COMPMID-1096 - Add fast_math flag to CLConvolutionLayer
COMPMID-1103 - CLWinogradConvolutionLayer mismatches
Change-Id: Iceaa9482a1790ec39d2720c220261aaea8043978
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/129398
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
diff --git a/tests/validation/CL/ConvolutionLayer.cpp b/tests/validation/CL/ConvolutionLayer.cpp
index 8685e5b..a2b55a8 100644
--- a/tests/validation/CL/ConvolutionLayer.cpp
+++ b/tests/validation/CL/ConvolutionLayer.cpp
@@ -73,44 +73,72 @@
TEST_SUITE(CL)
TEST_SUITE(ConvolutionLayer)
-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),
- TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
- TensorInfo(TensorShape(11U, 12U, 16U, 4U), 1, DataType::F32, 0)
- })),
- framework::dataset::make("ConvInfo", { PadStrideInfo(1, 2, 1, 1),
- PadStrideInfo(1, 2, 1, 1),
- PadStrideInfo(1, 1, 0, 0),
- PadStrideInfo(2, 1, 0, 0),
- PadStrideInfo(3, 2, 1, 0)
- })),
- framework::dataset::make("GpuTarget", { GPUTarget::BIFROST,
- GPUTarget::MIDGARD,
- GPUTarget::G71,
- GPUTarget::MIDGARD,
- GPUTarget::BIFROST
- })),
-
- 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)
+DATA_TEST_CASE(ValidateConvolutionMethod, framework::DatasetMode::ALL, zip(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),
+ TensorInfo(TensorShape(17U, 31U, 2U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(17U, 31U, 2U), 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),
+ TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(5U, 5U, 2U, 19U), 1, DataType::F32, 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),
+ TensorInfo(TensorShape(11U, 25U, 21U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(11U, 12U, 16U, 4U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(17U, 31U, 19U), 1, DataType::F32, 0),
+ TensorInfo(TensorShape(17U, 31U, 19U), 1, DataType::F32, 0)
+ })),
+ framework::dataset::make("ConvInfo", { PadStrideInfo(1, 2, 1, 1),
+ PadStrideInfo(1, 2, 1, 1),
+ PadStrideInfo(1, 1, 0, 0),
+ PadStrideInfo(2, 1, 0, 0),
+ PadStrideInfo(3, 2, 1, 0),
+ PadStrideInfo(1, 1, 2, 2),
+ PadStrideInfo(1, 1, 2, 2)
+ })),
+ framework::dataset::make("GpuTarget", { GPUTarget::BIFROST,
+ GPUTarget::MIDGARD,
+ GPUTarget::G71,
+ GPUTarget::MIDGARD,
+ GPUTarget::BIFROST,
+ GPUTarget::BIFROST,
+ GPUTarget::BIFROST
+ })),
+ framework::dataset::make("Dilation",
{
- ConvolutionMethod is_valid = CLConvolutionLayer::get_convolution_method(&input_info.clone()->set_is_resizable(false),
- &weights_info.clone()->set_is_resizable(false),
- &output_info.clone()->set_is_resizable(false), conv_info, WeightsInfo(), ActivationLayerInfo(), gpu_target);
+ Size2D(1U, 1U),
+ Size2D(1U, 1U),
+ Size2D(1U, 1U),
+ Size2D(1U, 1U),
+ Size2D(1U, 1U),
+ Size2D(1U, 1U),
+ Size2D(2U, 1U),
+})),
+framework::dataset::make("EnableFastMath", { false, false, false, false, false, true, true })),
+framework::dataset::make("Expected",
+{
+ ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::WINOGRAD, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::WINOGRAD, ConvolutionMethod::GEMM,
+})),
+input_info, weights_info, output_info, conv_info, gpu_target, dilation, enable_fast_math, expected)
+{
+ ConvolutionMethod is_valid = CLConvolutionLayer::get_convolution_method(&input_info.clone()->set_is_resizable(true),
