK if not */
Under ANSI/ISO, the result of these two definitions is a single definition for the object a. For most K&R
compilers, this sequence is illegal, because int a is defined twice.
• ANSI/ISO prohibits, but K&R allows objects with external linkage to be redeclared as static:
extern int a;
static int a; /* illegal unless --kr_compatible used */
• Unrecognized escape sequences in string and character constants are explicitly illegal under ANSI/ISO
but ignored under K&R:
char c = '\q'; /* same as 'q' if --kr_compatible used, error if not */
• ANSI/ISO specifies that bit fields must be of type int or unsigned. With --kr_compatible, bit fields can
be legally defined with any integral type. For example:
struct s
{
short f : 2; /* illegal unless --kr_compatible used */
};
• K&R syntax allows a trailing comma in enumerator lists:
enum { a, b, c, }; /* illegal unless --kr_compatible used */
• K&R syntax allows trailing tokens on preprocessor directives:
#endif NAME /* illegal unless --kr_compatible used */

6.14.2 Enabling Strict ANSI/ISO Mode and Relaxed ANSI/ISO Mode (--strict_ansi and --
relaxed_ansi Options)
Use the --strict_ansi option when you want to compile under strict ANSI/ISO mode. In this mode, error
messages are provided when non-ANSI/ISO features are used, and language extensions that could
invalidate a strictly conforming program are disabled. Examples of such extensions are the inline and asm
keywords.
Use the --relaxed_ansi option when you want the compiler to ignore strict ANSI/ISO violations rather than
emit a warning (as occurs in normal ANSI/ISO mode) or an error message (as occurs in strict ANSI/ISO
mode). In relaxed ANSI/ISO mode, the compiler accepts extensions to the ANSI/ISO C standard, even
when they conflict with ANSI/ISO C. The GCC language extensions described in Section 6.15 are
available in relaxed ANSI/ISO mode.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 177

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
GNU Language Extensions www.ti.com

6.14.3 Enabling Embedded C++ Mode (--embedded_cpp Option)

The compiler supports the compilation of embedded C++. In this mode, some features of C++ are
removed that are of less value or too expensive to support in an embedded system. When compiling for
embedded C++, the compiler generates diagnostics for the use of omitted features.
Embedded C++ is enabled by compiling with the --embedded_cpp option.
Embedded C++ omits these C++ features:
• Templates
• Exception handling
• Run-time type information
• The new cast syntax
• The keyword mutable
• Multiple inheritance
• Virtual inheritance
Under the standard definition of embedded C++, namespaces and using-declarations are not supported.
The C6000 compiler nevertheless allows these features under embedded C++ because the C++ run-time-
support library makes use of them. Furthermore, these features impose no run-time penalty.
The compiler does not support embedded C++ run-time-support libraries.

6.15 GNU Language Extensions

The GNU compiler collection (GCC) defines a number of language features not found in the ANSI/ISO C
and C++ standards. The definition and examples of these extensions (for GCC version 3.4) can be found
at the GNU web site, http://gcc.gnu.org/onlinedocs/gcc-3.4.6/gcc/C-Extensions.html.
Most of these extensions are also available for C++ source code.

6.15.1 Extensions
Most of the GCC language extensions are available in the TI compiler when compiling in relaxed ANSI
mode (--relaxed_ansi) or if the --gcc option is used.
The extensions that the TI compiler supports are listed in Table 6-4, which is based on the list of
extensions found at the GNU web site. The shaded rows describe extensions that are not supported.

Table 6-4. GCC Language Extensions

Extensions Descriptions
Statement expressions Putting statements and declarations inside expressions (useful for creating smart 'safe' macros)
Local labels Labels local to a statement expression
Labels as values Pointers to labels and computed gotos
Nested functions As in Algol and Pascal, lexical scoping of functions
Constructing calls Dispatching a call to another function
Naming types (1) Giving a name to the type of an expression
typeof operator typeof referring to the type of an expression
Generalized lvalues Using question mark (?) and comma (,) and casts in lvalues
Conditionals Omitting the middle operand of a ?: expression
long long Double long word integers and long long int type
Hex floats Hexadecimal floating-point constants
Complex Data types for complex numbers
Zero length Zero-length arrays
Variadic macros Macros with a variable number of arguments
Variable length Arrays whose length is computed at run time

(1)
Feature defined for GCC 3.0; definition and examples at http://gcc.gnu.org/onlinedocs/gcc-3.0.4/gcc/C-Extensions.html

178 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com GNU Language Extensions

Table 6-4. GCC Language Extensions (continued)

Extensions Descriptions
Empty structures Structures with no members
Subscripting Any array can be subscripted, even if it is not an lvalue.
Escaped newlines Slightly looser rules for escaped newlines
Multi-line strings (1) String literals with embedded newlines
Pointer arithmetic Arithmetic on void pointers and function pointers
Initializers Non-constant initializers
Compound literals Compound literals give structures, unions, or arrays as values
Designated initializers Labeling elements of initializers
Cast to union Casting to union type from any member of the union
Case ranges 'Case 1 ... 9' and such
Mixed declarations Mixing declarations and code
Function attributes Declaring that functions have no side effects, or that they can never return
Attribute syntax Formal syntax for attributes
Function prototypes Prototype declarations and old-style definitions
C++ comments C++ comments are recognized.
Dollar signs A dollar sign is allowed in identifiers.
Character escapes The character ESC is represented as \e
Variable attributes Specifying the attributes of variables
Type attributes Specifying the attributes of types
Alignment Inquiring about the alignment of a type or variable
Inline Defining inline functions (as fast as macros)
Assembly labels Specifying the assembler name to use for a C symbol
Extended asm Assembler instructions with C operands
Constraints Constraints for asm operands
Alternate keywords Header files can use __const__, __asm__, etc
Explicit reg vars Defining variables residing in specified registers
Incomplete enum types Define an enum tag without specifying its possible values
Function names Printable strings which are the name of the current function
Return address Getting the return or frame address of a function (limited support)
Other built-ins Other built-in functions (see Section 6.15.5)
Vector extensions Using vector instructions through built-in functions
Target built-ins Built-in functions specific to particular targets
Pragmas Pragmas accepted by GCC
Unnamed fields Unnamed struct/union fields within structs/unions
Thread-local Per-thread variables

6.15.2 Function Attributes

The following GCC function attributes are supported: always_inline, const, constructor, deprecated,
format, format_arg, malloc, noinline, noreturn, pure, section, unused, used and warn_unused_result.
In addition, the visibility function attribute is supported for EABI mode (--abi=eabi).
The format attribute is applied to the declarations of printf, fprintf, sprintf, snprintf, vprintf, vfprintf, vsprintf,
vsnprintf, scanf, fscanf, vfscanf, vscanf, vsscanf, and sscanf in stdio.h. Thus when GCC extensions are
enabled, the data arguments of these functions are type checked against the format specifiers in the
format string argument and warnings are issued when there is a mismatch. These warnings can be
suppressed in the usual ways if they are not desired.
The malloc attribute is applied to the declarations of malloc, calloc, realloc and memalign in stdlib.h.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 179

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
GNU Language Extensions www.ti.com

6.15.3 Variable Attributes

The following variable attributes are supported: aligned, deprecated, mode, packed, section,
transparent_union, unused, and used.
The used attribute is defined in GCC 4.2 (see http://gcc.gnu.org/onlinedocs/gcc-4.2.4/gcc/Variable-
Attributes.html#Variable-Attributes).
The packed attribute for structure and union types is available only when there is hardware support for
unaligned accesses. For C6000 this means C6400+, C6400, C6740, and C6600.
In addition, the weak variable attribute is supported for EABI mode (--abi=eabi).

6.15.4 Type Attributes

The following type attributes are supported: aligned, deprecated, packed, transparent_union, and unused.
In addition, the visibility type attribute is supported for EABI mode (--abi=eabi).
The packed attribute on struct and union types is available only for target architectures that have hardware
support for unaligned access (such as C64x+, C64x).
Members of a packed structure are stored as closely to each other as possible, omitting additional bytes of
padding usually added to preserve word-alignment. For example, assuming a word-size of 4 bytes
ordinarily has 3 bytes of padding between members c1 and i, and another 3 bytes of trailing padding after
member c2, leading to a total size of 12 bytes:
struct unpacked_struct { char c1; int i; char c2;};

However, the members of a packed struct are byte-aligned. Thus the following does not have any bytes of
padding between or after members and totals 6 bytes:
struct __attribute__((__packed__)) packed_struct { char c1; int i; char c2; };

Subsequently, packed structures in an array are packed together without trailing padding between array
elements.
Bit fields of a packed structure are bit-aligned. The byte alignment of adjacent struct members that are not
bit fields does not change. However, there are no bits of padding between adjacent bit fields.
The packed attribute can only be applied to the original definition of a structure or union type. It cannot be
applied with a typedef to a non-packed structure that has already been defined, nor can it be applied to
the declaration of a struct or union object. Therefore, any given structure or union type can only be packed
or non-packed, and all objects of that type will inherit its packed or non-packed attribute.
The packed attribute is not applied recursively to structure types that are contained within a packed
structure. Thus, in the following example the member s retains the same internal layout as in the first
example above. There is no padding between c and s, so s falls on an unaligned boundary:
struct __attribute__((__packed__)) outer_packed_struct { char c; struct unpacked_struct s; };

It is illegal to implicitly or explicitly cast the address of a packed struct member as a pointer to any non-
packed type except an unsigned char. In the following example, p1, p2, and the call to foo are all illegal.
void foo(int *param);
struct packed_struct ps;

int *p1 = &ps.i;

int *p2 = (int *)&ps.i;
foo(&ps.i);

However, it is legal to explicitly cast the address of a packed struct member as a pointer to an unsigned
char:
unsigned char *pc = (unsigned char *)&ps.i;

The TI compiler also supports an unpacked attribute for an enumeration type to allow you to indicate that
the representation is to be an integer type that is no smaller than int; in other words, it is not packed.

180 TMS320C6000C/C++ Language Implementation SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com GNU Language Extensions

6.15.5 Built-In Functions

The following builtin functions are supported: __builtin_abs, __builtin_classify_type, __builtin_constant_p,
__builtin_expect, __builtin_fabs, __builtin_fabsf, __builtin_frame_address, __builtin_labs,
__builtin_memcpy, and __builtin_return_address.
The __builtin_frame_address function returns zero unless the argument is a constant zero.
The __builtin_return_address function always returns zero.

SPRU187U – July 2012 TMS320C6000C/C++ Language Implementation 181

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Chapter 7
SPRU187U – July 2012

Run-Time Environment

This chapter describes the TMS320C6000 C/C++ run-time environment. To ensure successful execution
of C/C++ programs, it is critical that all run-time code maintain this environment. It is also important to
follow the guidelines in this chapter if you write assembly language functions that interface with C/C++
code.

Topic ........................................................................................................................... Page

7.1 Memory Model ................................................................................................. 183

7.2 Object Representation ...................................................................................... 188
7.3 Register Conventions ....................................................................................... 196
7.4 Function Structure and Calling Conventions ....................................................... 197
7.5 Interfacing C and C++ With Assembly Language ................................................. 200
7.6 Interrupt Handling ............................................................................................ 230
7.7 Run-Time-Support Arithmetic Routines .............................................................. 232
7.8 System Initialization ......................................................................................... 234
7.9 Support for Multi-Threaded Applications ............................................................ 245

182 Run-Time Environment SPRU187U – July 2012

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com Memory Model

7.1 Memory Model

The C6000 compiler treats memory as a single linear block that is partitioned into subblocks of code and
data. Each subblock of code or data generated by a C program is placed in its own continuous memory
space. The compiler assumes that a full 32-bit address space is available in target memory.

The Linker Defines the Memory Map

NOTE: The linker, not the compiler, defines the memory map and allocates code and data into
target memory. The compiler assumes nothing about the types of memory available, about
any locations not available for code or data (holes), or about any locations reserved for I/O or
control purposes. The compiler produces relocatable code that allows the linker to allocate
code and data into the appropriate memory spaces.
For example, you can use the linker to allocate global variables into on-chip RAM or to
allocate executable code into external ROM. You can allocate each block of code or data
individually into memory, but this is not a general practice (an exception to this is memory-
mapped I/O, although you can access physical memory locations with C/C++ pointer types).

7.1.1 Sections
The compiler produces relocatable blocks of code and data called sections. The sections are allocated
into memory in a variety of ways to conform to a variety of system configurations. For more information
about sections and allocating them, see the introductory object file information in the TMS320C6000
Assembly Language Tools User's Guide.
There are two basic types of sections:
• Initialized sections contain data or executable code. The C/C++ compiler creates the following
initialized sections:
– The .args section contains the command argument for a host-based loader. This section is read-
only. See the --arg_size option for details.
– For EABI only, the .binit section contains boot time copy tables. For details on BINIT, see the
TMS320C6000 Assembly Language Tools User's Guide for linker command file information.
– For COFF ABI only, the .cinit section contains tables for initializing variables and constants.
– The .pinit section for COFF ABI, or the .init_array section for EABI, contains the table for calling
global constructor tables.
– For EABI only, the .c6xabi.exidx section contains the index table for exception handling. The
.c6xabi.extab section contains un-winded instructions for exception handling. These sections are
read-only. See the --exceptions option for details.
– The .name.load section contains the compressed image of section name. This section is read-
only. See the TMS320C6000 Assembly Language Tools User's Guide for information on copy
tables.
– The .ppinfo section contains correlation tables and the .ppdata section contains data tables for
compiler-based profiling. See the --gen_profile_info option for details.
– The .const section contains string literals, floating-point constants, and data defined with the
C/C++ qualifiers far and const (provided the constant is not also defined as volatile).
– For EABI only, the .fardata section reserves space for non-const, initialized far global and static
variables.
– For EABI only, the .neardata section reserves space for non-const, initialized near global and
static variables.
– For EABI only, the .rodata section reserves space for const near global and static variables.
– The .switch section contains jump tables for large switch statements.
– The .text section contains all the executable code.

SPRU187U – July 2012 Run-Time Environment 183

• Uninitialized sections reserve space in memory (usually RAM). A program can use this space at run
time to create and store variables. The compiler creates the following uninitialized sections:
– For COFF ABI only, the .bss section reserves space for global and static variables. When you
specify the --rom_model linker option, at program startup, the C boot routine copies data out of the
.cinit section (which can be in ROM) and stores it in the .bss section. The compiler defines the
global symbol $bss and assigns $bss the value of the starting address of the .bss section.
– For EABI only, the .bss section reserves space for uninitialized global and static variables.
– The .far section reserves space for global and static variables that are declared far.
– The .stack section reserves memory for the system stack.
– The .sysmem section reserves space for dynamic memory allocation. The reserved space is used
by dynamic memory allocation routines, such as malloc, calloc, realloc, or new. If a C/C++ program
does not use these functions, the compiler does not create the .sysmem section.

Use Only Code in Program Memory

NOTE: With the exception of code sections, the initialized and uninitialized sections cannot be
allocated into internal program memory.

The assembler creates the default sections .text, .bss, and .data. The C/C++ compiler, however, does not
use the .data section. You can instruct the compiler to create additional sections by using the
CODE_SECTION and DATA_SECTION pragmas (see Section 6.9.3 and Section 6.9.6).

7.1.2 C/C++ System Stack

The C/C++ compiler uses a stack to:
• Save function return addresses
• Allocate local variables
• Pass arguments to functions
• Save temporary results

The run-time stack grows from the high addresses to the low addresses. The compiler uses the B15
register to manage this stack. B15 is the stack pointer (SP), which points to the next unused location on
the stack.
The linker sets the stack size, creates a global symbol, __TI_STACK_SIZE, and assigns it a value equal
to the stack size in bytes. The default stack size is 1K bytes. You can change the stack size at link time by
using the --stack_size option with the linker command. For more information on the --stack_size option,
see the linker description chapter in the TMS320C6000 Assembly Language Tools User's Guide.
At system initialization, SP is set to the first 8-byte aligned address before the end (highest numerical
address) of the .stack section. For C6600, SP is set to the first 16-byte aligned address. Since the position
of the stack depends on where the .stack section is allocated, the actual address of the stack is
determined at link time.
The C/C++ environment automatically decrements SP at the entry to a function to reserve all the space
necessary for the execution of that function. The stack pointer is incremented at the exit of the function to
restore the stack to the state before the function was entered. If you interface assembly language routines
to C/C++ programs, be sure to restore the stack pointer to the same state it was in before the function
was entered.
For more information about the stack and stack pointer, see Section 7.4.

Unaligned SP Can Cause Application Crash

NOTE: The HWI dispatcher uses SP during an interrupt call regardless of SP alignment. Therefore,
SP must never be misaligned, even for 1 cycle.

184 Run-Time Environment SPRU187U – July 2012

Stack Overflow
NOTE: The compiler provides no means to check for stack overflow during compilation or at run
time. A stack overflow disrupts the run-time environment, causing your program to fail. Be
sure to allow enough space for the stack to grow. You can use the --entry_hook option to
add code to the beginning of each function to check for stack overflow; see Section 2.17.

7.1.3 Dynamic Memory Allocation

The run-time-support library supplied with the C6000 compiler contains several functions (such as malloc,
calloc, and realloc) that allow you to allocate memory dynamically for variables at run time.
Memory is allocated from a global pool, or heap, that is defined in the .sysmem section. You can set the
size of the .sysmem section by using the --heap_size=size option with the linker command. The linker also
creates a global symbol, __TI_SYSMEM_SIZE, and assigns it a value equal to the size of the heap in
bytes. The default size is 1K bytes. For more information on the --heap_size option, see the linker
description chapter in the TMS320C6000 Assembly Language Tools User's Guide.
Dynamically allocated objects are not addressed directly (they are always accessed with pointers) and the
memory pool is in a separate section (.sysmem); therefore, the dynamic memory pool can have a size
limited only by the amount of available memory in your system. To conserve space in the .bss section,
you can allocate large arrays from the heap instead of defining them as global or static. For example,
instead of a definition such as:
struct big table[100];

Use a pointer and call the malloc function:

struct big *table
table = (struct big *)malloc(100*sizeof(struct big));

7.1.4 Initialization of Variables in COFF ABI

The C/C++ compiler produces code that is suitable for use as firmware in a ROM-based system. In such a
system, the initialization tables in the .cinit section are stored in ROM. At system initialization time, the
C/C++ boot routine copies data from these tables (in ROM) to the initialized variables in .bss (RAM).
In situations where a program is loaded directly from an object file into memory and run, you can avoid
having the .cinit section occupy space in memory. A loader can read the initialization tables directly from
the object file (instead of from ROM) and perform the initialization directly at load time instead of at run
time. You can specify this to the linker by using the --ram_model link option. For more information, see
Section 7.8.

7.1.5 Data Memory Models

Several options extend the C6000 data addressing model.

7.1.5.1 Determining the Data Address Model

As of the 5.1.0 version of the compiler tools, if a near or far keyword is not specified for an object, the
compiler generates far accesses to aggregate data and near accesses to all other data. This means that
structures, unions, C++ classes, and arrays are not accessed through the data-page (DP) pointer.
Non-aggregate data, by default, is placed in the .bss section and is accessed using relative-offset
addressing from the data page pointer (DP, which is B14). DP points to the beginning of the .bss section.
Accessing data via the data page pointer is generally faster and uses fewer instructions than the
mechanism used for far data accesses.
If you want to use near accesses to aggregate data, you must specify the --mem_model:data=near option,
or declare your data with the near keyword.
If you have too much static and extern data to fit within a 15-bit scaled offset from the beginning of the
.bss section, you cannot use --mem_model:data=near. The linker will issue an error message if there is a
DP-relative data access that will not reach.

SPRU187U – July 2012 Run-Time Environment 185

The --mem_model:data=type option controls how data is accessed:

--mem_model:data=near Data accesses default to near

--mem_model:data=far Data accesses default to far
--mem_model:data=far_aggregates Data accesses to aggregate data default to far, data
accesses to non-aggregate data default to near. This is
the default behavior.

The --mem_model:data options do not affect the access to objects explicitly declared with the near of far
keyword.
By default, all run-time-support data is defined as far.
For more information on near and far accesses to data, see Section 6.5.4.

7.1.5.2 Using DP-Relative Addressing

The default behavior of the compiler is to use DP-relative addressing for near (.bss) data, and absolute
addressing for all other (far) data. The --dprel option specifies that all data, including const data and far
data, is addressed using DP-relative addressing.
The purpose of the --dprel option is to support a shared object model so multiple applications running
simultaneously can share code, but each have their own copy of the data.
The --dprel option is supported for ELF only.

7.1.5.3 Const Objects as Far

The --mem_model:const option allows const objects to be made far independently of the --
mem_model:data option. This enables an application with a small amount of non-const data but a large
amount of const data to move the const data out of .bss. Also, since consts can be shared, but .bss
cannot, it saves memory by moving the const data into .const.
The --mem_model:const=type option has the following values:

--mem_model:const=data Const objects are placed according to the --

mem_model:data option. This is the default behavior.
--mem_model:const=far Const objects default to far independent of the --
mem_model:data option.
--mem_model:const=far_aggregates Const aggregate objects default to far, scalar consts
default to near.

Consts that are declared far, either explicitly through the far keyword or implicitly using --
mem_model:const are always placed in the .const section.

7.1.6 Trampoline Generation for Function Calls

The C6000 compiler generates trampolines by default. Trampolines are a method for modifying function
calls at link time to reach destinations that would normally be too far away. When a function call is more
than +/- 1M instructions away from its destination, the linker will generate an indirect branch (or
trampoline) to that destination, and will redirect the function call to point to the trampoline. The end result
is that these function calls branch to the trampoline, and then the trampoline branches to the final
destination. With trampolines, you no longer need to specify memory model options to generate far calls.

186 Run-Time Environment SPRU187U – July 2012

7.1.7 Position Independent Data

Near global and static data are stored in the .bss section. All near data for a program must fit within 32K
bytes of memory. This limit comes from the addressing mode used to access near data, which is limited to
a 15-bit unsigned offset from DP (B14), which is the data page pointer.
For some applications, it may be desirable to have multiple data pages with separate instances of near
data. For example, a multi-channel application may have multiple copies of the same program running
with different data pages. The functionality is supported by the C6000 compiler's memory model, and is
referred to as position independent data.
Position independent data means that all near data accesses are relative to the data page (DP) pointer,
allowing for the DP to be changed at run time. There are three areas where position independent data is
implemented by the compiler:
• Near direct memory access
STW B4,*DP(_a)
.global _a
.bss _a,4,4
All near direct accesses are relative to the DP.
• Near indirect memory access
MVK (_a - $bss),A0
ADD DP,A0,A0
The expression (_a - $bss) calculates the offset of the symbol _a from the start of the .bss section. The
compiler defines the global $bss in generated assembly code. The value of $bss is the starting
address of the .bss section.
• Initialized near pointers
The .cinit record for an initialized near pointer value is stored as an offset from the beginning of the
.bss section. During the autoinitialization of global variables, the data page pointer is added to these
offsets. (See Section 7.8.3.)

SPRU187U – July 2012 Run-Time Environment 187

7.2 Object Representation

This section explains how various data objects are sized, aligned, and accessed.

7.2.1 Data Type Storage

Table 7-1 lists register and memory storage for various data types:

Table 7-1. Data Representation in Registers and Memory

Data Type Register Storage Memory Storage
char Bits 0-7 of register 8 bits aligned to 8-bit boundary
unsigned char Bits 0-7 of register 8 bits aligned to 8-bit boundary
short Bits 0-15 of register 16 bits aligned to 16-bit boundary
unsigned short Bits 0-15 of register 16 bits aligned to 16-bit boundary
int Entire register 32 bits aligned to 32-bit boundary
unsigned int Entire register 32 bits aligned to 32-bit boundary
enum Entire register 32 bits aligned to 32-bit boundary
float Entire register 32 bits aligned to 32-bit boundary
long (EABI) Entire register 32 bits aligned to 32-bit boundary
unsigned long (EABI) Entire register 32 bits aligned to 32-bit boundary
long (COFF ABI) Bits 0-39 of even/odd register pair 64 bits aligned to 64-bit boundary
unsigned long (COFF ABI) Bits 0-39 of even/odd register pair 64 bits aligned to 64-bit boundary
__int40_t Even/odd register pair 64 bits aligned to 64-bit boundary
unsigned __int40_t Even/odd register pair 64 bits aligned to 64-bit boundary
long long Even/odd register pair 64 bits aligned to 64-bit boundary
unsigned long long Even/odd register pair 64 bits aligned to 64-bit boundary
double Even/odd register pair 64 bits aligned to 64-bit boundary
long double Even/odd register pair 64 bits aligned to 64-bit boundary
__x128_t (C6600 only) (1) Register quad 128-bits aligned to 128-bit boundary
struct Members are stored as their individual types Multiple of 8 bits aligned to boundary of largest
require. member type; members are stored and aligned
as their individual types require.
array Members are stored as their individual types Members are stored as their individual types
require. require. (2) All arrays inside a structure are
aligned according to the type of each element in
the array.
pointer to data member Bits 0-31 of register 32 bits aligned to 32-bit boundary
pointer to member function Components stored as their individual types 64 bits aligned to 32-bit boundary
require
(1)
For details on the __x128_t container type see Section 7.5.6.
(2)
For C6400, C6400+, C6740, and C6600, aligned to a 64-bit boundary. For C6200, C6700, and C6700+, aligned to a 32-bit
boundary for all types 32 bits and smaller, and to a 64-bit boundary for all types larger than 32 bits. For C6600, aligned to a 128-
bit boundary.

188 Run-Time Environment SPRU187U – July 2012

7.2.1.1 char and short Data Types (signed and unsigned)

The char and unsigned char data types are stored in memory as a single byte and are loaded to and
stored from bits 0-7 of a register (see Figure 7-1). Objects defined as short or unsigned short are stored in
memory as two bytes at a halfword (2 byte) aligned address and they are loaded to and stored from bits
0-15 of a register (see Figure 7-1).
In big-endian mode, 2-byte objects are loaded to registers by moving the first byte (that is, the lower
address) of memory to bits 8-15 of the register and moving the second byte of memory to bits 0-7. In little-
endian mode, 2-byte objects are loaded to registers by moving the first byte (that is, the lower address) of
memory to bits 0-7 of the register and moving the second byte of memory to bits 8-15.

Figure 7-1. Char and Short Data Storage Format

Signed 8-bit char
MS LS

S S S S S S S S S S S S S S S S S S S S S S S S S S I I I I I I
31 7 0

Unsigned 8-bit char

MS LS

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 U U U U U U U
31 7 0

Signed 16-bit short

MS LS

S S S S S S S S S S S S S S S S S I I I I I I I I I I I I I I I
31 15 0

Unsigned 16-bit short

MS LS

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 U U U U U U U U U U U U U U U U
31 15 0
LEGEND: S = sign, I = signed integer, U = unsigned integer, MS = most significant, LS = least significant

SPRU187U – July 2012 Run-Time Environment 189

7.2.1.2 enum, int, and long (EABI) Data Types (signed and unsigned)
The int, unsigned int, and enum data types are stored in memory as 32-bit objects (see Figure 7-2).
Objects of these types are loaded to and stored from bits 0-31 of a register. In big-endian mode, 4-byte
objects are loaded to registers by moving the first byte (that is, the lower address) of memory to bits 24-31
of the register, moving the second byte of memory to bits 16-23, moving the third byte to bits 8-15, and
moving the fourth byte to bits 0-7. In little-endian mode, 4-byte objects are loaded to registers by moving
the first byte (that is, the lower address) of memory to bits 0-7 of the register, moving the second byte to
bits 8-15, moving the third byte to bits 16-23, and moving the fourth byte to bits 24-31.

Figure 7-2. 32-Bit Data Storage Format

Signed 32-bit integer, or enum char
MS LS
S I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
31 0

Unsigned 32-bit integer

MS LS
U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U
31 0
LEGEND: S = sign, U = unsigned integer, I = signed integer, MS = most significant, LS = least significant

7.2.1.3 float Data Type

The float data type is stored in memory as 32-bit objects (see Figure 7-3). Objects defined as float are
loaded to and stored from bits 0-31 of a register. In big-endian mode, 4-byte objects are loaded to
registers by moving the first byte (that is, the lower address) of memory to bits 24-31 of the register,
moving the second byte of memory to bits 16-23, moving the third byte to bits 8-15, and moving the fourth
byte to bits 0-7. In little-endian mode, 4-byte objects are loaded to registers by moving the first byte (that
is, the lower address) of memory to bits 0-7 of the register, moving the second byte to bits 8-15, moving
the third byte to bits 16-23, and moving the fourth byte to bits 24-31.

Figure 7-3. Single-Precision Floating-Point Char Data Storage Format

MS LS
S E E E E E E E E M M M M M M M M M M M M M M M M M M M M M M M
31 23 0
LEGEND: S = sign, M = mantissa, E = exponent, MS = most significant, LS = least significant

190 Run-Time Environment SPRU187U – July 2012

7.2.1.4 __int40_t and COFF ABI long Data Types (signed and unsigned)
Long and unsigned long data types are stored in an odd/even pair of registers (see Figure 7-4) and are
always referenced as a pair in the format of odd register:even register (for example, A1:A0). In little-endian
mode, the lower address is loaded into the even register and the higher address is loaded into the odd
register; if data is loaded from location 0, then the byte at 0 is the lowest byte of the even register. In big-
endian mode, the higher address is loaded into the even register and the lower address is loaded into the
odd register; if data is loaded from location 0, then the byte at 0 is the highest byte of the odd register but
is ignored.

