IVGCVSW-5338 TfLiteDelegate: Implement the Convolution operators
* Add Convolution, DepthwiseConvolution and TransposeConvolution
Signed-off-by: Kevin May <kevin.may@arm.com>
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I797e42844dfee0cc80beb64eabc3111b96320daf
diff --git a/delegate/src/test/ConvolutionTestHelper.hpp b/delegate/src/test/ConvolutionTestHelper.hpp
new file mode 100644
index 0000000..b7705cc
--- /dev/null
+++ b/delegate/src/test/ConvolutionTestHelper.hpp
@@ -0,0 +1,504 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include <armnn_delegate.hpp>
+
+#include <flatbuffers/flatbuffers.h>
+#include <tensorflow/lite/interpreter.h>
+#include <tensorflow/lite/kernels/register.h>
+#include <tensorflow/lite/model.h>
+#include <tensorflow/lite/schema/schema_generated.h>
+#include <tensorflow/lite/version.h>
+
+#include <doctest/doctest.h>
+
+namespace
+{
+
+template <typename T, typename B = float>
+std::vector<char> CreateConv2dTfLiteModel(tflite::BuiltinOperator convolutionOperatorCode,
+ tflite::TensorType tensorType,
+ uint32_t strideX,
+ uint32_t strideY,
+ uint32_t dilationX,
+ uint32_t dilationY,
+ tflite::Padding padding,
+ tflite::ActivationFunctionType fused_activation_function,
+ const std::vector <int32_t>& inputTensorShape,
+ const std::vector <int32_t>& filterTensorShape,
+ const std::vector <int32_t>& biasTensorShape,
+ const std::vector <int32_t>& outputTensorShape,
+ const std::vector <T>& filterData,
+ const std::vector <B>& biasData,
+ float filterScale = 1.0f,
+ int filterOffset = 0,
+ float outputQuantScale = 2.0f,
+ int outputQuantOffset = 0,
+ float quantScale = 1.0f,
+ int quantOffset = 0,
+ int32_t depth_multiplier = 1)
+{
+ using namespace tflite;
+ flatbuffers::FlatBufferBuilder flatBufferBuilder;
+
+ std::array<flatbuffers::Offset<tflite::Buffer>, 3> buffers;
+ buffers[0] = CreateBuffer(flatBufferBuilder, flatBufferBuilder.CreateVector({}));
+ buffers[1] = CreateBuffer(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(filterData.data()),
+ sizeof(T) * filterData.size()));
+
+ buffers[2] = CreateBuffer(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(biasData.data()),
+ sizeof(B) * biasData.size()));
+
+ auto quantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ quantScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ quantOffset }));
+ auto outputQuantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ outputQuantScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ outputQuantOffset }));
+ auto filterQuantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ filterScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ filterOffset }));
+
+ std::array<flatbuffers::Offset<Tensor>, 4> tensors;
+ tensors[0] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(inputTensorShape.data(),
+ inputTensorShape.size()),
+ tensorType,
+ 0,
+ flatBufferBuilder.CreateString("input"),
+ quantizationParameters);
+ tensors[1] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(filterTensorShape.data(),
+ filterTensorShape.size()),
+ tensorType,
+ 1,
+ flatBufferBuilder.CreateString("filter"),
+ filterQuantizationParameters);
+
+ auto biasTensorType = ::tflite::TensorType_FLOAT32;
+ if (tensorType == ::tflite::TensorType_UINT8)
+ {
+ biasTensorType = ::tflite::TensorType_INT32;
+ }
+ tensors[2] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(biasTensorShape.data(), biasTensorShape.size()),
+ biasTensorType,
+ 2,
+ flatBufferBuilder.CreateString("bias"),
+ quantizationParameters);
+ tensors[3] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(outputTensorShape.data(),
