diff options
| author | Joel Liang <joel.liang@arm.com> | 2017-11-17 11:34:19 +0800 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:42:17 +0000 |
| commit | 07c37f9954555ae3523c85f16e46cf94e9a9e290 (patch) | |
| tree | 30b4fa60148b7a9a44e26b9488f3dcf482e11faa /src/runtime/GLES_COMPUTE | |
| parent | d9f8071de655fd23529588d312dd5624b2d43315 (diff) | |
| download | ComputeLibrary-07c37f9954555ae3523c85f16e46cf94e9a9e290.tar.gz | |
APPBROWSER-313: Performance improvement for softmax layer
Process 8 elements at one time for better performance
Change-Id: I90d31e5d0834c5096fdb82f174482ade762b63d2
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111840
Reviewed-by: Stephen Li <stephen.li@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'src/runtime/GLES_COMPUTE')
| -rw-r--r-- | src/runtime/GLES_COMPUTE/functions/GCSoftmaxLayer.cpp | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/src/runtime/GLES_COMPUTE/functions/GCSoftmaxLayer.cpp b/src/runtime/GLES_COMPUTE/functions/GCSoftmaxLayer.cpp index d7d47d2802..1db927c8ff 100644 --- a/src/runtime/GLES_COMPUTE/functions/GCSoftmaxLayer.cpp +++ b/src/runtime/GLES_COMPUTE/functions/GCSoftmaxLayer.cpp @@ -61,6 +61,8 @@ void GCSoftmaxLayer::configure(const IGCTensor *input, IGCTensor *output) void GCSoftmaxLayer::run() { GCScheduler::get().enqueue(_max_kernel, false); + GCScheduler::get().sync(); GCScheduler::get().enqueue(_shift_exp_sum_kernel, false); + GCScheduler::get().sync(); GCScheduler::get().enqueue(_norm_kernel); } |