Figure 7-4. 40-Bit Data Storage Format Signed __int40_t or 40-bit long
Odd register
MS
X X X X X X X X X X X X X X X X X X X X X X X X X S I I I I I I
31 8 7 6 0

Even register
LS
I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
31 0
LEGEND: S = sign, U = unsigned integer, I = signed integer, X = unused, MS = most significant, LS = least significant

Figure 7-5. Unsigned 40-bit __int40_t or long

Odd register
MS
X X X X X X X X X X X X X X X X X X X X X X X X X U U U U U U U
31 8 7 0

Even register
LS
U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U
31 0
LEGEND: S = sign, U = unsigned integer, I = signed integer, X = unused, MS = most significant, LS = least significant

SPRU187U – July 2012 Run-Time Environment 191

7.2.1.5 long long Data Types (signed and unsigned)

Long long and unsigned long long data types are stored in an odd/even pair of registers (see Figure 7-6)
and are always referenced as a pair in the format of odd register:even register (for example, A1:A0). In
little-endian mode, the lower address is loaded into the even register and the higher address is loaded into
the odd register; if data is loaded from location 0, then the byte at 0 is the lowest byte of the even register.
In big-endian mode, the higher address is loaded into the even register and the lower address is loaded
into the odd register; if data is loaded from location 0, then the byte at 0 is the highest byte of the odd
register.

Figure 7-6. 64-Bit Data Storage Format Signed 64-bit long

Odd register
MS
S I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
31 0

Figure 7-7. Unsigned 64-bit long

Odd register
MS
U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U U
31 0

192 Run-Time Environment SPRU187U – July 2012

7.2.1.6 double and long double Data Types

Double and long double data types are stored in an odd/even pair of registers (see Figure 7-8) and can
only exist in a register in one format: as a pair in the format of odd register:even register (for example,
A1:A0). The odd memory word contains the sign bit, exponent, and the most significant part of the
mantissa. The even memory word contains the least significant part of the mantissa. In little-endian mode,
the lower address is loaded into the even register and the higher address is loaded into the odd register.
In big-endian mode, the higher address is loaded into the even register and the lower address is loaded
into the odd register. In little-endian mode, if code is loaded from location 0, then the byte at 0 is the
lowest byte of the even register. In big-endian mode, if code is loaded from location 0, then the byte at 0 is
the highest byte of the odd register.

Figure 7-8. Double-Precision Floating-Point Data Storage Format

Odd register
MS
S E E E E E E E E E E E E M M M M M M M M M M M M M M M M M M M
31 20 0

Even register
LS
M M M M M M M M M M M M M M M M M M M M M M M M M M M M M M M M
31 0
LEGEND: S = sign, M = mantissa, E = exponent, MS = most significant, LS = least significant

7.2.1.7 Pointer to Data Member Types

Pointer to data member objects are stored in memory like an unsigned int (32 bit) integral type. Its value is
the byte offset to the data member in the class, plus 1. The zero value is reserved to represent the NULL
pointer.

7.2.1.8 Pointer to Member Function Types

Pointer to member function objects are stored as a structure with three members, and the layout is
equivalent to:
struct {
short int d;
short int i;
union {
void (*f) ();
int 0; }
};

The parameter d is the offset to be added to the beginning of the class object for this pointer. The
parameter I is the index into the virtual function table, offset by 1. The index enables the NULL pointer to
be represented. Its value is -1 if the function is nonvirtual. The parameter f is the pointer to the member
function if it is nonvirtual, when I is 0. The 0 is the offset to the virtual function pointer within the class
object.

7.2.1.9 Structures and Arrays

A nested structure is aligned to a boundary required by the largest type it contains. For example, if the
largest type in a nested structure is of type short, then the nested structure is aligned to a 2-byte
boundary. If the largest type in a nested structure is of type long, unsigned long, double, or long double,
then the nested structure is aligned to an 8-byte boundary.
Structures always reserve memory in multiples of the size of the largest element type. For example, if a
structure contains an int, unsigned int, or float, a multiple of 4 bytes of storage is reserved in memory.
Members of structures are stored in the same manner as if they were individual objects. An array member
in a struct is aligned to the natural boundary of its elements.

SPRU187U – July 2012 Run-Time Environment 193

Top-level arrays are aligned on an 8-byte boundary for C6400, C6400+, C6740, and C6600, and either a
4-byte (for all element types of 32 bits or smaller) or an 8-byte boundary for C6200, C6700, or C6700+.
Top-level arrays are aligned on a 16-byte boundary for C6600. Elements of arrays are stored in the same
manner as if they were individual objects.

7.2.2 Bit Fields

Bit fields are handled differently in COFF ABI and EABI modes. Section 7.2.2.1 details how bit fields are
handled in all modes. Section 7.2.2.2 details how bit fields differ in EABI mode.

7.2.2.1 Generic Bit Fields

Bit fields are the only objects that are packed within a byte. That is, two bit fields can be stored in the
same byte. Bit fields can range in size from 1 to 32 bits for COFF ABI, and 1 to 64 bits in C or larger in
C++ for EABI.
For big-endian mode, bit fields are packed into registers from most significant bit (MSB) to least significant
bit (LSB) in the order in which they are defined. Bit fields are packed in memory from most significant byte
(MSbyte) to least significant byte (LSbyte). For little-endian mode, bit fields are packed into registers from
the LSB to the MSB in the order in which they are defined, and packed in memory from LSbyte to MSbyte.
Figure 7-9 illustrates bit-field packing, using the following bit field definitions:
struct{
int A:7
int B:10
int C:3
int D:2
int E:9
}x;

A0 represents the least significant bit of the field A; A1 represents the next least significant bit, etc. Again,
storage of bit fields in memory is done with a byte-by-byte, rather than bit-by-bit, transfer.

Figure 7-9. Bit-Field Packing in Big-Endian and Little-Endian Formats

Big-endian register
MS LS
A A A A A A A B B B B B B B B B B C C C D D E E E E E E E E E X
6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 2 1 0 1 0 8 7 6 5 4 3 2 1 0 X
31 0

Big-endian memory
Byte 0 Byte 1 Byte 2 Byte 3
A A A A A A A B B B B B B B B B B C C C D D E E E E E E E E E X
6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 2 1 0 1 0 8 7 6 5 4 3 2 1 0 X

Little-endian register
MS LS
X E E E E E E E E E D D C C C B B B B B B B B B B A A A A A A A
X 8 7 6 5 4 3 2 1 0 1 0 2 1 0 9 8 7 6 5 4 3 2 1 0 6 5 4 3 2 1 0
31 0

Little-endian memory
Byte 0 Byte 1 Byte 2 Byte 3
B A A A A A A A B B B B B B B B E E D D C C C B X E E E E E E E
0 6 5 4 3 2 1 0 8 7 6 5 4 3 2 1 1 0 1 0 2 1 0 9 X 8 7 6 5 4 3 2
LEGEND: X = not used, MS = most significant, LS = least significant

194 Run-Time Environment SPRU187U – July 2012

7.2.2.2 EABI Bit Field Differences

Bit fields are handled differently in TIABI mode versus EABI mode in these ways:
• In COFF ABI, bit fields of type long long are not allowed. In EABI, long long bit fields are supported.
• In COFF ABI, all bit fields are treated as signed or unsigned int type. In EABI, bit fields are treated as
the declared type.
• In COFF ABI, the size and alignment a bit field contributes to the struct containing it depends on the
number of bits in the bit field. In EABI, the size and alignment of the struct containing the bit field
depends on the declared type of the bit field. For example, consider the struct:
struct st
{
int a:4
};
In COFF ABI, this struct takes up 1 byte and is aligned at 1 byte. In EABI, this struct uses up 4 bytes
and is aligned at 4 bytes.
• In COFF ABI, unnamed bit fields are zero-sized bit fields do not affect the struct or union alignment. In
EABI, such fields affect the alignment of the struct or union. For example, consider the struct:
struct st
{
char a:4;
int :22;
};
In COFF ABI, this struct uses 4 bytes and is aligned at a 1-byte boundary. In EABI, this struct uses 4
bytes and is aligned at a 4-byte boundary.
• With EABI, bit fields declared volatile are accessed according to the bit field's declared type. A volatile
bit field reference generates exactly one reference to its storage; multiple volatile bit field accesses are
not merged.

7.2.3 Character String Constants

In C, a character string constant is used in one of the following ways:
• To initialize an array of characters. For example:
char s[] = "abc";
When a string is used as an initializer, it is simply treated as an initialized array; each character is a
separate initializer. For more information about initialization, see Section 7.8.
• In an expression. For example:
strcpy (s, "abc");
When a string is used in an expression, the string itself is defined in the .const:string section with the
.string assembler directive, along with a unique label that points to the string; the terminating 0 byte is
explicitly added by the compiler. For example, the following lines define the string abc, and the
terminating 0 byte (the label SL5 points to the string):
.sect ".const:string"
$C$SL5: .string "abc",0
String labels have the form $C$SLn, where $C$ is the compiler-generated symbol prefix and n is a
number assigned by the compiler to make the label unique. The number begins at 0 and is increased
by 1 for each string defined. All strings used in a source module are defined at the end of the compiled
assembly language module.
The label $C$SLn represents the address of the string constant. The compiler uses this label to
reference the string expression.
Because strings are stored in the .const section (possibly in ROM) and shared, it is bad practice for a
program to modify a string constant. The following code is an example of incorrect string use:
const char *a = "abc"
a[1] = 'x'; /* Incorrect! */

SPRU187U – July 2012 Run-Time Environment 195

7.3 Register Conventions

Strict conventions associate specific registers with specific operations in the C/C++ environment. If you
plan to interface an assembly language routine to a C/C++ program, you must understand and follow
these register conventions.

The register conventions dictate how the compiler uses registers and how values are preserved across
function calls. Table 7-2 summarizes how the compiler uses the TMS320C6000 registers.
The registers in Table 7-2 are available to the compiler for allocation to register variables and temporary
expression results. If the compiler cannot allocate a register of a required type, spilling occurs. Spilling is
the process of moving a register's contents to memory to free the register for another purpose.
Objects of type double, long, long long, or long double are allocated into an odd/even register pair and are
always referenced as a register pair (for example, A1:A0). The odd register contains the sign bit, the
exponent, and the most significant part of the mantissa. The even register contains the least significant
part of the mantissa. The A4 register is used with A5 for passing the first argument if the first argument is
a double, long, long long, or long double. The same is true for B4 and B5 for the second parameter, and
so on. For more information about argument-passing registers and return registers, see Section 7.4.

Table 7-2. Register Usage

Function Function
Register Preserved By Special Uses Register Preserved By Special Uses
A0 Parent – B0 Parent –
A1 Parent – B1 Parent –
A2 Parent – B2 Parent –
A3 Parent Structure register (pointer to a B3 Parent Return register (address to return
returned structure) (1) to)
A4 Parent Argument 1 or return value B4 Parent Argument 2
A5 Parent Argument 1 or return value with A4 B5 Parent Argument 2 with B4 for doubles,
for doubles, longs and long longs longs and long longs
A6 Parent Argument 3 B6 Parent Argument 4
A7 Parent Argument 3 with A6 for doubles, B7 Parent Argument 4 with B6 for doubles,
longs, and long longs longs, and long longs
A8 Parent Argument 5 B8 Parent Argument 6
A9 Parent Argument 5 with A8 for doubles, B9 Parent Argument 6 with B8 for doubles,
longs, and long longs longs, and long longs
A10 Child Argument 7 B10 Child Argument 8
A11 Child Argument 7 with A10 for doubles, B11 Child Argument 8 with B10 for doubles,
longs, and long longs longs, and long longs
A12 Child Argument 9 B12 Child Argument 10
A13 Child Argument 9 with A12 for doubles, B13 Child Argument 10 with B12 for doubles,
longs, and long longs longs, and long longs
A14 Child – B14 Child Data page pointer (DP)
A15 Child Frame pointer (FP) B15 Child Stack pointer (SP)
A16-A31 Parent C6400, C6400+, C6700+, C6740, B16-B31 Parent C6400, C6400+, C6700+, C6740,
and C6600 only and C6600 only
ILC Child C6400+, C6740, and C6600 only, NRP Parent
loop buffer counter
IRP Parent RILC Child C6400+, C6740, and C6600 only,
loop buffer counter
(1)
For EABI, structs of size 64 or less are passed by value in registers instead of by reference using a pointer in A3.

196 Run-Time Environment SPRU187U – July 2012

All other control registers are not saved or restored by the compiler.
The compiler assumes that control registers not listed in Table 7-2 that can have an effect on compiled
code have default values. For example, the compiler assumes all circular addressing-enabled registers
are set for linear addressing (the AMR is used to enable circular addressing). Enabling circular addressing
and then calling a C/C++ function without restoring the AMR to a default setting violates the calling
convention. You must be certain that control registers which affect compiler-generated code have a default
value when calling a C/C++ function from assembly.
Assembly language programmers must be aware that the linker assumes B15 contains the stack pointer.
The linker needs to save and restore values on the stack in trampoline code that it generates. If you do
not use B15 as the stack pointer in assembly code, you should use the linker option that disables
trampolines, --trampolines=off. Otherwise, trampolines could corrupt memory and overwrite register
values.

7.4 Function Structure and Calling Conventions

The C/C++ compiler imposes a strict set of rules on function calls. Except for special run-time support
functions, any function that calls or is called by a C/C++ function must follow these rules. Failure to adhere
to these rules can disrupt the C/C++ environment and cause a program to fail.
For details on the calling conventions in EABI mode, refer to The C6000 Embedded Application Binary
Interface Application Report (SPRAB89).

7.4.1 How a Function Makes a Call

A function (parent function) performs the following tasks when it calls another function (child function).
1. Arguments passed to a function are placed in registers or on the stack.
A function (parent function) performs the following tasks when it calls another function (child function):
If arguments are passed to a function, up to the first ten arguments are placed in registers A4, B4, A6,
B6, A8, B8, A10, B10, A12, and B12. If longs, long longs, doubles, or long doubles are passed, they
are placed in register pairs A5:A4, B5:B4, A7:A6, and so on.
However, for C6600, if one or more __x128_t arguments are passed, the next __x128_t argument is
passed in the first available quad, where the list of available quads has the ordering: A7:A6:A5:A4,
B7:B6:B5:B4, A11:A10:A9:A8, B11:B10:B9:B8. If there are no more available quads, the __x128_t
goes onto the stack. A subsequent 32-bit, 40-bit, or 64-bit argument can take the first available register
or register pair even if an earlier __x128_t argument has been put on the stack.
Any remaining arguments are placed on the stack (that is, the stack pointer points to the next free
location; SP + offset points to the eleventh argument, and so on, assuming for C6600 an __x128_t is
not passed.) Arguments placed on the stack must be aligned to a value appropriate for their size. An
argument that is not declared in a prototype and whose size is less than the size of int is passed as an
int. An argument that is a float is passed as double if it has no prototype declared.
A structure argument is passed as the address of the structure. It is up to the called function to make a
local copy.
For a function declared with an ellipsis indicating that it is called with varying numbers of arguments,
the convention is slightly modified. The last explicitly declared argument is passed on the stack, so that
its stack address can act as a reference for accessing the undeclared arguments.
Figure 7-10 shows the register argument conventions.
2. The calling function must save registers A0 to A9 and B0 to B9 (and A16 to A31 and B16 to B31 for
C6400, C6400+, and C6700+), if their values are needed after the call, by pushing the values onto the
stack.
3. The caller (parent) calls the function (child).
4. Upon returning, the caller reclaims any stack space needed for arguments by adding to the stack
pointer. This step is needed only in assembly programs that were not compiled from C/C++ code. This
is because the C/C++ compiler allocates the stack space needed for all calls at the beginning of the
function and deallocates the space at the end of the function.

SPRU187U – July 2012 Run-Time Environment 197

Figure 7-10. Register Argument Conventions

int func1( int a, int b, int c);

A4 A4 B4 A6
int func2( int a, float b, int c) struct A d, float e, int f, int g);

A4 A4 B4 A6 B6 A8 B8 A10
int func3( int a, double b, float c) long double d);

A4 A4 B5:B4 A6 B7:B6
/*NOTE: The following function has a variable number of arguments. */
int vararg(int a, int b, int c, int d);

A4 A4 B4 A6 stack
struct A func4( int y);

A3 A4
__x128_t func5( __x128_t a);

A7:A6:A5:A4 A7:A6:A5:A4
void func6(int a, int b, __x128_t c);

A4 B4 A11:A10:A9:A8
void func7(int a, int b, __x128_t c, int d, int e, int f, __x128_t g, int h);

A4 B4 A11:A10:A9:A8 A6 B6 B8 stack B10

7.4.2 How a Called Function Responds

A called function (child function) must perform the following tasks:
1. The called function (child) allocates enough space on the stack for any local variables, temporary
storage areas, and arguments to functions that this function might call. This allocation occurs once at
the beginning of the function and may include the allocation of the frame pointer (FP).
The frame pointer is used to read arguments from the stack and to handle register spilling instructions.
If any arguments are placed on the stack or if the frame size exceeds 128K bytes, the frame pointer
(A15) is allocated in the following manner:
(a) The old A15 is saved on the stack.
(b) The new frame pointer is set to the current SP (B15).
(c) The frame is allocated by decrementing SP by a constant.
(d) Neither A15 (FP) nor B15 (SP) is decremented anywhere else within this function.
If the above conditions are not met, the frame pointer (A15) is not allocated. In this situation, the frame
is allocated by subtracting a constant from register B15 (SP). Register B15 (SP) is not decremented
anywhere else within this function.
2. If the called function calls any other functions, the return address must be saved on the stack.
Otherwise, it is left in the return register (B3) and is overwritten by the next function call.
3. If the called function modifies any registers numbered A10 to A15 or B10 to B15, it must save them,
either in other registers or on the stack. The called function can modify any other registers without
saving them.
4. If the called function expects a structure argument, it receives a pointer to the structure instead. If
writes are made to the structure from within the called function, space for a local copy of the structure
must be allocated on the stack and the local structure must be copied from the passed pointer to the
structure. If no writes are made to the structure, it can be referenced in the called function indirectly
through the pointer argument.

198 Run-Time Environment SPRU187U – July 2012

You must be careful to declare functions properly that accept structure arguments, both at the point
where they are called (so that the structure argument is passed as an address) and at the point where
they are declared (so the function knows to copy the structure to a local copy).
5. The called function executes the code for the function.
6. If the called function returns any integer, pointer, or float type, the return value is placed in the A4
register. If the function returns a double, long double, long, or long long type, the value is placed in the
A5:A4 register pair. For C6600 if the function returns a __x128_t, the value is placed in A7:A6:A5:A4.
If the function returns a structure, the caller allocates space for the structure and passes the address of
the return space to the called function in A3. To return a structure, the called function copies the
structure to the memory block pointed to by the extra argument.
In this way, the caller can be smart about telling the called function where to return the structure. For
example, in the statement s = f(x), where s is a structure and f is a function that returns a structure, the
caller can actually make the call as f(&s, x). The function f then copies the return structure directly into
s, performing the assignment automatically.
If the caller does not use the return structure value, an address value of 0 can be passed as the first
argument. This directs the called function not to copy the return structure.
You must be careful to declare functions properly that return structures, both at the point where they
are called (so that the extra argument is passed) and at the point where they are declared (so the
function knows to copy the result).
7. Any register numbered A10 to A15 or B10 to B15 that was saved in is restored.
8. If A15 was used as a frame pointer (FP), the old value of A15 is restored from the stack. The space
allocated for the function in is reclaimed at the end of the function by adding a constant to register B15
(SP).
9. The function returns by jumping to the value of the return register (B3) or the saved value of the return
register.

7.4.3 Accessing Arguments and Local Variables

A function accesses its stack arguments and local nonregister variables indirectly through register A15
(FP) or through register B15 (SP), one of which points to the top of the stack. Since the stack grows
toward smaller addresses, the local and argument data for a function are accessed with a positive offset
from FP or SP. Local variables, temporary storage, and the area reserved for stack arguments to functions
called by this function are accessed with offsets smaller than the constant subtracted from FP or SP at the
beginning of the function.
Stack arguments passed to this function are accessed with offsets greater than or equal to the constant
subtracted from register FP or SP at the beginning of the function. The compiler attempts to keep register
arguments in their original registers if optimization is used or if they are defined with the register keyword.
Otherwise, the arguments are copied to the stack to free those registers for further allocation.
For information on whether FP or SP is used to access local variables, temporary storage, and stack
arguments, see Section 7.4.2. For more information on the C/C++ System stack, see Section 7.1.2.

SPRU187U – July 2012 Run-Time Environment 199

7.5 Interfacing C and C++ With Assembly Language

The following are ways to use assembly language with C/C++ code:
• Use separate modules of assembled code and link them with compiled C/C++ modules (see
Section 7.5.1).
• Use assembly language variables and constants in C/C++ source (see Section 7.5.2).
• Use inline assembly language embedded directly in the C/C++ source (see Section 7.5.4).
• Use intrinsics in C/C++ source to directly call an assembly language statement (see Section 7.5.5).

7.5.1 Using Assembly Language Modules With C/C++ Code

Interfacing C/C++ with assembly language functions is straightforward if you follow the calling conventions
defined in Section 7.4, and the register conventions defined in Section 7.3. C/C++ code can access
variables and call functions defined in assembly language, and assembly code can access C/C++
variables and call C/C++ functions.
Follow these guidelines to interface assembly language and C:
• All functions, whether they are written in C/C++ or assembly language, must follow the register
conventions outlined in Section 7.3.
• You must preserve registers A10 to A15, B3, and B10 to B15, and you may need to preserve A3. If
you use the stack normally, you do not need to explicitly preserve the stack. In other words, you are
free to use the stack inside a function as long as you pop everything you pushed before your function
exits. You can use all other registers freely without preserving their contents.
• A10 to A15 and B10 to B15 need to be restored before a function returns, even if any of A10 to A13
and B10 to B13 are being used for passing arguments.
• Interrupt routines must save all the registers they use. For more information, see Section 7.6.
• When you call a C/C++ function from assembly language, load the designated registers with
arguments and push the remaining arguments onto the stack as described in Section 7.4.1.
Remember that only A10 to A15 and B10 to B15 are preserved by the C/C++ compiler. C/C++
functions can alter any other registers, save any other registers whose contents need to be preserved
by pushing them onto the stack before the function is called, and restore them after the function
returns.
• Functions must return values correctly according to their C/C++ declarations. Integers and 32-bit
floating-point (float) values are returned in A4. Doubles, long doubles, longs, and long longs are
returned in A5:A4. For C6600 __x128_t values are returned in A7:A6:A5:A4. Structures are returned by
copying them to the address in A3.
• No assembly module should use the .cinit section for any purpose other than autoinitialization of global
variables. The C/C++ startup routine assumes that the .cinit section consists entirely of initialization
tables. Disrupting the tables by putting other information in .cinit can cause unpredictable results.
• The compiler assigns linknames to all external objects. Thus, when you are writing assembly language
code, you must use the same linknames as those assigned by the compiler. See Section 6.12 for more
information.
• Any object or function declared in assembly language that is accessed or called from C/C++ must be
declared with the .def or .global directive in the assembly language modifier. This declares the symbol
as external and allows the linker to resolve references to it.
Likewise, to access a C/C++ function or object from assembly language, declare the C/C++ object with
the .ref or .global directive in the assembly language module. This creates an undeclared external
reference that the linker resolves.
• The SGIE bit of the TSR control register may need to be saved. Please see Section 7.6.1 for more
information.
• The compiler assumes that control registers not listed in Table 7-2 that can have an effect on compiled
code have default values. For example, the compiler assumes all circular-addressing-enabled registers
are set for linear addressing (the AMR is used to enable circular addressing). Enabling circular
addressing and then calling a C/C++ function without restoring the AMR to a default setting violates the
calling convention. Also, enabling circular addressing and having interrupts enabled violates the calling

200 Run-Time Environment SPRU187U – July 2012

convention. You must be certain that control registers that affect compiler-generated code have a
default value when calling a C/C++ function from assembly.
• Assembly language programmers must be aware that the linker assumes B15 contains the stack
pointer. The linker needs to save and restore values on the stack in trampoline code that it generates.
If you do not use B15 as the stack pointer in your assembly code, you should use the linker option that
disables trampolines, --trampolines=off. Otherwise, trampolines could corrupt memory and overwrite
register values.
• Assembly code that utilizes B14 and/or B15 for localized purposes other than the data-page pointer
and stack pointer may violate the calling convention. The assembly programmer needs to protect these
areas of non-standard use of B14 and B15 by turning off interrupts around this code. Because interrupt
handling routines need the stack (and thus assume the stack pointer is in B15) interrupts need to be
turned off around this code. Furthermore, because interrupt service routines may access global data
and may call other functions which access global data, this special treatment also applies to B14. After
the data-page pointer and stack pointer have been restored, interrupts may be turned back on.
Example 7-1 illustrates a C++ function called main, which calls an assembly language function called
asmfunc, Example 7-2. The asmfunc function takes its single argument, adds it to the C++ global variable
called gvar, and returns the result.

Example 7-1. Calling an Assembly Language Function From a C/C++ Program

extern "C" {
extern int asmfunc(int a); /* declare external asm function */
int gvar = 0; /* define global variable */
}

void main()
{
int I = 5;

I = asmfunc(I); /* call function normally */

Example 7-2. Assembly Language Program Called by Example 7-1

.global _asmfunc
.global _gvar
_asmfunc:
LDW *+b14(_gvar),A3
NOP 4
ADD a3,a4,a3
STW a3,*b14(_gvar)
MV a3,a4
B b3
NOP 5

In the C++ program in Example 7-1, the extern declaration of asmfunc is optional because the return type
is int. Like C/C++ functions, you need to declare assembly functions only if they return noninteger values
or pass noninteger parameters.

NOTE: SP Semantics
The stack pointer must always be 8-byte aligned. For C6600 the stack pointer must always
be 16-byte aligned. This is automatically performed by the C compiler and system
initialization code in the run-time-support libraries. Any hand assembly code that has
interrupts enabled or calls a function defined in C or linear assembly source should also
reserve a multiple of 8 bytes (multiple of 16 bytes for C6600) on the stack.

SPRU187U – July 2012 Run-Time Environment 201

NOTE: Stack Allocation

Even though the compiler guarantees a doubleword alignment of the stack and the stack
pointer (SP) points to the next free location in the stack space, there is only enough
guaranteed room to store one 32-bit word at that location. The called function must allocate
space to store the doubleword.

NOTE: Alignment of __x128_t Data Objects (C6600 Only)

The C6600 introduces the 128-bit container type __x128_t. Global data objects of this type
are aligned to a 16-byte boundary (128 bits). Local __x128_t variables are allocated on the
stack, but are not necessarily aligned on a 16-byte boundary since its actual alignment will
depend on the alignment of the stack and the SP-relative offset of the local __x128_t type
object. For the C6400/C6400+/C6740/C6600 ISAs, the compiler will always align the stack to
an 8-byte boundary.

7.5.2 Accessing Assembly Language Variables From C/C++

It is sometimes useful for a C/C++ program to access variables or constants defined in assembly
language. There are several methods that you can use to accomplish this, depending on where and how
the item is defined: a variable defined in the .bss section, a variable not defined in the .bss section, or a
constant.

7.5.2.1 Accessing Assembly Language Global Variables

Accessing uninitialized variables from the .bss section or a section named with .usect is straightforward:
1. Use the .bss or .usect directive to define the variable.
2. When you use .usect, the variable is defined in a section other than .bss and therefore must be
declared far in C.
3. Use the .def or .global directive to make the definition external.
4. Use the appropriate linkname in assembly language.
5. In C/C++, declare the variable as extern and access it normally.
Example 7-4 and Example 7-3 show how you can access a variable defined in .bss.

Example 7-3. Assembly Language Variable Program

* Note the use of underscores in the following lines

.bss _var1,4,4 ; Define the variable

.global var1 ; Declare it as external

_var2 .usect "mysect",4,4 ; Define the variable

.global _var2 ; Declare it as external

Example 7-4. C Program to Access Assembly Language From Example 7-3

extern int var1; /* External variable */

extern far int var2; /* External variable */
var1 = 1; /* Use the variable */
var2 = 1; /* Use the variable */

202 Run-Time Environment SPRU187U – July 2012

7.5.2.2 Accessing Assembly Language Constants

You can define global constants in assembly language by using the .set, .def, and .global directives, or
you can define them in a linker command file using a linker assignment statement. These constants are
accessible from C/C++ only with the use of special operators.
For normal variables defined in C/C++ or assembly language, the symbol table contains the address of
the value of the variable. For assembler constants, however, the symbol table contains the value of the
constant. The compiler cannot tell which items in the symbol table are values and which are addresses.
If you try to access an assembler (or linker) constant by name, the compiler attempts to fetch a value from
the address represented in the symbol table. To prevent this unwanted fetch, you must use the & (address
of) operator to get the value. In other words, if x is an assembly language constant, its value in C/C++ is
&x.
You can use casts and #defines to ease the use of these symbols in your program, as in Example 7-5 and
Example 7-6.

Example 7-5. Accessing an Assembly Language Constant From C

extern int table_size; /external ref /

#define TABLE_SIZE ((int) (&table_size))
. /* use cast to hide address-of */
.
.
for (I=0; i<TABLE_SIZE; ++I) /* use like normal symbol */

Example 7-6. Assembly Language Program for Example 7-5

_table_size .set 10000 ; define the constant

.global _table_size ; make it global

Because you are referencing only the symbol's value as stored in the symbol table, the symbol's declared
type is unimportant. In Example 7-5, int is used. You can reference linker-defined symbols in a similar
manner.

7.5.3 Sharing C/C++ Header Files With Assembly Source

You can use the .cdecls assembler directive to share C headers containing declarations and prototypes
between C and assembly code. Any legal C/C++ can be used in a .cdecls block and the C/C++
declarations will cause suitable assembly to be generated automatically, allowing you to reference the
C/C++ constructs in assembly code. For more information, see the C/C++ header files chapter in the
TMS320C6000 Assembly Language Tools User's Guide.

