diff options
author | Cathal Corbett <cathal.corbett@arm.com> | 2022-12-07 11:50:50 +0000 |
---|---|---|
committer | Cathal Corbett <cathal.corbett@arm.com> | 2022-12-12 20:09:36 +0000 |
commit | b30e6554ad41f21c8326e387aa2c1f8a5d4e6445 (patch) | |
tree | 7267ad8027a9eed45348b3808da5fcf901b0b767 /src/backends/tosaCommon/test | |
parent | ec67a0f08e0f96a5aebf3cac65331c67f6649f5e (diff) | |
download | armnn-b30e6554ad41f21c8326e387aa2c1f8a5d4e6445.tar.gz |
IVGCVSW-7174 Add Reshape support to TOSA Reference Backend
* Spelling corrections and code refactors added to TosaCommon
* TosaDTypeToString() implemented and used in TosaRef IsLayerSupported()
instead of enum integer.
* Using namespace armnn in TosaCommon OneToOneMappingTests and
TosaReference TosaRefLayerSupportTests instead of armnn::ClassName.
* Updated VerifyTosaAttribute() to also verify certain attributes
from input and output shapes.
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I71dfca404d081a665f748ab724153c6dc36b7eca
Diffstat (limited to 'src/backends/tosaCommon/test')
-rw-r--r-- | src/backends/tosaCommon/test/AvgPool2DIgnoreValueChecker.hpp | 18 | ||||
-rw-r--r-- | src/backends/tosaCommon/test/OneToOneMappingTests.cpp | 146 | ||||
-rw-r--r-- | src/backends/tosaCommon/test/TosaTestUtils.hpp | 52 |
3 files changed, 159 insertions, 57 deletions
diff --git a/src/backends/tosaCommon/test/AvgPool2DIgnoreValueChecker.hpp b/src/backends/tosaCommon/test/AvgPool2DIgnoreValueChecker.hpp index a38f66b466..6f57c4a61e 100644 --- a/src/backends/tosaCommon/test/AvgPool2DIgnoreValueChecker.hpp +++ b/src/backends/tosaCommon/test/AvgPool2DIgnoreValueChecker.hpp @@ -68,9 +68,11 @@ void VerifyAvgPool2DIgnoreValue(TosaSerializationBasicBlock* basicBlock, CHECK(padOp->GetAttributeType() == Attribute_PadAttribute); CHECK(padOp->GetOp() == Op_PAD); - VerifyTosaAttributeFromDescriptor(descriptor, - padOp->GetAttribute(), - LayerType::Pooling2d); + VerifyTosaAttribute(descriptor, + padOp->GetAttribute(), + inputShape[0], + outputShape[0], + LayerType::Pooling2d); // // Verify average pool operator second. @@ -115,9 +117,11 @@ void VerifyAvgPool2DIgnoreValue(TosaSerializationBasicBlock* basicBlock, CHECK(poolOp->GetAttributeType() == Attribute_PoolAttribute); CHECK(poolOp->GetOp() == Op_AVG_POOL2D); - VerifyTosaAttributeFromDescriptor(descriptor, - poolOp->GetAttribute(), - LayerType::Pooling2d, - 1); + VerifyTosaAttribute(descriptor, + poolOp->GetAttribute(), + inputShape[0], + outputShape[0], + LayerType::Pooling2d, + 1); }
\ No newline at end of file diff --git a/src/backends/tosaCommon/test/OneToOneMappingTests.cpp b/src/backends/tosaCommon/test/OneToOneMappingTests.cpp index af9f9e26df..b1fa6847bc 100644 --- a/src/backends/tosaCommon/test/OneToOneMappingTests.cpp +++ b/src/backends/tosaCommon/test/OneToOneMappingTests.cpp @@ -79,7 +79,7 @@ TEST_CASE("GetTosaMappingFromLayer_ConstantLayer") std::vector<std::vector<int32_t>> outputShape = {{ 1, 2, 4, 2 }}; std::vector<float> data = GenerateRandomData<float>(info.GetNumElements()); - armnn::ConstTensor constTensor(info, data); + ConstTensor constTensor(info, data); IConnectableLayer* constant = net->AddConstantLayer(constTensor, "constant"); IConnectableLayer* output = net->AddOutputLayer(0, "output"); @@ -95,7 +95,7 @@ TEST_CASE("GetTosaMappingFromLayer_ConstantLayer") TEST_CASE("GetTosaMapping_Conv2dLayer") { - armnn::Convolution2dDescriptor descriptor; + Convolution2dDescriptor descriptor; descriptor.m_PadLeft = 1; descriptor.m_PadRight = 1; descriptor.m_PadTop = 1; @@ -106,10 +106,10 @@ TEST_CASE("GetTosaMapping_Conv2dLayer") descriptor.m_DilationY = 2; descriptor.m_BiasEnabled = true; - const armnn::TensorInfo inputInfo ({ 1, 5, 5, 1 }, armnn::DataType::Float32); - const armnn::TensorInfo outputInfo({ 1, 3, 3, 1 }, armnn::DataType::Float32); - const armnn::TensorInfo weightsInfo({ 1, 3, 3, 1 }, armnn::DataType::Float32, 