You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

284 lines
13 KiB

/*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#define LOG_TAG "Operations"
#include <functional>
#include <vector>
#include "IndexedShapeWrapper.h"
#include "OperationResolver.h"
#include "OperationsUtils.h"
namespace android {
namespace nn {
namespace comparisons {
constexpr uint32_t kNumInputs = 2;
constexpr uint32_t kInputTensor1 = 0;
constexpr uint32_t kInputTensor2 = 1;
constexpr uint32_t kNumOutputs = 1;
constexpr uint32_t kOutputTensor = 0;
namespace {
template <typename DataType, typename ComparisonType>
bool compute(const std::function<bool(ComparisonType, ComparisonType)>& func, const DataType* aData,
const Shape& aShape, const DataType* bData, const Shape& bShape, bool8* outputData,
const Shape& outputShape) {
IndexedShapeWrapper aShapeIndexed(aShape);
IndexedShapeWrapper bShapeIndexed(bShape);
IndexedShapeWrapper outputShapeIndexed(outputShape);
std::vector<uint32_t> curIndex(outputShape.dimensions.size(), 0);
bool lastIndex = false;
do {
uint32_t outputFlatIndex;
NN_RET_CHECK(outputShapeIndexed.indexToFlatIndex(curIndex, &outputFlatIndex));
uint32_t aFlatIndex;
NN_RET_CHECK(aShapeIndexed.broadcastedIndexToFlatIndex(curIndex, &aFlatIndex));
uint32_t bFlatIndex;
NN_RET_CHECK(bShapeIndexed.broadcastedIndexToFlatIndex(curIndex, &bFlatIndex));
if (aShape.type == OperandType::TENSOR_QUANT8_ASYMM ||
aShape.type == OperandType::TENSOR_QUANT8_ASYMM_SIGNED) {
const float realA = (aData[aFlatIndex] - aShape.offset) * aShape.scale;
const float realB = (bData[bFlatIndex] - bShape.offset) * bShape.scale;
outputData[outputFlatIndex] = func(realA, realB);
} else {
outputData[outputFlatIndex] = func(aData[aFlatIndex], bData[bFlatIndex]);
}
NN_RET_CHECK(outputShapeIndexed.nextIndexInplace(&curIndex, &lastIndex));
} while (!lastIndex);
return true;
}
template <typename DataType, typename ComparisonType>
bool executeLessTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::less<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
template <typename DataType, typename ComparisonType>
bool executeLessEqualTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::less_equal<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
template <typename DataType, typename ComparisonType>
bool executeEqualTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::equal_to<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
template <typename DataType, typename ComparisonType>
bool executeNotEqualTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::not_equal_to<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
template <typename DataType, typename ComparisonType>
bool executeGreaterEqualTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::greater_equal<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
template <typename DataType, typename ComparisonType>
bool executeGreaterTyped(IOperationExecutionContext* context) {
return compute<DataType, ComparisonType>(
std::greater<ComparisonType>(), context->getInputBuffer<DataType>(kInputTensor1),
context->getInputShape(kInputTensor1), context->getInputBuffer<DataType>(kInputTensor2),
context->getInputShape(kInputTensor2), context->getOutputBuffer<bool8>(kOutputTensor),
context->getOutputShape(kOutputTensor));
}
} // namespace
Result<Version> validate(const IOperationValidationContext* context) {
NN_RET_CHECK_EQ(context->getNumInputs(), kNumInputs);
NN_RET_CHECK_EQ(context->getNumOutputs(), kNumOutputs);
OperandType inputType = context->getInputType(kInputTensor1);
NN_RET_CHECK(
inputType == OperandType::TENSOR_BOOL8 || inputType == OperandType::TENSOR_FLOAT16 ||
inputType == OperandType::TENSOR_FLOAT32 || inputType == OperandType::TENSOR_INT32 ||
inputType == OperandType::TENSOR_QUANT8_ASYMM ||
inputType == OperandType::TENSOR_QUANT8_ASYMM_SIGNED)
<< "Unsupported input operand type for comparison op: " << inputType;
NN_RET_CHECK(validateInputTypes(context, {inputType, inputType}));
NN_RET_CHECK(validateOutputTypes(context, {OperandType::TENSOR_BOOL8}));
if (inputType == OperandType::TENSOR_QUANT8_ASYMM_SIGNED) {
return Version::ANDROID_R;
} else {
return Version::ANDROID_Q;
}
}
bool prepare(IOperationExecutionContext* context) {
Shape input1 = context->getInputShape(kInputTensor1);
Shape input2 = context->getInputShape(kInputTensor2);
Shape output = context->getOutputShape(kOutputTensor);
NN_RET_CHECK(calculateBroadcastedShape(input1, input2, &output));
return context->setOutputShape(kOutputTensor, output);
}
bool executeLess(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeLessTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeLessTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeLessTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeLessTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeLessTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeLessTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
bool executeLessEqual(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeLessEqualTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeLessEqualTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeLessEqualTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeLessEqualTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeLessEqualTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeLessEqualTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
bool executeEqual(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeEqualTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeEqualTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeEqualTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeEqualTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeEqualTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeEqualTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
bool executeNotEqual(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeNotEqualTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeNotEqualTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeNotEqualTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeNotEqualTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeNotEqualTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeNotEqualTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
bool executeGreaterEqual(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeGreaterEqualTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeGreaterEqualTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeGreaterEqualTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeGreaterEqualTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeGreaterEqualTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeGreaterEqualTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
bool executeGreater(IOperationExecutionContext* context) {
switch (context->getInputType(kInputTensor1)) {
case OperandType::TENSOR_FLOAT16:
return executeGreaterTyped<_Float16, _Float16>(context);
case OperandType::TENSOR_FLOAT32:
return executeGreaterTyped<float, float>(context);
case OperandType::TENSOR_INT32:
return executeGreaterTyped<int32_t, int32_t>(context);
case OperandType::TENSOR_QUANT8_ASYMM:
return executeGreaterTyped<uint8_t, float>(context);
case OperandType::TENSOR_QUANT8_ASYMM_SIGNED:
return executeGreaterTyped<int8_t, float>(context);
case OperandType::TENSOR_BOOL8:
return executeGreaterTyped<bool8, bool8>(context);
default:
NN_RET_CHECK_FAIL() << "Unsupported tensor type for comparison";
}
}
} // namespace comparisons
NN_REGISTER_OPERATION(LESS, "LESS", comparisons::validate, comparisons::prepare,
comparisons::executeLess);
NN_REGISTER_OPERATION(LESS_EQUAL, "LESS_EQUAL", comparisons::validate, comparisons::prepare,
comparisons::executeLessEqual);
NN_REGISTER_OPERATION(EQUAL, "EQUAL", comparisons::validate, comparisons::prepare,
comparisons::executeEqual);
NN_REGISTER_OPERATION(NOT_EQUAL, "NOT_EQUAL", comparisons::validate, comparisons::prepare,
comparisons::executeNotEqual);
NN_REGISTER_OPERATION(GREATER_EQUAL, "GREATER_EQUAL", comparisons::validate, comparisons::prepare,
comparisons::executeGreaterEqual);
NN_REGISTER_OPERATION(GREATER, "GREATER", comparisons::validate, comparisons::prepare,
comparisons::executeGreater);
} // namespace nn
} // namespace android