SPRU187U – July 2012 Run-Time Environment 203

7.5.4 Using Inline Assembly Language

Within a C/C++ program, you can use the asm statement to insert a single line of assembly language into
the assembly language file created by the compiler. A series of asm statements places sequential lines of
assembly language into the compiler output with no intervening code. For more information, see
Section 6.8.
The asm statement is useful for inserting comments in the compiler output. Simply start the assembly
code string with a semicolon (;) as shown below:
asm(";*** this is an assembly language comment");

NOTE: Using the asm Statement

Keep the following in mind when using the asm statement:
• Be extremely careful not to disrupt the C/C++ environment. The compiler does not check
or analyze the inserted instructions.
• Avoid inserting jumps or labels into C/C++ code because they can produce
unpredictable results by confusing the register-tracking algorithms that the code
generator uses.
• Do not change the value of a C/C++ variable when using an asm statement. This is
because the compiler does not verify such statements. They are inserted as is into the
assembly code, and potentially can cause problems if you are not sure of their effect.
• Do not use the asm statement to insert assembler directives that change the assembly
environment.
• Avoid creating assembly macros in C code and compiling with the --symdebug:dwarf (or
-g) option. The C environment’s debug information and the assembly macro expansion
are not compatible.

7.5.5 Using Intrinsics to Access Assembly Language Statements

The C6000 compiler recognizes a number of intrinsic operators. Intrinsics allow you to express the
meaning of certain assembly statements that would otherwise be cumbersome or inexpressible in C/C++.
Intrinsics are used like functions; you can use C/C++ variables with these intrinsics, just as you would with
any normal function.
The intrinsics are specified with a leading underscore, and are accessed by calling them as you do a
function. For example:
int x1, x2, y;
y = _sadd(x1, x2);

Intrinsic Instructions in C Versus Assembly Language

NOTE: In some instances, an intrinsic’s exact corresponding assembly language instruction may not
be used by the compiler. When this is the case, the meaning of the program does not
change.

Table 7-4 provides a summary of the C6000 intrinsics clarifying which devices support which intrinsics.

204 Run-Time Environment SPRU187U – July 2012

Table 7-3. C6000 C/C++ Intrinsics Support by Device

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_abs Yes Yes Yes Yes Yes Yes
_abs2 Yes Yes Yes Yes
_add2 Yes Yes Yes Yes Yes Yes
_add4 Yes Yes Yes Yes
_addsub Yes Yes Yes
_addsub2 Yes Yes Yes
_amem2 Yes Yes Yes Yes Yes Yes
_amem2_const Yes Yes Yes Yes Yes Yes
_amem4 Yes Yes Yes Yes Yes Yes
_amem4_const Yes Yes Yes Yes Yes Yes
_amem8 Yes Yes Yes Yes
_amem8_const Yes Yes Yes Yes
_amem8_f2 Yes Yes Yes Yes
_amem8_f2_const Yes Yes Yes Yes
_amemd8 Yes Yes Yes Yes Yes Yes
_amemd8_const Yes Yes Yes Yes Yes Yes
_avg2 Yes Yes Yes Yes
_avgu4 Yes Yes Yes Yes
_bitc4 Yes Yes Yes Yes
_bitr Yes Yes Yes Yes
_ccmatmpy Yes
_ccmatmpyr1 Yes
_ccmpy32r1 Yes
_clr Yes Yes Yes Yes Yes Yes
_clrr Yes Yes Yes Yes Yes Yes
_cmatmpy Yes
_cmatmpyr1 Yes
_cmpeq2 Yes Yes Yes Yes
_cmpeg4 Yes Yes Yes Yes
_cmpgt2 Yes Yes Yes Yes
_cmpgtu4 Yes Yes Yes Yes
_cmplt2 Yes Yes Yes Yes
_cmpltu4 Yes Yes Yes Yes
_cmpy Yes Yes Yes
_cmpy32r1 Yes
_cmpyr Yes Yes Yes
_cmpyr1 Yes Yes Yes
_cmpysp Yes
_complex_conjugate_mpysp Yes
_complex_mpysp Yes
_crot270 Yes
_crot90 Yes
_dadd Yes
_dadd2 Yes
_daddsp Yes
_dadd_c Yes
_dapys2 Yes

SPRU187U – July 2012 Run-Time Environment 205

Table 7-3. C6000 C/C++ Intrinsics Support by Device (continued)

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_davg2 Yes
_davgnr2 Yes
_davgnru4 Yes
_davgu4 Yes
_dccmpyr1 Yes
_dcmpeq2 Yes
_dcmpeq4 Yes
_dcmpgt2 Yes
_dcmpgtu4 Yes
_dccmpy Yes
_dcmpy Yes
_dcmpyr1 Yes
_dcrot90 Yes
_dcrot270 Yes
_ddotp4 Yes Yes Yes
_ddotp4h Yes
_ddotph2 Yes Yes Yes
_ddotph2r Yes Yes Yes
_ddotpl2 Yes Yes Yes
_ddotpl2r Yes Yes Yes
_ddotpsu4h Yes
_deal Yes Yes Yes Yes
_dinthsp Yes
_dinthspu Yes
_dintsp Yes
_dintspu Yes
_dmax2 Yes
_dmaxu4 Yes
_dmin2 Yes
_dminu4 Yes
_dmpy2 Yes
_dmpysp Yes
_dmpysu4 Yes
_dmpyu2 Yes
_dmpyu4 Yes
_dmv Yes Yes Yes
_dmvd Yes
_dotp2 Yes Yes Yes Yes
_dotp4h Yes
_dotp4hll Yes
_dotpn2 Yes Yes Yes Yes
_dotpnrsu2 Yes Yes Yes Yes
_dotpnrus2 Yes Yes Yes Yes
_dotprsu2 Yes Yes Yes Yes
_dotpsu4 Yes Yes Yes Yes
_dotpus4 Yes Yes Yes Yes
_dotpsu4h Yes

206 Run-Time Environment SPRU187U – July 2012

Table 7-3. C6000 C/C++ Intrinsics Support by Device (continued)

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_dotpsu4hll Yes
_dotpu4 Yes Yes Yes Yes
_dpack2 Yes Yes Yes
_dpackh2 Yes
_dpackh4 Yes
_dpacklh2 Yes
_dpacklh4 Yes
_dpackl2 Yes
_dpackl4 Yes
_dpackx2 Yes Yes Yes
_dpint Yes Yes Yes
_dsadd Yes
_dsadd2 Yes
_dshl Yes
_dshl2 Yes
_dshr Yes
_dshr2 Yes
_dshru Yes
_dshru2 Yes
_dsmpy2 Yes
_dspacku4 Yes
_dspint Yes
_dspinth Yes
_dssub Yes
_dssub2 Yes
_dsub Yes
_dsub2 Yes
_dsubsp Yes
_dtol Yes Yes Yes Yes Yes Yes
_dtoll Yes Yes Yes Yes Yes Yes
_dxpnd2 Yes
_dxpnd4 Yes
_ext Yes Yes Yes Yes Yes Yes
_extr Yes Yes Yes Yes Yes Yes
_extu Yes Yes Yes Yes Yes Yes
_extur Yes Yes Yes Yes Yes Yes
_f2tol Yes Yes Yes
_f2toll Yes Yes Yes
_fabs Yes Yes Yes
_fabsf Yes Yes Yes
_fdmvd_f2 Yes
_fmdv_f2 Yes Yes Yes
_ftoi Yes Yes Yes Yes Yes Yes
_gmpy Yes Yes Yes
_gmpy4 Yes Yes Yes Yes
_hi Yes Yes Yes Yes Yes Yes
_hill Yes Yes Yes Yes Yes Yes

SPRU187U – July 2012 Run-Time Environment 207

Table 7-3. C6000 C/C++ Intrinsics Support by Device (continued)

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_itod Yes Yes Yes Yes Yes Yes
_itof Yes Yes Yes Yes Yes Yes
_itoll Yes Yes Yes Yes Yes Yes
_labs Yes Yes Yes Yes Yes Yes
_land Yes
_landn Yes
_ldotp2 Yes Yes Yes Yes
_lmbd Yes Yes Yes Yes Yes Yes
_lnorm Yes Yes Yes Yes Yes Yes
_lo Yes Yes Yes Yes Yes Yes
_loll Yes Yes Yes Yes Yes Yes
_lor Yes
_lsadd Yes Yes Yes Yes Yes Yes
_lssub Yes Yes Yes Yes Yes Yes
_ltod Yes Yes Yes Yes Yes Yes
_lltod Yes Yes Yes Yes Yes Yes
_lltof2 Yes Yes Yes
_ltof2 Yes Yes Yes
_max2 Yes Yes Yes Yes
_maxu4 Yes Yes Yes Yes
_mfence Yes
_min2 Yes Yes Yes Yes
_minu4 Yes Yes Yes Yes
_mem2 Yes Yes Yes Yes
_mem2_const Yes Yes Yes Yes
_mem4 Yes Yes Yes Yes
_mem4_const Yes Yes Yes Yes
_mem8 Yes Yes Yes Yes
_mem8_const Yes Yes Yes Yes
_mem8_f2 Yes Yes Yes
_mem8_f2_const Yes Yes Yes
_memd8 Yes Yes Yes Yes
_memd8_const Yes Yes Yes Yes
_mpy Yes Yes Yes Yes Yes Yes
_mpy2ir Yes Yes Yes
_mpy2ll Yes Yes Yes Yes
_mpy32 Yes Yes Yes
_mpy32ll Yes Yes Yes
_mpy32su Yes Yes Yes
_mpy32u Yes Yes Yes
_mpy32us Yes Yes Yes
_mpyh Yes Yes Yes Yes Yes Yes
_mpyhill Yes Yes Yes Yes
_mpyihll Yes Yes Yes Yes
_mpyilll Yes Yes Yes Yes
_mpyhir Yes Yes Yes Yes
_mpyihr Yes Yes Yes Yes

208 Run-Time Environment SPRU187U – July 2012

Table 7-3. C6000 C/C++ Intrinsics Support by Device (continued)

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_mpyilr Yes Yes Yes Yes
_mpyhl Yes Yes Yes Yes Yes Yes
_mpyhlu Yes Yes Yes Yes Yes Yes
_mpyhslu Yes Yes Yes Yes Yes Yes
_mpyhsu Yes Yes Yes Yes Yes Yes
_myphu Yes Yes Yes Yes Yes Yes
_mpyhuls Yes Yes Yes Yes Yes Yes
_mpyhus Yes Yes Yes Yes Yes Yes
_mpyidll Yes Yes Yes
_mpylh Yes Yes Yes Yes Yes Yes
_mpylhu Yes Yes Yes Yes Yes Yes
_mpylill Yes Yes Yes Yes
_mpylir Yes Yes Yes Yes
_mpylshu Yes Yes Yes Yes Yes Yes
_mpyluhs Yes Yes Yes Yes Yes Yes
_mpysp2dp Yes Yes Yes
_mpyspdp Yes Yes Yes
_mpysu Yes Yes Yes Yes Yes Yes
_mpysu4ll Yes Yes Yes Yes
_mpyus4ll Yes Yes Yes Yes
_mpyu Yes Yes Yes Yes Yes Yes
_mpyu2 Yes
_mpyu4ll Yes Yes Yes Yes
_mpyus Yes Yes Yes Yes Yes Yes
_mvd Yes Yes Yes Yes
_nassert Yes Yes Yes Yes Yes Yes
_norm Yes Yes Yes Yes Yes Yes
_pack2 Yes Yes Yes Yes
_packh2 Yes Yes Yes Yes
_packh4 Yes Yes Yes Yes
_packhl2 Yes Yes Yes Yes
_packl4 Yes Yes Yes Yes
_packlh2 Yes Yes Yes Yes
_qmpy32 Yes
_qmpysp Yes
_qsmpy32r1 Yes
_rcpdp Yes Yes Yes
_rcpsp Yes Yes Yes
_rsqrdp Yes Yes Yes
_rsqrsp Yes Yes Yes
_rotl Yes Yes Yes Yes
_rpack2 Yes Yes Yes
_sadd Yes Yes Yes Yes Yes Yes
_sadd2 Yes Yes Yes Yes
_saddsub Yes Yes Yes
_saddsub2 Yes Yes Yes
_saddu4 Yes Yes Yes Yes

SPRU187U – July 2012 Run-Time Environment 209

Table 7-3. C6000 C/C++ Intrinsics Support by Device (continued)

C6700/C6700
Intrinsic C6200 C6400 C6400+ C6600 C6740
+
_saddus2 Yes Yes Yes Yes
_saddsu2 Yes Yes Yes Yes
_sat Yes Yes Yes Yes Yes Yes
_set Yes Yes Yes Yes Yes Yes
_setr Yes Yes Yes Yes Yes Yes
_shfl Yes Yes Yes Yes
_shfl3 Yes Yes Yes
_shl2 Yes
_shlmb Yes Yes Yes Yes
_shr2 Yes Yes Yes Yes
_shrmb Yes Yes Yes Yes
_shru2 Yes Yes Yes Yes
_smpy Yes Yes Yes Yes Yes Yes
_smpy2ll Yes Yes Yes Yes
_smpy32 Yes Yes Yes
_smpyh Yes Yes Yes Yes Yes Yes
_smpyhl Yes Yes Yes Yes Yes Yes
_smpylh Yes Yes Yes Yes Yes Yes
_spack2 Yes Yes Yes Yes
_spacku4 Yes Yes Yes Yes
_spint Yes Yes Yes
_sshl Yes Yes Yes Yes Yes Yes
_sshvl Yes Yes Yes Yes
_sshvr Yes Yes Yes Yes
_ssub Yes Yes Yes Yes Yes Yes
_ssub2 Yes Yes Yes
_sub2 Yes Yes Yes Yes Yes Yes
_sub4 Yes Yes Yes Yes
_subabs4 Yes Yes Yes Yes
_subc Yes Yes Yes Yes Yes Yes
_swap4 Yes Yes Yes Yes
_swap2 Yes Yes Yes Yes
_unpkbu4 Yes
_unpkh2 Yes
_unpkhu2 Yes
_unpkhu4 Yes Yes Yes Yes
_unpklu4 Yes Yes Yes Yes
_xorll_c Yes
_xormpy Yes Yes Yes
_xpnd2 Yes Yes Yes Yes
_xpnd4 Yes Yes Yes Yes

210 Run-Time Environment SPRU187U – July 2012

The intrinsics listed in Table 7-4 can be used on all C6000 devices. They correspond to the indicated
C6000 assembly language instruction(s). See the TMS320C6000 CPU and Instruction Set Reference
Guide for more information.
See Table 7-5 for the listing of C6400-specific intrinsics, which are also compatible with C6400+, C6740,
and C6600. See Table 7-6 for the listing of C6400+-specific intrinsics, which are also compatible with
C6740 and C6600 devices. See Table 7-7 for the listing of C6700-specific intrinsics. See Table 7-8 for a
listing of C6600-specifiic intrinsics.

SPRU187U – July 2012 Run-Time Environment 211

Table 7-4. TMS320C6000 C/C++ Compiler Intrinsics

Assembly
C/C++ Compiler Intrinsic Instruction Description
int _abs (int src); ABS Returns the saturated absolute value of src
int _labs (__int40_t src);
int _add2 (int src1, int src2); ADD2 Adds the upper and lower halves of src1 to the upper and lower
halves of src2 and returns the result. Any overflow from the lower
half add does not affect the upper half add.
ushort & _amem2 (void *ptr); LDHU Allows aligned loads and stores of 2 bytes to memory. The pointer
STHU must be aligned to a two-byte boundary. (1)
const ushort & _amem2_const (const void *ptr); LDHU Allows aligned loads of 2 bytes from memory. The pointer must be
aligned to a two-byte boundary. (1)
unsigned & _amem4 (void *ptr); LDW Allows aligned loads and stores of 4 bytes to memory. The pointer
STW must be aligned to a four-byte boundary. (1)
const unsigned & _amem4_const (const void *ptr); LDW Allows aligned loads of 4 bytes from memory. The pointer must be
aligned to a four-byte boundary. (1)
double & _amemd8 (void *ptr); LDW/LDW Allows aligned loads and stores of 8 bytes to memory. The pointer
STW/STW must be aligned to an eight-byte boundary. (1) (2)
For C6400 _amemd8 corresponds to different assembly instructions
than when used with other C6000 devices; see Table 7-5 for
specifics.
const double & _amemd8_const (const void *ptr); LDDW Allows aligned loads of 8 bytes from memory. The pointer must be
aligned to an eight-byte boundary. (1) (2)
unsigned _clr (unsigned src2, unsigned csta, CLR Clears the specified field in src2. The beginning and ending bits of
unsigned cstb); the field to be cleared are specified by csta and cstb, respectively.
unsigned _clrr (unsigned src2, int src1); CLR Clears the specified field in src2. The beginning and ending bits of
the field to be cleared are specified by the lower 10 bits of src1.
__int40_t _dtol (double src); Reinterprets double register pair src as an __int40_t (stored as a
register pair).
long long _dtoll (double src); Reinterprets double register pair src as a long long register pair.
int _ext (int src2, unsigned csta, unsigned cstb); EXT Extracts the specified field in src2, sign-extended to 32 bits. The
extract is performed by a shift left followed by a signed shift right;
csta and cstb are the shift left and shift right amounts, respectively.
int _extr (int src2, int src1); EXT Extracts the specified field in src2, sign-extended to 32 bits. The
extract is performed by a shift left followed by a signed shift right;
the shift left and shift right amounts are specified by the lower 10
bits of src1.
unsigned _extu (unsigned src2, unsigned csta , EXTU Extracts the specified field in src2, zero-extended to 32 bits. The
unsigned cstb); extract is performed by a shift left followed by a unsigned shift right;
csta and cstb are the shift left and shift right amounts, respectively.
unsigned _extur (unsigned src2, int src1); EXTU Extracts the specified field in src2, zero-extended to 32 bits. The
extract is performed by a shift left followed by a unsigned shift right;
the shift left and shift right amounts are specified by the lower 10
bits of src1.
unsigned _ftoi (float src); Reinterprets the bits in the float as an unsigned. For example:
_ftoi (1.0) == 1065353216U
unsigned _hi (double src); Returns the high (odd) register of a double register pair
unsigned _hill (long long src); Returns the high (odd) register of a long long register pair
double _itod (unsigned src2, unsigned src1); Builds a new double register pair by reinterpreting two unsigned
values, where src2 is the high (odd) register and src1 is the low
(even) register
float _itof (unsigned src); Reinterprets the bits in the unsigned as a float. For example:
_itof (0x3f800000) = 1.0
long long _itoll (unsigned src2, unsigned src1); Builds a new long long register pair by reinterpreting two unsigned
values, where src2 is the high (odd) register and src1 is the low
(even) register
unsigned _lmbd (unsigned src1, unsigned src2); LMBD Searches for a leftmost 1 or 0 of src2 determined by the LSB of
src1. Returns the number of bits up to the bit change.
unsigned _lo (double src); Returns the low (even) register of a double register pair
(1)
See the TMS320C6000 Programmer's Guide for more information.
(2)
See Section 7.5.9 for details on manipulating 8-byte data quantities.

212 Run-Time Environment SPRU187U – July 2012

Table 7-4. TMS320C6000 C/C++ Compiler Intrinsics (continued)

Assembly
C/C++ Compiler Intrinsic Instruction Description
unsigned _loll (long long src); Returns the low (even) register of a long long register pair
double _ltod (__int40_t src); Reinterprets an __int40_t register pair src as a double register pair.
double _lltod (long long src); Reinterprets long long register pair src as a double register pair.
int _mpy (int src1, int src2); MPY Multiplies the 16 LSBs of src1 by the 16 LSBs of src2 and returns
int _mpyus (unsigned src1, int src2); MPYUS the result. Values can be signed or unsigned.
int _mpysu (int src1, unsigned src2); MPYSU
unsigned _mpyu (unsigned src1, unsigned src2); MPYU
int _mpyh (int src1, int src2); MPYH Multiplies the 16 MSBs of src1 by the 16 MSBs of src2 and returns
int _mpyhus (unsigned src1, int src2); MPYHUS the result. Values can be signed or unsigned.
int _mpyhsu (int src1, unsigned src2); MPYHSU
unsigned _mpyhu (unsigned src1, unsigned src2); MPYHU
int _mpyhl (int src1, int src2); MPYHL Multiplies the 16 MSBs of src1 by the 16 LSBs of src2 and returns
int _mpyhuls (unsigned src1, int src2); MPYHULS the result. Values can be signed or unsigned.
int _mpyhslu (int src1, unsigned src2); MPYHSLU
unsigned _mpyhlu (unsigned src1, unsigned src2); MPYHLU
int _mpylh (int src1, int src2); MPYLH Multiplies the 16 LSBs of src1 by the 16 MSBs of src2 and returns
int _mpyluhs (unsigned src1, int src2); MPYLUHS the result. Values can be signed or unsigned.
int _mpylshu (int src1, unsigned src2); MPYLSHU
unsigned _mpylhu (unsigned src1, unsigned src2); MPYLHU
void _nassert (int src); Generates no code. Tells the optimizer that the expression declared
with the assert function is true; this gives a hint to the optimizer as
to what optimizations might be valid.
unsigned _norm (int src); NORM Returns the number of bits up to the first nonredundant sign bit of
unsigned _lnorm (__int40_t src); src
int _sadd (int src1, int src2); SADD Adds src1 to src2 and saturates the result. Returns the result.
long _lsadd (int src1, __int40_t src2);
int _sat (__int40_t src2); SAT Converts a 40-bit long to a 32-bit signed int and saturates if
necessary.
unsigned _set (unsigned src2, unsigned csta , SET Sets the specified field in src2 to all 1s and returns the src2 value.
unsigned cstb); The beginning and ending bits of the field to be set are specified by
csta and cstb, respectively.
unsigned _setr (unit src2, int src1); SET Sets the specified field in src2 to all 1s and returns the src2 value.
The beginning and ending bits of the field to be set are specified by
the lower ten bits of src1.
int _smpy (int src1, int src2); SMPY Multiplies src1 by src2, left shifts the result by 1, and returns the
int _smpyh (int src1, int src2); SMPYH result. If the result is 0x80000000, saturates the result to
int _smpyhl (int src1, int src2); SMPYHL 0x7FFFFFFF
int _smpylh (int src1, int src2); SMPYLH
int _sshl (int src2, unsigned src1); SSHL Shifts src2 left by the contents of src1, saturates the result to 32
bits, and returns the result
int _ssub (int src1, int src2); SSUB Subtracts src2 from src1, saturates the result, and returns the result.
__int40_t _lssub (int src1, __int40_t src2);
unsigned _subc (unsigned src1, unsigned src2); SUBC Conditional subtract divide step
int _sub2 (int src1, int src2); SUB2 Subtracts the upper and lower halves of src2 from the upper and
lower halves of src1, and returns the result. Borrowing in the lower
half subtract does not affect the upper half subtract.

The intrinsics listed in Table 7-5 can be used for C6400, C6400+, C6740, and C6600 devices. The
intrinsics shown correspond to the indicated C6000 assembly language instruction(s). See the
TMS320C6000 CPU and Instruction Set Reference Guide for more information.
See Table 7-4 for the listing of generic C6000 intrinsics. See Table 7-6 for the listing of C6400+-, C6740-,
and C6600-specific intrinsics. See Table 7-7 for the listing of C6700-specific intrinsics. See Table 7-8 for
the listing of C6600-specific intrinsics.

SPRU187U – July 2012 Run-Time Environment 213

Table 7-5. TMS320C6400, C6400+, C6740, and C6600 C/C++ Compiler Intrinsics
Assembly
C/C++ Compiler Intrinsic Instruction Description
int _abs2 (int src); ABS2 Calculates the absolute value for each 16-bit value
int _add4 (int src1, int src2); ADD4 Performs 2s-complement addition to pairs of packed 8-bit numbers
long long & _amem8 (void *ptr); LDDW Allows aligned loads and stores of 8 bytes to memory. The pointer
STDW must be aligned to an eight-byte boundary.
const long long & _amem8_const (const void *ptr); LDDW Allows aligned loads of 8 bytes from memory. The pointer must be
aligned to an eight-byte boundary. (1)
__float2_t & _amem8_f2(void * ptr); LDDW Allows aligned loads and stores of 8 bytes to memory. The pointer
STDW must be aligned to an eight-byte boundary. You must include
c6x.h. (1) (2)
const __float2_t & _amem8_f2_const(void * ptr); LDDW Allows aligned loads of 8 bytes from memory. The pointer must be
aligned to an eight-byte boundary. You must include c6x.h. (1) (2)
double & _amemd8 (void *ptr); LDDW Allows aligned loads and stores of 8 bytes to memory. The pointer
STDW must be aligned to an eight-byte boundary. (1) (2)
For C6400 _amemd8 corresponds to different assembly instructions
than when used with other C6000 devices; see Table 7-4.
const double & _amemd8_const (const void *ptr); LDDW Allows aligned loads of 8 bytes from memory. The pointer must be
aligned to an eight-byte boundary. (1) (2)
int _avg2 (int src1, int src2); AVG2 Calculates the average for each pair of signed 16-bit values
unsigned _avgu4 (unsigned, unsigned); AVGU4 Calculates the average for each pair of signed 8-bit values
unsigned _bitc4 (unsigned src); BITC4 For each of the 8-bit quantities in src, the number of 1 bits is written
to the corresponding position in the return value
unsigned _bitr (unsigned src); BITR Reverses the order of the bits
int _cmpeq2 (int src1, int src2); CMPEQ2 Performs equality comparisons on each pair of 16-bit values.
Equality results are packed into the two least-significant bits of the
return value.
int _cmpeq4 (int src1, int src2); CMPEQ4 Performs equality comparisons on each pair of 8-bit values. Equality
results are packed into the four least-significant bits of the return
value.
int _cmpgt2 (int src1, int src2); CMPGT2 Compares each pair of signed 16-bit values. Results are packed
into the two least-significant bits of the return value.
unsigned _cmpgtu4 (unsigned src1, unsigned src2); CMPGTU4 Compares each pair of 8-bit values. Results are packed into the four
least-significant bits of the return value.
int _cmplt2 (int src1, int src2); CMPLT2 Swaps operands and calls _cmpgt2.
unsigned _cmpltu4 (unsigned src1, unsigned src2); CMPLTU4 Swaps operands and calls _cmpgtu4.
unsigned _deal (unsigned src ); DEAL The odd and even bits of src are extracted into two separate 16-bit
values.
int _dotp2 (int src1, int src2); DOTP2 The product of the signed lower 16-bit values of src1 and src2 is
__int40_t _ldotp2 (int src1, int src2); DOTP2 added to the product of the signed upper 16-bit values of src1 and
src2.The _lo and _hi intrinsics are needed to access each half of
the 64-bit integer result.
int _dotpn2 (int src1, int src2); DOTPN2 The product of the signed lower 16-bit values of src1 and src2 is
subtracted from the product of the signed upper 16-bit values of
src1 and src2.
int _dotpnrsu2 (int src1, unsigned src2); DOTPNRSU2 The product of the lower 16-bit values of src1 and src2 is subtracted
from the product of the upper 16-bit values of src1 and src2. The
values in src1 are treated as signed packed quantities; the values in
src2 are treated as unsigned packed quantities. 2^15 is added and
the result is sign shifted right by 16.
int _dotpnrus2 (unsigned src1, int src2); DOTPNRUS2 Swaps the operands and calls _dotpnrsu2.
int _dotprsu2 (int src1, unsigned src2); DOTPRSU2 The product of the lower 16-bit values of src1 and src2 is added to
the product of the upper 16-bit values of src1 and src2. The values
in src1 are treated as signed packed quantities; the values in src2
are treated as unsigned packed quantities. 2^15 is added and the
result is sign shifted by 16.

(1)
See Section 7.5.9 for details on manipulating 8-byte data quantities.
(2)
See the TMS320C6000 Programmer's Guide for more information.

214 Run-Time Environment SPRU187U – July 2012

Table 7-5. TMS320C6400, C6400+, C6740, and C6600 C/C++ Compiler Intrinsics (continued)
Assembly
C/C++ Compiler Intrinsic Instruction Description
int _dotpsu4 (int src1, unsigned src2); DOTPSU4 For each pair of 8-bit values in src1 and src2, the 8-bit value from
int _dotpus4 (unsigned src1, int src2); DOTPUS4 src1 is multiplied with the 8-bit value from src2. The four products
unsigned _dotpu4 (unsigned src1, unsigned src2); DOTPU4 are summed together.
int _gmpy4 (int src1, int src2); GMPY4 Performs the Galois Field multiply on four values in src1 with four
parallel values in src2. The four products are packed into the return
value.
int _max2 (int src1, int src2); MAX2 Places the larger/smaller of each pair of values in the corresponding
int _min2 (int src1, int src2); MIN2 position in the return value. Values can be 16-bit signed or 8-bit
unsigned _maxu4 (unsigned src1, unsigned src2); MAX4 unsigned.
unsigned _minu4 (unsigned src1, unsigned src2); MINU4
ushort & _mem2 (void * ptr); LDB/LDB Allows unaligned loads and stores of 2 bytes to memory (2)
STB/STB
const ushort & _mem2_const (const void * ptr); LDB/LDB Allows unaligned loads of 2 bytes to memory (2)
unsigned & _mem4 (void * ptr); LDNW Allows unaligned loads and stores of 4 bytes to memory (2)
STNW
const unsigned & _mem4_const (const void * ptr); LDNW Allows unaligned loads of 4 bytes from memory (2)
long long & _mem8 (void * ptr); LDNDW Allows unaligned loads and stores of 8 bytes to memory (2)
STNDW
const long long & _mem8_const (const void * ptr); LDNDW Allows unaligned loads of 8 bytes from memory (3)
double & _memd8 (void * ptr); LDNDW Allows unaligned loads and stores of 8 bytes to memory (4) (3)
STNDW
const double & _memd8_const (const void * ptr); LDNDW Allows unaligned loads of 8 bytes from memory (4) (3)
long long _mpy2ll (int src1, int src2); MPY2 Returns the products of the lower and higher 16-bit values in src1
and src2
long long _mpyhill (int src1, int src2); MPYHI Produces a 16 by 32 multiply. The result is placed into the lower 48
long long _mpylill (int src1, int src2); MPYLI bits of the return type. Can use the upper or lower 16 bits of src1.
long long _mpyihll (int src1, int src2); MPYIH Swaps operands and calls _mpyhill.
long long _mpyilll (int src1, int src2); MPYIL Swaps operands and calls _mpylill.
int _mpyhir (int src1, int src2); MPYHIR Produces a signed 16 by 32 multiply. The result is shifted right by
int _mpylir (int src1, int src2); MPYLIR 15 bits. Can use the upper or lower 16 bits of src1.
int _mpyihr (int src1, int src2); MPYIHR Swaps operands and calls _mpyhir.
int _mpyilr (int src1, int src2); MPYILR Swaps operands and calls _mpylir.
long long _mpysu4ll (int src1, unsigned src2); MPYSU4 For each 8-bit quantity in src1 and src2, performs an 8-bit by 8-bit
long long _mpyus4ll (unsigned src1, int src2); MPYUS4 multiply. The four 16-bit results are packed into a 64-bit result. The
long long _mpyu4ll (unsigned src1, unsigned src2); MPYU4 results can be signed or unsigned.
int _mvd (int src2 ); MVD Moves the data from src2 to the return value over four cycles using
the multiplier pipeline
unsigned _pack2 (unsigned src1, unsigned src2); PACK2 The lower/upper halfwords of src1 and src2 are placed in the return
unsigned _packh2 (unsigned src1, unsigned src2); PACKH2 value.
unsigned _packh4 (unsigned src1, unsigned src2); PACKH4 Packs alternate bytes into return value. Can pack high or low bytes.
unsigned _packl4 (unsigned src1, unsigned src2); PACKL4
unsigned _packhl2 (unsigned src1, unsigned src2); PACKHL2 The upper/lower halfword of src1 is placed in the upper halfword the
unsigned _packlh2 (unsigned src1, unsigned src2); PACKLH2 return value. The lower/upper halfword of src2 is placed in the lower
halfword the return value.
unsigned _rotl (unsigned src1, unsigned src2); ROTL Rotates src2 to the left by the amount in src1
int _sadd2 (int src1, int src2); SADD2 Performs saturated addition between pairs of 16-bit values in src1
int _saddus2 (unsigned src1, int src2); SADDUS2 and src2. Values for src1 can be signed or unsigned.
int _saddsu2 (int src1, unsigned src2); SADDSU2
unsigned _saddu4 (unsigned src1, unsigned src2); SADDU4 Performs saturated addition between pairs of 8-bit unsigned values
in src1 and src2.
unsigned _shfl (unsigned src2); SHFL The lower 16 bits of src2 are placed in the even bit positions, and
the upper 16 bits of src are placed in the odd bit positions.
unsigned _shlmb (unsigned src1, unsigned src2); SHLMB Shifts src2 left/right by one byte, and the most/least significant byte
unsigned _shrmb (unsigned src1, unsigned src2); SHRMB of src1 is merged into the least/most significant byte position.

