Raghuveer Devulapalli

@serge-sans-paille: Reformatted your C++ code and added 16-bit and 64-bit sorting.

Added benchmark numbers. Need to fix the CI errors.

> The [cygwin failure](https://github.com/numpy/numpy/actions/runs/3148780125/jobs/5119674733#step:11:15) is strange: "Permission denied" when importing the c-extension. Is there a compilation problem? Doesn't look like it from logs. It was able to build and install...

> The segfault in SIMD reminds me of [this problem](https://github.com/numpy/numpy/pull/18330#issuecomment-820767278), which was fixed in #18795; maybe that came up again? That's possible. What would be the easiest way to reproduce...

@DWesl Thanks for that! And yup, its the same problem as you mentioned! Bunch of aligned mov instructions to unaligned stack pointer. ``` 0x5af801157 vmovaps %zmm1,0x120(%rsp) 0x5af801162 vzeroupper 0x5af801165 call...

I'm trying to reproduce the Windows Python38-32bit-fast CI failure (followed the steps listed in `azure-steps-windows.yml`). I can't reproduce the failing tests but I see a different test fail: ``` FAILED...

Yes looks like it. Doesn't the CI use the python.exe downloaded from here:? https://github.com/numpy/numpy/blob/3cf2ca1611ecaf1d13c41075a7025b96e54e3ee3/azure-steps-windows.yml#L11 And that is a 64-bit one: ``` PS C:\Users\rdevulap\numpy> ..\pypy3\python.exe --version Python 3.8.12 (0089b4a7ab2306925a251b35912885d52ead1aba, Mar 16...

Might need some help reproducing and figuring out why the tests fail on WIN32 ~and MacOS~. Disabling it for now. Will open an issue so we keep track of it....

The ever elusive "All checks have passed" 🤦‍♂️But I think the last CI [failure ](https://app.travis-ci.com/github/numpy/numpy/jobs/585480747) has nothing to do with this patch.

ping .. any idea why the two failures? one is on Arm64 and other one might also not be related (this patch is disabled on 32-bit windows).