+ outputTensorShape.size()),
+ tensorType,
+ 0,
+ flatBufferBuilder.CreateString("output"),
+ outputQuantizationParameters);
+
+ flatbuffers::Offset<void> operatorBuiltinOptions;
+ tflite::BuiltinOptions operatorBuiltinOptionsType;
+
+ if(convolutionOperatorCode == tflite::BuiltinOperator_DEPTHWISE_CONV_2D)
+ {
+ operatorBuiltinOptionsType = tflite::BuiltinOptions_DepthwiseConv2DOptions;
+ operatorBuiltinOptions = CreateDepthwiseConv2DOptions(flatBufferBuilder,
+ padding,
+ strideX,
+ strideY,
+ depth_multiplier,
+ fused_activation_function,
+ dilationX,
+ dilationY).Union();
+ }
+ if(convolutionOperatorCode == tflite::BuiltinOperator_CONV_2D)
+ {
+ operatorBuiltinOptionsType = tflite::BuiltinOptions_Conv2DOptions;
+ operatorBuiltinOptions = CreateConv2DOptions(flatBufferBuilder,
+ padding,
+ strideX,
+ strideY,
+ fused_activation_function,
+ dilationX,
+ dilationY).Union();
+ }
+
+ // create operator
+ const std::vector<int> operatorInputs{{0, 1, 2}};
+ const std::vector<int> operatorOutputs{{3}};
+ flatbuffers::Offset <Operator> convolutionOperator =
+ CreateOperator(flatBufferBuilder,
+ 0,
+ flatBufferBuilder.CreateVector<int32_t>(operatorInputs.data(), operatorInputs.size()),
+ flatBufferBuilder.CreateVector<int32_t>(operatorOutputs.data(), operatorOutputs.size()),
+ operatorBuiltinOptionsType,
+ operatorBuiltinOptions);
+
+ const std::vector<int> subgraphInputs{ {0, 1, 2} };
+ const std::vector<int> subgraphOutputs{{3}};
+ flatbuffers::Offset <SubGraph> subgraph =
+ CreateSubGraph(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(tensors.data(), tensors.size()),
+ flatBufferBuilder.CreateVector<int32_t>(subgraphInputs.data(), subgraphInputs.size()),
+ flatBufferBuilder.CreateVector<int32_t>(subgraphOutputs.data(), subgraphOutputs.size()),
+ flatBufferBuilder.CreateVector(&convolutionOperator, 1));
+
+ flatbuffers::Offset <flatbuffers::String> modelDescription =
+ flatBufferBuilder.CreateString("ArmnnDelegate: Convolution2d Operator Model");
+ flatbuffers::Offset <OperatorCode> operatorCode = CreateOperatorCode(flatBufferBuilder, convolutionOperatorCode);
+
+ flatbuffers::Offset <Model> flatbufferModel =
+ CreateModel(flatBufferBuilder,
+ TFLITE_SCHEMA_VERSION,
+ flatBufferBuilder.CreateVector(&operatorCode, 1),
+ flatBufferBuilder.CreateVector(&subgraph, 1),
+ modelDescription,
+ flatBufferBuilder.CreateVector(buffers.data(), buffers.size()));
+
+ flatBufferBuilder.Finish(flatbufferModel);
+
+ return std::vector<char>(flatBufferBuilder.GetBufferPointer(),
+ flatBufferBuilder.GetBufferPointer() + flatBufferBuilder.GetSize());
+}
+
+template <typename T, typename B = float>
+void ConvolutionTest(tflite::BuiltinOperator convolutionOperatorCode,
+ tflite::TensorType tensorType,
+ uint32_t strideX,
+ uint32_t strideY,
+ uint32_t dilationX,
+ uint32_t dilationY,
+ tflite::Padding padding,
+ tflite::ActivationFunctionType fused_activation_function,
+ std::vector<armnn::BackendId>& backends,
+ std::vector<int32_t>& inputShape,
+ std::vector<int32_t>& filterShape,
+ std::vector<int32_t>& outputShape,
+ std::vector<T>& inputValues,
+ std::vector<T>& filterValues,
+ std::vector<T>& expectedOutputValues,
+ const std::vector<int32_t>& biasShape = {},
+ const std::vector<B>& biasValues = {},
+ float filterScale = 1.0f,
+ int filterOffset = 0,
+ float outputQuantScale = 2.0f,
+ int outputQuantOffset = 0,
+ float quantScale = 1.0f,
+ int quantOffset = 0,
+ int32_t depth_multiplier = 1)
+
+{
+ using namespace tflite;
+
+ std::vector<char> modelBuffer;
+ modelBuffer = CreateConv2dTfLiteModel(convolutionOperatorCode,
+ tensorType,
+ strideX,
+ strideY,
