Add VPCLMULQDQ crc32 tests to Google benchmarks

[nmoinvaz]

No description provided.

[Dead2]

LGTM 👍

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (1aa53f4) 83.13% compared to head (f49ddf8) 83.15%.

Files	Patch %	Lines
test/benchmarks/benchmark_crc32.cc	0.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1651      +/-   ##
===========================================
+ Coverage    83.13%   83.15%   +0.01%     
===========================================
  Files          135      135              
  Lines        10895    10898       +3     
  Branches      2816     2817       +1     
===========================================
+ Hits          9058     9062       +4     
  Misses        1127     1127              
+ Partials       710      709       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[nmoinvaz]

Strange, it seems this potentially discovered a crash on GCC benchmark with VPCLMULQDQ. @KungFuJesus maybe you can take a look?

[Dead2]

Strange, it seems this potentially discovered a crash on GCC benchmark with VPCLMULQDQ.

The test machine doesn't have AVX512, so it shouldn't try to use VPCLMULQDQ at all. So looks like the cpu check is possibly at fault somehow.

[nmoinvaz]

Good observation I didn't think about that possibility.

[nmoinvaz]

I think I missed some of the logic for the benchmark condition. I found existing logic elsewhere and just copied it.

[nmoinvaz]

That doesn't seem to have fixed it. So maybe there is a bug in CPU runtime detection..

[nmoinvaz]

For example, here is what running the same crc32 tests on my desktop (without AVX512) using MSVC shows:

[nmoinvaz]

Looks like that fixed it.

[nmoinvaz]

I'm not sure why the code coverage would be less...

Seems ok?

[mtl1979]

I'm not sure why the code coverage would be less... Seems ok?

There is a lot of partial coverage on the lines with preprocessor macros... This suggests that some of the expanded code is not reached.

[nmoinvaz]

Ah yea those Google testing/benchmark macros have a lot of boilerplate.