(3)
See the TMS320C6000 Programmer's Guide for more information.
(4)
See Section 7.5.9 for details on manipulating 8-byte data quantities.

SPRU187U – July 2012 Run-Time Environment 215

Table 7-5. TMS320C6400, C6400+, C6740, and C6600 C/C++ Compiler Intrinsics (continued)
Assembly
C/C++ Compiler Intrinsic Instruction Description
int _shr2 (int src1, unsigned src2); SHR2 For each 16-bit quantity in src2, the quantity is arithmetically or
unsigned shru2 (unsigned src1, unsigned src2); SHRU2 logically shifted right by src1 number of bits. src2 can contain signed
or unsigned values
long long _smpy2ll (int src1, int src2); SMPY2 Performs 16-bit multiplication between pairs of signed packed 16-bit
values, with an additional 1 bit left-shift and saturate into a 64-bit
result.
int _spack2 (int src1, int src2); SPACK2 Two signed 32-bit values are saturated to 16-bit values and packed
into the return value
unsigned _spacku4 (int src1 , int src2); SPACKU4 Four signed 16-bit values are saturated to 8-bit values and packed
into the return value
int _sshvl (int src2, int src1); SSHVL Shifts src2 to the left/right src1 bits. Saturates the result if the
int _sshvr (int src2, int src1); SSHVR shifted value is greater than MAX_INT or less than MIN_INT.
int _sub4 (int src1, int src2); SUB4 Performs 2s-complement subtraction between pairs of packed 8-bit
values
int _subabs4 (int src1, int src2); SUBABS4 Calculates the absolute value of the differences for each pair of
packed 8-bit values
unsigned _swap4 (unsigned src); SWAP4 Exchanges pairs of bytes (an endian swap) within each 16-bit value
unsigned _swap2 (unsigned src); SWAP2 Calls _packlh2.
unsigned _unpkhu4 (unsigned src); UNPKHU4 Unpacks the two high unsigned 8-bit values into unsigned packed
16-bit values
unsigned _unpklu4 (unsigned src); UNPKLU4 Unpacks the two low unsigned 8-bit values into unsigned packed
16-bit values
unsigned _xpnd2 (unsigned src); XPND2 Bits 1 and 0 of src are replicated to the upper and lower halfwords
of the result, respectively.
unsigned _xpnd4 (unsigned src); XPND4 Bits 3 and 0 of src are replicated to bytes 3 through 0 of the result.

The intrinsics listed in Table 7-6 are included only for C6400+, C6740, and C6600 devices. The intrinsics
shown correspond to the indicated C6000 assembly language instruction(s). See the TMS320C6000 CPU
and Instruction Set Reference Guide for more information.
See Table 7-4 for the listing of generic C6000 intrinsics. See Table 7-5 for the general listing of intrinsics
for C6400 devices, which includes C6400, C6400+, C6740 and C6600. See Table 7-7 for the listing of
C6700-specific intrinsics. See Table 7-8 for a listing of additional intrinsics only for C6600.

Table 7-6. TMS320C6400+, C6740, and C6600 C/C++ Compiler Intrinsics

Assembly
C/C++ Compiler Intrinsic Instruction Description
long long _addsub (int src1, int src2); ADDSUB Performs an addition and subtraction in parallel.
long long _addsub2 (int src1, int src2); ADDSUB2 Performs an ADD2 and SUB2 in parallel.
long long _cmpy (unsigned src1, unsigned src2); CMPY Performs various complex multiply operations.
unsigned _cmpyr (unsigned src1, unsigned src2); CMPYR
unsigned _cmpyr1 (unsigned src1, unsigned src2 ); CMPYR1
long long _ddotp4 (unsigned src1, unsigned src2); DDOTP4 Performs two DOTP2 operations simultaneously.
long long _ddotph2 (long long src1, unsigned src2); DDOTPH2 Performs various dual dot-product operations between two pairs
long long _ddotpl2 (long long src1, unsigned src2); DDOTPL2 of signed, packed 16-bit values.
unsigned _ddotph2r (long long src1, unsigned src2); DDOTPH2R
unsigned _ddotpl2r (long long src1, unsigned src2); DDOTPL2
long long _dmv (int src1, int src2); DMV Places src1 in the 32 LSBs of the long long and src2 in the 32
MSBs of the long long. See also _itoll().
long long _dpack2 (unsigned src1, unsigned src2); DPACK2 PACK2 and PACKH2 operations performed in parallel.
long long _dpackx2 (unsigned src1, unsigned src2); DPACKX2 PACKLH2 and PACKX2 operations performed in parallel.
__float2_t _fmdv_f2(float src1, float src2) DMV Places src1 in the 32 LSBs of the __float2_t and src2 in the 32
MSBs of the __float2_t. See also _itoll().

216 Run-Time Environment SPRU187U – July 2012

Table 7-6. TMS320C6400+, C6740, and C6600 C/C++ Compiler Intrinsics (continued)
Assembly
C/C++ Compiler Intrinsic Instruction Description
unsigned _gmpy (unsigned src1, unsigned src2); GMPY Performs the Galois Field multiply.
long long _mpy2ir (int src1, int src2); MPY2IR Performs two 16 by 32 multiplies. Both results are shifted right by
15 bits to produce a rounded result.
int _mpy32 (int src1, int src2); MPY32 Returns the 32 LSBs of a 32 by 32 multiply.
long long _mpy32ll (int src1, int src2); MPY32 Returns all 64 bits of a 32 by 32 multiply. Values can be signed or
long long _mpy32su (int src1, int src2); MPY32SU unsigned.
long long _mpy32us (unsigned src1, int src2); MPY32US
long long _mpy32u (unsigned src1, unsigned src2); MPY32U
int _rpack2 (int src1, int src2); RPACK2 Shifts src1 and src2 left by 1 with saturation. The 16 MSBs of the
shifted src1 is placed in the 16 MSBs of the long long. The 16
MSBs of the shifted src2 is placed in the 16 LSBs of the long
long.
long long _saddsub (unsigned src1, unsigned src2); SADDSUB Performs a saturated addition and a saturated subtraction in
parallel.
long long _saddsub2 (unsigned src1, unsigned src2); SADDSUB2 Performs a SADD2 and a SSUB2 in parallel.
long long _shfl3 (unsigned src1, unsigned src2); SHFL3 Takes two 16-bit values from src1 and 16 LSBs from src2 to
perform a 3-way interleave, creating a 48-bit result.
int _smpy32 (int src1, int src2); SMPY32 Returns the 32 MSBs of a 32 by 32 multiply shifted left by 1.
int _ssub2 (unsigned src1, unsigned src2); SSUB2 Subtracts the upper and lower halves of src2 from the upper and
lower halves of src1 and saturates each result.
unsigned _xormpy (unsigned src1, unsigned src2); XORMPY Performs a Galois Field multiply

The intrinsics listed in Table 7-7 can be used for C6700, C6700+, C6740, and C6600 devices. The
intrinsics shown correspond to the indicated C6000 assembly language instruction(s). See the
TMS320C6000 CPU and Instruction Set Reference Guide for more information.
See Table 7-4 for the listing of generic C6000 intrinsics. See Table 7-5 for the listing of C6400-specific
intrinsics, which are also compatible with C6400+, C6740 and C6600. See Table 7-6 for the listing of
C6400+-specific intrinsics, which are also compatible with C6740 and C6600 devices. See Table 7-8 for
the listing of C6600-specific intrinsics.

Table 7-7. TMS320C6700, C6700+, C6740, and C6600 C/C++ Compiler Intrinsics
Assembly
C/C++ Compiler Intrinsic Instruction Description
int _dpint (double src); DPINT Converts 64-bit double to 32-bit signed integer, using the rounding
mode set by the CSR register
__int40_t _f2tol(__float2_t src); Reinterprets a __float2_t register pair src as an __int40_t (stored as a
register pair). You must include c6x.h.
__float2_t _f2toll(__float2_t src); Reinterprets a __float2_t register pair as a long long register pair. You
must include c6x.h.
double _fabs (double src); ABSDP Returns absolute value of src
float _fabsf (float src); ABSSP
__float2_t _lltof2(long long src); Reinterprets a long long register pair as a __float2_t register pair. You
must include c6x.h.
__float2_t _ltof2(__int40_t src); Reinterprets an __int40_t register pair as a __float2_t register pair. You
must include c6x.h.
__float2_t & _mem8_f2(void * ptr); LDNDW Allows unaligned loads and stores of 8 bytes to memory (1)
STNDW
const __float2_t & _mem8_f2_const(void * ptr); LDNDW Allows unaligned loads of 8 bytes from memory (1)
STNDW
long long _mpyidll (int src1, int src2); MPYID Produces a signed integer multiply. The result is placed in a register
pair.

(1)
See the TMS320C6000 Programmer's Guide for more information.

SPRU187U – July 2012 Run-Time Environment 217

Table 7-7. TMS320C6700, C6700+, C6740, and C6600 C/C++ Compiler Intrinsics (continued)
Assembly
C/C++ Compiler Intrinsic Instruction Description
double_mpysp2dp (float src1, float src2); MPYSP2DP (C6700+, C6740, and C6600 only) Produces a double-precision
floating-point multiply. The result is placed in a register pair.
double_mpyspdp (float src1, double src2); MPYSPDP (C6700+, C6740, and C6600 only) Produces a double-precision
floating-point multiply. The result is placed in a register pair.
double _rcpdp (double src); RCPDP Computes the approximate 64-bit double reciprocal
float _rcpsp (float src); RCPSP Computes the approximate 32-bit float reciprocal
double _rsqrdp (double src); RSQRDP Computes the approximate 64-bit double square root reciprocal
float _rsqrsp (float src); RSQRSP Computes the approximate 32-bit float square root reciprocal
int _spint (float); SPINT Converts 32-bit float to 32-bit signed integer, using the rounding mode
set by the CSR register

The intrinsics listed in Table 7-8 are included only for C6600 devices. These intrinsics are in addition to
those listed in Table 7-5 and Table 7-6. The intrinsics shown correspond to the indicated C6000 assembly
language instruction(s). See the TMS320C6000 CPU and Instruction Set Reference Guide for more
information.
See Table 7-4 for the listing of generic C6000 intrinsics. See Table 7-5 for the listing of C6400-specific
intrinsics, which are also compatible with C6400+, C6740 and C6600. See Table 7-6 for the listing of
C6400+-specific intrinsics, which are also compatible with C6740 and C6600 devices. See Table 7-7 for
the listing of C6700-specific intrinsics.

Table 7-8. TMS320C6600 C/C++ Compiler Intrinsics

Assembly
C/C++ Compiler Intrinsic Instruction Description
ADDDP No intrinsic. Use native C: a + b where a and b are
doubles
ADDSP No intrinsic. Use native C: a + b where a and b are floats.
AND No intrinsic: Use native C: "a & b" where a and b are long
longs
ANDN No intrinsic: Use native C: "a & ~b" where a and b are long
longs
MPYSP No intrinsic. Use native C: a * b where a and b are doubles
OR No intrinsic: Use native C: "a | b" where a and b are long
longs
SUBDP No intrinsic. Use native C: a - b where a and b are doubles
SUBSP No intrinsic. Use native C: a - b where a and b are floats
XOR No intrinsic: Use native C: "a ^ b" where a and b are long
longs. See also _xorll_c().
__x128_t _ccmatmpy (long long src1, __x128_t src2); CMATMPY Multiply the conjugate of 1x2 complex vector by a 2x2
complex matrix, producing two 64-bit results.
long long _ccmatmpyr1 (long long src1, CCMATCMPYR1 Multiply the complex conjugate of a 1x2 complex vector by
__x128_t src2); a 2x2 complex matrix, producing two 32-bit complex
results.
long long _ccmpy32r1 (long long src1, long long src2); CCMPY32R1 32-bit complex conjugate multiply of Q31 numbers with
rounding
__x128_t _cmatmpy (long long src1, __x128_t src2); CMATMPY Multiply a 1x2 vector by a 2x2 complex matrix, producing
two 64-bit complex results.
long long _cmatmpyr1 (long long src1, __x128_t src2); CMATMPYR1 Multiply a 1x2 complex vector by a 2x2 complex matrix,
producing two 32-bit complex results.
long long _cmpy32r1 (long long src1, long long src2); CMPY32R1 32-bit complex multiply of Q31 numbers with rounding
__x128_t _cmpysp (__float2_t src1, __float2_t src2); CMPYSP Perform the multiply operations for a complex multiply of
two complex numbers (See also _complex_mpysp and
_complex_conjugate_mpysp.)

218 Run-Time Environment SPRU187U – July 2012

Table 7-8. TMS320C6600 C/C++ Compiler Intrinsics (continued)

Assembly
C/C++ Compiler Intrinsic Instruction Description
double _complex_conjugate_mpysp (double src1, CMPYSP Performs a complex conjugate multiply by performing a
double src2); DSUBSP CMPYSP and DSUBSP
double _complex_mpysp (double src1, double src2); CMPYSP Performs a complex multiply by performing a CMPYSP
DADDSP and DADDSP
int _crot90 (int src); CROT90 Rotate complex number by 90 degrees
int _crot270 (int src); CROT270 Rotate complex number by 270 degrees
long long _dadd (long long src1, long long src2); DADD Two-way SIMD addition of signed 32-bit values producing
two signed 32-bit results.
long long _dadd2 (long long src1, long long src2); DADD2 Four-way SIMD addition of signed 16-bit values producing
four signed 32-bit results. (Two-way _add2)
__float2_t _daddsp (__float2_t src1, __float2_t src2); DADDSP Two-way SIMD addition of 32-bit single precision numbers
long long _dadd_c (scst5 immediate src1, DADD Addition of two signed 32-bit values by a single constant in
long long src2); src2 (-16 to 15) producing two signed 32-bit results.
long long _dapys2 (long long src1, long long src2); DAPYS2 Use the sign bit of src1 to determine whether to multiply
the four 16-bit values in src2 by 1 or -1. Yields four signed
16-bit results. (If src1 and src2 are the same register pair,
it is equivalent to a two-way _abs2).
long long _davg2 (long long src1, long long src2); DAVG2 Four-way SIMD average of signed 16-bit values, with
rounding. (Two-way _avg2
long long _davgnr2 (long long src1, long long src2); DAVGNR2 Four-way SIMD average of signed 16-bit values, without
rounding
long long _davgnru4 (long long src1, long long src2); DAVGNRU4 Eight-way SIMD average of unsigned 8-bit values, without
rounding
long long _davgu4 (long long src1, long long src2); DAVGU4 Eight-way SIMD average of unsigned 8-bit values, with
rounding. (Two-way _avgu4)
long long _dccmpyr1 (long long src1, long long src2); DCCMPYR1 Two-way SIMD complex multiply with rounding (_cmpyr1)
with complex conjugate of src2
unsigned _dcmpeq2 (long long src1, long long src2); DCMPEQ2 Four-way SIMD comparison of signed 16-bit values.
Results are packed into the four least-significant bits of the
return value. (Two-way _cmpeq2)
unsigned _dcmpeq4 (long long src1, long long src2); DCMPEQ4 Eight-way SIMD comparison of unsigned 8-bit values.
Results are packed into the eight least-significant bits of
the return value. (Two-way _cmpeq4)
unsigned _dcmpgt2 (long long src1, long long src2); DCMPGT2 Four-way SIMD comparison of signed 16-bit values.
Results are packed into the four least-significant bits of the
return value. (Two-way _cmpgt2)
unsigned _dcmpgtu4 (long long src1, long long src2); DCMPGTU4 Eight-way SIMD comparison of unsigned 8-bit values.
Results are packed into the eight least-significant bits of
the return value. (Two-way _cmpgtu4)
__x128_t _dccmpy (long long src1, long long src2); DCCMPY Two complex multiply operations on two sets of packed
complex numbers, with complex conjugate of src2.
__x128_t _dcmpy (long long src1, long long src2); DCMPY Performs two complex multiply operations on two sets of
packed complex numbers. (Two-way SIMD _cmpy).
long long _dcmpyr1 (long long src1, long long src2); DCMPYR1 Two-way SIMD complex multiply with rounding (_cmpyr1)
long long _dcrot90 (long long src); DCROT90 Two-way SIMD version of _crot90
long long _dcrot270 (long long src); DCROT270 Two-way SIMD version of _crot270
long long _ddotp4h (__x128_t src1, __x128_t src2 ); DDOTP4H Performs two dot-products between four sets of packed
16-bit values. (Two-way _dotp4h)
long long _ddotpsu4h (__x128_t src1, __x128_t src2 ); DDOTPSU4H Performs two dot-products between four sets of packed
16-bit values. (Two-way _dotpsu4h)
__float2_t _dinthsp (int src); DINTHSP Converts two packed signed 16-bit values into two single-
precision floating point values
__float2_t _dinthspu (unsigned src); DINTHSPU Converts two packed unsigned 16-bit values into two
single-precision float point values

SPRU187U – July 2012 Run-Time Environment 219

Table 7-8. TMS320C6600 C/C++ Compiler Intrinsics (continued)

Assembly
C/C++ Compiler Intrinsic Instruction Description
__float2_t _dintsp(long long src); DINTSP Converts two 32-bit signed integers to two single-precision
float point values.
__float2_t _dintspu(long long src); DINTSPU Converts two 32-bit unsigned integers to two single-
precision float point values.
long long _dmax2 (long long src1, long long src2); DMAX2 Four-way SIMD maximum of 16-bit signed values
producing four signed 16-bit results. (Two-way _max2)
long long _dmaxu4 (long long src1, long long src2); DMAXU4 8-way SIMD maximum of unsigned 8-bit values producing
eight unsigned 8-bit results. (Two-way _maxu4)
long long _dmin2 (long long src1, long long src2); DMIN2 Four-way SIMD minimum of signed 16-bit values
producing four signed 16-bit results. (Two-way _min2)
long long _dminu4 (long long src1, long long src2); DMINU4 8-way SIMD minimum of unsigned 8-bit values producing
eight unsigned 8-bit results. (Two-way _minu4)
__x128_t _dmpy2 (long long src1, long long src2); DMPY2 Four-way SIMD multiply of signed 16-bit values producing
four signed 32-bit results. (Two-way _mpy2)
__float2_t _dmpysp (__float2_t src1, __float2_t src2); DMPYSP Two-way single precision floating point multiply producing
two single-precision results
__x128_t _dmpysu4 (long long src1, long long src2); DMPYSU4 Eight-way SIMD multiply of signed 8-bit values by
unsigned 8-bit values producing eight signed 16-bit results.
(Two-way _mpysu4)
__x128_t _dmpyu2 (long long src1, long long src2); DMPYU2 Four-way SIMD multiply of unsigned 16-bit values
producing four unsigned 32-bit results. (Two-way _mpyu2)
__x128_t _dmpyu4 (long long src1, long long src2); DMPYU4 Eight-way SIMD multiply of signed 8-bit values producing
eight signed 16-bit results. (Two-way _mpyu4)
long long _dmvd (long long src1, unsigned src2 ); DMVD Places src1 in the low register of the long long and src2 in
the high register of the long long. Takes four cycles. See
also _dmv(), _fdmv_f2, and _itoll().
int _dotp4h (long long src1, long long src2 ); DDOTP4H Multiply two sets of four signed 16-bit values and return
the 32-bit sum.
long long _dotp4hll (long long src1, long long src2 ); DOTP4H Multiply two sets of four signed 16-bit values and return
the 64-bit sum.
int _dotpsu4h (long long src1, long long src2); DOTPSU4H Multiply four signed 16-bit values by four unsigned 16-bit
values and return the 32-bit sum.
long long _dotspu4hll (long long src1, long long src2); DOTPSU4H Multiply four signed 16-bit values by four unsigned 16-bit
values and return the 64-bit sum.
long long _dpackh2 (long long src1, long long src2); DPACKH2 Two-way _packh2
long long _dpackh4 (long long src1, long long src2); DPACKH4 Two-way _packh4
long long _dpacklh2 (long long src1, long long src2); DPACKLH2 Two-way _packlh2
long long _dpacklh4 (unsigned src1, unsigned src2); DPACKLH4 Performs a _packl4 and a _packh4. The output of the
_packl4 is in the low register of the result and the output of
the _packh4 is in the high register of the result.
long long _dpackl2 (long long src1, long long src2); DPACKL2 Two-way _packl2
long long _dpackl4 (long long src1, long long src2); DPACKL4 Two-way _packl4
long long _dsadd (long long src1, long long src2); DSADD Two-way SIMD saturated addition of signed 32-bit values
producing two signed 32-bit results. (Two-way _sadd)
long long _dsadd2 (long long src1, long long src2); DSADD2 Four-way SIMD saturated addition of signed 16-bit values
producing four signed 16-bit results. (Two-way _sadd2)
long long _dshl (long long src1, unsigned src2); DSHL Shift-left of two signed 32-bit values by a single value in
the src2 argument.
long long _dshl2 (long long src1, unsigned src2); DSHL2 Shift-left of four signed 16-bit values by a single value in
the src2 argument. (Two-way _shl2)
long long _dshr (long long src1, unsigned src2); DSHR Shift-right of two signed 32-bit values by a single value in
the src2 argument.
long long _dshr2 (long long src1, unsigned src2); DSHR2 Shift-right of four signed 16-bit values by a single value in
the src2 argument. (Two-way _shr2)

220 Run-Time Environment SPRU187U – July 2012

Table 7-8. TMS320C6600 C/C++ Compiler Intrinsics (continued)

Assembly
C/C++ Compiler Intrinsic Instruction Description
long long _dshru (long long src1, unsigned src2); DSHRU Shift-right of two unsigned 32-bit values by a single value
in the src2 argument.
long long _dshru2 (long long src1, unsigned src2); DSHRU2 Shift-right of four unsigned 16-bit values by a single value
in the src2 argument. (Two-way _shru2)
__x128_t _dsmpy2 (long long src1, long long src2); DSMPY2 Four-way SIMD multiply of signed 16-bit values with 1-bit
left-shift and saturate producing four signed 32-bit results.
(Two-way _smpy2)
long long _dspacku4 (long long src1, long long src2); DSPACKU4 Two-way _spacku4
long long _dspint (__float2_t src); DSPINT Converts two packed single-precision floating point values
to two signed 32-bit values
unsigned _dspinth (__float2_t src); DSPINTH Converts two packed single-precision floating point values
to two packed signed 16-bit values
long long _dssub (long long src1, long long src2); DSSUB Two-way SIMD saturated subtraction of 32-bit signed
values producing two signed 32-bit results.
long long _dssub2 (long long src1, long long src2); DSSUB2 Four-way SIMD saturated subtraction of signed 16-bit
values producing four signed 16-bit results. (Two-way
_ssub2)
long long _dsub (long long src1, long long src2); DSUB Two-way SIMD subtraction of 32-bit signed values
producing two signed 32-bit results.
long long _dsub2 (long long src1, long long src2); DSUB2 Four-way SIMD subtraction of signed 16-bit values
producing four signed 16-bit results. (Two-way _sub2)
__float2_t _dsubsp (__float2_t src1, __float2_t src2); DSUBSP Two-way SIMD subtraction of 32-bit single precision
numbers
long long _dxpnd2 (unsigned src); DXPND2 Expand four lower bits to four 16-bit fields.
long long _dxpnd4 (unsigned src); DXPND4 Expand eight lower bits to eight 8-bit fields.
__float2_t _fdmvd_f2(float src1, float src2); DMVD Places src1 in the low register of the __float2_t and src2 in
the high register of the __float2_t. Takes four cycles. See
also _dmv(), _dmvd(), and _itoll(). You must include c6x.h.
int _land (int src1, int src2); LAND Logical AND of src1 and src2
int _landn (int src1, int src2); LANDN Logical AND of src1 and NOT of src2; i.e. src1 AND ~src2
int _lor (int src1, int src2); LOR Logical OR of src1 and src2
void _mfence(); MFENCE Stall CPU while memory system is busy
double_mpysp2dp (float src1, float src2); MPYSP2DP (C6600 and C6700+ only) Produces a double-precision
floating-point multiply. The result is placed in a register
pair.
double_mpyspdp (float src1, double src2); MPYSPDP (C6600 and C6700+ only) Produces a double-precision
floating-point multiply. The result is placed in a register
pair.
long long _mpyu2 (unsigned src1, unsigned src2 ); MPYU2 Two-way SIMD multiply of unsigned 16-bit values
producing two unsigned 32-bit results.
__x128_t _qmpy32 (__x128_t src1, __x128_t src2); QMPY32 Four-way SIMD multiply of signed 32-bit values producing
four 32-bit results. (Four-way _mpy32)
__x128_t _qmpysp (__x128_t src1, __x128_t src2); QMPYSP Four-way SIMD 32-bit single precision multiply producing
four 32-bit single precision results
__x128_t _qsmpy32r1 (__x128_t src1, __x128_t src2); QSMPY32R1 4-way SIMD fractional 32-bit by 32-bit multiply where each
result value is shifted right by 31 bits and rounded. This
normalizes the result to lie within -1 and 1 in a Q31
fractional number system.
unsigned _shl2 (unsigned src1, unsigned src2); SHL2 Shift-left of two signed 16-bit values by a single value in
the src2 argument.
long long _unpkbu4 (unsigned src); UNPKBU4 Unpack four unsigned 8-bit values into four unsigned 16-bit
values. (See also _unpklu4 and _unpkhu4)
long long _unpkh2 (unsigned src); UNPKH2 Unpack two signed 16-bit values to two signed 32-bit
values

SPRU187U – July 2012 Run-Time Environment 221

Table 7-8. TMS320C6600 C/C++ Compiler Intrinsics (continued)

Assembly
C/C++ Compiler Intrinsic Instruction Description
long long _unpkhu2 (unsigned src); UNPKHU2 Unpack two unsigned 16-bit values to two unsigned 32-bit
values
long long _xorll_c (scst5 immediate src1, long long src2); XOR XOR src1 with the upper and lower 32-bit portions of src2
(SIMD XOR by constant)

For details on the __x128_t container type see Section 7.5.6.

222 Run-Time Environment SPRU187U – July 2012

7.5.6 The __x128_t Container Type

The __x128_t container type is for storing 128-bits of data and its use is necessary when performing
certain SIMD operations on C6600. This type can only be used when compiling for C6600. Also, note the
leading double underscore. When using the __x128_t container type, you must include c6x.h.
This type can be used to define objects that can be used with certain C6600 intrinsics. (See Table 7-8.)
The object can be filled and manipulated using various intrinsics. The type is not a full-fledged built-in type
(like long long), and so various native C operations are not allowed. Think of this type as a struct with
private members and special manipulation functions.
When the compiler puts a __x128_t object in the register file, the __x128_t object takes four registers (a
register quad).
Objects of type __x128_t are aligned to a 128-bit boundary in memory.
The following operations are supported:
• Declare a __x128_t global object (for example: __x128_t a;). By default, it will be put in the .far section.
• Declare a __x128_t local object (for example: __x128_t a;). It will be put on the stack.
• Declare a __x128_t global/local pointer (for example: __x128_t *a;).
• Declare an array of __x128_t objects (for example: __x128_t a[10];).
• Declare a __x128_t type as a member of a struct, class, or union.
• Assign a __x128_t object to another __x128_t object.
• Pass a __x128_t object to a function (including variadic argument functions). (Pass by value.)
• Return a __x128_t object from a function.
• Use 128-bit manipulation intrinsics to set and extract contents (see Table 7-9).
The following operations are not supported:
• Native-type operations on __x128_t objects, such as +, -, *, etc.
• Cast an object to a __x128_t type.
• Access the elements of a __x128_t using array or struct notation.
• Pass a __x128_t object to I/O functions like printf. Instead, extract the values from the __x128_t object
by using appropriate intrinsics.