+ dilationX,
+ dilationY,
+ padding,
+ fused_activation_function,
+ inputShape,
+ filterShape,
+ biasShape,
+ outputShape,
+ filterValues,
+ biasValues,
+ filterScale,
+ filterOffset,
+ outputQuantScale,
+ outputQuantOffset,
+ quantScale,
+ quantOffset,
+ depth_multiplier);
+
+
+ const Model* tfLiteModel = GetModel(modelBuffer.data());
+ // Create TfLite Interpreters
+ std::unique_ptr<Interpreter> armnnDelegateInterpreter;
+ CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+ (&armnnDelegateInterpreter) == kTfLiteOk);
+ CHECK(armnnDelegateInterpreter != nullptr);
+ CHECK(armnnDelegateInterpreter->AllocateTensors() == kTfLiteOk);
+
+ std::unique_ptr<Interpreter> tfLiteInterpreter;
+ CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+ (&tfLiteInterpreter) == kTfLiteOk);
+ CHECK(tfLiteInterpreter != nullptr);
+ CHECK(tfLiteInterpreter->AllocateTensors() == kTfLiteOk);
+
+ // Create the ArmNN Delegate
+ armnnDelegate::DelegateOptions delegateOptions(backends);
+ std::unique_ptr<TfLiteDelegate, decltype(&armnnDelegate::TfLiteArmnnDelegateDelete)>
+ theArmnnDelegate(armnnDelegate::TfLiteArmnnDelegateCreate(delegateOptions),
+ armnnDelegate::TfLiteArmnnDelegateDelete);
+ CHECK(theArmnnDelegate != nullptr);
+ // Modify armnnDelegateInterpreter to use armnnDelegate
+ CHECK(armnnDelegateInterpreter->ModifyGraphWithDelegate(theArmnnDelegate.get()) == kTfLiteOk);
+
+ // Set input data
+ auto tfLiteDelegateInputId = tfLiteInterpreter->inputs()[0];
+ auto tfLiteDelageInputData = tfLiteInterpreter->typed_tensor<T>(tfLiteDelegateInputId);
+ for (unsigned int i = 0; i < inputValues.size(); ++i)
+ {
+ tfLiteDelageInputData[i] = inputValues[i];
+ }
+
+ auto armnnDelegateInputId = armnnDelegateInterpreter->inputs()[0];
+ auto armnnDelegateInputData = armnnDelegateInterpreter->typed_tensor<T>(armnnDelegateInputId);
+ for (unsigned int i = 0; i < inputValues.size(); ++i)
+ {
+ armnnDelegateInputData[i] = inputValues[i];
+ }
+ // Run EnqueueWorkload
+ CHECK(tfLiteInterpreter->Invoke() == kTfLiteOk);
+ CHECK(armnnDelegateInterpreter->Invoke() == kTfLiteOk);
+
+ // Compare output data
+ auto tfLiteDelegateOutputId = tfLiteInterpreter->outputs()[0];
+ auto tfLiteDelagateOutputData = tfLiteInterpreter->typed_tensor<T>(tfLiteDelegateOutputId);
+ auto armnnDelegateOutputId = armnnDelegateInterpreter->outputs()[0];
+ auto armnnDelegateOutputData = armnnDelegateInterpreter->typed_tensor<T>(armnnDelegateOutputId);
+ for (size_t i = 0; i < expectedOutputValues.size(); i++)
+ {
+ CHECK(tfLiteDelagateOutputData[i] == armnnDelegateOutputData[i]);
+ CHECK(doctest::Approx(tfLiteDelagateOutputData[i]).epsilon(0.000001f) == expectedOutputValues[i]);
+ CHECK(doctest::Approx(armnnDelegateOutputData[i]).epsilon(0.000001f) == expectedOutputValues[i]);
+ }
+}
+
+template <typename T>
+std::vector<char> CreateTransposeConvTfLiteModel(tflite::TensorType tensorType,
+ uint32_t strideX,
+ uint32_t strideY,
+ tflite::Padding padding,
+ const std::vector <int32_t>& transposeTensorShape,
+ const std::vector <int32_t>& filterTensorShape,
+ const std::vector <int32_t>& inputTensorShape,
+ const std::vector <int32_t>& outputTensorShape,
+ const std::vector <int32_t>& transposeData,
+ const std::vector <T>& filterData,
+ float filterScale = 1.0f,
+ int filterOffset = 0,
+ float outputQuantScale = 2.0f,
+ int outputQuantOffset = 0,
+ float quantScale = 1.0f,
+ int quantOffset = 0)
+{
+ using namespace tflite;
+ flatbuffers::FlatBufferBuilder flatBufferBuilder;
+
+ std::array<flatbuffers::Offset<tflite::Buffer>, 3> buffers;
+ buffers[0] = CreateBuffer(flatBufferBuilder, flatBufferBuilder.CreateVector({}));
