diff options
| author | Daya S Khudia <dskhudia@fb.com> | 2019-03-21 20:03:36 +0300 |
|---|---|---|
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2019-03-21 20:07:54 +0300 |
| commit | f65f0ebe54f0512d8f42ee10025b596e3f42e0b8 (patch) | |
| tree | 8a80b9de7c8d5ae034d707b27ac7c84cecd83d0d /src/ExecuteKernelU8S8.cc | |
| parent | 452627c5f29412528c26b57880f27914b1068d6e (diff) | |
Improves small N cases back to what they were
Summary: In D14507536 and D14516232 small N cases suffered if we increased the NR. This fixes those cases.
Reviewed By: jianyuh
Differential Revision: D14529494
fbshipit-source-id: 6f53797948de760d6ed24b767cbbe8d27768660f
Diffstat (limited to 'src/ExecuteKernelU8S8.cc')
| -rw-r--r-- | src/ExecuteKernelU8S8.cc | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/ExecuteKernelU8S8.cc b/src/ExecuteKernelU8S8.cc index f2b028d..9b0ea41 100644 --- a/src/ExecuteKernelU8S8.cc +++ b/src/ExecuteKernelU8S8.cc @@ -51,10 +51,10 @@ ExecuteKernel< int8_t, typename packingAMatrix::accType, inst_set_t::avx512>::NCB; - nrSize_ = PackingTraits< + nrMinSize_ = PackingTraits< int8_t, typename packingAMatrix::accType, - inst_set_t::avx512>::NR; + inst_set_t::avx512>::NR_MIN; } else if (fbgemmHasAvx2Support()) { mbSize_ = PackingTraits< int8_t, @@ -64,7 +64,7 @@ ExecuteKernel< int8_t, typename packingAMatrix::accType, inst_set_t::avx2>::NCB; - nrSize_ = PackingTraits< + nrMinSize_ = PackingTraits< int8_t, typename packingAMatrix::accType, inst_set_t::avx2>::NR; @@ -132,7 +132,7 @@ void ExecuteKernel< for (int jb = 0; jb < bColBlocks; ++jb) { if (jb == bColBlocks - 1) { - int nc = ((packedB_.lastBcol() - 1) / nrSize_ + 1) * nrSize_; + int nc = ((packedB_.lastBcol() - 1) / nrMinSize_ + 1) * nrMinSize_; if (nc != nbSize_) { if (cpuinfo_initialize()) { if (fbgemmHasAvx512Support()) { |