From 1257131193fdb9b6940055a41691320e37a208b5 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Thu, 25 Aug 2022 12:25:44 +0100 Subject: 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 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8147 Tested-by: Arm Jenkins Reviewed-by: Gunes Bayir Benchmark: Arm Jenkins --- src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.cpp') 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(); -- cgit v1.2.1