From 45a9b775bf63283320315d90e4e9a6c641df6e20 Mon Sep 17 00:00:00 2001 From: James Conroy Date: Wed, 31 Oct 2018 11:47:53 +0000 Subject: IVGCVSW-2102: Fix Pooling2D CpuRef indexing bug * Fixes bug when calcuating indexes for NHWC in Pooling2D CpuRef implementation, it now uses TensorBufferArrayView. * Adds 2-Channel unit tests for Pooling2d on CpuRef, Cl and Neon. The single channel tests were not properly exercising Pooling2d using NHWC data layout. * Refactors Pooling2D NHWC tests so that the input and output data are permuted to NHWC when necessary, instead of hard coding the data in NHWC format. Change-Id: I5b9d41ed425ff283ea8c8ef6b1266ae0bc80f43b --- src/backends/test/Pooling2dTestImpl.hpp | 215 ++++++++++++++++++++++---------- 1 file changed, 146 insertions(+), 69 deletions(-) (limited to 'src/backends/test/Pooling2dTestImpl.hpp') diff --git a/src/backends/test/Pooling2dTestImpl.hpp b/src/backends/test/Pooling2dTestImpl.hpp index 90be2897e8..eea423275c 100644 --- a/src/backends/test/Pooling2dTestImpl.hpp +++ b/src/backends/test/Pooling2dTestImpl.hpp @@ -4,6 +4,7 @@ // #pragma once +#include #include #include @@ -13,6 +14,8 @@ #include #include #include +#include "Permute.hpp" +#include template LayerTestResult SimplePooling2dTestImpl(armnn::IWorkloadFactory& workloadFactory, @@ -22,9 +25,10 @@ LayerTestResult SimplePooling2dTestImpl(armnn::IWorkloadFactory& workloadF const boost::multi_array& input, const boost::multi_array& outputExpected) { - const unsigned int channelsIndex = descriptor.m_DataLayout.GetChannelsIndex(); - const unsigned int heightIndex = descriptor.m_DataLayout.GetHeightIndex(); - const unsigned int widthIndex = descriptor.m_DataLayout.GetWidthIndex(); + const armnn::DataLayoutIndexed dataLayout = descriptor.m_DataLayout; + auto heightIndex = dataLayout.GetHeightIndex(); + auto widthIndex = dataLayout.GetWidthIndex(); + auto channelsIndex = dataLayout.GetChannelsIndex(); unsigned int inputHeight = boost::numeric_cast(input.shape()[heightIndex]); unsigned int inputWidth = boost::numeric_cast(input.shape()[widthIndex]); @@ -36,23 +40,10 @@ LayerTestResult SimplePooling2dTestImpl(armnn::IWorkloadFactory& workloadF unsigned int outputChannels = boost::numeric_cast(outputExpected.shape()[channelsIndex]); unsigned int outputBatchSize = boost::numeric_cast(outputExpected.shape()[0]); - armnn::TensorShape inputTensorShape; - armnn::TensorShape outputTensorShape; - - switch (descriptor.m_DataLayout.GetDataLayout()) - { - case armnn::DataLayout::NHWC: - inputTensorShape = { inputBatchSize, inputHeight, inputWidth, inputChannels }; - outputTensorShape = { outputBatchSize, outputHeight, outputWidth, outputChannels }; - break; - case armnn::DataLayout::NCHW: - default: - inputTensorShape = { inputBatchSize, inputChannels, inputHeight, inputWidth }; - outputTensorShape = { outputBatchSize, outputChannels, outputHeight, outputWidth }; - } - - armnn::TensorInfo inputTensorInfo(inputTensorShape, armnn::GetDataType()); - armnn::TensorInfo