Make stripslashes() only dependent on SSE2 configuration.

[nielsdos]

Alex Dowad noticed[1] that the SIMD stripslashes implementation actually only depended on SSE2, and not on SSE4.2 instructions. Remove the checking for SSE4.2 and only check for SSE2. This also greatly simplifies the supporting code.

I thought I'd make a PR since @alexdowad mentioned he still has a lot of other stuff to do [2].

[1] #10313 (comment)
[2] #10313 (comment)

[alexdowad]

Thanks so much for working on this, @nielsdos.

[alexdowad]

The code looks good to me. It is indeed a big simplification.

I am interested in learning more about what __attribute__((ifunc("resolve_stripslashes"))) was for. I can see that we had a global function pointer to the implementation being used, this pointer was set in the module init function, then calls to the function go through that pointer. Why also have the attribute as well?

[nielsdos]

It was necessary to resolve which function to use at link time: either the SSE4.2 one or the normal one. You can read more about it here: https://gcc.gnu.org/onlinedocs/gcc-4.7.2/gcc/Function-Attributes.html (CTRL+F for ifunc ("resolver")).

[alexdowad]

It was necessary to resolve which function to use at link time: either the SSE4.2 one or the normal one. You can read more about it here: https://gcc.gnu.org/onlinedocs/gcc-4.7.2/gcc/Function-Attributes.html (CTRL+F for ifunc ("resolver")).

But then why also have the global function pointer which is set in the module init function? Is it so that the best host-specific implementation will be chosen even if PHP is linked against a different libc, or against an old version of glibc which doesn't support ifunc?

[nielsdos]

But then why also have the global function pointer which is set in the module init function? Is it so that the best host-specific implementation will be chosen even if PHP is linked against a different libc, or against an old version of glibc which doesn't support ifunc?

Yes. In case the dynamic linker / libc does not support the link-time resolution method, then it can fallback to the global function pointer.

[Girgias]

Thank you!