experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Use a new bug for the TODO

Browse Source 
PiperOrigin-RevId: 221238785
This commit is contained in:
Adrian Kuegel 2018-11-13 02:58:11 -08:00 committed by TensorFlower Gardener
parent a40c6e323b
commit 6a16d5ee56
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/compiler/xla/service/gpu/cudnn_conv_rewriter.cc
View File

@ -254,7 +254,7 @@ MatchBackwardInput(HloInstruction* conv) {
const auto no_match_result =
std::make_tuple(false, Window(), ConvolutionDimensionNumbers(), nullptr);
// TODO(b/31709653): Theoretically cuDNN supports grouped convolutions also
// TODO(b/119479517): Theoretically cuDNN supports grouped convolutions also
// for the backward input convolution, but at least for now with version 7.1.4
// it is slower. This needs to be re-evaluated for future cuDNN versions.
// Note that we already have the necessary code down below, the only thing to