outputTensorInfo(outputTensorShape, armnn::GetDataType()); + armnn::TensorInfo inputTensorInfo = GetTensorInfo(inputBatchSize, inputChannels, inputHeight, + inputWidth, dataLayout); + armnn::TensorInfo outputTensorInfo = GetTensorInfo(outputBatchSize, outputChannels, outputHeight, + outputWidth, dataLayout); // Set quantization parameters if the requested type is a quantized type. if(armnn::IsQuantizedType()) @@ -70,7 +61,7 @@ LayerTestResult SimplePooling2dTestImpl(armnn::IWorkloadFactory& workloadF armnn::Pooling2dQueueDescriptor queueDescriptor; queueDescriptor.m_Parameters = descriptor; - queueDescriptor.m_Parameters.m_DataLayout = descriptor.m_DataLayout; + queueDescriptor.m_Parameters.m_DataLayout = dataLayout; armnn::WorkloadInfo workloadInfo; AddInputToWorkload(queueDescriptor, workloadInfo, inputTensorInfo, inputHandle.get()); @@ -234,26 +225,20 @@ LayerTestResult SimpleMaxPooling2dSize3x3Stride2x4TestCommon(armnn::IWorkl } template -LayerTestResult SimpleAveragePooling2dTestCommon(armnn::IWorkloadFactory& workloadFactory, - const armnn::TensorShape& inputTensorShape, - const armnn::TensorShape& outputTensorShape, - armnn::DataLayout dataLayout, - float qScale = 1.0f, - int32_t qOffset = 0) +LayerTestResult SimpleMaxPooling2dTestCommon(armnn::IWorkloadFactory& workloadFactory, + const armnn::DataLayoutIndexed& dataLayout = armnn::DataLayout::NCHW, + float qScale = 1.0f, + int32_t qOffset = 0) { armnn::Pooling2dDescriptor descriptor; - descriptor.m_PoolType = armnn::PoolingAlgorithm::Average; + descriptor.m_PoolType = armnn::PoolingAlgorithm::Max; descriptor.m_PoolWidth = descriptor.m_PoolHeight = 2; descriptor.m_StrideX = descriptor.m_StrideY = 2; - descriptor.m_PadLeft = 1; - descriptor.m_PadRight = 1; - descriptor.m_PadTop = 1; - descriptor.m_PadBottom = 1; descriptor.m_PaddingMethod = armnn::PaddingMethod::Exclude; descriptor.m_DataLayout = dataLayout; - armnn::TensorInfo inputTensorInfo(inputTensorShape, armnn::GetDataType()); - armnn::TensorInfo outputTensorInfo(outputTensorShape, armnn::GetDataType()); + armnn::TensorInfo inputTensorInfo = GetTensorInfo(1, 2, 4, 4, dataLayout); + armnn::TensorInfo outputTensorInfo = GetTensorInfo(1, 2, 2, 2, dataLayout); // Set quantization parameters if the requested type is a quantized type. if(armnn::IsQuantizedType()) @@ -264,46 +249,111 @@ LayerTestResult SimpleAveragePooling2dTestCommon(armnn::IWorkloadFactory& outputTensorInfo.SetQuantizationOffset(qOffset); } - auto input = MakeTensor(inputTensorInfo, + std::vector inputData( QuantizedVector(qScale, qOffset, { - 1.0f, 2.0f, 3.0f, 4.0f, - 1.0f, 2.0f, 3.0f, 4.0f, - 1.0f, 2.0f, 3.0f, 4.0f, - 1.0f, 2.0f, 3.0f, 4.0f, + 1.0f, 2.0f, 5.0f, 6.0f, + 3.0f, 4.0f, 7.0f, 8.0f, + 9.0f, 10.0f, 13.0f, 14.0f, + 11.0f, 12.0f, 15.0f, 16.0f, + + 17.0f, 18.0f, 21.0f, 22.0f, + 19.0f, 20.0f, 23.0f, 24.0f, + 25.0f, 26.0f, 29.0f, 30.0f, + 27.0f, 28.0f, 31.0f, 32.0f, })); - auto outputExpected = MakeTensor(outputTensorInfo, + std::vector