Example 7-7. The __x128_t Container Type

#include <c6x.h>
#include <stdio.h>
__x128_t mpy_four_way_example(__x128_t s, int a, int b, int c, int d)
{
__x128_t t = _ito128(a, b, c, d); // Pack values into a __x128_t
__x128_t results = _qmpy32(s, t); // Perform a four-way SIMD multiply

int lowest32 = _get32_128(results, 0); // Extract lowest reg of __x128_t

int highest32 = _get32_128(results, 3); // Extract highest reg of __x128_t

printf("lowest = %d\n", lowest32);

printf("highest = %d\n", highest32);

return results;
}

Include c6x.h With Type __x128_t or __float2_t

NOTE: When using the __x128_t container type, or __float2_t typedef, or any intrinsics involving
__float2_t, you must include c6x.h.

SPRU187U – July 2012 Run-Time Environment 223

Table 7-9. Vector-in-Scalar Support C/C++ Compiler v7.2 Intrinsics

C/C++ Compiler Intrinsic Description
Creation
__x128_t _ito128 (unsigned src1, unsigned src2, unsigned src3, Creates __x128_t from (u)int (reg+3, reg+2, reg+1, reg+0)
unsigned src4);
__x128_t _fto128 (float src1, float src2, float src3, float src4); Creates __x128_t from float (reg+3, reg+2, reg+1, reg+0)
__x128_t _llto128 (long long src1, long long src2); Creates __x128_t from two long longs
__x128_t _dto128 (double src1, double src2); Creates __x128_t from two doubles
__x128_t _f2to128(__float2_t src1, __float2_t src2); Creates __x128_t from two __float2_t objects
__x128_t _dup32_128 (int src); Creates __x128_t from duplicating src1
__float2_t _ftof2(float src1, float src2); Creates __float2_t from two floats
Extraction
float _hif (double src); Extracts upper float from double
float _lof (double src); Extracts lower float from double
float _hif2(__float2_t src); Extracts upper float from __float2_t
float _lof2(__float2_t src); Extracts lower float from __float2_t
long long _hi128 (__x128_t src); Extracts upper two registers of register quad
double _hid128 (__x128_t src); Extracts upper two registers of register quad
__float2_t _hif2_128(__x128_t src); Extracts upper two registers of register quad
long long _lo128 (__x128_t src); Extracts lower two registers of register quad
double _lod128 (__x128_t src); Extracts lower two registers of register quad
__float2_t _lof2_128(__x128_t src); Extracts lower two registers of register quad
unsigned _get32_128 (__x128_t src, 0); Extracts first register of register quad (base reg + 0)
unsigned _get32_128 (__x128_t src, 1); Extracts second register of register quad (base reg + 1)
unsigned _get32_128 (__x128_t src, 2); Extracts third register of register quad (base reg + 2)
unsigned _get32_128 (__x128_t src, 3); Extracts fourth register of register quad (base reg + 3)
float _get32f_128 (__x128_t src, 0); Extracts first register of register quad (base reg + 0)
float _get32f_128 (__x128_t src, 1); Extracts second register of register quad (base reg + 1)
float _get32f_128 (__x128_t src, 2); Extracts third register of register quad (base reg + 2)
float _get32f_128 (__x128_t src, 3); Extracts fourth register of register quad (base reg + 3)

7.5.7 The __float2_t Container Type

The __float2_t container type should be used (instead of double) to store two floats. There are
manipulation intrinsics to create and manipulate objects with the __float2_t type (see Table 7-9). The run-
time-support file, c6x.h, must be included when using __float2_t or when using any of the __float2_t
manipulation intrinsics.
Recommendations for using the __float2_t type:
• Use __float2_t to store two floats. Do not use double.
• Use long long to store 64-bit packed integer data. Do not use double or __float2_t for packed integer
data.

224 Run-Time Environment SPRU187U – July 2012

7.5.8 Using Intrinsics for Interrupt Control and Atomic Sections

The C/C++ compiler supports three intrinsics for enabling, disabling, and restoring interrupts. The
syntaxes are:

unsigned int _disable_interrupts ( );

unsigned int _enable_interrupts ( );
void _restore_interrupts (unsigned int);

The _disable_interrupts() and _enable_interrupts( ) intrinsics both return an unsigned int that can be
subsequently passed to _restore_interrupts( ) to restore the previous interrupt state. These intrinsics
provide a barrier to optimization and are therefore appropriate for implementing a critical (or atomic)
section. For example,
unsigned int restore_value;

restore_value = _disable_interrupts();
if (sem) sem--;
_restore_interrupts(restore_value);

The example code disables interrupts so that the value of sem read for the conditional clause does not
change before the modification of sem in the then clause. The intrinsics are barriers to optimization, so the
memory reads and writes of sem do not cross the _disable_interrupts or _restore_interrupts locations.

Overwrites CSR
NOTE: The _restore_interrupts( ) intrinsic overwrites the CSR control register with the value in the
argument. Any CSR bits changed since the _disable_interrupts( ) intrinsic or
_enable_interrupts( ) intrinsic will be lost.

On C6400+, C6740, and C6600, the _restore_interrupts( ) intrinsic does not use the RINT instruction.

7.5.9 Using Unaligned Data and 64-Bit Values

The C6400, C6400+, C6740, and C6600 families have support for unaligned loads and stores of 64-bit
and 32-bit values via the use of the _mem8, _memd8, and _mem4 intrinsics. The _lo and _hi intrinsics are
useful for extracting the two 32-bit portions from a 64-bit double. The _loll and _hill intrinsics are useful for
extracting the two 32-bit portions from a 64-bit long long.
For the C6400+, C6740, and C6600 intrinsics that use 64-bit types, the equivalent C type is long long. Do
not use the C type double or the compiler performs a call to a run-time-support math function to do the
floating-point conversion. Here are ways to access 64-bit and 32-bit values:
• To get the upper 32 bits of a long long in C code, use >> 32 or the _hill( ) intrinsic.
• To get the lower 32 bits of a long long in C code, use a cast to int or unsigned, or use the _loll intrinsic.
• To get the upper 32 bits of a double (interpreted as an int), use _hi( ).
• To get the lower 32 bits of a double (interpreted as an int), use _lo( ).
• To get the upper 32 bits of a __float2_t, use _hif2().
• To get the lower 32 bits of a __float2_t, use _lof2().
• To create a long long value, use the _itoll(int high32bits, int low32bits) intrinsic.
• To create a __float2_t value, use the _ftof2(float high32bits, float low32bits) intrinsic.
Example 7-8 shows the usage of the _mem8 intrinsic.

SPRU187U – July 2012 Run-Time Environment 225

Example 7-8. Using the _mem8 Intrinsic

void alt_load_longlong_unaligned(void a, int high, int *low)

{
long long p = _mem8(a);
*high = p >> 32;
*low = (unsigned int) p;
}

7.5.10 Using MUST_ITERATE and _nassert to Enable SIMD and Expand Compiler
Knowledge of Loops
Through the use of MUST_ITERATE and _nassert, you can guarantee that a loop executes a certain
number of times.
This example tells the compiler that the loop is guaranteed to run exactly 10 times:
#pragma MUST_ITERATE(10,10);
for (I = 0; I < trip_count; I++) { ...

MUST_ITERATE can also be used to specify a range for the trip count as well as a factor of the trip count.
For example:
#pragma MUST_ITERATE(8,48,8);
for (I = 0; I < trip; I++) { ...

This example tells the compiler that the loop executes between 8 and 48 times and that the trip variable is
a multiple of 8 (8, 16, 24, 32, 40, 48). The compiler can now use all this information to generate the best
loop possible by unrolling better even when the --interrupt_thresholdn option is used to specify that
interrupts do occur every n cycles.
The TMS320C6000 Programmer's Guide states that one of the ways to refine C/C++ code is to use word
accesses to operate on 16-bit data stored in the high and low parts of a 32-bit register. Examples using
casts to int pointers are shown with the use of intrinsics to use certain instructions like _mpyh. This can be
automated by using the _nassert(); intrinsic to specify that 16-bit short arrays are aligned on a 32-bit
(word) boundary.
The following examples generate the same assembly code:
• Example 1
int dot_product(short *x, short *y, short z)
{
int *w_x = (int *)x;
int *w_y = (int *)y;
int sum1 = 0, sum2 = 0, I;
for (I = 0; I < z/2; I++)
{
sum1 += _mpy(w_x[i], w_y[i]);
sum2 += _mpyh(w_x[i], w_y[i]);
}
return (sum1 + sum2);
}
• Example 2
int dot_product(short *x, short *y, short z)
{
int sum = 0, I;

_nassert (((int)(x) & 0x3) == 0);

_nassert (((int)(y) & 0x3) == 0);
#pragma MUST_ITERATE(20, , 4);
for (I = 0; I < z; I++) sum += x[i] * y[i];
return sum;
}

226 Run-Time Environment SPRU187U – July 2012

C++ Syntax for _nassert

NOTE: In C++ code, _nassert is part of the standard namespace. Thus, the correct syntax is
std::_nassert().

7.5.11 Methods to Align Data

In the following code, the _nassert tells the compiler, for every invocation of f(), that ptr is aligned to an 8-
byte boundary. Such an assertion often leads to the compiler producing code which operates on multiple
data values with a single instruction, also known as SIMD (single instruction multiple data) optimization.
void f(short *ptr)
{
_nassert((int) ptr % 8 == 0)

; a loop operating on data accessed by ptr

}

The following subsections describe methods you can use to ensure the data referenced by ptr is aligned.
You have to employ one of these methods at every place in your code where f() is called.

7.5.11.1 Base Address of an Array

An argument such as ptr is most commonly passed the base address of an array, for example:
short buffer[100];
...
f(buffer);

When compiling for C6600 devices, such an array is automatically aligned to a 16-byte boundary. When
compiling for C6400, C6400+, C6740, and C6600 devices, such an array is automatically aligned to an 8-
byte boundary. When compiling for C6200 or C6700, such an array is automatically aligned to 4-byte
boundary, or, if the base type requires it, an 8-byte boundary. This is true whether the array is global,
static, or local. This automatic alignment is all that is required to achieve SIMD optimization on those
respective devices. You still need to include the _nassert because, in the general case, the compiler
cannot guarantee that ptr holds the address of a properly aligned array.
If you always pass the base address of an array to pointers like ptr, then you can use the following macro
to reflect that fact.
#if defined(_TMS320C6600)
#define ALIGNED_ARRAY(ptr) _nassert((int) ptr % 16 == 0)
#elif defined(_TMS320C6400)
#define ALIGNED_ARRAY(ptr) _nassert((int) ptr % 8 == 0)
#elif defined(_TMS320C6200) || defined(_TMS320C6700)
#define ALIGNED_ARRAY(ptr) _nassert((int) ptr % 4 == 0)
#else
#define ALIGNED_ARRAY(ptr) /* empty */
#endif

void f(short *ptr)

{
ALIGNED_ARRAY(ptr);
; a loop operating on data accessed by ptr
}

The macro works regardless of which C6000 device you build for, or if you port the code to another target.

7.5.11.2 Offset from the Base of an Array

A more rare case is to pass the address of an offset from an array, for example:
f(&buffer[3]);

This code passes an unaligned address to ptr, thus violating the presumption coded in the _nassert().
There is no direct remedy for this case. Avoid this practice whenever possible.

SPRU187U – July 2012 Run-Time Environment 227

7.5.11.3 Dynamic Memory Allocation

Ordinary dynamic memory allocation guarantees that the allocated memory is properly aligned for any
scalar object of a native type (for instance, it is correctly aligned for a long double or long long int), but
does not guarantee any larger alignment. For example:
buffer = calloc(100, sizeof(short))

To get a stricter alignment, use the function memalign with the desired alignment. To get an alignment of
256 bytes for example:
buffer = memalign(256, 100 * sizeof(short);

If you are using BIOS memory allocation routines, be sure to pass the alignment factor as the last
argument using the syntax that follows:

buffer = MEM_alloc( segid , 100 * sizeof(short), 256);

See the TMS320C6000 DSP/BIOS Help for more information about BIOS memory allocation routines and
the segid parameter in particular.

7.5.11.4 Member of a Structure or Class

Arrays which are members of a structure or a class are aligned only as the base type of the array
requires. The automatic alignment described in Section 7.5.11.1 does not occur.

Example 7-9. An Array in a Structure

struct s
{
...
short buf1[50];
...
} g;

...

f(g.buf1);

Example 7-10. An Array in a Class

class c
{
public :
short buf1[50];
void mfunc(void);
...
};

void c::mfunc()
{
f(buf1);
...
}

228 Run-Time Environment SPRU187U – July 2012

To align an array in a structure, place it inside a union with a dummy object that has the desired
alignment. If you want 8 byte alignment, use a "long long" dummy field. For example:
struct s
{
union u
{ long long dummy; /* 8-byte alignment */
short buffer[50]; /* also 8-byte alignment */
} u;
...
};

If you want to declare several arrays contiguously, and maintain a given alignment, you can do so by
keeping the array size, measured in bytes, an even multiple of the desired alignment. For example:
struct s
{
long long dummy; /* 8-byte alignment */
short buffer[50]; /* also 8-byte alignment */
short buf2[50]; /* 4-byte alignment */
...
};

Because the size of buf1 is 50 * 2-bytes per short = 100 bytes, and 100 is an even multiple of 4, not 8,
buf2 is only aligned on a 4-byte boundary. Padding buf1 out to 52 elements makes buf2 8-byte aligned.
Within a structure or class, there is no way to enforce an array alignment greater than 8. For the purposes
of SIMD optimization, this is not necessary.

Alignment With Program-Level Optimization

NOTE: In most cases program-level optimization (see Section 3.7) entails compiling all of your
source files with a single invocation of the compiler, while using the -pm -o3 options. This
allows the compiler to see all of your source code at once, thus enabling optimizations that
are rarely applied otherwise. Among these optimizations is seeing that, for instance, all of the
calls to the function f() are passing the base address of an array to ptr, and thus ptr is always
correctly aligned for SIMD optimization. In such a case, the _nassert() is not required. The
compiler automatically determines that ptr must be aligned, and produces the optimized
SIMD instructions.

7.5.12 SAT Bit Side Effects

The saturated intrinsic operations define the SAT bit if saturation occurs. The SAT bit can be set and
cleared from C/C++ code by accessing the control status register (CSR). The compiler uses the following
steps for generating code that accesses the SAT bit:
1. The SAT bit becomes undefined by a function call or a function return. This means that the SAT bit in
the CSR is valid and can be read in C/C++ code until a function call or until a function returns.
2. If the code in a function accesses the CSR, then the compiler assumes that the SAT bit is live across
the function, which means:
• The SAT bit is maintained by the code that disables interrupts around software pipelined loops.
• Saturated instructions cannot be speculatively executed.
3. If an interrupt service routine modifies the SAT bit, then the routine should be written to save and
restore the CSR.

SPRU187U – July 2012 Run-Time Environment 229

7.5.13 IRP and AMR Conventions

There are certain assumptions that the compiler makes about the IRP and AMR control registers. The
assumptions should be enforced in all programs and are as follows:
1. The AMR must be set to 0 upon calling or returning from a function. A function does not have to save
and restore the AMR, but must ensure that the AMR is 0 before returning.
2. The AMR must be set to 0 when interrupts are enabled, or the SAVE_AMR and STORE_AMR macros
should be used in all interrupts (see Section 7.6.3).
3. The IRP can be safely modified only when interrupts are disabled.
4. The IRP's value must be saved and restored if you use the IRP as a temporary register.

7.5.14 Floating Point Control Register Side Effects

When performing floating point operations on a floating-point architecture, status bits in certain control
registers may be set. In particular, status bits may be set in the FADCR, FAUCR, and FMCR registers,
hereafter referred to as the "floating point control registers". These bits can be set and cleared from C/C++
code by writing to or reading from these registers, as shown in example 6-1.
In compiler versions released after July of 2009, the compiler uses the following steps for generating code
that accesses any of the floating point control registers.
1. The floating point control registers become undefined by a function call or a function return. The means
the data in the floating point control registers is valid and can be read in C/C++ code until a function
call or a function returns.
2. If the code in a function accesses any of the floating point control registers, the compiler assumes that
those registers are live across the function, which means that floating point instructions that may set
bits in those floating point control registers cannot be speculatively executed.
3. If an interrupt service routine modifies any of the bits in a floating point control register, the interrupt
service routine should be written to save and restore that floating point control register.

7.6 Interrupt Handling

As long as you follow the guidelines in this section, you can interrupt and return to C/C++ code without
disrupting the C/C++ environment. When the C/C++ environment is initialized, the startup routine does not
enable or disable interrupts. If the system is initialized by way of a hardware reset, interrupts are disabled.
If your system uses interrupts, you must handle any required enabling or masking of interrupts. Such
operations have no effect on the C/C++ environment and are easily incorporated with asm statements or
calling an assembly language function.

7.6.1 Saving the SGIE Bit

When compiling for C6400+, C6740, and C6600, the compiler may use the C6400+-, C6740-, and C6600-
specific instructions DINT and RINT to disable and restore interrupts around software-pipelined loops.
These instructions utilize the CSR control register as well as the SGIE bit in the TSR control register.
Therefore, the SGIE bit is considered to be save-on-call. If you have assembly code that calls compiler-
generated code, the SGIE bit should be saved (e.g. to the stack) if it is needed later. The SGIE bit should
then be restored upon return from compiler generated code.

7.6.2 Saving Registers During Interrupts

When C/C++ code is interrupted, the interrupt routine must preserve the contents of all machine registers
that are used by the routine or by any functions called by the routine. The compiler handles register
preservation if the interrupt service routine is written in C/C++ and declared with the interrupt keyword. For
C6400+, C6740, and C6600, the compiler will save and restore the ILC and RILC control registers if
needed.

230 Run-Time Environment SPRU187U – July 2012

7.6.3 Using C/C++ Interrupt Routines

A C/C++ interrupt routine is like any other C/C++ function in that it can have local variables and register
variables; however, it should be declared with no arguments and should return void. C/C++ interrupt
routines can allocate up to 32K on the stack for local variables. For example:
interrupt void example (void)
{
...
}

If a C/C++ interrupt routine does not call any other functions, only those registers that the interrupt handler
attempts to define are saved and restored. However, if a C/C++ interrupt routine does call other functions,
these functions can modify unknown registers that the interrupt handler does not use. For this reason, the
routine saves all usable registers if any other functions are called. Interrupts branch to the interrupt return
pointer (IRP). Do not call interrupt handling functions directly.
Interrupts can be handled directly with C/C++ functions by using the interrupt pragma or the interrupt
keyword. For more information, see Section 6.9.18 and Section 6.5.3, respectively.
You are responsible for handling the AMR control register and the SAT bit in the CSR correctly inside an
interrupt. By default, the compiler does not do anything extra to save/restore the AMR and the SAT bit.
Macros for handling the SAT bit and the AMR register are included in the c6x.h header file.
For example, you are using circular addressing in some hand assembly code (that is, the AMR does not
equal 0). This hand assembly code can be interrupted into a C code interrupt service routine. The C code
interrupt service routine assumes that the AMR is set to 0. You need to define a local unsigned int
temporary variable and call the SAVE_AMR and RESTORE_AMR macros at the beginning and end of
your C interrupt service routine to correctly save/restore the AMR inside the C interrupt service routine.

Example 7-11. AMR and SAT Handling

#include <c6x.h>

interrupt void interrupt_func()

{
unsigned int temp_amr;
/* define other local variables used inside interrupt */

/* save the AMR to a temp location and set it to 0 */

SAVE_AMR(temp_amr);

/* code and function calls for interrupt service routine */

...

/* restore the AMR for you hand assembly code before exiting */
RESTORE_AMR(temp_amr);
}

If you need to save/restore the SAT bit (i.e. you were performing saturated arithmetic when interrupted
into the C interrupt service routine which may also perform some saturated arithmetic) in your C interrupt
service routine, it can be done in a similar way as the above example using the SAVE_SAT and
RESTORE_SAT macros.
For C6400+, C6740, and C6600, the compiler saves and restores the ILC and RILC control registers if
needed.
For floating point architectures, you are responsible for handling the floating point control registers
FADCR, FAUCR and FMCR. If you are reading bits out of the floating pointer control registers, and if the
interrupt service routine (or any called function) performs floating point operations, then the relevant
floating point control registers should be saved and restored. No macros are provided for these registers,
as simple assignment to and from an unsigned int temporary will suffice.

SPRU187U – July 2012 Run-Time Environment 231

7.6.4 Using Assembly Language Interrupt Routines

You can handle interrupts with assembly language code as long as you follow the same register
conventions the compiler does. Like all assembly functions, interrupt routines can use the stack, access
global C/C++ variables, and call C/C++ functions normally. When calling C/C++ functions, be sure that
any registers listed in Table 7-2 are saved, because the C/C++ function can modify them.

7.7 Run-Time-Support Arithmetic Routines

The run-time-support library contains a number of assembly language functions that provide arithmetic
routines for C/C++ math operations that the C6000 instruction set does not provide, such as integer
division, integer remainder, and floating-point operations.
These routines follow the standard C/C++ calling sequence. The compiler automatically adds these
routines when appropriate; they are not intended to be called directly by your programs.
The source code for these functions is in the source library rts.src. The source code has comments that
describe the operation of the functions. You can extract, inspect, and modify any of the math functions. Be
sure, however, that you follow the calling conventions and register-saving rules outlined in this chapter.
Table 7-10 summarizes the run-time-support functions used for arithmetic.

Table 7-10. Summary of Run-Time-Support Arithmetic Functions

Type Function in COFF ABI Function in EABI Description
float _cvtdf (double) __c6xabi_cvtdf (double) Convert double to float
int _fixdi (double) __c6xabi_fixdi (double) Convert double to signed integer
long _fixdi (double) __c6xabi_fixdi (double) Convert double to long
long long _fixdlli (double) __c6xabi_fixdlli (double) Convert double to long long
uint _fixdlli (double) __c6xabi_fixdlli (double) Convert double to unsigned integer
ulong _fixdul (double) __c6xabi_fixdul (double) Convert double to unsigned long
ulong long _fixdull (double) __c6xabi_fixdull (double) Convert double to unsigned long long
double _cvtfd (float) __c6xabi_cvtfd (float) Convert float to double
int _fixfi (float) __c6xabi_fixfi (float) Convert float to signed integer
long _fixfli (float) __c6xabi_fixfli (float) Convert float to long
long long _fixflli (float) __c6xabi_fixflli (float) Convert float to long long
uint _fixfu (float) __c6xabi_fixfu (float) Convert float to unsigned integer
ulong _fixful (float) __c6xabi_fixful (float) Convert float to unsigned long
ulong long _fixfull (float) __c6xabi_fixfull (float) Convert float to unsigned long long
double _fltid (int) __c6xabi_fltid (int) Convert signed integer to double
float _fltif (int) __c6xabi_fltif (int) Convert signed integer to float
double _fltud (uint) __c6xabi_fltud (uint) Convert unsigned integer to double
float _fltuf (uint) __c6xabi_fltuf (uint) Convert unsigned integer to float
double _fltlid (long) __c6xabi_fltlid (long) Convert signed long to double
float _fltlif (long) __c6xabi_fltlif (long) Convert signed long to float
double _fltuld (ulong) __c6xabi_fltuld (ulong) Convert unsigned long to double
float _fltulf (ulong) __c6xabi_fltulf (ulong) Convert unsigned long to float
double _fltllid (long long) __c6xabi_fltllid (long long) Convert signed long long to double
float _fltllif (long long) __c6xabi_fltllif (long long) Convert signed long long to float
double _fltulld (ulong long) __c6xabi_fltulld (ulong long) Convert unsigned long long to double
float _fltullf (ulong long) __c6xabi_fltullf (ulong long) Convert unsigned long long to float
double _absd (double) __c6xabi_absd (double) Double absolute value
float _absf (float) __c6xabi_absf (float) Float absolute value
long _labs (long) __c6xabi_labs (long) Long absolute value
long long _llabs (long long) __c6xabi_llabs (long long) Long long absolute value
double _negd (double) __c6xabi_negd (double) Double negate value

232 Run-Time Environment SPRU187U – July 2012

Table 7-10. Summary of Run-Time-Support Arithmetic Functions (continued)

Type Function in COFF ABI Function in EABI Description
float _negf (float) __c6xabi_negf (float) Float negate value
long long _negll (long) __c6xabi_negll (long) Long long negate value
long long _llshl (long long) __c6xabi_llshl (long long) Long long shift left
long long _llshr (long long) __c6xabi_llshr (long long) Long long shift right
ulong long _llshru (ulong long) __c6xabi_llshru (ulong long) Unsigned long long shift right
double _addd (double, double) __c6xabi_addd (double, double) Double addition
double _cmpd (double, double) __c6xabi_cmpd (double, double) Double comparison
double _divd (double, double) __c6xabi_divd (double, double) Double division
double _mpyd (double, double) __c6xabi_mpyd (double, double) Double multiplication
double _subd (double, double) __c6xabi_subd (double, double) Double subtraction
float _addf (float, float) __c6xabi_addf (float, float) Float addition
float _cmpf (float, float) __c6xabi_cmpf (float, float) Float comparison
float _divf (float, float) __c6xabi_divf (float, float) Float division
float _mpyf (float, float) __c6xabi_mpyf (float, float) Float multiplication
float _subf (float, float) __c6xabi_subf (float, float) Float subtraction
int _divi (int, int) __c6xabi_divi (int, int) Signed integer division
int _remi (int, int) __c6xabi_remi (int, int) Signed integer remainder
uint _divu (uint, uint) __c6xabi_divu (uint, uint) Unsigned integer division
uint _remu (uint, uint) __c6xabi_remu (uint, uint) Unsigned integer remainder
long _divli (long, long) __c6xabi_divli (long, long) Signed long division
long _remli (long, long) __c6xabi_remli (long, long) Signed long remainder
ulong _divul (ulong, ulong) __c6xabi_divul (ulong, ulong) Unsigned long division
ulong _remul (ulong, ulong) __c6xabi_remul (ulong, ulong) Unsigned long remainder
long long _divlli (long long, long long) __c6xabi_divlii (long long, long long) Signed long long division
long long _remlli (long long, long long) __c6xabi_remlli (long long, long long) Signed long long remainder
ulong long _mpyll(ulong long, ulong long) __c6xabi_mpyll (long long, long long) Unsigned long long multiplication
ulong long _divull (ulong long, ulong long) __c6xabi_divull (ulong long, ulong long) Unsigned long long division
ulong long _remull (ulong long, ulong long) __c6xabi_remull (ulong long, ulong long) Unsigned long long remainder

SPRU187U – July 2012 Run-Time Environment 233

7.8 System Initialization

Before you can run a C/C++ program, you must create the C/C++ run-time environment. The C/C++ boot
routine performs this task using a function called c_int00 (or _c_int00). The run-time-support source
library, rts.src, contains the source to this routine in a module named boot.c (or boot.asm).
To begin running the system, the c_int00 function can be branched to or called, but it is usually vectored
to by reset hardware. You must link the c_int00 function with the other object files. This occurs
automatically when you use the --rom_model or --ram_model link option and include a standard run-time-
support library as one of the linker input files.
When C/C++ programs are linked, the linker sets the entry point value in the executable output file to the
symbol c_int00. This does not, however, set the hardware to automatically vector to c_int00 at reset (see
the TMS320C2x DSP CPU and Instruction Set Reference Guide, the TMS320C64x/C64x+ DSP CPU and
Instruction Set Reference Guide, the TMS320C67x/C67x+ DSP CPU and Instruction Set Reference
Guide, or the TMS320C66x+ DSP CPU and Instruction Set Reference Guide).
The c_int00 function performs the following tasks to initialize the environment:
1. Defines a section called .stack for the system stack and sets up the initial stack pointers
2. Performs C autoinitialization of global/static variables. For more information, see Section 7.8.1 for
COFF ABI mode and Section 7.8.2 for EABI mode.
3. Initializes global variables by copying the data from the initialization tables to the storage allocated for
the variables in the .bss section. If you are initializing variables at load time (--ram_model option), a
loader performs this step before the program runs (it is not performed by the boot routine). For more
information, see Section 7.8.1.
4. Calls C++ initialization routines for file scope construction from the global constructor table. For more
information, see Section 7.8.2.6 for EABI mode and Section 7.8.1.3 for COFF ABI mode.
5. Calls the function main to run the C/C++ program
You can replace or modify the boot routine to meet your system requirements. However, the boot routine
must perform the operations listed above to correctly initialize the C/C++ environment.

7.8.1 COFF ABI Automatic Initialization of Variables

Some global variables must have initial values assigned to them before a C/C++ program starts running.
The process of retrieving these variables' data and initializing the variables with the data is called
autoinitialization.
The COFF ABI compiler builds tables in a special section called .cinit that contains data for initializing
global and static variables. Each compiled module contains these initialization tables. The linker combines
them into a single table (a single .cinit section). The boot routine or a loader uses this table to initialize all
the system variables.

Initializing Variables
NOTE: In ANSI/ISO C, global and static variables that are not explicitly initialized must be set to 0
before program execution. The COFF ABI C/C++ compiler does not perform any
preinitialization of uninitialized variables. Explicitly initialize any variable that must have an
initial value of 0.

Global variables are either autoinitialized at run time or at load time; see Section 7.8.1.1 and
Section 7.8.1.2. Also see Section 6.13. In EABI mode, the compiler automatically zero initializes the
uninitialized variables. See Section 7.8.2 for details.

234 Run-Time Environment SPRU187U – July 2012

7.8.1.1 Autoinitialization of Variables at Run Time

Autoinitializing variables at run time is the default method of autoinitialization. To use this method, invoke
the linker with the --rom_model option.
Using this method, the .cinit section is loaded into memory along with all the other initialized sections, and
global variables are initialized at run time. The linker defines a special symbol called cinit that points to the
beginning of the initialization tables in memory. When the program begins running, the C/C++ boot routine
copies data from the tables (pointed to by .cinit) into the specified variables in the .bss section. This allows
initialization data to be stored in ROM and copied to RAM each time the program starts.
Figure 7-11 illustrates autoinitialization at run time. Use this method in any system where your application
runs from code burned into ROM.

