Use aligned_alloc when the platform we use supports it

[AZero13]

aligned_alloc is the preferred way to do AlignedAlloc when we can support that, with the other code being a fallback if that is not the case.

[AZero13]

@ahoppen Could you please initiate the test?

[ahoppen]

I’m not at all familiar with this code. @jckar Looks like you wrote it originally 10 years ago, could you take a look?

[AZero13]

@compnerd can we please test this

[compnerd]

@AtariDreams I like the premise of the change, but I do not believe that the change is correct. The second condition for Windows needs to be changed to #elif. I would also recommend that you double check if MSVCRT supports aligned_alloc (it does support most of C11).

[AZero13]

@AtariDreams I like the premise of the change, but I do not believe that the change is correct. The second condition for Windows needs to be changed to #elif. I would also recommend that you double check if MSVCRT supports aligned_alloc (it does support most of C11).

Fixed!

[AZero13]

@AtariDreams I like the premise of the change, but I do not believe that the change is correct. The second condition for Windows needs to be changed to #elif. I would also recommend that you double check if MSVCRT supports aligned_alloc (it does support most of C11).

Just checked, and Microsoft's C runtime does not support it, so thank you.

[compnerd]

@swift-ci please test

[compnerd]

@swift-ci please benchmark

[AZero13]

@compnerd Thoughts? Things passed but benchmarks are wishy-washy

[compnerd]

I think that we can run them once again to see if they are stable or just noise

[compnerd]

@swift-ci please benchmark

[compnerd]

FWIW, I do not expect this to impact the benchmarks in any meaningful way (if there is a bit of noise, it is likely for cache line behaviour).

[AZero13]

@compnerd benchmarks seem fine

[compnerd]

Yeah, I agree, sounds like it is mostly noise. Going to merge this.