IVGCVSW-5393 'TfLiteDelegate: Implement the split operators'

* Added SPLIT and SPLIT_V support to armnn_delegate

Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I2def9b8be783b25ef17a997e521c6027553035d3

1 files changed, 368 insertions, 0 deletions

diff --git a/delegate/src/test/SplitTestHelper.hpp b/delegate/src/test/SplitTestHelper.hpp
new file mode 100644
index 0000000000..31fc7d5e46
--- /dev/null
+++ b/delegate/src/test/SplitTestHelper.hpp
@@ -0,0 +1,368 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "TestUtils.hpp"
+
+#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>
+
+#include <string>
+
+namespace
+{
+
+std::vector<char> CreateSplitTfLiteModel(tflite::TensorType tensorType,
+                                         std::vector<int32_t>& axisTensorShape,
+                                         std::vector<int32_t>& inputTensorShape,
+                                         const std::vector<std::vector<int32_t>>& outputTensorShapes,
+                                         std::vector<int32_t>& axisData,
+                                         const int32_t numSplits,
+                                         float quantScale = 1.0f,
+                                         int quantOffset  = 0)
+{
+    using namespace tflite;
+    flatbuffers::FlatBufferBuilder flatBufferBuilder;
+
+    std::array<flatbuffers::Offset<tflite::Buffer>, 2> buffers;
+    buffers[0] = CreateBuffer(flatBufferBuilder, flatBufferBuilder.CreateVector({}));
+    buffers[1] = CreateBuffer(flatBufferBuilder,
+                              flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(axisData.data()),
+                                                             sizeof(int32_t) * axisData.size()));
+
+    auto quantizationParameters =
+            CreateQuantizationParameters(flatBufferBuilder,
+                                         0,
+                                         0,
+                                         flatBufferBuilder.CreateVector<float>({ quantScale }),
+                                         flatBufferBuilder.CreateVector<int64_t>({ quantOffset }));
+
+    std::array<flatbuffers::Offset<Tensor>, 4> tensors;
+    tensors[0] = CreateTensor(flatBufferBuilder,
+                              flatBufferBuilder.CreateVector<int32_t>(axisTensorShape.data(),
+                                                                      axisTensorShape.size()),
+                              ::tflite::TensorType_INT32,
+                              1,
+                              flatBufferBuilder.CreateString("axis"),
+                              quantizationParameters);
+    tensors[1] = CreateTensor(flatBufferBuilder,
+                              flatBufferBuilder.CreateVector<int32_t>(inputTensorShape.data(),
+                                                                      inputTensorShape.size()),
+                              tensorType,
+                              0,
+                              flatBufferBuilder.CreateString("input"),
+                              quantizationParameters);
+
+    // Create output tensor
+    for (unsigned int i = 0; i < outputTensorShapes.size(); ++i)
+    {
+        tensors[i + 2] = CreateTensor(flatBufferBuilder,
+                                  flatBufferBuilder.CreateVector<int32_t>(outputTensorShapes[i].data(),
+                                                                          outputTensorShapes[i].size()),
+                                  tensorType,
+                                  0,
+                                  flatBufferBuilder.CreateString("output"),
+                                  quantizationParameters);
+    }
+
+    // create operator. Mean uses ReducerOptions.