+ buffers[1] = CreateBuffer(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(transposeData.data()),
+ sizeof(int32_t) * transposeData.size()));
+ buffers[2] = CreateBuffer(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(filterData.data()),
+ sizeof(T) * filterData.size()));
+
+ auto quantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ quantScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ quantOffset }));
+ auto outputQuantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ outputQuantScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ outputQuantOffset }));
+ auto filterQuantizationParameters =
+ CreateQuantizationParameters(flatBufferBuilder,
+ 0,
+ 0,
+ flatBufferBuilder.CreateVector<float>({ filterScale }),
+ flatBufferBuilder.CreateVector<int64_t>({ filterOffset }));
+
+ std::array<flatbuffers::Offset<Tensor>, 4> tensors;
+ tensors[0] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(transposeTensorShape.data(),
+ transposeTensorShape.size()),
+ tflite::TensorType_INT32,
+ 1);
+ tensors[1] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(filterTensorShape.data(),
+ filterTensorShape.size()),
+ tensorType,
+ 2,
+ flatBufferBuilder.CreateString("filter"),
+ filterQuantizationParameters);
+ tensors[2] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(inputTensorShape.data(),
+ inputTensorShape.size()),
+ tensorType,
+ 0,
+ flatBufferBuilder.CreateString("input"),
+ quantizationParameters);
+ tensors[3] = CreateTensor(flatBufferBuilder,
+ flatBufferBuilder.CreateVector<int32_t>(outputTensorShape.data(),
+ outputTensorShape.size()),
+ tensorType,
+ 0,
+ flatBufferBuilder.CreateString("output"),
+ outputQuantizationParameters);
+
+ tflite::BuiltinOptions operatorBuiltinOptionsType = tflite::BuiltinOptions_TransposeConvOptions;
+ flatbuffers::Offset<void> operatorBuiltinOptions =
+ CreateTransposeConvOptions(flatBufferBuilder, padding, strideX, strideY).Union();
+
+ // create operator
+ const std::vector<int> operatorInputs{{0, 1, 2}};
+ const std::vector<int> operatorOutputs{{3}};
+ flatbuffers::Offset <Operator> convolutionOperator =
+ CreateOperator(flatBufferBuilder,
+ 0,
+ flatBufferBuilder.CreateVector<int32_t>(operatorInputs.data(), operatorInputs.size()),
+ flatBufferBuilder.CreateVector<int32_t>(operatorOutputs.data(), operatorOutputs.size()),
+ operatorBuiltinOptionsType,
+ operatorBuiltinOptions);
+
+ const std::vector<int> subgraphInputs{ {0, 1, 2} };
+ const std::vector<int> subgraphOutputs{{3}};
+ flatbuffers::Offset <SubGraph> subgraph =
+ CreateSubGraph(flatBufferBuilder,
+ flatBufferBuilder.CreateVector(tensors.data(), tensors.size()),
+ flatBufferBuilder.CreateVector<int32_t>(subgraphInputs.data(), subgraphInputs.size()),
+ flatBufferBuilder.CreateVector<int32_t>(subgraphOutputs.data(), subgraphOutputs.size()),
+ flatBufferBuilder.CreateVector(&convolutionOperator, 1));
+
+ flatbuffers::Offset <flatbuffers::String> modelDescription =
+ flatBufferBuilder.CreateString("ArmnnDelegate: TransposeConv Operator Model");
+ flatbuffers::Offset <OperatorCode> operatorCode =
+ CreateOperatorCode(flatBufferBuilder, tflite::BuiltinOperator_TRANSPOSE_CONV);
+
+ flatbuffers::Offset <Model> flatbufferModel =
+ CreateModel(flatBufferBuilder,
+ TFLITE_SCHEMA_VERSION,
+ flatBufferBuilder.CreateVector(&operatorCode, 1),
+ flatBufferBuilder.CreateVector(&subgraph, 1),
+ modelDescription,
+ flatBufferBuilder.CreateVector(buffers.data(), buffers.size()));
+
+ flatBufferBuilder.Finish(flatbufferModel);
+
+ return std::vector<char>(flatBufferBuilder.GetBufferPointer(),
