diff options
| author | Jongsoo Park <jongsoo@fb.com> | 2018-11-29 06:41:59 +0300 |
|---|---|---|
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2018-11-29 06:44:08 +0300 |
| commit | 027de07a11a0460fd1daffb026d50dba0e56eb79 (patch) | |
| tree | 5e0f497059d6a22b18de8508c2031ffdbc9f52d3 /src/ExecuteKernelU8S8.cc | |
| parent | 90535d3da35f9d3da6a8dbd62da0c68d01696924 (diff) | |
sparse convolution output processing (#27)
Summary:
Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/27
DoSpmdmOnInpBuffer can't be used together with PackAWithIm2Col because DoSpmdmOnInpBuffer expects im2col'ed A matrix. This diff implements DoSConvOnInpBuffer that does sparse convolution directly on A input without im2col. The performance is well optimized and need to see if this implementation is good enough to get good resnet50 performance.
Reviewed By: dskhudia
Differential Revision: D13192336
fbshipit-source-id: 2076555ba9749e111afbaec408a2bfa0f55bd5bc
Diffstat (limited to 'src/ExecuteKernelU8S8.cc')
| -rw-r--r-- | src/ExecuteKernelU8S8.cc | 18 |
1 files changed, 18 insertions, 0 deletions
diff --git a/src/ExecuteKernelU8S8.cc b/src/ExecuteKernelU8S8.cc index f1ec882..152d7f1 100644 --- a/src/ExecuteKernelU8S8.cc +++ b/src/ExecuteKernelU8S8.cc @@ -381,6 +381,24 @@ INSTANTIATE_Q_GRANS(true); #undef INSTANTIATE_Q_GRANS #undef INSTANTIATE_BASE +#define INSTANTIATE_BASE(RELU, Q_GRAN) \ + template class ExecuteKernel< \ + PackAWithIm2Col<uint8_t, int16_t>, \ + PackBMatrix<int8_t, int16_t>, \ + uint8_t, \ + DoSConvOnInpBuffer<uint8_t, int32_t, ReQuantizeOutput<RELU, Q_GRAN>>>; + +#define INSTANTIATE_Q_GRANS(RELU) \ + INSTANTIATE_BASE(RELU, QuantizationGranularity::TENSOR); \ + INSTANTIATE_BASE(RELU, QuantizationGranularity::GROUP); \ + INSTANTIATE_BASE(RELU, QuantizationGranularity::OUT_CHANNEL); + +INSTANTIATE_Q_GRANS(false); +INSTANTIATE_Q_GRANS(true); + +#undef INSTANTIATE_Q_GRANS +#undef INSTANTIATE_BASE + template class ExecuteKernel< PackAWithRowOffset<uint8_t, int16_t>, PackBMatrix<int8_t, int16_t>, |