diff options
| author | Jongsoo Park <jongsoo@fb.com> | 2019-02-13 01:35:32 +0300 |
|---|---|---|
| committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2019-02-13 01:48:03 +0300 |
| commit | 86eeae2f917b92126af5cb1d37336f7d503292ee (patch) | |
| tree | cb7af7e0fc924fb27d673aa296c6dd0af3d60e1e /bench | |
| parent | df7b1c1237c2f4274294ad9136861f30a7234c14 (diff) | |
group conv optimized for 16 channels per group (#68)
Summary:
Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/68
Continuing optimizations for group convolution. Even though op-level speedup for 16 channels per group is lower compared to 4 or 8-channel cases, we have a nice overall speedup in resnext101-32x4d because it has many Conv operators with 16 channels per group.
Reviewed By: protonu
Differential Revision: D13949873
fbshipit-source-id: 1dff4b1acfdabe23616e7df365daf2b7f6e8aea9
Diffstat (limited to 'bench')
| -rw-r--r-- | bench/GroupwiseConvRequantizeBenchmark.cc | 13 |
1 files changed, 9 insertions, 4 deletions
diff --git a/bench/GroupwiseConvRequantizeBenchmark.cc b/bench/GroupwiseConvRequantizeBenchmark.cc index 158ca4f..4c93f23 100644 --- a/bench/GroupwiseConvRequantizeBenchmark.cc +++ b/bench/GroupwiseConvRequantizeBenchmark.cc @@ -59,10 +59,15 @@ void performance_test() { // conv_param_t<>(2, 128, 128, {56, 48}, 32, {3, 3}, {1, 1}, {1, 1, 1, // 1}), - conv_param_t<>(1, 256, 256, {56, 48}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), - conv_param_t<>(1, 256, 256, {48, 56}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), - conv_param_t<>(1, 256, 256, {56, 56}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), - conv_param_t<>(2, 256, 256, {56, 56}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(1, 256, 256, {28, 24}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(1, 256, 256, {24, 28}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(1, 256, 256, {28, 28}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(2, 256, 256, {28, 28}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + + conv_param_t<>(1, 512, 512, {14, 12}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(1, 512, 512, {12, 14}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(1, 512, 512, {14, 14}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), + conv_param_t<>(2, 512, 512, {14, 14}, 32, {3, 3}, {1, 1}, {1, 1, 1, 1}), }; bool flush = true; |