+ &weights_info.clone()->set_is_resizable(true),
+ &output_info.clone()->set_is_resizable(true), conv_info,
+ WeightsInfo(),
+ ActivationLayerInfo(),
+ gpu_target,
+ dilation,
+ enable_fast_math);
ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
}
TEST_SUITE_END()
diff --git a/tests/validation/CL/DilatedConvolutionLayer.cpp b/tests/validation/CL/DilatedConvolutionLayer.cpp
index e6a765b..9ee002c 100644
--- a/tests/validation/CL/DilatedConvolutionLayer.cpp
+++ b/tests/validation/CL/DilatedConvolutionLayer.cpp
@@ -104,9 +104,9 @@
framework::dataset::make("Expected", { ConvolutionMethod::GEMM, ConvolutionMethod::GEMM, ConvolutionMethod::WINOGRAD, ConvolutionMethod::GEMM, ConvolutionMethod::GEMM })),
input_info, weights_info, output_info, conv_info, gpu_target, dilation, expected)
{
- ConvolutionMethod is_valid = CLConvolutionLayer::get_convolution_method(&input_info.clone()->set_is_resizable(false),
- &weights_info.clone()->set_is_resizable(false),
- &output_info.clone()->set_is_resizable(false), conv_info, WeightsInfo(), ActivationLayerInfo(), gpu_target, dilation);
+ ConvolutionMethod is_valid = CLConvolutionLayer::get_convolution_method(&input_info.clone()->set_is_resizable(true),
+ &weights_info.clone()->set_is_resizable(true),
+ &output_info.clone()->set_is_resizable(true), conv_info, WeightsInfo(), ActivationLayerInfo(), gpu_target, dilation);
ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
}
TEST_SUITE_END()
diff --git a/tests/validation/CL/Winograd.cpp b/tests/validation/CL/Winograd.cpp
index 30d8d75..d892c9f 100644
--- a/tests/validation/CL/Winograd.cpp
+++ b/tests/validation/CL/Winograd.cpp
@@ -51,7 +51,8 @@
{
namespace
{
-constexpr AbsoluteTolerance<float> tolerance_f32(0.001f);
+constexpr AbsoluteTolerance<float> tolerance_f32(0.0001f);
+constexpr AbsoluteTolerance<float> tolerance_fast_math_f32(0.1f);
} // namespace
using namespace arm_compute::misc::shape_calculator;
@@ -379,6 +380,27 @@
// Validate output
validate(CLAccessor(_target), _reference, tolerance_f32);
}
+TEST_SUITE(EnableFastMath)
+using CLWinogradConvolutionLayerFastMathFixture = WinogradConvolutionLayerFastMathValidationFixture<CLTensor, CLAccessor, CLWinogradConvolutionLayer, float>;
+FIXTURE_DATA_TEST_CASE(RunSmall, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::PRECOMMIT,
+ combine(combine(framework::dataset::concat(datasets::SmallWinogradConvolutionLayer3x3Dataset(), datasets::SmallWinogradConvolutionLayer5x5Dataset()),
+ framework::dataset::make("DataType", { DataType::F32 })),
+ framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })))
+{
+ // Validate output
+ validate(CLAccessor(_target), _reference, tolerance_fast_math_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge, CLWinogradConvolutionLayerFastMathFixture, framework::DatasetMode::NIGHTLY,
+ combine(combine(framework::dataset::concat(datasets::LargeWinogradConvolutionLayer3x3Dataset(), datasets::LargeWinogradConvolutionLayer5x5Dataset()),
+ framework::dataset::make("DataType", { DataType::F32 })),
+ framework::dataset::make("ActivationLayerInfo", { ActivationLayerInfo() })))
+{
+ // Validate output
+ validate(CLAccessor(_target), _reference, tolerance_fast_math_f32);
+}
+
+TEST_SUITE_END() // EnableFastMath
TEST_SUITE_END() // ConvolutionLayer
TEST_SUITE_END() // Winograd