outputData( QuantizedVector(qScale, qOffset, { - 1.0f, 2.5f, 4.0f, - 1.0f, 2.5f, 4.0f, - 1.0f, 2.5f, 4.0f, + 4.0f, 8.0f, + 12.0f, 16.0f, + + 20.0f, 24.0f, + 28.0f, 32.0f, })); + const armnn::PermutationVector NCHWToNHWC = { 0, 3, 1, 2 }; + if (dataLayout.GetDataLayout() == armnn::DataLayout::NHWC) + { + std::vector tmp(inputData.size()); + armnnUtils::Permute(inputTensorInfo.GetShape(), NCHWToNHWC, inputData.data(), tmp.data()); + inputData = tmp; + + std::vector tmp1(outputData.size()); + armnnUtils::Permute(outputTensorInfo.GetShape(), NCHWToNHWC, outputData.data(), tmp1.data()); + outputData = tmp1; + } + + auto input = MakeTensor(inputTensorInfo, inputData); + + auto outputExpected = MakeTensor(outputTensorInfo, outputData); + return SimplePooling2dTestImpl(workloadFactory, descriptor, qScale, qOffset, input, outputExpected); } template -LayerTestResult SimpleAveragePooling2dTest(armnn::IWorkloadFactory& workloadFactory, - float qScale = 1.0f, - int32_t qOffset = 0) +LayerTestResult SimpleAveragePooling2dTestCommon(armnn::IWorkloadFactory& workloadFactory, + armnn::DataLayoutIndexed dataLayout = armnn::DataLayout::NCHW, + float qScale = 1.0f, + int32_t qOffset = 0) { - const armnn::TensorShape inputTensorShape { 1, 1, 4, 4 }; - const armnn::TensorShape outputTensorShape { 1, 1, 3, 3 }; + armnn::Pooling2dDescriptor descriptor; + descriptor.m_PoolType = armnn::PoolingAlgorithm::Average; + descriptor.m_PoolWidth = descriptor.m_PoolHeight = 2; + descriptor.m_StrideX = descriptor.m_StrideY = 2; + descriptor.m_PaddingMethod = armnn::PaddingMethod::Exclude; + descriptor.m_DataLayout = dataLayout; - return SimpleAveragePooling2dTestCommon(workloadFactory, inputTensorShape, outputTensorShape, - armnn::DataLayout::NCHW, qScale, qOffset); -} + armnn::TensorInfo inputTensorInfo = GetTensorInfo(1, 2, 4, 4, dataLayout); + armnn::TensorInfo outputTensorInfo = GetTensorInfo(1, 2, 2, 2, dataLayout); -template -LayerTestResult SimpleAveragePooling2dNhwcTest(armnn::IWorkloadFactory& workloadFactory, - float qScale = 1.0f, - int32_t qOffset = 0) -{ - const armnn::TensorShape inputTensorShape { 1, 4, 4, 1 }; - const armnn::TensorShape outputTensorShape { 1, 3, 3, 1 }; + // Set quantization parameters if the requested type is a quantized type. + if(armnn::IsQuantizedType()) + { + inputTensorInfo.SetQuantizationScale(qScale); + inputTensorInfo.SetQuantizationOffset(qOffset); + outputTensorInfo.SetQuantizationScale(qScale); + outputTensorInfo.SetQuantizationOffset(qOffset); + } + + std::vector inputData( + QuantizedVector(qScale, qOffset, { + 2.0f, 2.0f, 6.0f, 6.0f, + 4.0f, 4.0f, 8.0f, 8.0f, + 10.0f, 12.0f, 14.0f, 16.0f, + 10.0f, 12.0f, 16.0f, 14.0f, + + 18.0f, 20.0f, 24.0f, 22.0f, + 20.0f, 18.0f, 22.0f, 24.0f, + 26.0f, 28.0f, 0.0f, 0.0f, + 26.0f, 28.0f, 0.0f, 0.0f, + })); + + std::vector outputData( + QuantizedVector(qScale, qOffset, { + 3.0f, 7.0f, + 11.0f, 15.0f, + + 19.0f, 23.0f, + 27.0f, 0.0f, + })); + + const armnn::PermutationVector NCHWToNHWC = { 0, 3, 1, 2 }; + if (dataLayout.GetDataLayout() == armnn::DataLayout::NHWC) + { + std::vector tmp(inputData.size()); + armnnUtils::Permute(inputTensorInfo.GetShape(), NCHWToNHWC, inputData.data(), tmp.data()); + inputData = tmp; - return SimpleAveragePooling2dTestCommon(workloadFactory, inputTensorShape, outputTensorShape, - armnn::DataLayout::NHWC, qScale, qOffset); + std::vector tmp1(outputData.size()); + armnnUtils::Permute(outputTensorInfo.GetShape(), NCHWToNHWC, outputData.data(), tmp1.data()); + outputData = tmp1; + } + + auto input = MakeTensor(inputTensorInfo, inputData); + + auto outputExpected = MakeTensor(outputTensorInfo, outputData); + + return SimplePooling2dTestImpl(workloadFactory, descriptor, qScale, qOffset, input, outputExpected); } template @@ -356,6 +406,7 @@ LayerTestResult LargeTensorsAveragePooling2dTestCommon(armnn::IWorkloadFac template LayerTestResult SimpleL2Pooling2dTestCommon(armnn::IWorkloadFactory& workloadFactory, + armnn::DataLayoutIndexed dataLayout = armnn::DataLayout::NCHW, float qScale = 1.0f, int32_t qOffset = 0) { @@ -364,23 +415,49 @@ LayerTestResult SimpleL2Pooling2dTestCommon(armnn::IWorkloadFactory& workl descriptor.m_PoolWidth = descriptor.m_PoolHeight = 2; descriptor.m_StrideX = descriptor.m_StrideY = 2; descriptor.m_PaddingMethod = armnn::PaddingMethod::Exclude; + descriptor.m_DataLayout = dataLayout; - armnn::TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, armnn::GetDataType()); - auto input = MakeTensor(inputTensorInfo, + armnn::TensorInfo inputTensorInfo = GetTensorInfo(1, 2, 4, 4, dataLayout); + armnn::TensorInfo outputTensorInfo = GetTensorInfo(1, 2, 2, 2, dataLayout); + + std::vector inputData( QuantizedVector(qScale, qOffset, { - 1.0f, 7.0f, 1.0f, 7.0f, - 1.0f, 7.0f, 1.0f, 7.0f, - 1.0f, 7.0f, 1.0f, 7.0f, - 1.0f, 7.0f, 1.0f, 7.0f, + 1.0f, 7.0f, 5.0f, 5.0f, + 1.0f, 7.0f, 5.0f, 5.0f, + 3.0f, 3.0f, 1.0f, 1.0f, + 3.0f, 3.0f, 1.0f, 1.0f, + + 1.0f, 7.0f, 0.0f, 0.0f, + 1.0f, 7.0f, 2.0f, 0.0f, + 0.0f, 2.0f, 1.0f, 1.0f, + 0.0f, 0.0f, 1.0f, 1.0f, })); - armnn::TensorInfo outputTensorInfo({ 1, 1, 2, 2 }, armnn::GetDataType()); - auto outputExpected = MakeTensor(outputTensorInfo, + std::vector outputData( QuantizedVector(qScale, qOffset, { 5.0f, 5.0f, - 5.0f, 5.0f, + 3.0f, 1.0f, + + 5.0f, 1.0f, + 1.0f, 1.0f, })); + const armnn::PermutationVector NCHWToNHWC = { 0, 3, 1, 2 }; + if (dataLayout.GetDataLayout() == armnn::DataLayout::NHWC) + { + std::vector tmp(inputData.size()); + armnnUtils::Permute(inputTensorInfo.GetShape(), NCHWToNHWC, inputData.data(), tmp.data()); + inputData = tmp; + + std::vector tmp1(outputData.size()); + armnnUtils::Permute(outputTensorInfo.GetShape(), NCHWToNHWC, outputData.data(), tmp1.data()); + outputData = tmp1; + } + + auto input = MakeTensor(inputTensorInfo, inputData); + + auto outputExpected = MakeTensor(outputTensorInfo, outputData); + return SimplePooling2dTestImpl(workloadFactory, descriptor, qScale, qOffset, input, outputExpected); } -- cgit v1.2.1