Scribd
Upload
32 views

Homework 5

This document contains details about a homework assignment involving computer architecture concepts like virtual memory, caches, and processors. It asks the student to calculate the size of page tables for a virtual memory system, the maximum size of a virtually indexed physically tagged cache, the CPI of a memory system, analyze dependencies in MIPS code, and compare the performance of a superscalar CPU and SMT CPU in issuing instructions from multiple threads.

Uploaded by

chientv28
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
32 views

Homework 5

This document contains details about a homework assignment involving computer architecture concepts like virtual memory, caches, and processors. It asks the student to calculate the size of page tables for a virtual memory system, the maximum size of a virtually indexed physically tagged cache, the CPI of a memory system, analyze dependencies in MIPS code, and compare the performance of a superscalar CPU and SMT CPU in issuing instructions from multiple threads.

Uploaded by

chientv28
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

CSE 141 Summer 2016 Homework 5

PID:
Name:
1. Assume that you have a computer with 4KB pages and a 4-entry full-associative
TLB that uses LRU replacement policy. If page must be brought into main
memory, increment the largest page number. If the current TLB content is
Valid tag Physical page number
0 1 0xB 12
1 1 0x7 4
2 1 0x3 6
3 0 0x4 9
and the current page table is
Valid Physical page or in disk
0 1 0x5
1 0 Disk
2 0 Disk
3 1 0x6
4 1 0x9
5 1 0xB
6 0 Disk
7 1 0x4
8 0 Disk
9 0 Disk
10 1 0x3
11 1 0xC
Please identify how many TLB misses and page faults in the following address stream:
0x123D, 0x8B3, 0x365C, 0x871B, 0xBEE6, 0x3140, 0xC049
Answer:
Assume index 0 is the most recent used entry.
TLB {TAG [PPN]} | PT update
0x123D PF 1[D], B[C], 7[4], 3[6]
0x08B3 M 0[5], 1[D], B[C], 7[4]
0x365C M 3[6], 0[5], 1[D], B[C]
0x871B PF 8[E], 3[6], 0[5], 1[D]
0xBEE6 M B[C], 8[E], 3[6], 0[5] |1[D]
0x3140 H 3[6], B[C], 8[E], 0[5]
0xC049 PF C[F], 3[6], B[C], 8[E]
3 Page faults and 3(6) TLB Miss // Page faults happens after TLB misses
2. Assume the virtual address space of the computer is 64 bits, each page is 8KB in
size, each page table entry occupies 8 bytes memory and the system is running 6
processes concurrently.
(1) If the computer uses conventional page table, what’s the total size of page
tables in the system?
VA space: 264
Page size: 213
Page table entries (per process): 264 / 213 = 251

Total Page Tables size = #process * Page table entries * entry size
= 6 * 251 * 8
= 3 * 255 bytes (96PB)

(2) If we are building a 4-way set associative, virtually indexed, physically


tagged cache, what’s the maximum available cache size?
A= 4
lg(S) + lg(B) = 13
C <= ABS = 4 * 8K = 32KB
3. You are building a system around a single-issue in-order processor running at 2
GHz and the processor has a base CPI of 1 if all memory accesses are hits. The
only instructions that read or write data from memory are loads (20% of all
instructions) and stores (5% of all instructions). The processor uses
virtually-indexed, physically-tagged caches with no penalty in address translation
if the TLB access is a hit. However, if the TLB misses, the system needs 120ns to
finish the address translation and TLB updates. The TLB miss rate is 2%. The L1
cache is split into I-cache and D-cache with no penalty on hits. Both the I-cache
and D-cache are direct mapped and hold 32KB each. You may assume the caches
use write-allocate and write-back policies. The L1 I-cache has a 2% miss rate and
the L1 D-cache has a 5% miss rate. Also, 50% of all blocks replaced from L1
D-cache are dirty. The 512KB write-back, unified L2 cache has an access time of
10ns. Of all memory references sent to the L2 cache in this system, 80% are
satisfied without going to main memory. Also 25% of all blocks replaced are dirty.
The main memory has an access latency of 60ns. What is the overall CPI,
including memory accesses?
Cycle time = 1 / 2Ghz = 0.5 ns
TLB miss: 120ns = 240 cycles
L2 access: 10ns = 20 cycles
Memory access: 60ns = 120 cycles

L2 miss penalty = Memory access + L2 write back penalty


= 120 + (5% + 20%) * 120
= 150 cycles
L1-I miss penalty = L2 access latency + L2 miss rate * L2 miss penalty
= 20 + (1-80%) * 150
= 50 cycles
L1-D miss penalty = L2 access latency + L2 miss rate * L2 miss penalty
+ L1 write back penalty
= (20+0.2*150)*(1+50%)
= 75 cycles
Overall CPI = Basic CPI + CPI due to I cache miss +
CP I due to D cache miss + CPI due to TLB misses
= 1 + 2% * 50 + (20% + 5%) * 5% * 75
+ 75% * 2% * 240 + 2 * (20% + 5%) * 2% * 240
= 8.94
5. Consider the following execution sequence of MIPS instructions:
1: LOOP: lw $t1, 0($a0)
2: addi $a0, $a0, 4
3: add $v0, $v0, $t1
4: bne $a0, $t0, LOOP
5: LOOP: lw $t1, 0($a0)
6: addi $a0, $a0, 4
7: add $v0, $v0, $t1
8: bne $a0, $t0, LOOP
9: sw $v0, 0($a1)
(1) Please draw the data dependency graph

(2) Please identify the false dependencies (WAW and WAR) in the given code
(3) The processor that you’re given is a 2-issue out-of-order processor with
unlimited physical registers and it takes 3 cycles to finish a load/store, 1
cycle to finish other instructions and with a perfect branch predictor.
Assume all instructions are now already in the instruction window. How
many cycles it takes to issue (moving out from schedule to execution) all
the dynamic instructions?

It takes 6 cycles to issue all instructions.


6. Consider the following two processors:
I. CPU CMP: 2-core superscalar processor with out-of-order issue capabilities.
Each core has two functional units and these functional units can be pipelined if it
needs more cycles to execute. Only a single thread can run on each core at a time.
II. CPU SMT: An SMT processor that allows instructions from two threads to be
run concurrently on the two functional units, and instructions from either or both
threads can be issued to run on any cycle.

Assume we have two threads A and B to run on these CPUs that include the
following operations (You may assume these processors have full capability of
eliminating false dependencies and all the instructions are already in instruction
queues. If not specified, the instruction will take one cycle to execute):
Thread A Thread B
A1: takes 2 cycles to execute B1: no dependencies
A2: depends on the result of A1 B2: conflicts for a functional unit with
A3: conflicts for a functional unit with B1 (They must use the same one)
A2 (They must use the same one) B3: no dependencies
A4: depends on the result of A2 B4: depends on the result of B2
(1) How many cycles will it take to issue these two threads on each processor?
CPU CMP CPU SMT
Core #1 Core #2
A1 A3
A1 A3 B1 B3
B1 B3
B2
B2 A2
A2 B4
B4 A4
A4
4 cycles for thread A
4 cycles for thread A
4cycles for thread B
3 cycles for thread B

(2) How many issue slots are wasted due to hazards?

For CPU CMP: 8 slots


For CPU SMT: 0 slots

You might also like