Figure 7-11. Autoinitialization at Run Time

Object file Memory

cint Initialization
.cinit
Loader tables
section
(EXT_MEM)
Boot
routine
.bss
section
(D_MEM)

7.8.1.2 Initialization of Variables at Load Time

Initialization of variables at load time enhances performance by reducing boot time and by saving the
memory used by the initialization tables. To use this method, invoke the linker with the --ram_model
option.
When you use the --ram_model link option, the linker sets the STYP_COPY bit in the .cinit section's
header. This tells the loader not to load the .cinit section into memory. (The .cinit section occupies no
space in the memory map.) The linker also sets the cinit symbol to -1 (normally, cinit points to the
beginning of the initialization tables). This indicates to the boot routine that the initialization tables are not
present in memory; accordingly, no run-time initialization is performed at boot time.
A loader (which is not part of the compiler package) must be able to perform the following tasks to use
initialization at load time:
• Detect the presence of the .cinit section in the object file
• Determine that STYP_COPY is set in the .cinit section header, so that it knows not to copy the .cinit
section into memory
• Understand the format of the initialization tables
Figure 7-12 illustrates the initialization of variables at load time.

SPRU187U – July 2012 Run-Time Environment 235

Figure 7-12. Initialization at Load Time

Object file Memory

.cinit Loader

.bss

Regardless of the use of the --rom_model or --ram_model options, the .pinit section is always loaded and
processed at run time.

7.8.1.3 Global Constructors for COFF

All global C++ variables that have constructors must have their constructor called before main. The
compiler builds a table in a section called .pinit of global constructor addresses that must be called, in
order, before main. The linker combines the .pinit section form each input file to form a single table in the
.pinit section. The boot routine uses this table to execute the constructors. (See Section 7.8.2.6 for global
constructor details for EABI.)

7.8.2 EABI Automatic Initialization of Variables

Any global variables declared as preinitialized must have initial values assigned to them before a C/C++
program starts running. The process of retrieving these variables' data and initializing the variables with
the data is called autoinitialization.

7.8.2.1 Zero Initializing Variables

In ANSI C, global and static variables that are not explicitly initialized, must be set to 0 before program
execution. The C/C++ EABI compiler supports preinitialization of uninitialized variables by default. This
can be turned off by specifying the linker option --zero_init=off. COFF ABI does not support zero
initialization.

7.8.2.2 EABI Direct Initialization

The EABI compiler uses direct initialization to initialize global variables. For example, consider the
following C code:
int i = 23;
int a[5] = { 1, 2, 3, 4, 5 };

The compiler allocates the variables 'i' and 'a[] to .data section and the initial values are placed directly.
.global i
.data
.align 4
i:
.field 23,32 ; i @ 0

.global a
.data
.align 4
a:
.field 1,32 ; a[0] @ 0
.field 2,32 ; a[1] @ 32
.field 3,32 ; a[2] @ 64

236 Run-Time Environment SPRU187U – July 2012

Each compiled module that defines static or global variables contains these .data sections. The linker
treats the .data section like any other initialized section and creates an output section. In the load-time
initialization model, the sections are loaded into memory and used by the program. See Section 7.8.2.5.
In the run-time initialization model, the linker uses the data in these sections to create initialization data
and an additional initialization table. The boot routine processes the initialization table to copy data from
load addresses to run addresses. See Section 7.8.2.3.

7.8.2.3 Autoinitialization of Variables at Run Time in EABI Mode

Autoinitializing variables at run time is the default method of autoinitialization. To use this method, invoke
the linker with the --rom_model option.
Using this method, the linker creates an initialization table and initialization data from the direct initialized
sections in the compiled module. The table and data are used by the C/C++ boot routine to initialize
variables in RAM using the table and data in ROM.
Figure 7-13 illustrates autoinitialization at run time in EABI Mode. Use this method in any system where
your application runs from code burned into ROM.

Figure 7-13. Autoinitialization at Run Time in EABI Mode

Object file Memory

C auto init
table and data C auto init
(ROM) Loader table and data
(.cinit section) (ROM)
Boot
routine
.data
uninitialized
(RAM)

7.8.2.4 Autoinitialization Tables

In EABI mode, the compiled object files do not have initialization tables. The variables are initialized
directly . The linker, when the --rom_model option is specified, creates C auto initialization table and the
initialization data. The linker creates both the table and the initialization data in an output section named
.cinit.

Migration from COFF to ELF Initialization

NOTE: The name .cinit is used primarily to simplify migration from COFF to ELF format and the .cinit
section created by the linker has nothing in common (except the name) with the COFF cinit
records.

The autoinitialization table has the following format:

SPRU187U – July 2012 Run-Time Environment 237

_TI_CINIT_Base:
32-bit load address 32-bit run address

32-bit load address 32-bit run address

_TI_CINT_Limit:

The linker defined symbols __TI_CINIT_Base and __TI_CINIT_Limit point to the start and end of the
table, respectively. Each entry in this table corresponds to one output section that needs to be initialized.
The initialization data for each output section could be encoded using different encoding.
The load address in the C auto initialization record points to initialization data with the following format:

8-bit index Encoded data

The first 8-bits of the initialization data is the handler index. It indexes into a handler table to get the
address of a handler function that knows how to decode the following data.
The handler table is a list of 32-bit function pointers.
_TI_Handler_Table_Base:
32-bit handler 1 address

32-bit handler n address

_TI_Handler_Table_Limit:

The encoded data that follows the 8-bit index can be in one of the following format types. For clarity the 8-
bit index is also depicted for each format.

7.8.2.4.1 Length Followed by Data Format

8-bit index 24-bit padding 32-bit length (N) N byte initialization data (not compressed)

The compiler uses 24-bit padding to align the length field to a 32-bit boundary. The 32-bit length field
encodes the length of the initialization data in bytes (N). N byte initialization data is not compressed and is
copied to the run address as is.
The run-time support library has a function __TI_zero_init() to process this type of initialization data. The
first argument to this function is the address pointing to the byte after the 8-bit index. The second
argument is the run address from the C auto initialization record.

7.8.2.4.2 Zero Initialization Format

8-bit index 24-bit padding 32-bit length (N)

The compiler uses 24-bit padding to align the length field to a 32-bit boundary. The 32-bit length field
encodes the number of bytes to be zero initialized.
The run-time support library has a function __TI_zero_init() to process the zero initialization. The first
argument to this function is the address pointing to the byte after the 8-bit index. The second argument is
the run address from the C auto initialization record.

238 Run-Time Environment SPRU187U – July 2012

7.8.2.4.3 Run Length Encoded (RLE) Format

8-bit index Initialization data compressed using run length encoding

The data following the 8-bit index is compressed using Run Length Encoded (RLE) format. uses a simple
run length encoding that can be decompressed using the following algorithm:
1. Read the first byte, Delimiter (D).
2. Read the next byte (B).
3. If B != D, copy B to the output buffer and go to step 2.
4. Read the next byte (L).
(a) If L == 0, then length is either a 16-bit, a 24-bit value, or we’ve reached the end of the data, read
next byte (L).
(i) If L == 0, length is a 24-bit value or the end of the data is reached, read next byte (L).
(i) If L == 0, the end of the data is reached, go to step 7.
(ii) Else L <<= 16, read next two bytes into lower 16 bits of L to complete 24-bit value for L.
(ii) Else L <<= 8, read next byte into lower 8 bits of L to complete 16-bit value for L.
(b) Else if L > 0 and L < 4, copy D to the output buffer L times. Go to step 2.
(c) Else, length is 8-bit value (L).
5. Read the next byte (C); C is the repeat character.
6. Write C to the output buffer L times; go to step 2.
7. End of processing.
The run-time support library has a routine __TI_decompress_rle24() to decompress data compressed
using RLE. The first argument to this function is the address pointing to the byte after the 8-bit index. The
second argument is the run address from the C auto initialization record.

RLE Decompression Routine

NOTE: The previous decompression routine, __TI_decompress_rle(), is included in the run-time-
support library for decompressing RLE encodings that are generated by older versions of the
linker.

7.8.2.4.4 Lempel-Ziv-Storer-Szymanski Compression (LZSS) Format

8-bit index Initialization data compressed using LZSS

The data following the 8-bit index is compressed using LZSS compression. The run-time support library
has the routine __TI_decompress_lzss() to decompress the data compressed using LZSS. The first
argument to this function is the address pointing to the byte after the 8-bit index. The second argument is
the run address from the C auto initialization record.

7.8.2.4.5 Sample C Code to Process the C Autoinitialization Table

The run-time support boot routine has code to process the C autoinitialization table. The following C code
illustrates how the autoinitialization table can be processed on the target.

Example 7-12. Processing the C Autoinitialization Table

typedef void (handler_fptr)(const unsigned char in,

unsigned char *out);

#define HANDLER_TABLE __TI_Handler_Table_Base

#pragma WEAK(HANDLER_TABLE)
extern unsigned int HANDLER_TABLE;

SPRU187U – July 2012 Run-Time Environment 239

Example 7-12. Processing the C Autoinitialization Table (continued)

extern unsigned char *__TI_CINIT_Base;
extern unsigned char *__TI_CINIT_Limit;

void auto_initialize()
{
unsigned char **table_ptr;
unsigned char **table_limit;

/*--------------------------------------------------------------*/
/* Check if Handler table has entries. */
/*--------------------------------------------------------------*/
if (&__TI_Handler_Table_Base >= &__TI_Handler_Table_Limit)
return;

/*---------------------------------------------------------------*/
/* Get the Start and End of the CINIT Table. */
/*---------------------------------------------------------------*/
table_ptr = (unsigned char **)&__TI_CINIT_Base;
table_limit = (unsigned char **)&__TI_CINIT_Limit;
while (table_ptr < table_limit)
{

/*-------------------------------------------------------------*/
/* 1. Get the Load and Run address. */
/* 2. Read the 8-bit index from the load address. */
/* 3. Get the handler function pointer using the index from */
/* handler table. */
/*-------------------------------------------------------------*/
unsigned char *load_addr = *table_ptr++;
unsigned char *run_addr = *table_ptr++;
unsigned char handler_idx = *load_addr++;
handler_fptr handler =
(handler_fptr)(&HANDLER_TABLE)[handler_idx];

/*-------------------------------------------------------------*/
/* 4. Call the handler and pass the pointer to the load data */
/* after index and the run address. */
/*-------------------------------------------------------------*/
(*handler)((const unsigned char *)load_addr, run_addr);
}
}

7.8.2.5 Initialization of Variables at Load Time in EABI Mode

240 Run-Time Environment SPRU187U – July 2012

Figure 7-14. Initialization at Load Time in EABI Mode

Object file Memory

.data
Loader
section

.data section
(initialized)
(RAM)

7.8.2.6 Global Constructors

All global C++ variables that have constructors must have their constructor called before main. The
compiler builds a table of global constructor addresses that must be called, in order, before main in a
section called .init_array. The linker combines the .init_array section form each input file to form a single
table in the .init_array section. The boot routine uses this table to execute the constructors. The linker
defines two symbols to identify the combined .init_array table as shown below. This table is not null
terminated by the linker.

Figure 7-15. Constructor Table for EABI Mode

__TI_INITARRAY_Base:

Address of constructor 1
Address of constructor 2

Address of constructor n

__TI_INITARRAY_Limit:

7.8.3 Initialization Tables

The tables in the .cinit section consist of variable-size initialization records. Each variable that must be
autoinitialized has a record in the .cinit section. Figure 7-16 shows the format of the .cinit section and the
initialization records.

SPRU187U – July 2012 Run-Time Environment 241

Figure 7-16. Format of Initialization Records in the .cinit Section

.cinit section
Initialization record 1
Initialization record 2
Initialization record
Initialization record 3
Size in Pointer to Initialization
? bytes .bss area data

Initialization record n

The fields of an initialization record contain the following information:

• The first field of an initialization record contains the size (in bytes) of the initialization data.
• The second field contains the starting address of the area within the .bss section where the
initialization data must be copied.
• The third field contains the data that is copied into the .bss section to initialize the variable.
Each variable that must be autoinitialized has an initialization record in the .cinit section.
Example 7-13 shows initialized global variables defined in C. Example 7-14 shows the corresponding
initialization table. The section .cinit:c is a subsection in the .cinit section that contains all scalar data. The
subsection is handled as one record during initialization, which minimizes the overall size of the .cinit
section.

Example 7-13. Initialized Variables Defined in C

int x;
short i = 23;
int *p =
int a[5] = {1,2,3,4,5};

242 Run-Time Environment SPRU187U – July 2012

Example 7-14. Initialized Information for Variables Defined in Example 7-13

.global _x
.bss _x,4,4

.sect ".cinit:c"
.align 8
.field (CIR - $) - 8, 32
.field _I+0,32
.field 23,16 ; _I @ 0

.sect ".text"
.global _I
_I: .usect ".bss:c",2,2

.sect ".cinit:c"
.align 4
.field _x,32 ; _p @ 0

.sect ".text"
.global _p
_p: .usect ".bss:c",4,4

.sect ".cinit"
.align 8
.field IR_1,32
.field _a+0,32
.field 1,32 ; _a[0] @ 0
.field 2,32 ; _a[1] @ 32
.field 3,32 ; _a[2] @ 64
.field 4,32 ; _a[3] @ 96
.field 5,32 ; _a[4] @ 128
IR_1: .set 20

.sect ".text"
.global _a
.bss _a,20,4
;**********************************************************************
;* MARK THE END OF THE SCALAR INIT RECORD IN CINIT:C *
;**********************************************************************

CIR: .sect ".cinit:c"

The .cinit section must contain only initialization tables in this format. When interfacing assembly language
modules, do not use the .cinit section for any other purpose.
The table in the .pinit section simply consists of a list of addresses of constructors to be called (see
Figure 7-17). The constructors appear in the table after the .cinit initialization.

SPRU187U – July 2012 Run-Time Environment 243

Figure 7-17. Format of Initialization Records in the .pinit Section

.pinit section

Address of constructor 1
Address of constructor 2
Address of constructor 3

•
•
Address of constructor n

When you use the --rom_model or --ram_model option, the linker combines the .cinit sections from all the
C/C++ modules and appends a null word to the end of the composite .cinit section. This terminating record
appears as a record with a size field of 0 and marks the end of the initialization tables.
Likewise, the --rom_model or --ram_model link option causes the linker to combine all of the .pinit sections
from all C/C++ modules and append a null word to the end of the composite .pinit section. The boot
routine knows the end of the global constructor table when it encounters a null constructor address.
The const-qualified variables are initialized differently; see Section 6.5.1.

244 Run-Time Environment SPRU187U – July 2012

7.9 Support for Multi-Threaded Applications

The compiler supports various features for multi-threaded applications. These features assume an
underlying runtime operating system that provides thread management services.

7.9.1 Compiling with OpenMP

The compiler implements support for the OpenMP 3.0 API. To enable support for OpenMP use the --
openmp or --omp compiler option.
OpenMP is the industry standard for shared memory parallel programming. It provides portable high-level
programming constructs that enable users to easily expose a program's task and loop level parallelism in
an incremental fashion. With OpenMP, users specify the parallelization strategy for a program at a high
level by annotating the program code with compiler directives that specify how a region of code is
executed by a team of threads. The compiler works out the detailed mapping of the computation to the
machine. The OpenMP programming API enables the programmer to perform the following:
• Create and manage threads
• Assign and distribute work (tasks) to threads
• Specify which data is shared among threads and which data is private
• Coordinate thread access to shared data
OpenMP is a thread-based programming language. The master thread executes the sequential parts of a
program. When the master thread encounters a parallel region, it forks a team of worker threads that
along with the master thread execute in parallel.
The OpenMP API is made up of directives(#pragmas), function calls, and environment variables. The
compiler translates the OpenMP API into multi-threaded code with calls to a custom runtime library that
implements support for thread management, shared memory and synchronization. For further details on
the OpenMP API (including the API specification), please refer to http://www.openmp.org.
The OpenMP runtime for SYS/BIOS (OMP) library implements the OpenMP solution stack. Currently,
OpenMP is supported on TI DSPs only for SYS/BIOS operating system. All OpenMP programs must be
linked with the OMP run-time library found in the BIOS-MCSDK 2.1.

7.9.2 Multi-Threading Runtime Support

7.9.2.1 Runtime Thread Safety

Thread safety involves creation, initialization, maintenance, and destruction of thread-private data. It also
requires that accesses to data that is shared between threads must be protected. That is, only one thread
should be allowed to access a piece of shared data at a given time. An additional issue that needs to be
addressed on C6000 multi-core devices that have a private data cache (L1D cache on C6600 devices, for
example) is that copies of shared data that exist among private data caches and shared memory must be
kept coherent. This means that if a thread reads a piece of shared data into the private data cache on a
processor, it must invalidate any local copies of that data that exist in the local data cache before
accessing or modifying the data. This ensures that the currently executing thread will only access the
latest available copy of the shared data.

7.9.2.1.1 Thread Creation, Initialization, and Termination

Thread libraries are responsible for allocating a thread-local area of memory when a thread is created,
then initializing any thread-private data objects that reside in that thread-local area, and when the thread is
terminated, the thread library must free the thread-local area that was allocated for thread-private data.

SPRU187U – July 2012 Run-Time Environment 245

7.9.2.1.2 Thread Local Storage (TLS)

The compiler now supports the __thread qualifier to identify a data object that is to be given thread-private
storage. To access a variable that has been identified as thread-local, the compiler will rely on support of
the runtime operating system's thread library to find the run-time location of a thread-local variable.
Specifically, the thread library must provide an implementation of the __c6xabi_get_tp() function. The
runtime operating system's thread library will provide the address of a TLS block that has been allocated
on behalf of the currently executed thread and then the compiler can access data in the TLS block with
knowledge of where a given thread-local variable exists within the TLS block. Further information about
TLS data objects can be found in the "Thread Local Storage Specification for C6000 EABI" document
(C6000_TLS.docx in the 'doc' sub-directory where your C6000 CGT is installed).

7.9.2.1.3 Accessing Shared Data

Accesses to shared data objects must be protected within a critical region which prevents a second thread
from entering the critical region of code while the first thread is accessing a shared data object. We also
need to be concerned about data coherency between copies of shared data that may exist in both shared
memory and in private data caches as mentioned above in Section 7.9.2.1.

246 Run-Time Environment SPRU187U – July 2012

Using Run-Time-Support Functions and Building Libraries

Some of the features of C/C++ (such as I/O, dynamic memory allocation, string operations, and
trigonometric functions) are provided as an ANSI/ISO C/C++ standard library, rather than as part of the
compiler itself. The TI implementation of this library is the run-time-support library (RTS). The C/C++
compiler implements the complete ISO standard library except for those facilities that handle signal and
locale issues (properties that depend on local language, nationality, or culture). Using the ANSI/ISO
standard library ensures a consistent set of functions that provide for greater portability.
In addition to the ANSI/ISO-specified functions, the run-time-support library includes routines that give you
processor-specific commands and direct C language I/O requests. These are detailed inSection 8.1 and
Section 8.2.
A library-build utility is provided with the code generation tools that lets you create customized run-time-
support libraries. This process is described in Section 8.5 .

Topic ........................................................................................................................... Page

8.1 C and C++ Run-Time Support Libraries .............................................................. 248

8.2 The C I/O Functions ......................................................................................... 250
8.3 Handling Reentrancy (_register_lock() and _register_unlock() Functions) .............. 262
8.4 C6700 FastMath Library .................................................................................... 263
8.5 Library-Build Process ...................................................................................... 264

SPRU187U – July 2012 Using Run-Time-Support Functions and Building Libraries 247
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
C and C++ Run-Time Support Libraries www.ti.com

8.1 C and C++ Run-Time Support Libraries

TMS320C6000 compiler releases include pre-built run-time libraries that provide all the standard
capabilities. Separate libraries are provided for each target CPU version, big and little endian support,
each ABI, and C++ exception support. See Section 8.5 for information on the library-naming conventions.
The run-time-support library contains the following:
• ANSI/ISO C/C++ standard library
• C I/O library
• Low-level support functions that provide I/O to the host operating system
• Fundamental arithmetic routines
• System startup routine, _c_int00
• Functions and macros that allow C/C++ to access specific instructions
The run-time-support libraries do not contain functions involving signals and locale issues.
The C++ library supports wide chars, in that template functions and classes that are defined for char are
also available for wide char. For example, wide char stream classes wios, wiostream, wstreambuf and so
on (corresponding to char classes ios, iostream, streambuf) are implemented. However, there is no low-
level file I/O for wide chars. Also, the C library interface to wide char support (through the C++ headers
<cwchar> and <cwctype>) is limited as described in Section 6.1.
The C++ library included with the compiler is licensed from Dinkumware, Ltd. The Dinkumware C++ library
is a fully conforming, industry-leading implementation of the standard C++ library.
TI does not provide documentation that covers the functionality of the C++ library. TI suggests referring to
one of the following sources:
• The Standard C++ Library: A Tutorial and Reference,Nicolai M. Josuttis, Addison-Wesley, ISBN 0-201-
37926-0
• The C++ Programming Language (Third or Special Editions), Bjarne Stroustrup, Addison-Wesley,
ISBN 0-201-88954-4 or 0-201-70073-5
• Dinkumware's online reference at http://dinkumware.com/manuals

8.1.1 Linking Code With the Object Library

When you link your program, you must specify the object library as one of the linker input files so that
references to the I/O and run-time-support functions can be resolved. You can either specify the library or
allow the compiler to select one for you. See Section 5.3.1 for further information.
When a library is linked, the linker includes only those library members required to resolve undefined
references. For more information about linking, see the TMS320C6000 Assembly Language Tools User's
Guide.
C, C++, and mixed C and C++ programs can use the same run-time-support library. Run-time-support
functions and variables that can be called and referenced from both C and C++ will have the same
linkage.

248 Using Run-Time-Support Functions and Building Libraries SPRU187U – July 2012
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com C and C++ Run-Time Support Libraries

8.1.2 Header Files

You must use the header files provided with the compiler run-time support when using functions from
C/C++ standard library. Set the C6X_C_DIR environment variable to the include directory where the tools
are installed.

8.1.3 Modifying a Library Function

You can inspect or modify library functions by unzipping the source file (rtssrc.zip), changing the specific
function file, and rebuilding the library. When extracted (with any standard unzip tool on windows, linux, or
unix), this zip file recreates the run-time source tree for the run-time library.
The source for the libraries is included in the rtssrc.zip file. See Section 8.5 for details on rebuilding.
You can also build a new library this way, rather than rebuilding into rts6200.lib. See Section 8.5.

8.1.4 Minimal Support for Internationalization

The library now includes the header files <locale.h>, <wchar.h>, and <wctype.h>, which provide APIs to
support non-ASCII character sets and conventions. Our implementation of these APIs is limited in the
following ways:
• The library has minimal support for wide and multi-byte characters. The type wchar_t is implemented
as int. The wide character set is equivalent to the set of values of type char. The library includes the
header files <wchar.h> and <wctype.h> but does not include all the functions specified in the standard.
So-called multi-byte characters are limited to single characters. There are no shift states. The mapping
between multi-byte characters and wide characters is simple equivalence; that is, each wide character
maps to and from exactly a single multi-byte character having the same value.
• The C library includes the header file <locale.h> but with a minimal implementation. The only
supported locale is the C locale. That is, library behavior that is specified to vary by locale is hard-
coded to the behavior of the C locale, and attempting to install a different locale via a call to setlocale()
will return NULL.

8.1.5 Allowable Number of Open Files

In the <stdio.h> header file, the value for the macro FOPEN_MAX has been changed from 12 to the value
of the macro _NFILE, which is set to 10. The impact is that you can only have 10 files simultaneously
open at one time (including the pre-defined streams - stdin, stdout, stderr).
The C standard requires that the minimum value for the FOPEN_MAX macro is 8. The macro determines
the maximum number of files that can be opened at one time. The macro is defined in the stdio.h header
file and can be modified by changing the value of the _NFILE macro.

8.1.6 Library Naming Conventions

By default, the linker uses automatic library selection to select the correct run-time-support library (see
Section 5.3.1.1) for your application. If you select the library manually, you must select the matching
library according to the following naming scheme:

rtstrg[endian][abi][eh].lib

trg The device family of the C6000 architecture that the library was built for. This can be one
of the following: 6200, 6400, 64plus, 6600, 6700, 6740, 67plus.
endian Indicates endianness:
(blank) Little-endian library
e Big-endian library
abi Indicates the application binary interface (ABI) used:
(blank) COFF ABI
_elf EABI
eh Indicates whether the library has exception handling support
(blank) exception handling not supported
_eh exception handling support

For information on the C6700 FastMath source library, fastmathc67x.src, see Section 8.4.

8.2 The C I/O Functions

The C I/O functions make it possible to access the host's operating system to perform I/O. The capability
to perform I/O on the host gives you more options when debugging and testing code.
The I/O functions are logically divided into layers: high level, low level, and device-driver level.
With properly written device drivers, the C-standard high-level I/O functions can be used to perform I/O on
custom user-defined devices. This provides an easy way to use the sophisticated buffering of the high-
level I/O functions on an arbitrary device.

Debugger Required for Default HOST

NOTE: For the default HOST device to work, there must be a debugger to handle the C I/O
requests; the default HOST device cannot work by itself in an embedded system. To work in
an embedded system, you will need to provide an appropriate driver for your system.

NOTE: C I/O Mysteriously Fails

If there is not enough space on the heap for a C I/O buffer, operations on the file will silently
fail. If a call to printf() mysteriously fails, this may be the reason. The heap needs to be at
least large enough to allocate a block of size BUFSIZ (defined in stdio.h) for every file on
which I/O is performed, including stdout, stdin, and stderr, plus allocations performed by the
user's code, plus allocation bookkeeping overhead. Alternately, declare a char array of size
BUFSIZ and pass it to setvbuf to avoid dynamic allocation. To set the heap size, use the --
heap_size option when linking (refer to the Linker Description chapter in theTMS320C6000
Assembly Language Tools User's Guide).

NOTE: Open Mysteriously Fails

The run-time support limits the total number of open files to a small number relative to
general-purpose processors. If you attempt to open more files than the maximum, you may
find that the open will mysteriously fail. You can increase the number of open files by
extracting the source code from rts.src and editing the constants controlling the size of some
of the C I/O data structures. The macro _NFILE controls how many FILE (fopen) objects can
be open at one time (stdin, stdout, and stderr count against this total). (See also
FOPEN_MAX.) The macro _NSTREAM controls how many low-level file descriptors can be
open at one time (the low-level files underlying stdin, stdout, and stderr count against this
todal). The macro _NDEVICE controls how many device drivers are installed at one time (the
HOST device counts against this total).

8.2.1 High-Level I/O Functions

The high-level functions are the standard C library of stream I/O routines (printf, scanf, fopen, getchar, and
so on). These functions call one or more low-level I/O functions to carry out the high-level I/O request. The
high-level I/O routines operate on FILE pointers, also called streams.
Portable applications should use only the high-level I/O functions.
To use the high-level I/O functions, include the header file stdio.h, or cstdio for C++ code, for each module
that references a C I/O function.
For example, given the following C program in a file named main.c:
#include <stdio.h>;

void main()
{
FILE *fid;

fid = fopen("myfile","w");
fprintf(fid,"Hello, world\n");
fclose(fid);

printf("Hello again, world\n");

}

Issuing the following compiler command compiles, links, and creates the file main.out from the run-time-
support library:
cl6x main.c -z --heap_size=1000 --output_file=main.out

Executing main.out results in

Hello, world

being output to a file and

Hello again, world

being output to your host's stdout window.

8.2.2 Overview of Low-Level I/O Implementation

The low-level functions are comprised of seven basic I/O functions: open, read, write, close, lseek,
rename, and unlink. These low-level routines provide the interface between the high-level functions and
the device-level drivers that actually perform the I/O command on the specified device.
The low-level functions are designed to be appropriate for all I/O methods, even those which are not
actually disk files. Abstractly, all I/O channels can be treated as files, although some operations (such as
lseek) may not be appropriate. See Section 8.2.3 for more details.
The low-level functions are inspired by, but not identical to, the POSIX functions of the same names.
The low-level functions operate on file descriptors. A file descriptor is an integer returned by open,
representing an opened file. Multiple file descriptors may be associated with a file; each has its own
independent file position indicator.

open Open File for I/O

Syntax #include <file.h>

int open (const char * path , unsigned flags , int file_descriptor );

Description The open function opens the file specified by path and prepares it for I/O.
• The path is the filename of the file to be opened, including an optional directory path
and an optional device specifier (see Section 8.2.5).
• The flags are attributes that specify how the file is manipulated. The flags are
specified using the following symbols:
O_RDONLY (0x0000) /* open for reading */
O_WRONLY (0x0001) /* open for writing */
O_RDWR (0x0002) /* open for read & write */
O_APPEND (0x0008) /* append on each write */
O_CREAT (0x0200) /* open with file create */
O_TRUNC (0x0400) /* open with truncation */
O_BINARY (0x8000) /* open in binary mode */
Low-level I/O routines allow or disallow some operations depending on the flags used
when the file was opened. Some flags may not be meaningful for some devices,
depending on how the device implements files.
• The file_descriptor is assigned by open to an opened file.
The next available file descriptor is assigned to each new file opened.

Return Value The function returns one of the following values:

non-negative file descriptor if successful
-1 on failure

close Close File for I/O

Syntax #include <file.h>

int close (int file_descriptor );

Description The close function closes the file associated with file_descriptor.
The file_descriptor is the number assigned by open to an opened file.

Return Value The return value is one of the following:

0 if successful
-1 on failure

read Read Characters from a File

Syntax #include <file.h>

int read (int file_descriptor , char * buffer , unsigned count );

Description The read function reads count characters into the buffer from the file associated with
file_descriptor.
• The file_descriptor is the number assigned by open to an opened file.
• The buffer is where the read characters are placed.
• The count is the number of characters to read from the file.