0.0f, 0, true); - const armnn::TensorInfo biasesInfo ({ 1 }, armnn::DataType::Float32, 0.0f, 0, true); + const TensorInfo inputInfo ({ 1, 5, 5, 1 }, DataType::Float32); + const TensorInfo outputInfo({ 1, 3, 3, 1 }, DataType::Float32); + const TensorInfo weightsInfo({ 1, 3, 3, 1 }, DataType::Float32, 0.0f, 0, true); + const TensorInfo biasesInfo ({ 1 }, DataType::Float32, 0.0f, 0, true); std::vector<std::vector<int32_t>> inputShape = {{ 1, 5, 5, 1 }, { 1, 3, 3, 1 }, { 1 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 3, 3, 1 }}; @@ -131,7 +131,7 @@ TEST_CASE("GetTosaMappingFromLayer_Conv2dLayer") // Builds up the structure of the network. INetworkPtr net(INetwork::Create()); - armnn::Convolution2dDescriptor descriptor; + Convolution2dDescriptor descriptor; descriptor.m_PadLeft = 1; descriptor.m_PadRight = 1; descriptor.m_PadTop = 1; @@ -142,25 +142,25 @@ TEST_CASE("GetTosaMappingFromLayer_Conv2dLayer") descriptor.m_DilationY = 2; descriptor.m_BiasEnabled = true; - const armnn::TensorInfo inputInfo ({ 1, 5, 5, 1 }, armnn::DataType::Float32); - const armnn::TensorInfo outputInfo({ 1, 3, 3, 1 }, armnn::DataType::Float32); - const armnn::TensorInfo weightsInfo({ 1, 3, 3, 1 }, armnn::DataType::Float32, 0.0f, 0, true); - const armnn::TensorInfo biasesInfo ({ 1 }, armnn::DataType::Float32, 0.0f, 0, true); + const TensorInfo inputInfo ({ 1, 5, 5, 1 }, DataType::Float32); + const TensorInfo outputInfo({ 1, 3, 3, 1 }, DataType::Float32); + const TensorInfo weightsInfo({ 1, 3, 3, 1 }, DataType::Float32, 0.0f, 0, true); + const TensorInfo biasesInfo ({ 1 }, DataType::Float32, 0.0f, 0, true); std::vector<std::vector<int32_t>> inputShape = {{ 1, 5, 5, 1 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 3, 3, 1 }}; std::vector<float> weightsData = GenerateRandomData<float>(weightsInfo.GetNumElements()); - armnn::ConstTensor weights(weightsInfo, weightsData); + ConstTensor weights(weightsInfo, weightsData); std::vector<float> biasesData = GenerateRandomData<float>(biasesInfo.GetNumElements()); - armnn::ConstTensor biases(biasesInfo, biasesData); + ConstTensor biases(biasesInfo, biasesData); - armnn::IConnectableLayer* const inputLayer = net->AddInputLayer(0, "input0"); - armnn::IConnectableLayer* const weightsLayer = net->AddConstantLayer(weights, "weights"); - armnn::IConnectableLayer* const biasesLayer = net->AddConstantLayer(biases, "biases"); - armnn::IConnectableLayer* const convLayer = net->AddConvolution2dLayer(descriptor, "conv2d"); - armnn::IConnectableLayer* const outputLayer = net->AddOutputLayer(0); + IConnectableLayer* const inputLayer = net->AddInputLayer(0, "input0"); + IConnectableLayer* const weightsLayer = net->AddConstantLayer(weights, "weights"); + IConnectableLayer* const biasesLayer = net->AddConstantLayer(biases, "biases"); + IConnectableLayer* const convLayer = net->AddConvolution2dLayer(descriptor, "conv2d"); + IConnectableLayer* const outputLayer = net->AddOutputLayer(0); inputLayer->GetOutputSlot(0).Connect(convLayer->GetInputSlot(0)); weightsLayer->GetOutputSlot(0).Connect(convLayer->GetInputSlot(1)); @@ -179,18 +179,18 @@ TEST_CASE("GetTosaMappingFromLayer_Conv2dLayer") TEST_CASE("GetTosaMapping_MaxPool2DLayer") { - armnn::Pooling2dDescriptor descriptor; - descriptor.m_PoolType = armnn::PoolingAlgorithm::Max; + Pooling2dDescriptor descriptor; + descriptor.m_PoolType = 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_PaddingMethod = PaddingMethod::Exclude; - armnn::TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); - armnn::TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); + TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); + TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); std::vector<std::vector<int32_t>> inputShape = {{ 1, 1, 4, 4 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 1, 3, 3 }}; @@ -209,30 +209,30 @@ TEST_CASE("GetTosaMappingFromLayer_MaxPool2DLayer") // Builds