Excessive check for CLMUL on "Westmere" path about simdutf

[lemire]

You are correct. I invite a pull request.

[easyaspi314]

Should I change the references to Nehalem or just leave the technical misnomer as an implementation detail?

[easyaspi314]

Actually. I found a few other bugs with the CPUID check, it doesn't check if the OS enabled AVX support and it doesn't explicitly zero ECX for the second CPUID. Might as well fix both.

[lemire]

Fixed as per #236 by @easyaspi314

Will be part of the next release.

[easyaspi314]

It seems that the commit was immediately reverted but the PR was locked.

[lemire]

Indeed. I was told that the merged PR was causing problems to one corporate user whereas things worked fine for them before.

[easyaspi314]

Do you want me to add the part without the macOS check? That will cause macOS to be locked to AVX2 (due to the wonky AVX512 not being saved until a fault), but it will prevent operating systems that don't support AVX2/AVX512 from crashing.

[lemire]

We will go with the current approach which is well tested and works for our users. If problems arise in real-world scenarios, we will take measures at that time.