+    tflite::BuiltinOptions operatorBuiltinOptionsType = tflite::BuiltinOptions_SplitOptions;
+    flatbuffers::Offset<void> operatorBuiltinOptions = CreateSplitOptions(flatBufferBuilder, numSplits).Union();
+
+    const std::vector<int> operatorInputs{ {0, 1} };
+    const std::vector<int> operatorOutputs{ {2, 3} };
+    flatbuffers::Offset <Operator> controlOperator =
+            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} };
+    const std::vector<int> subgraphOutputs{ {2, 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(&controlOperator, 1));
+
+    flatbuffers::Offset <flatbuffers::String> modelDescription =
+            flatBufferBuilder.CreateString("ArmnnDelegate: SPLIT Operator Model");
+    flatbuffers::Offset <OperatorCode> operatorCode = CreateOperatorCode(flatBufferBuilder, BuiltinOperator_SPLIT);
+
+    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 SplitTest(tflite::TensorType tensorType,
+               std::vector<armnn::BackendId>& backends,
+               std::vector<int32_t>& axisTensorShape,
+               std::vector<int32_t>& inputTensorShape,
+               std::vector<std::vector<int32_t>>& outputTensorShapes,
+               std::vector<int32_t>& axisData,
+               std::vector<T>& inputValues,
+               std::vector<std::vector<T>>& expectedOutputValues,
+               const int32_t numSplits,
+               float quantScale = 1.0f,
+               int quantOffset  = 0)
+{
+    using namespace tflite;
+    std::vector<char> modelBuffer = CreateSplitTfLiteModel(tensorType,
+                                                           axisTensorShape,
+                                                           inputTensorShape,
+                                                           outputTensorShapes,
+                                                           axisData,
+                                                           numSplits,
+                                                           quantScale,
+                                                           quantOffset);
+    const Model* tfLiteModel = GetModel(modelBuffer.data());
+
+    // Create TfLite Interpreters
+    std::unique_ptr<Interpreter> armnnDelegate;
+    CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+              (&armnnDelegate) == kTfLiteOk);
+    CHECK(armnnDelegate != nullptr);
+    CHECK(armnnDelegate->AllocateTensors() == kTfLiteOk);
+
+    std::unique_ptr<Interpreter> tfLiteDelegate;
+    CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+              (&tfLiteDelegate) == kTfLiteOk);
+    CHECK(tfLiteDelegate != nullptr);
+    CHECK(tfLiteDelegate->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(armnnDelegate->ModifyGraphWithDelegate(theArmnnDelegate.get()) == kTfLiteOk);
+
+    // Set input data
+    armnnDelegate::FillInput<T>(tfLiteDelegate, 1, inputValues);
+    armnnDelegate::FillInput<T>(armnnDelegate, 1, inputValues);
+
+    // Run EnqueWorkload
+    CHECK(tfLiteDelegate->Invoke() == kTfLiteOk);
+    CHECK(armnnDelegate->Invoke() == kTfLiteOk);
+
+    // Compare output data
+    for (unsigned int i = 0; i < expectedOutputValues.size(); ++i)
+    {
+        armnnDelegate::CompareOutputData<T>(tfLiteDelegate,
+                                            armnnDelegate,
+                                            outputTensorShapes[i],
+                                            expectedOutputValues[i],
+                                            i);
+    }
+
+    tfLiteDelegate.reset(nullptr);
+    armnnDelegate.reset(nullptr);
+} // End of SPLIT Test
+
+std::vector<char> CreateSplitVTfLiteModel(tflite::TensorType tensorType,
+                                          std::vector<int32_t>& inputTensorShape,
+                                          std::vector<int32_t>& splitsTensorShape,
+                                          std::vector<int32_t>& axisTensorShape,
+                                          const std::vector<std::vector<int32_t>>& outputTensorShapes,
+                                          std::vector<int32_t>& splitsData,
+                                          std::vector<int32_t>& axisData,
+                                          const int32_t numSplits,
+                                          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*>(splitsData.data()),
+                                                             sizeof(int32_t) * splitsData.size()));
+    buffers[2] = CreateBuffer(flatBufferBuilder,
+                              flatBufferBuilder.CreateVector(reinterpret_cast<const uint8_t*>(axisData.data()),
+                                                             sizeof(int32_t) * axisData.size()));
+
+    auto quantizationParameters =
+        CreateQuantizationParameters(flatBufferBuilder,
+                                     0,
+                                     0,
+                                     flatBufferBuilder.CreateVector<float>({ quantScale }),
+                                     flatBufferBuilder.CreateVector<int64_t>({ quantOffset }));
+
+    std::array<flatbuffers::Offset<Tensor>, 5> 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>(splitsTensorShape.data(),
+                                                                      splitsTensorShape.size()),
+                              ::tflite::TensorType_INT32,
+                              1,
+                              flatBufferBuilder.CreateString("splits"),
+                              quantizationParameters);
+    tensors[2] = CreateTensor(flatBufferBuilder,
+                              flatBufferBuilder.CreateVector<int32_t>(axisTensorShape.data(),
+                                                                      axisTensorShape.size()),
+                              ::tflite::TensorType_INT32,
+                              2,
+                              flatBufferBuilder.CreateString("axis"),
+                              quantizationParameters);
+
+    // Create output tensor
+    for (unsigned int i = 0; i < outputTensorShapes.size(); ++i)
+    {
+        tensors[i + 3] = CreateTensor(flatBufferBuilder,
+                                      flatBufferBuilder.CreateVector<int32_t>(outputTensorShapes[i].data(),
+                                                                              outputTensorShapes[i].size()),
+                                      tensorType,
+                                      0,
+                                      flatBufferBuilder.CreateString("output"),
+                                      quantizationParameters);
+    }
+
+    // create operator. Mean uses ReducerOptions.
