diff options
author | Jongsoo Park <jongsoo@fb.com> | 2019-02-13 05:59:00 +0300 |
---|---|---|
committer | Facebook Github Bot <facebook-github-bot@users.noreply.github.com> | 2019-02-13 06:08:54 +0300 |
commit | 64269506c9bab424cfd808d4fa398574ac38cd5f (patch) | |
tree | c397705af6b54d1821643f7189ce4d1c9409ef07 /include/fbgemm/Fbgemm.h | |
parent | c4e50e0ed5db7c9aadb6df47faa4227f5286293c (diff) |
optimize gconv for b symmetric quantization (#70)
Summary:
Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/70
Skip row offset computation if B_zero_point == 0 .
Reviewed By: jianyuh
Differential Revision: D14020675
fbshipit-source-id: 88a6e225671762c67afefc15538b79f879d125a6
Diffstat (limited to 'include/fbgemm/Fbgemm.h')
-rw-r--r-- | include/fbgemm/Fbgemm.h | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/include/fbgemm/Fbgemm.h b/include/fbgemm/Fbgemm.h index a3c5ab7..2ac10f7 100644 --- a/include/fbgemm/Fbgemm.h +++ b/include/fbgemm/Fbgemm.h @@ -1177,6 +1177,7 @@ FBGEMM_API void fbgemmPacked( /** * @brief Perform small-channels-per-group groupwise convolution * + * @params rowOffsetBuf nullptr if B uses symmetric quantization */ template < @@ -1196,6 +1197,9 @@ FBGEMM_API void fbgemmGroupwiseConv( int thread_id, int num_threads); +/** + * @params rowOffsetBuf nullptr if B uses symmetric quantization + */ template < typename packed_W, typename outType, |