From e52a3000d2c13bc1b66ca66b3d12b6b836982394 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Wed, 11 Apr 2018 15:59:10 +0100 Subject: COMPMID-1026 - Add support for 4x4 output tile in CLWinogradConvolutionLayer The performance achieved can be found at the following confluence page: https://confluence.arm.com/display/MLENG/GEMM-based+convolution+vs+Winograd-based+convolution+on+OpenCL Change-Id: I4b690cfdd4eb4ff0cd17b14fdd49ccaa1d1dc85c Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/127729 Tested-by: Jenkins Reviewed-by: Georgios Pinitas --- tests/validation/CL/ConvolutionLayer.cpp | 37 +++++++++++++------------------- 1 file changed, 15 insertions(+), 22 deletions(-) (limited to 'tests/validation/CL/ConvolutionLayer.cpp') diff --git a/tests/validation/CL/ConvolutionLayer.cpp b/tests/validation/CL/ConvolutionLayer.cpp index b50bb94bbb..52ad417cd0 100644 --- a/tests/validation/CL/ConvolutionLayer.cpp +++ b/tests/validation/CL/ConvolutionLayer.cpp @@ -72,25 +72,19 @@ const auto ActivationFunctionsDataset = framework::dataset::make("ActivationInfo TEST_SUITE(CL) TEST_SUITE(ConvolutionLayer) -DATA_TEST_CASE(ValidateConvolutionMethod, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip( - framework::dataset::make("InputInfo", { TensorInfo(TensorShape(17U, 31U, 2U), 1, DataType::F32, 0), - TensorInfo(TensorShape(17U, 31U, 2U), 1, DataType::F32, 0), - TensorInfo(TensorShape(23U, 27U, 5U, 4U), 1, DataType::F32, 0), - TensorInfo(TensorShape(3U, 3U, 2U, 1U), 1, DataType::F32, 0), - TensorInfo(TensorShape(33U, 27U, 7U, 4U), 1, DataType::F32, 0) - }), - framework::dataset::make("WeightsInfo", { TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 0), - TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 0), - TensorInfo(TensorShape(3U, 3U, 5U, 21U), 1, DataType::F32, 0), - TensorInfo(TensorShape(3U, 3U, 5U, 21U), 1, DataType::F32, 0), - TensorInfo(TensorShape(5U, 5U, 7U, 16U), 1, DataType::F16, 0) - })), - framework::dataset::make("BiasesInfo", { TensorInfo(TensorShape(19U), 1, DataType::F32, 0), - TensorInfo(TensorShape(19U), 1, DataType::F32, 0), - TensorInfo(TensorShape(21U), 1, DataType::F32, 0), - TensorInfo(TensorShape(21U), 1, DataType::F32, 0), - TensorInfo(TensorShape(16U), 1, DataType::F32, 0) - })), +DATA_TEST_CASE(ValidateConvolutionMethod, framework::DatasetMode::ALL, zip(zip(zip(zip(zip( + framework::dataset::make("InputInfo", { TensorInfo(TensorShape(17U, 31U, 2U), 1, DataType::F32, 0), + TensorInfo(TensorShape(17U, 31U, 2U), 1, DataType::F32, 0), + TensorInfo(TensorShape(23U, 27U, 5U, 4U), 1, DataType::F32, 0), + TensorInfo(TensorShape(3U, 3U, 2U, 1U), 1, DataType::F32, 0), + TensorInfo(TensorShape(33U, 27U, 7U, 4U), 1, DataType::F32, 0) + }), + framework::dataset::make("WeightsInfo", { TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 0), + TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 0), + TensorInfo(TensorShape(3U, 3U, 5U, 21U), 1, DataType::F32, 0), + TensorInfo(TensorShape(3U, 3U, 5U, 21U), 1, DataType::F32, 0), + TensorInfo(TensorShape(5U, 5U, 7U, 16U), 1, DataType::F16, 0) + })), framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(15U, 15U, 19U), 1, DataType::F32, 0), TensorInfo(TensorShape(15U, 15U, 19U), 1, DataType::F32, 0), TensorInfo(TensorShape(21U, 25U, 21U, 4U), 1, DataType::F32, 0), @@ -110,12 +104,11 @@ DATA_TEST_CASE(ValidateConvolutionMethod, framework::DatasetMode::ALL, zip(zip(z GPUTarget::BIFROST })), - framework::dataset::make("Expected", { ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM })), - input_info, weights_info, biases_info, output_info, conv_info, gpu_target, expected) + framework::dataset::make("Expected", { ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::WINOGRAD, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM })), + input_info, weights_info, output_info, conv_info, gpu_target, expected) { ConvolutionMethod is_valid = CLConvolutionLayer::get_convolution_method(&input_info.clone()->set_is_resizable(false), &weights_info.clone()->set_is_resizable(false), - &biases_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), conv_info, WeightsInfo(), ActivationLayerInfo(), gpu_target); ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS); } -- cgit v1.2.1