diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2022-08-25 12:25:44 +0100 |
|---|---|---|
| committer | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2022-09-02 08:33:21 +0000 |
| commit | 1257131193fdb9b6940055a41691320e37a208b5 (patch) | |
| tree | ea48d445c6be8e258d742a67fe41a65472d51963 /src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | |
| parent | 16789a14afc27c1a77c8ca1e3d04b79cda6c833b (diff) | |
| download | ComputeLibrary-1257131193fdb9b6940055a41691320e37a208b5.tar.gz | |
Enable Winograd-based conv2d when IFM>=8 on Gpu
From an internal performance evaluation, it seems that Winograd-based
Conv2D offers better performance than alternative methods such as direct
convolution and gemm-based conv already from IFM=8. Before the condition
was for IFM>=16
Resolves COMPMID-5532
Change-Id: I9ff04835d6fd07f5f0abeec9645c9d9cc913b6b7
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8147
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp index d1f0338739..732d768308 100644 --- a/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp +++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp @@ -211,7 +211,7 @@ void CLDepthwiseConvolutionLayerNativeKernel::configure(const CLCompileContext & arm_compute::opencl::kernels::gemm::update_padding_for_cl_image(weights->info()); } - // Conditions of -cl-fast-relaxed-math causing accuracy issues can be traced from COMPMID-5324 + // Conditions of -cl-fast-relaxed-math causing accuracy issues can be traced from COMPMID-5324 const GPUTarget gpu_target = get_target(); const auto act_function = conv_info.act_info.activation(); const auto dst_data_type = _output->info()->data_type(); |