diff options
author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2017-08-16 18:38:32 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
commit | 1246b63ca04cb067f26ae860688647224d6ba24e (patch) | |
tree | a805ff1faa1de1b06f3569926ec2f09b63ecdb5f /src/runtime/CL/functions/CLConvolutionLayer.cpp | |
parent | f583fb74ecaad9d57672e1422d68566a78bb503e (diff) | |
download | ComputeLibrary-1246b63ca04cb067f26ae860688647224d6ba24e.tar.gz |
COMPMID-477 - Optimized Direct Convolution 3x3 and 5x5 (f32) for Bifrost.
Each work-item computes 4x3 output elements in case of 3x3 convolution and 4x2 in case of 5x5 convolution
Change-Id: I6ebbaff8b7e971c1f90d5845c0b58d2a40f39df5
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/84345
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Diffstat (limited to 'src/runtime/CL/functions/CLConvolutionLayer.cpp')
-rw-r--r-- | src/runtime/CL/functions/CLConvolutionLayer.cpp | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/src/runtime/CL/functions/CLConvolutionLayer.cpp b/src/runtime/CL/functions/CLConvolutionLayer.cpp index ff94e9d7a2..b1b83985d0 100644 --- a/src/runtime/CL/functions/CLConvolutionLayer.cpp +++ b/src/runtime/CL/functions/CLConvolutionLayer.cpp @@ -113,6 +113,9 @@ void CLConvolutionLayer::configure(const ICLTensor *input, const ICLTensor *weig const DataType dt = input->info()->data_type(); const int fixed_point_position = input->info()->fixed_point_position(); + // Set the GPU target for matrix multiply + _mm_kernel.set_target(CLScheduler::get().target()); + _has_bias = (biases != nullptr); _are_weights_reshaped = weights_info.are_reshaped(); |