up the structure of the network. INetworkPtr net(INetwork::Create()); - armnn::Pooling2dDescriptor descriptor; - descriptor.m_PoolType = armnn::PoolingAlgorithm::Max; + Pooling2dDescriptor descriptor; + descriptor.m_PoolType = 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_PaddingMethod = PaddingMethod::Exclude; - IConnectableLayer* input0 = net->AddInputLayer(0, "input0"); - IConnectableLayer* pool = net->AddPooling2dLayer(descriptor, "pool"); - IConnectableLayer* output = net->AddOutputLayer(0, "output"); + IConnectableLayer* input = net->AddInputLayer(0, "input0"); + IConnectableLayer* pool = net->AddPooling2dLayer(descriptor, "pool"); + IConnectableLayer* output = net->AddOutputLayer(0, "output"); - input0->GetOutputSlot(0).Connect(pool->GetInputSlot(0)); + input->GetOutputSlot(0).Connect(pool->GetInputSlot(0)); pool->GetOutputSlot(0).Connect(output->GetInputSlot(0)); - armnn::TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); - armnn::TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); + TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); + TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); std::vector<std::vector<int32_t>> inputShape = {{ 1, 1, 4, 4 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 1, 3, 3 }}; - input0->GetOutputSlot(0).SetTensorInfo(inputTensorInfo); + input->GetOutputSlot(0).SetTensorInfo(inputTensorInfo); pool->GetOutputSlot(0).SetTensorInfo(outputTensorInfo); TosaSerializationBasicBlock* basicBlock = @@ -243,18 +243,18 @@ TEST_CASE("GetTosaMappingFromLayer_MaxPool2DLayer") TEST_CASE("GetTosaMapping_AvgPool2DLayer") { - armnn::Pooling2dDescriptor descriptor; - descriptor.m_PoolType = armnn::PoolingAlgorithm::Average; + Pooling2dDescriptor descriptor; + descriptor.m_PoolType = PoolingAlgorithm::Average; 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_PaddingMethod = PaddingMethod::Exclude; - armnn::TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); - armnn::TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); + TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); + TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); std::vector<std::vector<int32_t>> inputShape = {{ 1, 1, 4, 4 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 1, 3, 3 }}; @@ -278,15 +278,15 @@ TEST_CASE("GetTosaMappingFromLayer_AvgPool2DLayer") // Builds up the structure of the network. INetworkPtr net(INetwork::Create()); - armnn::Pooling2dDescriptor descriptor; - descriptor.m_PoolType = armnn::PoolingAlgorithm::Average; + Pooling2dDescriptor descriptor; + descriptor.m_PoolType = PoolingAlgorithm::Average; 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_PaddingMethod = PaddingMethod::Exclude; IConnectableLayer* input0 = net->AddInputLayer(0, "input0"); IConnectableLayer* pool = net->AddPooling2dLayer(descriptor, "pool"); @@ -295,8 +295,8 @@ TEST_CASE("GetTosaMappingFromLayer_AvgPool2DLayer") input0->GetOutputSlot(0).Connect(pool->GetInputSlot(0)); pool->GetOutputSlot(0).Connect(output->GetInputSlot(0)); - armnn::TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); - armnn::TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); + TensorInfo inputTensorInfo({ 1, 1, 4, 4 }, DataType::Float32); + TensorInfo outputTensorInfo({ 1, 1, 3, 3 }, DataType::Float32); std::vector<std::vector<int32_t>> inputShape = {{ 1, 1, 4, 4 }}; std::vector<std::vector<int32_t>> outputShape = {{ 1, 1, 3, 3 }}; @@ -315,6 +315,66 @@ TEST_CASE("GetTosaMappingFromLayer_AvgPool2DLayer") LayerType::Pooling2d); } +TEST_CASE("GetTosaMapping_ReshapeLayer") +{ + TensorInfo inputInfo = TensorInfo({ 2, 3 }, DataType::Float32); + TensorInfo outputInfo = TensorInfo({ 6 }, DataType::Float32); + + std::vector<std::vector<int32_t>> inputShape = {{ 2, 3 }}; + std::vector<std::vector<int32_t>> outputShape = {{ 6 }}; + + ReshapeDescriptor descriptor; + descriptor.m_TargetShape = { 