Return Value The function returns one of the following values:

0 if EOF was encountered before any characters were read
# number of characters read (may be less than count)
-1 on failure

write Write Characters to a File

Syntax #include <file.h>

int write (int file_descriptor , const char * buffer , unsigned count );

Description The write function writes the number of characters specified by count from the buffer to
the file associated with file_descriptor.
• The file_descriptor is the number assigned by open to an opened file.
• The buffer is where the characters to be written are located.
• The count is the number of characters to write to the file.

Return Value The function returns one of the following values:

# number of characters written if successful (may be less than count)
-1 on failure

SPRU187U – July 2012 Using Run-Time-Support Functions and Building Libraries 253
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
lseek — Set File Position Indicator www.ti.com

lseek Set File Position Indicator

Syntax for C #include <file.h>

off_t lseek (int file_descriptor , off_t offset , int origin );

Description The lseek function sets the file position indicator for the given file to a location relative to
the specified origin. The file position indicator measures the position in characters from
the beginning of the file.
• The file_descriptor is the number assigned by open to an opened file.
• The offset indicates the relative offset from the origin in characters.
• The origin is used to indicate which of the base locations the offset is measured from.
The origin must be one of the following macros:
SEEK_SET (0x0000) Beginning of file
SEEK_CUR (0x0001) Current value of the file position indicator
SEEK_END (0x0002) End of file
Return Value The return value is one of the following:
# new value of the file position indicator if successful
(off_t)-1 on failure

unlink Delete File

Syntax #include <file.h>

int unlink (const char * path );

Description The unlink function deletes the file specified by path. Depending on the device, a deleted
file may still remain until all file descriptors which have been opened for that file have
been closed. See Section 8.2.3.
The path is the filename of the file, including path information and optional device prefix.
(See Section 8.2.5.)

Return Value The function returns one of the following values:

0 if successful
-1 on failure

rename Rename File

Syntax for C #include {<stdio.h> | <file.h>}

int rename (const char * old_name , const char * new_name );

Syntax for C++ #include {<cstdio> | <file.h>}

int std::rename (const char * old_name , const char * new_name );

Description The rename function changes the name of a file.

• The old_name is the current name of the file.
• The new_name is the new name for the file.

NOTE: The optional device specified in the new name must match the device of
the old name. If they do not match, a file copy would be required to
perform the rename, and rename is not capable of this action.

Return Value The function returns one of the following values:

0 if successful
-1 on failure

NOTE: Although rename is a low-level function, it is defined by the C standard

and can be used by portable applications.

8.2.3 Device-Driver Level I/O Functions

At the next level are the device-level drivers. They map directly to the low-level I/O functions. The default
device driver is the HOST device driver, which uses the debugger to perform file operations. The HOST
device driver is automatically used for the default C streams stdin, stdout, and stderr.
The HOST device driver shares a special protocol with the debugger running on a host system so that the
host can perform the C I/O requested by the program. Instructions for C I/O operations that the program
wants to perform are encoded in a special buffer named _CIOBUF_ in the .cio section. The debugger
halts the program at a special breakpoint (C$$IO$$), reads and decodes the target memory, and performs
the requested operation. The result is encoded into _CIOBUF_, the program is resumed, and the target
decodes the result.
The HOST device is implemented with seven functions, HOSTopen, HOSTclose, HOSTread, HOSTwrite,
HOSTlseek, HOSTunlink, and HOSTrename, which perform the encoding. Each function is called from the
low-level I/O function with a similar name.
A device driver is composed of seven required functions. Not all function need to be meaningful for all
devices, but all seven must be defined. Here we show the names of all seven functions as starting with
DEV, but you may chose any name except for HOST.

DEV_open Open File for I/O

Syntax int DEV_open (const char * path , unsigned flags , int llv_fd );

Description This function finds a file matching path and opens it for I/O as requested by flags.
• The path is the filename of the file to be opened. If the name of a file passed to open
has a device prefix, the device prefix will be stripped by open, so DEV_open will not
see it. (See Section 8.2.5 for details on the device prefix.)
• The flags are attributes that specify how the file is manipulated. The flags are
specified using the following symbols:
O_RDONLY (0x0000) /* open for reading */
O_WRONLY (0x0001) /* open for writing */
O_RDWR (0x0002) /* open for read & write */
O_APPEND (0x0008) /* append on each write */
O_CREAT (0x0200) /* open with file create */
O_TRUNC (0x0400) /* open with truncation */
O_BINARY (0x8000) /* open in binary mode */
See POSIX for further explanation of the flags.
• The llv_fd is treated as a suggested low-level file descriptor. This is a historical
artifact; newly-defined device drivers should ignore this argument. This differs from
the low-level I/O open function.
This function must arrange for information to be saved for each file descriptor, typically
including a file position indicator and any significant flags. For the HOST version, all the
bookkeeping is handled by the debugger running on the host machine. If the device uses
an internal buffer, the buffer can be created when a file is opened, or the buffer can be
created during a read or write.

Return Value This function must return -1 to indicate an error if for some reason the file could not be
opened; such as the file does not exist, could not be created, or there are too many files
open. The value of errno may optionally be set to indicate the exact error (the HOST
device does not set errno). Some devices might have special failure conditions; for
instance, if a device is read-only, a file cannot be opened O_WRONLY.
On success, this function must return a non-negative file descriptor unique among all
open files handled by the specific device. It need not be unique across devices. Only the
low-level I/O functions will see this device file descriptor; the low-level function open will
assign its own unique file descriptor.

256 Using Run-Time-Support Functions and Building Libraries SPRU187U – July 2012
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com DEV_close — Close File for I/O

DEV_close Close File for I/O

Syntax int DEV_close (int dev_fd );

Description This function closes a valid open file descriptor.

On some devices, DEV_close may need to be responsible for checking if this is the last
file descriptor pointing to a file that was unlinked. If so, it is responsible for ensuring that
the file is actually removed from the device and the resources reclaimed, if appropriate.

Return Value This function should return -1 to indicate an error if the file descriptor is invalid in some
way, such as being out of range or already closed, but this is not required. The user
should not call close() with an invalid file descriptor.

DEV_read Read Characters from a File

Syntax int DEV_read (int dev_fd , char * bu , unsigned count );

Description The read function reads count bytes from the input file associated with dev_fd.
• The dev_fd is the number assigned by open to an opened file.
• The buf is where the read characters are placed.
• The count is the number of characters to read from the file.

Return Value This function must return -1 to indicate an error if for some reason no bytes could be
read from the file. This could be because of an attempt to read from a O_WRONLY file,
or for device-specific reasons.
If count is 0, no bytes are read and this function returns 0.
This function returns the number of bytes read, from 0 to count. 0 indicates that EOF
was reached before any bytes were read. It is not an error to read less than count bytes;
this is common if the are not enough bytes left in the file or the request was larger than
an internal device buffer size.

DEV_write Write Characters to a File

Syntax int DEV_write (int dev_fd , const char * buf , unsigned count );

Description This function writes count bytes to the output file.

• The dev_fd is the number assigned by open to an opened file.
• The buffer is where the write characters are placed.
• The count is the number of characters to write to the file.

Return Value This function must return -1 to indicate an error if for some reason no bytes could be
written to the file. This could be because of an attempt to read from a O_RDONLY file,
or for device-specific reasons.

SPRU187U – July 2012 Using Run-Time-Support Functions and Building Libraries 257
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
DEV_lseek — Set File Position Indicator www.ti.com

DEV_lseek Set File Position Indicator

Syntax off_t lseek (int dev_fd , off_t offset , int origin );

Description This function sets the file's position indicator for this file descriptor as lseek.
If lseek is supported, it should not allow a seek to before the beginning of the file, but it
should support seeking past the end of the file. Such seeks do not change the size of
the file, but if it is followed by a write, the file size will increase.

Return Value If successful, this function returns the new value of the file position indicator.
This function must return -1 to indicate an error if for some reason no bytes could be
written to the file. For many devices, the lseek operation is nonsensical (e.g. a computer
monitor).

DEV_unlink Delete File

Syntax int DEV_unlink (const char * path );

Description Remove the association of the pathname with the file. This means that the file may no
longer by opened using this name, but the file may not actually be immediately removed.
Depending on the device, the file may be immediately removed, but for a device which
allows open file descriptors to point to unlinked files, the file will not actually be deleted
until the last file descriptor is closed. See Section 8.2.3.

Return Value This function must return -1 to indicate an error if for some reason the file could not be
unlinked (delayed removal does not count as a failure to unlink.)
If successful, this function returns 0.

DEV_rename Rename File

Syntax int DEV_rename (const char * old_name , const char * new_name );

Description This function changes the name associated with the file.
• The old_name is the current name of the file.
• The new_name is the new name for the file.
Return Value This function must return -1 to indicate an error if for some reason the file could not be
renamed, such as the file doesn't exist, or the new name already exists.

NOTE: It is inadvisable to allow renaming a file so that it is on a different device.

In general this would require a whole file copy, which may be more
expensive than you expect.

If successful, this function returns 0.

8.2.4 Adding a User-Defined Device Driver for C I/O

The function add_device allows you to add and use a device. When a device is registered with
add_device, the high-level I/O routines can be used for I/O on that device.
You can use a different protocol to communicate with any desired device and install that protocol using
add_device; however, the HOST functions should not be modified. The default streams stdin, stdout, and
stderr can be remapped to a file on a user-defined device instead of HOST by using freopen() as in
Example 8-1. If the default streams are reopened in this way, the buffering mode will change to _IOFBF
(fully buffered). To restore the default buffering behavior, call setvbuf on each reopened file with the
appropriate value (_IOLBF for stdin and stdout, _IONBF for stderr).
The default streams stdin, stdout, and stderr can be mapped to a file on a user-defined device instead of
HOST by using freopen() as shown in Example 8-1. Each function must set up and maintain its own data
structures as needed. Some function definitions perform no action and should just return.

Example 8-1. Mapping Default Streams to Device

#include <stdio.h>
#include <file.h>
#include "mydevice.h"

void main()
{
add_device("mydevice", _MSA,
MYDEVICE_open, MYDEVICE_close,
MYDEVICE_read, MYDEVICE_write,
MYDEVICE_lseek, MYDEVICE_unlink, MYDEVICE_rename);

/*-----------------------------------------------------------------------*/
/* Re-open stderr as a MYDEVICE file */
/*-----------------------------------------------------------------------*/
if (!freopen("mydevice:stderrfile", "w", stderr))
{
puts("Failed to freopen stderr");
exit(EXIT_FAILURE);
}

/*-----------------------------------------------------------------------*/
/* stderr should not be fully buffered; we want errors to be seen as */
/* soon as possible. Normally stderr is line-buffered, but this example */
/* doesn't buffer stderr at all. This means that there will be one call */
/* to write() for each character in the message. */
/*-----------------------------------------------------------------------*/
if (setvbuf(stderr, NULL, _IONBF, 0))
{
puts("Failed to setvbuf stderr");
exit(EXIT_FAILURE);
}

/*-----------------------------------------------------------------------*/
/* Try it out! */
/*-----------------------------------------------------------------------*/
printf("This goes to stdout\n");
fprintf(stderr, "This goes to stderr\n"); }

NOTE: Use Unique Function Names

The function names open, read, write, close, lseek, rename, and unlink are used by the low-
level routines. Use other names for the device-level functions that you write.

SPRU187U – July 2012 Using Run-Time-Support Functions and Building Libraries 259
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
add_device — Add Device to Device Table www.ti.com

Use the low-level function add_device() to add your device to the device_table. The device table is a
statically defined array that supports n devices, where n is defined by the macro _NDEVICE found in
stdio.h/cstdio.
The first entry in the device table is predefined to be the host device on which the debugger is running.
The low-level routine add_device() finds the first empty position in the device table and initializes the
device fields with the passed-in arguments. For a complete description, see the add_device function.

8.2.5 The device Prefix

A file can be opened to a user-defined device driver by using a device prefix in the pathname. The device
prefix is the device name used in the call to add_device followed by a colon. For example:
FILE *fptr = fopen("mydevice:file1", "r");
int fd = open("mydevice:file2, O_RDONLY, 0);

If no device prefix is used, the HOST device will be used to open the file.

add_device Add Device to Device Table

Syntax for C #include <file.h>

int add_device(char * name,
unsigned flags ,
int (* dopen )(const char *path, unsigned flags, int llv_fd),
int (* dclose )( int dev_fd),
int (* dread )(intdev_fd, char *buf, unsigned count),
int (* dwrite )(int dev_fd, const char *buf, unsigned count),
off_t (* dlseek )(int dev_fd, off_t ioffset, int origin),
int (* dunlink )(const char * path),
int (* drename )(const char *old_name, const char *new_name));

Defined in lowlev.c in rtssrc.zip

Description The add_device function adds a device record to the device table allowing that device to
be used for I/O from C. The first entry in the device table is predefined to be the HOST
device on which the debugger is running. The function add_device() finds the first empty
position in the device table and initializes the fields of the structure that represent a
device.
To open a stream on a newly added device use fopen( ) with a string of the format
devicename : filename as the first argument.
• The name is a character string denoting the device name. The name is limited to 8
characters.
• The flags are device characteristics. The flags are as follows:
_SSA Denotes that the device supports only one open stream at a time
_MSA Denotes that the device supports multiple open streams
More flags can be added by defining them in file.h.
• The dopen, dclose, dread, dwrite, dlseek, dunlink, and drename specifiers are
function pointers to the functions in the device driver that are called by the low-level
functions to perform I/O on the specified device. You must declare these functions
with the interface specified in Section 8.2.2. The device driver for the HOST that the
TMS320C6000 debugger is run on are included in the C I/O library.

Return Value The function returns one of the following values:

0 if successful
-1 on failure

260 Using Run-Time-Support Functions and Building Libraries SPRU187U – July 2012
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
www.ti.com add_device — Add Device to Device Table

Example Example 8-2 does the following:

• Adds the device mydevice to the device table
• Opens a file named test on that device and associates it with the FILE pointer fid
• Writes the string Hello, world into the file
• Closes the file

Example 8-2 illustrates adding and using a device for C I/O:

Example 8-2. Program for C I/O Device

#include <file.h>
#include <stdio.h>
/****************************************************************************/
/* Declarations of the user-defined device drivers */
/****************************************************************************/
extern int MYDEVICE_open(const char *path, unsigned flags, int fno);
extern int MYDEVICE_close(int fno);
extern int MYDEVICE_read(int fno, char *buffer, unsigned count);
extern int MYDEVICE_write(int fno, const char *buffer, unsigned count);
extern off_t MYDEVICE_lseek(int fno, off_t offset, int origin);
extern int MYDEVICE_unlink(const char *path);
extern int MYDEVICE_rename(const char *old_name, char *new_name);
main()
{
FILE *fid;
add_device("mydevice", _MSA, MYDEVICE_open, MYDEVICE_close, MYDEVICE_read,
MYDEVICE_write, MYDEVICE_lseek, MYDEVICE_unlink, MYDEVICE_rename);
fid = fopen("mydevice:test","w");
fprintf(fid,"Hello, world\n");

fclose(fid);
}

SPRU187U – July 2012 Using Run-Time-Support Functions and Building Libraries 261
Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
Handling Reentrancy (_register_lock() and _register_unlock() Functions) www.ti.com

8.3 Handling Reentrancy (_register_lock() and _register_unlock() Functions)

The C standard assumes only one thread of execution, with the only exception being extremely narrow
support for signal handlers. The issue of reentrancy is avoided by not allowing you to do much of anything
in a signal handler. However, BIOS applications have multiple threads which need to modify the same
global program state, such as the CIO buffer, so reentrancy is a concern.
Part of the problem of reentrancy remains your responsibility, but the run-time-support environment does
provide rudimentary support for multi-threaded reentrancy by providing support for critical sections. This
implementation does not protect you from reentrancy issues such as calling run-time-support functions
from inside interrupts; this remains your responsibility.
The run-time-support environment provides hooks to install critical section primitives. By default, a single-
threaded model is assumed, and the critical section primitives are not employed. In a multi-threaded
system such as BIOS, the kernel arranges to install semaphore lock primitive functions in these hooks,
which are then called when the run-time-support enters code that needs to be protected by a critical
section.
Throughout the run-time-support environment where a global state is accessed, and thus needs to be
protected with a critical section, there are calls to the function _lock(). This calls the provided primitive, if
installed, and acquires the semaphore before proceeding. Once the critical section is finished, _unlock() is
called to release the semaphore.
Usually BIOS is responsible for creating and installing the primitives, so you do not need to take any
action. However, this mechanism can be used in multi-threaded applications which do not use the BIOS
LCK mechanism.
You should not define the functions _lock() and _unlock() functions directly; instead, the installation
functions are called to instruct the run-time-support environment to use these new primitives:
void _register_lock (void ( *lock)());

void _register_unlock(void (*unlock)());

The arguments to _register_lock() and _register_unlock() should be functions which take no arguments
and return no values, and which implement some sort of global semaphore locking:
extern volatile sig_atomic_t *sema = SHARED_SEMAPHORE_LOCATION;
static int sema_depth = 0;
static void my_lock(void)
{
while (ATOMIC_TEST_AND_SET(sema, MY_UNIQUE_ID) != MY_UNIQUE_ID);
sema_depth++;
}
static void my_unlock(void)
{
if (!--sema_depth) ATOMIC_CLEAR(sema);
}

The run-time-support nests calls to _lock(), so the primitives must keep track of the nesting level.

8.4 C6700 FastMath Library

The C6700 FastMath Library provides hand-coded assembly-optimized versions of certain math functions.
These implementations are two to three times faster than those found in the standard run-time-support
library. However, these functions gain speed improvements at the cost of accuracy in the result.
The C6700 FastMath library contains these files:
• fastmath67x.lib—object library for use with little-endian C/C++ code
• fastmath67xe.lib—object library for use with big-endian C/C++ code
• fastmath67x.h—header file to be included with C/C++ code
To use the C67x FastMath library, specify it before the standard run-time-support library when linking your
program. For example:
cl6x -mv6700 --run_linker myprogram.obj --library=lnk.cmd --library=fastmath67x.lib --
library=rts6700.lib

If you are using Code Composer Studio, include the C6700 FastMath library in your project, and ensure it
appears before the standard run-time-support library in the Link Order tab in the Build Options dialog box.
For details, refer to the TMS320C67x FastRTS Library Programmer's Reference (SPRU100).

8.5 Library-Build Process

When using the C/C++ compiler, you can compile your code under a large number of different
configurations and options that are not necessarily compatible with one another. Because it would be
infeasible to include all possible run-time-support library variants, compiler releases pre-build only a small
number of very commonly-used libraries such as rts6200.lib.
To provide maximum flexibility, the run-time-support source code is provided as part of each compiler
release. You can build the missing libraries as desired. The linker can also automatically build missing
libraries. This is accomplished with a new library build process, the core of which is the executable mklib,
which is available beginning with CCS 5.1

8.5.1 Required Non-Texas Instruments Software

To use the self-contained run-time-support build process to rebuild a library with custom options, the
following are required:
• sh (Bourne shell)
• unzip (InfoZIP unzip 5.51 or later, or equivalent)
You can download the software from http://www.info-zip.org.
• gmake (GNU make 3.81 or later)
More information is available from GNU at http://www.gnu.org/software/make. GNU make (gmake) is
also available in earlier versions of Code Composer Studio. GNU make is also included in some Unix
support packages for Windows, such as the MKS Toolkit, Cygwin, and Interix. The GNU make used on
Windows platfortms should expliclty report "This program build for Windows32" when the following is
executed from the Command Prompt window:
gmake -h

All three of these programs are provided as a non-optional feature of CCS 5.1. They are also available as
part of the optional XDC Tools feature if you are using an earlier version of CCS.
The mklib program looks for these executables in the following order:
1. in your PATH
2. in the directory getenv("CCS_UTILS_DIR")/cygwin
3. in the directory getenv("CCS_UTILS_DIR")/bin
4. in the directory getenv("XDCROOT")
5. in the directory getenv("XDCROOT")/bin
If you are invoking mklib from the command line, and these executables are not in your path, you must set
the environment variable CCS_UTILS_DIR such that getenv("CCS_UTILS_DIR")/bin contains the correct
programs.

8.5.2 Using the Library-Build Process

You should normally let the linker automatically rebuild libraries as needed. If necessary, you can run
mklib directly to populate libraries. See Section 8.5.2.2 for situations when you might want to do this.

8.5.2.1 Automatic Standard Library Rebuilding by the Linker

The linker looks for run-time-support libraries primarily through the C6X_C_DIR environment variable.
Typically, one of the pathnames in C6X_C_DIR is your install directory/lib, which contains all of the pre-
built libraries, as well as the index library libc.a. The linker looks in C6X_C_DIR to find a library that is the
best match for the build attributes of the application. The build attributes are set indirectly according to the
command-line options used to build the application. Build attributes include things like CPU revision. If the
library is explicitly named (e.g. rts6200.lib), run-time support looks for that library exactly; otherwise, it
uses the index library libc.a to pick an appropriate library.

The index library describes a set of libraries with different build attributes. The linker will compare the build
attributes for each potential library with the build attributes of the application and will pick the best fit. For
details on the index library, see the archiver chapter in the TMS320C6000 Assembly Language Tools
User's Guide.
Now that the linker has decided which library to use, it checks whether the run-time-support library is
present in C6X_C_DIR . The library must be in exactly the same directory as the index library libc.a. If the
library is not present, the linker will will invoke mklib to build it. This happens when the library is missing,
regardless of whether the user specified the name of the library directly or allowed the linker to pick the
best library from the index library.
The mklib program builds the requested library and places it in 'lib' directory part of C6X_C_DIR in the
same directory as the index library, so it is available for subsequent compilations.
Things to watch out for:
• The linker invokes mklib and waits for it to finish before finishing the link, so you will experience a one-
time delay when an uncommonly-used library is built for the first time. Build times of 1-5 minutes have
been observed. This depends on the power of the host (number of CPUs, etc).
• In a shared installation, where an installation of the compiler is shared among more than one user, it is
possible that two users might cause the linker to rebuild the same library at the same time. The mklib
program tries to minimize the race condition, but it is possible one build will corrupt the other. In a
shared environment, all libraries which might be needed should be built at install time; see
Section 8.5.2.2 for instructions on invoking mklib directly to avoid this problem.
• The index library must exist, or the linker is unable to rebuild libraries automatically.
• The index library must be in a user-writable directory, or the library is not built. If the compiler
installation must be installed read-only (a good practice for shared installation), any missing libraries
must be built at installation time by invoking mklib directly.
• The mklib program is specific to a certain version of a certain library; you cannot use one compiler
version's run-time support's mklib to build a different compiler version's run-time support library.

8.5.2.2 Invoking mklib Manually

You may need to invoke mklib directly in special circumstances:
• The compiler installation directory is read-only or shared.
• You want to build a variant of the run-time-support library that is not pre-configured in the index library
libc.a or known to mklib. (e.g. a variant with source-level debugging turned on.)

8.5.2.2.1 Building Standard Libraries

You can invoke mklib directly to build any or all of the libraries indexed in the index library libc.a. The
libraries are built with the standard options for that library; the library names and the appropriate standard
option sets are known to mklib.
This is most easily done by changing the working directory to be the compiler run-time-support library
directory 'lib' and invoking the mklib executable there:
mklib --pattern=rts6200.lib

8.5.2.2.2 Shared or Read-Only Library Directory

If the compiler tools are to be installed in shared or read-only directory, mklib cannot build the standard
libraries at link time; the libraries must be built before the library directory is made shared or read-only.
At installation time, the installing user must build all of the libraries which will be used by any user. To
build all possible libraries, change the working directory to be the compiler RTS library directory 'lib' and
invoke the mklib executable there:
mklib --all

Some targets have many libraries, so this step can take a long time. To build a subset of the libraries,
invoke mklib individually for each desired library.

8.5.2.2.3 Building Libraries With Custom Options

You can build a library with any extra custom options desired. This is useful for building a debugging
version of the library, or with silicon exception workarounds enabled. The generated library is not a
standard library, and must not be placed in the 'lib' directory. It should be placed in a directory local to the
project which needs it. To build a debugging version of the library rts6200.lib, change the working directory
to the 'lib' directory and run the command:
mklib --pattern=rts6200.lib --name=rts6200_debug.lib --install_to=$Project/Debug --
extra_options="-g"

8.5.2.2.4 The mklib Program Option Summary

Run the following command to see the full list of options. These are described in Table 8-1.
mklib --help

Table 8-1. The mklib Program Options

Option Effect
--index=filename The index library (libc.a) for this release. Used to find a template library for custom builds, and to find the
source files (rtssrc.zip). REQUIRED.
--pattern=filename Pattern for building a library. If neither --extra_options nor --options are specified, the library will be the
standard library with the standard options for that library. If either --extra_options or --options are
specified, the library is a custom library with custom options. REQUIRED unless --all is used.
--all Build all standard libraries at once.
--install_to=directory The directory into which to write the library. For a standard library, this defaults to the same directory as
the index library (libc.a). For a custom library, this option is REQUIRED.
--compiler_bin_dir= The directory where the compiler executables are. When invoking mklib directly, the executables should
directory be in the path, but if they are not, this option must be used to tell mklib where they are. This option is
primarily for use when mklib is invoked by the linker.
--name=filename File name for the library with no directory part. Only useful for custom libraries.
--options='str' Options to use when building the library. The default options (see below) are replaced by this string. If
this option is used, the library will be a custom library.
--extra_options='str' Options to use when building the library. The default options (see below) are also used. If this option is
used, the library will be a custom library.
--list_libraries List the libraries this script is capable of building and exit. ordinary system-specific directory.
--log=filename Save the build log as filename.
--tmpdir=directory Use directory for scratch space instead of the ordinary system-specific directory.
--gmake=filename Gmake-compatible program to invoke instead of "gmake"
--parallel=N Compile N files at once ("gmake -j N").
--query=filename Does this script know how to build FILENAME?
--help or --h Display this help.
--quiet or --q Operate silently.
--verbose or --v Extra information to debug this executable.

Examples:
To build all standard libraries and place them in the compiler's library directory:
mklib --all --index=$C_DIR/lib

To build one standard library and place it in the compiler's library directory:
mklib --pattern=rts6200.lib --index=$C_DIR/lib

To build a custom library that is just like rts6200.lib, but has symbolic debugging support enabled:
mklib --pattern=rts6200.lib --extra_options="-g" --index=$C_DIR/lib --install_to=$Project/Debug --
name=rts6200_debug.lib

8.5.3 Extending mklib

The mklib API is a uniform interface that allows Code Composer Studio to build libraries without needing
to know exactly what underlying mechanism is used to build it. Each library vendor (e.g. the TI compiler)
provides a library-specific copy of 'mklib' in the library directory that can be invoked, which understands a
standardized set of options, and understands how to build the library. This allows the linker to
automatically build application-compatible versions of any vendor's library without needing to register the
library in advance, as long as the vendor supports mklib.

8.5.3.1 Underlying Mechanism

The underlying mechanism can be anything the vendor desires. For the compiler run-time-support
libraries, mklib is just a wrapper which knows how to unpack Makefile from rtssrc.zip and invoke gmake
with the appropriate options to build each library. If necessary, mklib can be bypassed and Makefile used
directly, but this mode of operation is not supported by TI, and the you are responsible for any changes to
Makefile. The format of the Makefile and the interface between mklib and the Makefile is subject to change
without notice. The mklib program is the forward-compatible path.

8.5.3.2 Libraries From Other Vendors

Any vendor who wishes to distribute a library that can be rebuilt automatically by the linker must provide:
• An index library (like 'libc.a', but with a different name)
• A copy of mklib specific to that library
• A copy of the library source code (in whatever format is convenient)
These things must be placed together in one directory that is part of the linker's library search path
(specified either in C6X_C_DIR or with the linker --search_path option).
If mklib needs extra information that is not possible to pass as command-line options to the compiler, the
vendor will need to provide some other means of discovering the information (such as a configuration file
written by a wizard run from inside CCS).
The vendor-supplied mklib must at least accept all of the options listed in Table 8-1 without error, even if
they do not do anything.

C++ Name Demangler

The C++ compiler implements function overloading, operator overloading, and type-safe linking by
encoding a function's prototype and namespace in its link-level name. The process of encoding the
prototype into the linkname is often referred to as name mangling. When you inspect mangled names,
such as in assembly files, disassembler output, or compiler or linker diagnostics, it can be difficult to
associate a mangled name with its corresponding name in the C++ source code. The C++ name
demangler is a debugging aid that translates each mangled name it detects to its original name found in
the C++ source code.
These topics tell you how to invoke and use the C++ name demangler. The C++ name demangler reads
in input, looking for mangled names. All unmangled text is copied to output unaltered. All mangled names
are demangled before being copied to output.

Topic ........................................................................................................................... Page

9.1 Invoking the C++ Name Demangler .................................................................... 269

9.2 C++ Name Demangler Options .......................................................................... 269
9.3 Sample Usage of the C++ Name Demangler ........................................................ 270

268 C++ Name Demangler SPRU187U – July 2012

9.1 Invoking the C++ Name Demangler

The syntax for invoking the C++ name demangler is:

dem6x [options ] [filenames]

dem6x Command that invokes the C++ name demangler.

options Options affect how the name demangler behaves. Options can appear anywhere on the
command line. (Options are discussed in Section 9.2.)
filenames Text input files, such as the assembly file output by the compiler, the assembler listing file,
the disassembly file, and the linker map file. If no filenames are specified on the command
line, dem6x uses standard input.

By default, the C++ name demangler outputs to standard output. You can use the -o file option if you want
to output to a file.