+ flatBufferBuilder.GetBufferPointer() + flatBufferBuilder.GetSize());
+}
+
+template <typename T>
+void TransposeConvTest(std::vector<armnn::BackendId>& backends,
+ tflite::TensorType tensorType,
+ uint32_t strideX,
+ uint32_t strideY,
+ tflite::Padding padding,
+ const std::vector <int32_t>& transposeTensorShape,
+ const std::vector <int32_t>& filterTensorShape,
+ const std::vector <int32_t>& inputTensorShape,
+ const std::vector <int32_t>& outputTensorShape,
+ const std::vector <int32_t>& transposeData,
+ const std::vector <T>& filterData,
+ std::vector<T>& inputValues,
+ std::vector<T>& expectedOutputValues,
+ float filterScale = 1.0f,
+ int filterOffset = 0,
+ float outputQuantScale = 1.0f,
+ int outputQuantOffset = 0,
+ float quantScale = 1.0f,
+ int quantOffset = 0)
+{
+ using namespace tflite;
+
+ std::vector<char> modelBuffer;
+ modelBuffer = CreateTransposeConvTfLiteModel<T>(tensorType,
+ strideX,
+ strideY,
+ padding,
+ transposeTensorShape,
+ filterTensorShape,
+ inputTensorShape,
+ outputTensorShape,
+ transposeData,
+ filterData,
+ filterScale,
+ filterOffset,
+ outputQuantScale,
+ outputQuantOffset,
+ quantScale,
+ quantOffset);
+
+
+ const Model* tfLiteModel = GetModel(modelBuffer.data());
+ // Create TfLite Interpreters
+ std::unique_ptr<Interpreter> armnnDelegateInterpreter;
+ CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+ (&armnnDelegateInterpreter) == kTfLiteOk);
+ CHECK(armnnDelegateInterpreter != nullptr);
+ CHECK(armnnDelegateInterpreter->AllocateTensors() == kTfLiteOk);
+
+ std::unique_ptr<Interpreter> tfLiteInterpreter;
+ CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+ (&tfLiteInterpreter) == kTfLiteOk);
+ CHECK(tfLiteInterpreter != nullptr);
+ CHECK(tfLiteInterpreter->AllocateTensors() == kTfLiteOk);
+
+ // Create the ArmNN Delegate
+ armnnDelegate::DelegateOptions delegateOptions(backends);
+ std::unique_ptr<TfLiteDelegate, decltype(&armnnDelegate::TfLiteArmnnDelegateDelete)>
+ theArmnnDelegate(armnnDelegate::TfLiteArmnnDelegateCreate(delegateOptions),
+ armnnDelegate::TfLiteArmnnDelegateDelete);
+ CHECK(theArmnnDelegate != nullptr);
+ // Modify armnnDelegateInterpreter to use armnnDelegate
+ CHECK(armnnDelegateInterpreter->ModifyGraphWithDelegate(theArmnnDelegate.get()) == kTfLiteOk);
+
+ // Set input data
+ auto tfLiteDelegateInputId = tfLiteInterpreter->inputs()[2];
+ auto tfLiteDelageInputData = tfLiteInterpreter->typed_tensor<T>(tfLiteDelegateInputId);
+ for (unsigned int i = 0; i < inputValues.size(); ++i)
+ {
+ tfLiteDelageInputData[i] = inputValues[i];
+ }
+
+ auto armnnDelegateInputId = armnnDelegateInterpreter->inputs()[2];
+ auto armnnDelegateInputData = armnnDelegateInterpreter->typed_tensor<T>(armnnDelegateInputId);
+ for (unsigned int i = 0; i < inputValues.size(); ++i)
+ {
+ armnnDelegateInputData[i] = inputValues[i];
+ }
+ // Run EnqueueWorkload
+ CHECK(tfLiteInterpreter->Invoke() == kTfLiteOk);
+ CHECK(armnnDelegateInterpreter->Invoke() == kTfLiteOk);
+
+ // Compare output data
+ auto tfLiteDelegateOutputId = tfLiteInterpreter->outputs()[0];
+ auto tfLiteDelagateOutputData = tfLiteInterpreter->typed_tensor<T>(tfLiteDelegateOutputId);
+ auto armnnDelegateOutputId = armnnDelegateInterpreter->outputs()[0];
+ auto armnnDelegateOutputData = armnnDelegateInterpreter->typed_tensor<T>(armnnDelegateOutputId);
+ for (size_t i = 0; i < expectedOutputValues.size(); i++)
+ {
+ CHECK(armnnDelegateOutputData[i] == expectedOutputValues[i]);
+ CHECK(tfLiteDelagateOutputData[i] == expectedOutputValues[i]);
+ CHECK(tfLiteDelagateOutputData[i] == armnnDelegateOutputData[i]);
+ }
+}
+
+} // anonymous namespace
+
+
+
+