6 }; + + TosaSerializationBasicBlock* basicBlock = + GetTosaMapping(nullptr, LayerType::Reshape, {&inputInfo}, {&outputInfo}, descriptor); + AssertTosaOneToOneMappingBasicBlock(basicBlock, + inputShape, + outputShape, + Op_RESHAPE, + Attribute_ReshapeAttribute, + descriptor, + LayerType::Reshape); +} + +TEST_CASE("GetTosaMappingFromLayer_ReshapeLayer") +{ + IRuntime::CreationOptions options; + IRuntimePtr runtime(IRuntime::Create(options)); + + // Builds up the structure of the network. + INetworkPtr net(INetwork::Create()); + + ReshapeDescriptor descriptor; + descriptor.m_TargetShape = { 6 }; + + IConnectableLayer* input = net->AddInputLayer(0, "input"); + IConnectableLayer* reshape = net->AddReshapeLayer(descriptor, "reshape"); + IConnectableLayer* output = net->AddOutputLayer(0, "output"); + + input->GetOutputSlot(0).Connect(reshape->GetInputSlot(0)); + reshape->GetOutputSlot(0).Connect(output->GetInputSlot(0)); + + TensorInfo inputInfo = TensorInfo({ 2, 3 }, DataType::Float32); + TensorInfo outputInfo = TensorInfo({ 6 }, DataType::Float32); + + input->GetOutputSlot(0).SetTensorInfo(inputInfo); + reshape->GetOutputSlot(0).SetTensorInfo(outputInfo); + + std::vector<std::vector<int32_t>> inputShape = {{ 2, 3 }}; + std::vector<std::vector<int32_t>> outputShape = {{ 6 }}; + + TosaSerializationBasicBlock* basicBlock = + GetTosaMappingFromLayer(PolymorphicDowncast<Layer*>(reshape)); + AssertTosaOneToOneMappingBasicBlock(basicBlock, + inputShape, + outputShape, + Op_RESHAPE, + Attribute_ReshapeAttribute, + descriptor, + LayerType::Reshape); +} + TEST_CASE("GetTosaMapping_Unimplemented") { TosaSerializationBasicBlock* basicBlock = diff --git a/src/backends/tosaCommon/test/TosaTestUtils.hpp b/src/backends/tosaCommon/test/TosaTestUtils.hpp index dd63c0efdf..5c10a6d638 100644 --- a/src/backends/tosaCommon/test/TosaTestUtils.hpp +++ b/src/backends/tosaCommon/test/TosaTestUtils.hpp @@ -8,16 +8,20 @@ #include <Layer.hpp> #include <tosaCommon/TosaMappings.hpp> +#include <tosaCommon/operatorMappings/TosaOperatorUtils.hpp> #include <doctest/doctest.h> +#include <numeric> using namespace armnn; using namespace tosa; -inline void VerifyTosaAttributeFromDescriptor(const BaseDescriptor& descriptor, - const TosaAttributeBase* attribute, - LayerType type, - uint32_t mappingOpNumber = 0) +inline void VerifyTosaAttribute(const BaseDescriptor& descriptor, + const TosaAttributeBase* attribute, + std::vector<int32_t> inputShape, + std::vector<int32_t> outputShape, + LayerType type, + uint32_t mappingOpNumber = 0) { switch (type) { @@ -100,6 +104,25 @@ inline void VerifyTosaAttributeFromDescriptor(const BaseDescriptor& descriptor, CHECK(stride == poolAttribute.stride()); break; } + case LayerType::Reshape: + { + auto reshapeDesc = PolymorphicDowncast<const ReshapeDescriptor*>(&descriptor); + TosaReshapeAttribute reshapeAttribute(attribute); + std::vector<int32_t> shapeAttrib = reshapeAttribute.new_shape(); + + CHECK(GetTosaTensorShape(reshapeDesc->m_TargetShape) == shapeAttrib); + CHECK(outputShape == shapeAttrib); + + auto numInputElements = std::accumulate(std::begin(inputShape), + std::end(inputShape), + 1, + std::multiplies<int32_t>()); + auto numAttributeShapeElements = std::accumulate(std::begin(shapeAttrib), + std::end(shapeAttrib), + 1, + std::multiplies<int32_t>()); + CHECK(numInputElements == numAttributeShapeElements); + } default: break; } @@ -195,7 +218,22 @@ inline void AssertTosaOneToOneMappingBasicBlock(TosaSerializationBasicBlock* bas } } - VerifyTosaAttributeFromDescriptor(descriptor, - op->GetAttribute(), - type); + std::vector<int32_t> input = {}; + std::vector<int32_t> output = {}; + + if (!inputShape.empty()) + { + input = inputShape[0]; + } + + if (!outputShape.empty()) + { + output = outputShape[0]; + } + + VerifyTosaAttribute(descriptor, + op->GetAttribute(), + input, + output, + type); }
\ No newline at end of file |