9.2 C++ Name Demangler Options

The following options apply only to the C++ name demangler:

--abi=eabi Demangles EABI identifiers

-h Prints a help screen that provides an online summary of the C++ name demangler
options
-o file Outputs to the given file rather than to standard out
-u Specifies that external names do not have a C++ prefix
-v Enables verbose mode (outputs a banner)

SPRU187U – July 2012 C++ Name Demangler 269

9.3 Sample Usage of the C++ Name Demangler

The examples in this section illustrate the demangling process. Example 9-1 shows a sample C++
program. Example 9-2 shows the resulting assembly that is output by the compiler. In this example, the
linknames of all the functions are mangled; that is, their signature information is encoded into their names.

Example 9‑1. C++ Code for calories_in_a_banana

class banana {

public:

int calories(void);

banana();

~banana();

};

int calories_in_a_banana(void)
{
banana x;

return x.calories();
}

Example 9‑2. Resulting Assembly for calories_in_a_banana

_calories_in_a_banana__Fv:
;** ----------------------------------------------------------------------*
CALL .S1 ___ct__6bananaFv ; |10|
STW .D2T2 B3,*SP--(16) ; |9|
MVKL .S2 RL0,B3 ; |10|
MVKH .S2 RL0,B3 ; |10|
ADD .S1X 8,SP,A4 ; |10|
NOP 1
RL0: ; CALL OCCURS ; |10|
CALL .S1 _calories__6bananaFv ; |12|
MVKL .S2 RL1,B3 ; |12|
ADD .S1X 8,SP,A4 ; |12|
MVKH .S2 RL1,B3 ; |12|
NOP 2
RL1: ; CALL OCCURS ; |12|
CALL .S1 ___dt__6bananaFv ; |13|
STW .D2T1 A4,*+SP(4) ; |12|
ADD .S1X 8,SP,A4 ; |13|
MVKL .S2 RL2,B3 ; |13|
MVK .S2 0x2,B4 ; |13|
MVKH .S2 RL2,B3 ; |13|
RL2: ; CALL OCCURS ; |13|
LDW .D2T1 *+SP(4),A4 ; |12|
LDW .D2T2 *++SP(16),B3 ; |13|
NOP 4
RET .S2 B3 ; |13|
NOP 5
; BRANCH OCCURS ; |13|

Executing the C++ name demangler demangles all names that it believes to be mangled. Enter:
dem6x calories_in_a_banana.asm

270 C++ Name Demangler SPRU187U – July 2012

The result is shown in Example 9-3. The linknames in Example 9-2 ___ct__6bananaFv,
_calories__6bananaFv, and ___dt__6bananaFv are demangled.

Example 9‑3. Result After Running the C++ Name Demangler

calories_in_a_banana():
;** ----------------------------------------------------------------------*
CALL .S1 banana::banana() ; |10|
STW .D2T2 B3,*SP--(16) ; |9|
MVKL .S2 RL0,B3 ; |10|
MVKH .S2 RL0,B3 ; |10|
ADD .S1X 8,SP,A4 ; |10|
NOP 1
RL0: ; CALL OCCURS ; |10|
CALL .S1 banana::calories() ; |12|
MVKL .S2 RL1,B3 ; |12|
ADD . S1X 8,SP,A4 ; |12|
MVKH .S2 RL1,B3 ; |12|
NOP 2
RL1: ; CALL OCCURS ; |12|
CALL .S1 banana::~banana() ; |13|
STW .D2T1 A4,*+SP(4) ; |12|
ADD .S1X 8,SP,A4 ; |13|
MVKL .S2 RL2,B3 ; |13|
MVK . S2 0x2,B4 ; |13|
MVKH . S2 RL2,B3 ; |13|
RL2: ; CALL OCCURS ; |13|
LDW .D2T1 *+SP(4),A4 ; |12|
LDW .D2T2 *++SP(16),B3 ; |13|
NOP 4
RET .S2 B3 ; |13|
NOP 5
; BRANCH OCCURS ; |13|

SPRU187U – July 2012 C++ Name Demangler 271

Glossary

absolute lister— A debugging tool that allows you to create assembler listings that contain absolute
addresses.
assignment statement— A statement that initializes a variable with a value.
autoinitialization— The process of initializing global C variables (contained in the .cinit section) before
program execution begins.
autoinitialization at run time— An autoinitialization method used by the linker when linking C code. The
linker uses this method when you invoke it with the --rom_model link option. The linker loads the
.cinit section of data tables into memory, and variables are initialized at run time.
alias disambiguation— A technique that determines when two pointer expressions cannot point to the
same location, allowing the compiler to freely optimize such expressions.
aliasing— The ability for a single object to be accessed in more than one way, such as when two
pointers point to a single object. It can disrupt optimization, because any indirect reference could
refer to any other object.
allocation— A process in which the linker calculates the final memory addresses of output sections.
ANSI— American National Standards Institute; an organization that establishes standards voluntarily
followed by industries.
archive library— A collection of individual files grouped into a single file by the archiver.
archiver— A software program that collects several individual files into a single file called an archive
library. With the archiver, you can add, delete, extract, or replace members of the archive library.
assembler— A software program that creates a machine-language program from a source file that
contains assembly language instructions, directives, and macro definitions. The assembler
substitutes absolute operation codes for symbolic operation codes and absolute or relocatable
addresses for symbolic addresses.
assignment statement— A statement that initializes a variable with a value.
autoinitialization— The process of initializing global C variables (contained in the .cinit section) before
program execution begins.
autoinitialization at run time— An autoinitialization method used by the linker when linking C code. The
linker uses this method when you invoke it with the --rom_model link option. The linker loads the
.cinit section of data tables into memory, and variables are initialized at run time.
big endian— An addressing protocol in which bytes are numbered from left to right within a word. More
significant bytes in a word have lower numbered addresses. Endian ordering is hardware-specific
and is determined at reset. See also little endian
block— A set of statements that are grouped together within braces and treated as an entity.
.bss section— One of the default object file sections. You use the assembler .bss directive to reserve a
specified amount of space in the memory map that you can use later for storing data. The .bss
section is uninitialized.
byte— Per ANSI/ISO C, the smallest addressable unit that can hold a character.

272 Glossary SPRU187U – July 2012

C/C++ compiler— A software program that translates C source statements into assembly language
source statements.
code generator— A compiler tool that takes the file produced by the parser or the optimizer and
produces an assembly language source file.
COFF— Common object file format; a system of object files configured according to a standard
developed by AT&T. These files are relocatable in memory space.
command file— A file that contains options, filenames, directives, or commands for the linker or hex
conversion utility.
comment— A source statement (or portion of a source statement) that documents or improves
readability of a source file. Comments are not compiled, assembled, or linked; they have no effect
on the object file.
compiler program— A utility that lets you compile, assemble, and optionally link in one step. The
compiler runs one or more source modules through the compiler (including the parser, optimizer,
and code generator), the assembler, and the linker.
configured memory— Memory that the linker has specified for allocation.
constant— A type whose value cannot change.
cross-reference listing— An output file created by the assembler that lists the symbols that were
defined, what line they were defined on, which lines referenced them, and their final values.
.data section— One of the default object file sections. The .data section is an initialized section that
contains initialized data. You can use the .data directive to assemble code into the .data section.
direct call— A function call where one function calls another using the function's name.
directives— Special-purpose commands that control the actions and functions of a software tool (as
opposed to assembly language instructions, which control the actions of a device).
disambiguation— See alias disambiguation
dynamic memory allocation— A technique used by several functions (such as malloc, calloc, and
realloc) to dynamically allocate memory for variables at run time. This is accomplished by defining a
large memory pool (heap) and using the functions to allocate memory from the heap.
ELF— Executable and Linkable Format; a system of object files configured according to the System V
Application Binary Interface specification.
emulator— A hardware development system that duplicates the TMS320C6000 operation.
entry point— A point in target memory where execution starts.
environment variable— A system symbol that you define and assign to a string. Environmental variables
are often included in Windows batch files or UNIX shell scripts such as .cshrc or .profile.
epilog— The portion of code in a function that restores the stack and returns. See also pipelined-loop
epilog.
executable object file— A linked, executable object file that is downloaded and executed on a target
system.
expression— A constant, a symbol, or a series of constants and symbols separated by arithmetic
operators.
external symbol— A symbol that is used in the current program module but defined or declared in a
different program module.

SPRU187U – July 2012 Glossary 273

file-level optimization— A level of optimization where the compiler uses the information that it has about
the entire file to optimize your code (as opposed to program-level optimization, where the compiler
uses information that it has about the entire program to optimize your code).
function inlining— The process of inserting code for a function at the point of call. This saves the
overhead of a function call and allows the optimizer to optimize the function in the context of the
surrounding code.
global symbol— A symbol that is either defined in the current module and accessed in another, or
accessed in the current module but defined in another.
high-level language debugging— The ability of a compiler to retain symbolic and high-level language
information (such as type and function definitions) so that a debugging tool can use this
information.
indirect call— A function call where one function calls another function by giving the address of the
called function.
initialization at load time— An autoinitialization method used by the linker when linking C/C++ code. The
linker uses this method when you invoke it with the --ram_model link option. This method initializes
variables at load time instead of run time.
initialized section— A section from an object file that will be linked into an executable object file.
input section— A section from an object file that will be linked into an executable object file.
integrated preprocessor— A C/C++ preprocessor that is merged with the parser, allowing for faster
compilation. Stand-alone preprocessing or preprocessed listing is also available.
interlist feature— A feature that inserts as comments your original C/C++ source statements into the
assembly language output from the assembler. The C/C++ statements are inserted next to the
equivalent assembly instructions.
intrinsics— Operators that are used like functions and produce assembly language code that would
otherwise be inexpressible in C, or would take greater time and effort to code.
ISO— International Organization for Standardization; a worldwide federation of national standards
bodies, which establishes international standards voluntarily followed by industries.
kernel— The body of a software-pipelined loop between the pipelined-loop prolog and the pipelined-loop
epilog.
K&R C— Kernighan and Ritchie C, the de facto standard as defined in the first edition of The C
Programming Language (K&R). Most K&R C programs written for earlier, non-ISO C compilers
should correctly compile and run without modification.
label— A symbol that begins in column 1 of an assembler source statement and corresponds to the
address of that statement. A label is the only assembler statement that can begin in column 1.
linker— A software program that combines object files to form an executable object file that can be
allocated into system memory and executed by the device.
listing file— An output file, created by the assembler, that lists source statements, their line numbers,
and their effects on the section program counter (SPC).
little endian— An addressing protocol in which bytes are numbered from right to left within a word. More
significant bytes in a word have higher numbered addresses. Endian ordering is hardware-specific
and is determined at reset. See also big endian
loader— A device that places an executable object file into system memory.
loop unrolling— An optimization that expands small loops so that each iteration of the loop appears in
your code. Although loop unrolling increases code size, it can improve the performance of your
code.

274 Glossary SPRU187U – July 2012

macro— A user-defined routine that can be used as an instruction.

macro call— The process of invoking a macro.
macro definition— A block of source statements that define the name and the code that make up a
macro.
macro expansion— The process of inserting source statements into your code in place of a macro call.
map file— An output file, created by the linker, that shows the memory configuration, section
composition, section allocation, symbol definitions and the addresses at which the symbols were
defined for your program.
memory map— A map of target system memory space that is partitioned into functional blocks.
name mangling— A compiler-specific feature that encodes a function name with information regarding
the function's arguments return types.
object file— An assembled or linked file that contains machine-language object code.
object library— An archive library made up of individual object files.
operand— An argument of an assembly language instruction, assembler directive, or macro directive
that supplies information to the operation performed by the instruction or directive.
optimizer— A software tool that improves the execution speed and reduces the size of C programs. See
also assembly optimizer.
options— Command-line parameters that allow you to request additional or specific functions when you
invoke a software tool.
output section— A final, allocated section in a linked, executable module.
parser— A software tool that reads the source file, performs preprocessing functions, checks the syntax,
and produces an intermediate file used as input for the optimizer or code generator.
partitioning— The process of assigning a data path to each instruction.
pipelining— A technique where a second instruction begins executing before the first instruction has
been completed. You can have several instructions in the pipeline, each at a different processing
stage.
pop— An operation that retrieves a data object from a stack.
pragma— A preprocessor directive that provides directions to the compiler about how to treat a particular
statement.
preprocessor— A software tool that interprets macro definitions, expands macros, interprets header
files, interprets conditional compilation, and acts upon preprocessor directives.
program-level optimization— An aggressive level of optimization where all of the source files are
compiled into one intermediate file. Because the compiler can see the entire program, several
optimizations are performed with program-level optimization that are rarely applied during file-level
optimization.
prolog— The portion of code in a function that sets up the stack. See also pipelined-loop prolog.
push— An operation that places a data object on a stack for temporary storage.
quiet run— An option that suppresses the normal banner and the progress information.
raw data— Executable code or initialized data in an output section.
relocation— A process in which the linker adjusts all the references to a symbol when the symbol's
address changes.

SPRU187U – July 2012 Glossary 275

run-time environment— The run time parameters in which your program must function. These
parameters are defined by the memory and register conventions, stack organization, function call
conventions, and system initialization.
run-time-support functions— Standard ISO functions that perform tasks that are not part of the C
language (such as memory allocation, string conversion, and string searches).
run-time-support library— A library file, rts.src, that contains the source for the run time-support
functions.
section— A relocatable block of code or data that ultimately will be contiguous with other sections in the
memory map.
sign extend— A process that fills the unused MSBs of a value with the value's sign bit.
simulator— A software development system that simulates TMS320C6000 operation.
source file— A file that contains C/C++ code or assembly language code that is compiled or assembled
to form an object file.
stand-alone preprocessor— A software tool that expands macros, #include files, and conditional
compilation as an independent program. It also performs integrated preprocessing, which includes
parsing of instructions.
static variable— A variable whose scope is confined to a function or a program. The values of static
variables are not discarded when the function or program is exited; their previous value is resumed
when the function or program is reentered.
storage class— An entry in the symbol table that indicates how to access a symbol.
string table— A table that stores symbol names that are longer than eight characters (symbol names of
eight characters or longer cannot be stored in the symbol table; instead they are stored in the string
table). The name portion of the symbol's entry points to the location of the string in the string table.
structure— A collection of one or more variables grouped together under a single name.
subsection— A relocatable block of code or data that ultimately will occupy continuous space in the
memory map. Subsections are smaller sections within larger sections. Subsections give you tighter
control of the memory map.
symbol— A string of alphanumeric characters that represents an address or a value.
symbolic debugging— The ability of a software tool to retain symbolic information that can be used by a
debugging tool such as a simulator or an emulator.
target system— The system on which the object code you have developed is executed.
.text section— One of the default object file sections. The .text section is initialized and contains
executable code. You can use the .text directive to assemble code into the .text section.
trigraph sequence— A 3-character sequence that has a meaning (as defined by the ISO 646-1983
Invariant Code Set). These characters cannot be represented in the C character set and are
expanded to one character. For example, the trigraph ??' is expanded to ^.
trip count— The number of times that a loop executes before it terminates.
unconfigured memory— Memory that is not defined as part of the memory map and cannot be loaded
with code or data.
uninitialized section— A object file section that reserves space in the memory map but that has no
actual contents. These sections are built with the .bss and .usect directives.
unsigned value— A value that is treated as a nonnegative number, regardless of its actual sign.
variable— A symbol representing a quantity that can assume any of a set of values.

276 Glossary SPRU187U – July 2012

word— A 32-bit addressable location in target memory

SPRU187U – July 2012 Glossary 277

Submit Documentation Feedback
Copyright © 2012, Texas Instruments Incorporated
IMPORTANT NOTICE
Texas Instruments Incorporated and its subsidiaries (TI) reserve the right to make corrections, enhancements, improvements and other
changes to its semiconductor products and services per JESD46C and to discontinue any product or service per JESD48B. Buyers should
obtain the latest relevant information before placing orders and should verify that such information is current and complete. All
semiconductor products (also referred to herein as “components”) are sold subject to TI’s terms and conditions of sale supplied at the time
of order acknowledgment.
TI warrants performance of its components to the specifications applicable at the time of sale, in accordance with the warranty in TI’s terms
and conditions of sale of semiconductor products. Testing and other quality control techniques are used to the extent TI deems necessary
to support this warranty. Except where mandated by applicable law, testing of all parameters of each component is not necessarily
performed.
TI assumes no liability for applications assistance or the design of Buyers’ products. Buyers are responsible for their products and
applications using TI components. To minimize the risks associated with Buyers’ products and applications, Buyers should provide
adequate design and operating safeguards.
TI does not warrant or represent that any license, either express or implied, is granted under any patent right, copyright, mask work right, or
other intellectual property right relating to any combination, machine, or process in which TI components or services are used. Information
published by TI regarding third-party products or services does not constitute a license to use such products or services or a warranty or
endorsement thereof. Use of such information may require a license from a third party under the patents or other intellectual property of the
third party, or a license from TI under the patents or other intellectual property of TI.
Reproduction of significant portions of TI information in TI data books or data sheets is permissible only if reproduction is without alteration
and is accompanied by all associated warranties, conditions, limitations, and notices. TI is not responsible or liable for such altered
documentation. Information of third parties may be subject to additional restrictions.
Resale of TI components or services with statements different from or beyond the parameters stated by TI for that component or service
voids all express and any implied warranties for the associated TI component or service and is an unfair and deceptive business practice.
TI is not responsible or liable for any such statements.
Buyer acknowledges and agrees that it is solely responsible for compliance with all legal, regulatory and safety-related requirements
concerning its products, and any use of TI components in its applications, notwithstanding any applications-related information or support
that may be provided by TI. Buyer represents and agrees that it has all the necessary expertise to create and implement safeguards which
anticipate dangerous consequences of failures, monitor failures and their consequences, lessen the likelihood of failures that might cause
harm and take appropriate remedial actions. Buyer will fully indemnify TI and its representatives against any damages arising out of the use
of any TI components in safety-critical applications.
In some cases, TI components may be promoted specifically to facilitate safety-related applications. With such components, TI’s goal is to
help enable customers to design and create their own end-product solutions that meet applicable functional safety standards and
requirements. Nonetheless, such components are subject to these terms.
No TI components are authorized for use in FDA Class III (or similar life-critical medical equipment) unless authorized officers of the parties
have executed a special agreement specifically governing such use.
Only those TI components which TI has specifically designated as military grade or “enhanced plastic” are designed and intended for use in
military/aerospace applications or environments. Buyer acknowledges and agrees that any military or aerospace use of TI components
which have not been so designated is solely at the Buyer's risk, and that Buyer is solely responsible for compliance with all legal and
regulatory requirements in connection with such use.
TI has specifically designated certain components which meet ISO/TS16949 requirements, mainly for automotive use. Components which
have not been so designated are neither designed nor intended for automotive use; and TI will not be responsible for any failure of such
components to meet such requirements.
Products Applications
Audio www.ti.com/audio Automotive and Transportation www.ti.com/automotive
Amplifiers amplifier.ti.com Communications and Telecom www.ti.com/communications
Data Converters dataconverter.ti.com Computers and Peripherals www.ti.com/computers
DLP® Products www.dlp.com Consumer Electronics www.ti.com/consumer-apps
DSP dsp.ti.com Energy and Lighting www.ti.com/energy
Clocks and Timers www.ti.com/clocks Industrial www.ti.com/industrial
Interface interface.ti.com Medical www.ti.com/medical
Logic logic.ti.com Security www.ti.com/security
Power Mgmt power.ti.com Space, Avionics and Defense www.ti.com/space-avionics-defense
Microcontrollers microcontroller.ti.com Video and Imaging www.ti.com/video
RFID www.ti-rfid.com
OMAP Mobile Processors www.ti.com/omap TI E2E Community e2e.ti.com
Wireless Connectivity www.ti.com/wirelessconnectivity

Intrusion Detection Honeypots
From Everand
Intrusion Detection Honeypots
Chris Sanders
3/5 (2)
Hi-Tech C User Manual
100% (2)
Hi-Tech C User Manual
460 pages
Audio, Video, and Media in the Ministry
From Everand
Audio, Video, and Media in the Ministry
Clarence Floyd Richmond
No ratings yet
TMS320C6000 Optimizing Compiler
No ratings yet
TMS320C6000 Optimizing Compiler
243 pages
Instruction Intrinseque
No ratings yet
Instruction Intrinseque
275 pages
Optimizing C
No ratings yet
Optimizing C
189 pages
User S Guide: TMS320C28x Optimizing C/C++ Compiler v6.0
No ratings yet
User S Guide: TMS320C28x Optimizing C/C++ Compiler v6.0
176 pages
Spru 187 o
No ratings yet
Spru 187 o
229 pages
Spru 514 H
No ratings yet
Spru 514 H
193 pages
TMS320C28x Optimizing C-C++ Compiler
No ratings yet
TMS320C28x Optimizing C-C++ Compiler
237 pages
Sprui04b - TMS320C6000 Optimizing Compiler v8.2.x
No ratings yet
Sprui04b - TMS320C6000 Optimizing Compiler v8.2.x
290 pages
Gray Hat Hacking the Ethical Hacker's
From Everand
Gray Hat Hacking the Ethical Hacker's
Çağatay Şanlı
5/5 (1)
Optimizing C&C++ Compiler V 4.1 Beta
No ratings yet
Optimizing C&C++ Compiler V 4.1 Beta
160 pages
CCOMPILADOR
No ratings yet
CCOMPILADOR
155 pages
Tms320c6000-Optimizing Compiler Users Guide
No ratings yet
Tms320c6000-Optimizing Compiler Users Guide
427 pages
Software Patterns Made Easy
From Everand
Software Patterns Made Easy
Justice Nanhou
No ratings yet
Spnu 151 W
No ratings yet
Spnu 151 W
211 pages
Blog Smarter, Not Harder: SEO, Blogging, and AI Strategies to Skyrocket Your Traffic
From Everand
Blog Smarter, Not Harder: SEO, Blogging, and AI Strategies to Skyrocket Your Traffic
Jay Nans
No ratings yet
9 Compiler C Spru514j PDF
No ratings yet
9 Compiler C Spru514j PDF
207 pages
The Linux Terminal for Advanced Users - The Command Line Made Easy: First Edition
From Everand
The Linux Terminal for Advanced Users - The Command Line Made Easy: First Edition
Michael Basler
No ratings yet
Content Creation Revolution with chatGPT
From Everand
Content Creation Revolution with chatGPT
Maria Cowen
No ratings yet
ChatGPT for Business: Strategies for Success
From Everand
ChatGPT for Business: Strategies for Success
Matthew C. Smith
1/5 (1)
Securing ChatGPT: Best Practices for Protecting Sensitive Data in AI Language Models
From Everand
Securing ChatGPT: Best Practices for Protecting Sensitive Data in AI Language Models
Matthew C. Smith
No ratings yet
Plain JavaScript: Learning the Front-End
From Everand
Plain JavaScript: Learning the Front-End
Roger Beans-Rivet
No ratings yet
Mastering Python Advanced Concepts and Practical Applications
From Everand
Mastering Python Advanced Concepts and Practical Applications
Aissa Younes
No ratings yet
A To Z of Internet: Everything You Wanted to Know
From Everand
A To Z of Internet: Everything You Wanted to Know
Bittu Kumar
No ratings yet
Slau132f PDF
No ratings yet
Slau132f PDF
166 pages
Spru 103 F
No ratings yet
Spru 103 F
299 pages
C6x Programmer's Guide (198d)
No ratings yet
C6x Programmer's Guide (198d)
398 pages
Slau 132 y
No ratings yet
Slau 132 y
187 pages
Cybersecurity for Executives: A Guide to Protecting Your Business
From Everand
Cybersecurity for Executives: A Guide to Protecting Your Business
Matthew C. Smith
No ratings yet
Tms320 DSP Bios User's Guide
No ratings yet
Tms320 DSP Bios User's Guide
286 pages
Advanced Multiplayer Game Development with Ureal Engine 5: A Comprehensive Guide to C++ Scripting
From Everand
Advanced Multiplayer Game Development with Ureal Engine 5: A Comprehensive Guide to C++ Scripting
Vladimir Kiselev
No ratings yet
Memory Management Spru281c
No ratings yet
Memory Management Spru281c
304 pages
CCS Tutorial
No ratings yet
CCS Tutorial
126 pages
Chapter3 C
No ratings yet
Chapter3 C
38 pages
CAN Bus for Beginners: A Practical Guide to Automotive Networking
From Everand
CAN Bus for Beginners: A Practical Guide to Automotive Networking
Mohamad Charara
No ratings yet
Spnu151j PDF
No ratings yet
Spnu151j PDF
202 pages
ARM Optimizing C/C++ Compiler
No ratings yet
ARM Optimizing C/C++ Compiler
210 pages
Unlocking Statistics for the Social Sciences
From Everand
Unlocking Statistics for the Social Sciences
Norma Sinclair
No ratings yet
10K Blueprint
From Everand
10K Blueprint
Cian O Farrell
5/5 (2)
ARM Optimizing C/C++
No ratings yet
ARM Optimizing C/C++
203 pages
MSP430 Optimizing C - C++ Compiler
No ratings yet
MSP430 Optimizing C - C++ Compiler
184 pages
Spruig 5 D
No ratings yet
Spruig 5 D
12 pages
Tms Assembler Directives
No ratings yet
Tms Assembler Directives
523 pages
Web Video Business
From Everand
Web Video Business
MUHAMMAD NUR WAHID ANUAR
No ratings yet
H8S, H8/300 Series C/C++ Compiler
No ratings yet
H8S, H8/300 Series C/C++ Compiler
635 pages
Tms320c54x DSP Ccs Tutorial
No ratings yet
Tms320c54x DSP Ccs Tutorial
126 pages
Optimizing C C++ Compiler
No ratings yet
Optimizing C C++ Compiler
201 pages
Human Nature Potential in Nurture
From Everand
Human Nature Potential in Nurture
David L. Hawk
No ratings yet
Grow with Python Programming: From Basics to Advanced
From Everand
Grow with Python Programming: From Basics to Advanced
Mark Fliks
No ratings yet
General Purpose Applications Spru 164
No ratings yet
General Purpose Applications Spru 164
167 pages
TMS320C62xx CPU and Instruction Set Reference Guide
No ratings yet
TMS320C62xx CPU and Instruction Set Reference Guide
209 pages
Hi-Tech Picc-18 STD
No ratings yet
Hi-Tech Picc-18 STD
490 pages
SC Compiler Users Guide
No ratings yet
SC Compiler Users Guide
263 pages
Tasking C196 Manual
No ratings yet
Tasking C196 Manual
389 pages
Manual PDF
100% (1)
Manual PDF
442 pages
CSL
No ratings yet
CSL
1,108 pages
Tms 320 C 5 X
100% (2)
Tms 320 C 5 X
764 pages
MW Enterprise Compiler
No ratings yet
MW Enterprise Compiler
380 pages
Program L
No ratings yet
Program L
20 pages
Separate Odd Even Numbers From Given Array
No ratings yet
Separate Odd Even Numbers From Given Array
26 pages
Automating The Construction of Compiler Heuristics Using Machine Learning
No ratings yet
Automating The Construction of Compiler Heuristics Using Machine Learning
162 pages
Android Assignment
No ratings yet
Android Assignment
19 pages
Common Path Pessimism Removal An Industry Perspective
No ratings yet
Common Path Pessimism Removal An Industry Perspective
4 pages
AT&CD Unit 5
No ratings yet
AT&CD Unit 5
13 pages
Apache Spark - Optimization Techniques
No ratings yet
Apache Spark - Optimization Techniques
7 pages
Embedded and Real Time Systems - Lecture Notes, Study Material and Important Questions, Answers
No ratings yet
Embedded and Real Time Systems - Lecture Notes, Study Material and Important Questions, Answers
6 pages
The_fundamental_role_of_artificial_intelligence_in_the_IT_industry
No ratings yet
The_fundamental_role_of_artificial_intelligence_in_the_IT_industry
6 pages
Compperf PDF
No ratings yet
Compperf PDF
398 pages
Reducing Boot Time: Techniques For Fast Booting
No ratings yet
Reducing Boot Time: Techniques For Fast Booting
27 pages
Compiler Construction Week 14
No ratings yet
Compiler Construction Week 14
23 pages
Course Code Course Title L T P C 1153EC103 Embedded System Design 3 0 0 3
No ratings yet
Course Code Course Title L T P C 1153EC103 Embedded System Design 3 0 0 3
3 pages
Dripax - Process Optimization For Direct Reduction Plants
No ratings yet
Dripax - Process Optimization For Direct Reduction Plants
2 pages
INFORMATICA-Performance Tuning
100% (9)
INFORMATICA-Performance Tuning
21 pages
Python Astronomy
0% (1)
Python Astronomy
44 pages
Architecture ! Verilog An Extended Example: Always @ (Posedge CLK) Begin
No ratings yet
Architecture ! Verilog An Extended Example: Always @ (Posedge CLK) Begin
16 pages
Ocean Manual
No ratings yet
Ocean Manual
29 pages
Managing Credit Line
No ratings yet
Managing Credit Line
5 pages
Automated Translation of MATLAB Code To C++ With Performance and Traceability
No ratings yet
Automated Translation of MATLAB Code To C++ With Performance and Traceability
6 pages
Modern GPU
100% (1)
Modern GPU
221 pages
5.3 Principal Sources of Optimization
No ratings yet
5.3 Principal Sources of Optimization
22 pages
ABB - Benefits of State Based Control White Paper
No ratings yet
ABB - Benefits of State Based Control White Paper
32 pages
Ec8791 Erts Syllabus
No ratings yet
Ec8791 Erts Syllabus
2 pages
MineStar Fleet - Benefit Summary
No ratings yet
MineStar Fleet - Benefit Summary
19 pages
DS 18031 Designer To Analyst Topology Whitepaper
No ratings yet
DS 18031 Designer To Analyst Topology Whitepaper
8 pages
Programmer Competency Matrix
No ratings yet
Programmer Competency Matrix
23 pages
Date: - Lab Manual Chemical Process Optimization (Ch-404) Lab No: - Roll No
No ratings yet
Date: - Lab Manual Chemical Process Optimization (Ch-404) Lab No: - Roll No
2 pages
docker-aws
No ratings yet
docker-aws
9 pages
Synthesis Optimization Techniques 7
75% (4)
Synthesis Optimization Techniques 7
11 pages