diff options
| author | Mike Tsai <miketsai@fb.com> | 2019-06-15 03:04:25 +0300 |
|---|---|---|
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2019-06-15 03:10:29 +0300 |
| commit | 604575ff5de717b2ee712190634840981a9c8fba (patch) | |
| tree | 198817e0992810a7dff5ac0ed2a99a9b08834346 /src | |
| parent | 5e71d2c304663f3b4e50cee723b8e98a867d11ca (diff) | |
Update the logic of checking valid parameters.
Summary: Add the check on NR_MIN and fix ymm/zmm register checks.
Reviewed By: dskhudia
Differential Revision: D15772144
fbshipit-source-id: 11e2c67fb3d47c5570b38ceaf9828ced0e60e65b
Diffstat (limited to 'src')
| -rw-r--r-- | src/GenerateKernelU8S8S32ACC16Avx512.cc | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/src/GenerateKernelU8S8S32ACC16Avx512.cc b/src/GenerateKernelU8S8S32ACC16Avx512.cc index 505fec1..e5687eb 100644 --- a/src/GenerateKernelU8S8S32ACC16Avx512.cc +++ b/src/GenerateKernelU8S8S32ACC16Avx512.cc @@ -201,9 +201,10 @@ CodeGenBase<uint8_t, int8_t, int32_t, int16_t>::getOrCreate<inst_set_t::avx512>( int maxMRegs = mRegBlockSize; int maxNRegs = nRegBlockSize * row_interleave / VLEN_; assert( - maxMRegs * maxNRegs <= 24 && - "MR*(NR*ROW_INTERLEAVE*8/512) \ - must be <= 24(available registers constraint)"); + (maxMRegs+1) * maxNRegs <= 28 && + "number of zmm registers for C + one row for loading B: \ + MR*(NR*ROW_INTERLEAVE*8/512) + (NR*ROW_INTERLEAVE*8/512) \ + must be <= 28(available registers constraint)"); int mRegBlocks = mc / mRegBlockSize; int mRegBlocksRem = mc % mRegBlockSize; |