+    tflite::BuiltinOptions operatorBuiltinOptionsType = tflite::BuiltinOptions_SplitVOptions;
+    flatbuffers::Offset<void> operatorBuiltinOptions = CreateSplitVOptions(flatBufferBuilder, numSplits).Union();
+
+    const std::vector<int> operatorInputs{ {0, 1, 2} };
+    const std::vector<int> operatorOutputs{ {3, 4} };
+    flatbuffers::Offset <Operator> controlOperator =
+        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, 4} };
+    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(&controlOperator, 1));
+
+    flatbuffers::Offset <flatbuffers::String> modelDescription =
+        flatBufferBuilder.CreateString("ArmnnDelegate: SPLIT_V Operator Model");
+    flatbuffers::Offset <OperatorCode> operatorCode = CreateOperatorCode(flatBufferBuilder, BuiltinOperator_SPLIT_V);
+
+    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 SplitVTest(tflite::TensorType tensorType,
+                std::vector<armnn::BackendId>& backends,
+                std::vector<int32_t>& inputTensorShape,
+                std::vector<int32_t>& splitsTensorShape,
+                std::vector<int32_t>& axisTensorShape,
+                std::vector<std::vector<int32_t>>& outputTensorShapes,
+                std::vector<T>& inputValues,
+                std::vector<int32_t>& splitsData,
+                std::vector<int32_t>& axisData,
+                std::vector<std::vector<T>>& expectedOutputValues,
+                const int32_t numSplits,
+                float quantScale = 1.0f,
+                int quantOffset  = 0)
+{
+    using namespace tflite;
+    std::vector<char> modelBuffer = CreateSplitVTfLiteModel(tensorType,
+                                                            inputTensorShape,
+                                                            splitsTensorShape,
+                                                            axisTensorShape,
+                                                            outputTensorShapes,
+                                                            splitsData,
+                                                            axisData,
+                                                            numSplits,
+                                                            quantScale,
+                                                            quantOffset);
+    const Model* tfLiteModel = GetModel(modelBuffer.data());
+
+    // Create TfLite Interpreters
+    std::unique_ptr<Interpreter> armnnDelegate;
+    CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+                  (&armnnDelegate) == kTfLiteOk);
+    CHECK(armnnDelegate != nullptr);
+    CHECK(armnnDelegate->AllocateTensors() == kTfLiteOk);
+
+    std::unique_ptr<Interpreter> tfLiteDelegate;
+    CHECK(InterpreterBuilder(tfLiteModel, ::tflite::ops::builtin::BuiltinOpResolver())
+                  (&tfLiteDelegate) == kTfLiteOk);
+    CHECK(tfLiteDelegate != nullptr);
+    CHECK(tfLiteDelegate->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(armnnDelegate->ModifyGraphWithDelegate(theArmnnDelegate.get()) == kTfLiteOk);
+
+    // Set input data
+    armnnDelegate::FillInput<T>(tfLiteDelegate, 0, inputValues);
+    armnnDelegate::FillInput<T>(armnnDelegate, 0, inputValues);
+
+    // Run EnqueWorkload
+    CHECK(tfLiteDelegate->Invoke() == kTfLiteOk);
+    CHECK(armnnDelegate->Invoke() == kTfLiteOk);
+
+    // Compare output data
+    for (unsigned int i = 0; i < expectedOutputValues.size(); ++i)
+    {
+        armnnDelegate::CompareOutputData<T>(tfLiteDelegate,
+                                            armnnDelegate,
+                                            outputTensorShapes[i],
+                                            expectedOutputValues[i],
+                                            i);
+    }
+
+    tfLiteDelegate.reset(nullptr);
+    armnnDelegate.reset(nullptr);
+} // End of SPLIT_V Test
+
+} // anonymous namespace
\ No newline at end of file