diff options
| author | Jianyu Huang <jianyuhuang@fb.com> | 2019-08-06 21:55:17 +0300 |
|---|---|---|
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2019-08-06 21:59:00 +0300 |
| commit | cf34b9a26b609109b18d6498f0608faddb7a911b (patch) | |
| tree | 1ceaddaf942edb9debcafad7491b750fc3a5f066 /src/PackWeightMatrixForGConv.cc | |
| parent | d8b3323668fdd15dc70e9cb43ab16e96f4846eeb (diff) | |
Back out "[fbgemm] Integrate VNNI into FBGEMM master branch"
Summary:
Original commit changeset: fcaa13cc3159
ASMJIT requires the CMake version to be 3.8
However, FBGEMM and PyTorch only need the CMake version to be 3.5+.
This caused the build failure in FBGEMM:
https://circleci.com/gh/pytorch/FBGEMM/122#build-timing/containers/0
Reviewed By: dskhudia
Differential Revision: D16670547
fbshipit-source-id: 506714c3db1cb82cf98895f58f82f235128f5285
Diffstat (limited to 'src/PackWeightMatrixForGConv.cc')
| -rw-r--r-- | src/PackWeightMatrixForGConv.cc | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/src/PackWeightMatrixForGConv.cc b/src/PackWeightMatrixForGConv.cc index f6ad59e..ba6adf3 100644 --- a/src/PackWeightMatrixForGConv.cc +++ b/src/PackWeightMatrixForGConv.cc @@ -106,7 +106,7 @@ inline int PackWeightMatrixForGConv<T, accT, SPATIAL_DIM>::packed_index_( * on 2 groups at a time and full SIMD width can be efficiently utilized even * while working on 1 group at a time. * In this case, the layout is G (C/4) R S K 4 - */ +*/ template <typename T, typename accT, int SPATIAL_DIM> void PackWeightMatrixForGConv<T, accT, SPATIAL_DIM>::pack_unpack_( @@ -148,9 +148,9 @@ void PackWeightMatrixForGConv<T, accT, SPATIAL_DIM>::pack_unpack_( if (ispack) { transposeConvWeights(conv_param_, src, dst); } else { - // TODO: Wrap this as a inverseTransposeConvWeights()? - // For unpack & transposed, call transposeConvWeights() - // G (R S C/G) K/G => G K/G (R S C/G) + // TODO: Wrap this as a inverseTransposeConvWeights()? + // For unpack & transposed, call transposeConvWeights() + // G (R S C/G) K/G => G K/G (R S C/G) for (int r = 0; r < R; ++r) { for (int s = 0; s < S; ++s) { for (int k = 0; k < OC_per_G; ++k) { |