diff options
| author | zhenglin <zhenglin.li@arm.com> | 2017-12-05 11:30:51 +0800 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:42:17 +0000 |
| commit | 923241eb998ad031f4cce7b12d8c24a0b6c80be8 (patch) | |
| tree | 9cec664884d324c6a5cd8ebc5816071380417de2 /src/core/GLES_COMPUTE/kernels | |
| parent | 0bc5a258903bbbf8623b6e1c4253d958551293ed (diff) | |
| download | ComputeLibrary-923241eb998ad031f4cce7b12d8c24a0b6c80be8.tar.gz | |
APPBROWSER-314: Performance optimazation for BatchNormalizationLayer
Change-Id: Ie3ad9abb64e90720609bb6e67662eaf9dd4f3689
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111826
Reviewed-by: Joel Liang <joel.liang@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
(cherry picked from commit 02c1fa663926cc4fcd1995d4d18d7528e0c85d94)
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111834
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'src/core/GLES_COMPUTE/kernels')
| -rw-r--r-- | src/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.cpp | 14 |
1 files changed, 9 insertions, 5 deletions
diff --git a/src/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.cpp index 982143f0b2..dee2a5579b 100644 --- a/src/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.cpp +++ b/src/core/GLES_COMPUTE/kernels/GCBatchNormalizationLayerKernel.cpp @@ -64,7 +64,11 @@ void GCBatchNormalizationLayerKernel::configure(const IGCTensor *input, IGCTenso _gamma = gamma; _epsilon = epsilon; - const unsigned int num_elems_processed_per_iteration = 4 / input->info()->element_size(); + unsigned int num_elems_processed_per_iteration = 1; + if(input->info()->data_type() == DataType::F16) + { + num_elems_processed_per_iteration = 4; + } // Set build options std::set<std::string> build_opts; @@ -83,10 +87,10 @@ void GCBatchNormalizationLayerKernel::configure(const IGCTensor *input, IGCTenso AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration); AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration); - AccessWindowStatic mean_access(mean->info(), 0, 0, mean->info()->dimension(0) + 1, mean->info()->dimension(1)); - AccessWindowStatic var_access(var->info(), 0, 0, var->info()->dimension(0) + 1, var->info()->dimension(1)); - AccessWindowStatic beta_access(beta->info(), 0, 0, beta->info()->dimension(0) + 1, beta->info()->dimension(1)); - AccessWindowStatic gamma_access(gamma->info(), 0, 0, gamma->info()->dimension(0) + 1, gamma->info()->dimension(1)); + AccessWindowStatic mean_access(mean->info(), 0, 0, mean->info()->dimension(0) + 3, mean->info()->dimension(1)); + AccessWindowStatic var_access(var->info(), 0, 0, var->info()->dimension(0) + 3, var->info()->dimension(1)); + AccessWindowStatic beta_access(beta->info(), 0, 0, beta->info()->dimension(0) + 3, beta->info()->dimension(1)); + AccessWindowStatic gamma_access(gamma->info(), 0, 0, gamma->info()->dimension(0) + 3, gamma->info()->dimension(1)); update_window_and_padding(win, input_access, output_access, mean_access, var_access, beta_access, gamma_access); output_access.set_valid_region(win, input->info()->valid_region()); |