diff options
author | Jongsoo Park <jongsoo@fb.com> | 2018-11-16 05:21:15 +0300 |
---|---|---|
committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2018-11-16 05:24:41 +0300 |
commit | 8392eca198742b949529e18619a7ec9a25f4b399 (patch) | |
tree | da6f827a09b6a90d6294ca566f073ff387225a3b /bench | |
parent | aa5b56a0de8b8f19047a9d797bf2d70cca4bc1f0 (diff) |
grouped (batched) gemm (#7)
Summary:
Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/7
This diff allows groups > 1 . Will have a separate diff for im2col + gemm fusion and conv with group > 1 .
Reviewed By: jianyuh
Differential Revision: D13039210
fbshipit-source-id: f7b3b0dbdb67fc6bc865de88292f034b252d029d
Diffstat (limited to 'bench')
-rw-r--r-- | bench/Im2ColFusedRequantizeAcc16Benchmark.cc | 6 | ||||
-rw-r--r-- | bench/Im2ColFusedRequantizeAcc32Benchmark.cc | 6 |
2 files changed, 10 insertions, 2 deletions
diff --git a/bench/Im2ColFusedRequantizeAcc16Benchmark.cc b/bench/Im2ColFusedRequantizeAcc16Benchmark.cc index 48b744a..8827b4c 100644 --- a/bench/Im2ColFusedRequantizeAcc16Benchmark.cc +++ b/bench/Im2ColFusedRequantizeAcc16Benchmark.cc @@ -189,7 +189,11 @@ void performance_test() { PackAWithIm2Col<uint8_t, int16_t>::rowOffsetBufferSize()); PackAWithIm2Col<uint8_t, int16_t> packA( - conv_p, Aint8.data(), nullptr, Aint8_zero_point, row_offset_buf.data()); + conv_p, + Aint8.data(), + nullptr, + Aint8_zero_point, + row_offset_buf.data()); PackBMatrix<int8_t, int16_t> packedB( matrix_op_t::NoTranspose, KDim, NDim, Bint8.data(), NDim); diff --git a/bench/Im2ColFusedRequantizeAcc32Benchmark.cc b/bench/Im2ColFusedRequantizeAcc32Benchmark.cc index 9201e52..b87f7d7 100644 --- a/bench/Im2ColFusedRequantizeAcc32Benchmark.cc +++ b/bench/Im2ColFusedRequantizeAcc32Benchmark.cc @@ -191,7 +191,11 @@ void performance_test() { PackAWithIm2Col<uint8_t, int32_t>::rowOffsetBufferSize()); PackAWithIm2Col<uint8_t, int32_t> packA( - conv_p, Aint8.data(), nullptr, Aint8_zero_point, row_offset_buf.data()); + conv_p, + Aint8.data(), + nullptr, + Aint8_zero_point, + row_offset_buf.data()); PackBMatrix<int8_t, int32_t> packedB( matrix_op_t::NoTranspose, KDim, NDim, Bint8.data(), NDim); |