Closed vankxr closed 1 year ago
Base: 80.88% // Head: 80.87% // Decreases project coverage by -0.00%
:warning:
Coverage data is based on head (
268fee4
) compared to base (2dfe2ef
). Patch coverage: 97.77% of modified lines in pull request are covered.
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
Added AVX implementations of the existent SIMD accelerated modules (
dotprod
andsumsq
). Also updated the autoconf script which detects CPU features, to be able to detect more recent features.Benchmark results are attached, both for the current HEAD (SSE4.1) and my AVX implementation. A quick comparation of the results (benchmarks whose performance increase is below 1.2x are ommited).
Benchmarked on an i7-8750H.
AVX-512 should also be possible this way, but may have to look deeper into it, since there is no such thing as
__mm512_dp_ps
, for example.