SquaredDifference.cpp

Go to the documentation of this file.

/*
 * Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
 *
 * 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.
 */
 
#include "OMStatus.h"
 
#include "core/OMUtils.h"
#include "core/OMRuntimeShape.h"
 
#include "execute/OMUtils.h"
#include "execute/OMKernelExecutionBuilder.h"
#include "execute/OMRuntimeKernel.h"
 
#include "PALSquaredDifference.h"
 
using namespace onert_micro;
using namespace onert_micro::execute;
 
namespace
{
 
constexpr uint32_t numInput = 2;
constexpr uint32_t numOutput = 1;
 
constexpr uint32_t input1TensorIdx = 0;
constexpr uint32_t input2TensorIdx = 1;
constexpr uint32_t outputTensorIdx = 0;
 
} // namespace
 
namespace onert_micro
{
namespace execute
{
 
// NOTE: doesnt currently support dynamic shapes
// TODO: reduce code duplication with Add, Mul
OMStatus execute_kernel_CircleSquaredDifference(const OMExecuteArgs &execute_args)
{
  core::OMRuntimeContext &runtime_context = execute_args.runtime_context;
  core::OMRuntimeStorage &runtime_storage = execute_args.runtime_storage;
  uint16_t op_index = execute_args.kernel_index;
 
  const circle::Tensor *input1;
  const circle::Tensor *input2;
  const circle::Tensor *output;
 
  uint8_t *input1_data;
  uint8_t *input2_data;
  uint8_t *output_data;
 
  OMStatus status = Ok;
 
  // Read kernel
  {
    execute::OMRuntimeKernel runtime_kernel;
    status = runtime_kernel.readKernel(op_index, runtime_context);
    if (status != Ok)
      return status;
 
    input1 = runtime_kernel.inputs[input1TensorIdx];
    input2 = runtime_kernel.inputs[input2TensorIdx];
    output = runtime_kernel.outputs[outputTensorIdx];
    assert(input1 != nullptr);
    assert(input2 != nullptr);
    assert(output != nullptr);
 
    status = runtime_kernel.getDataFromStorage(op_index, runtime_storage, runtime_context);
    if (status != Ok)
      return status;
 
    input1_data = runtime_kernel.inputs_data[input1TensorIdx];
    input2_data = runtime_kernel.inputs_data[input2TensorIdx];
    output_data = runtime_kernel.outputs_data[outputTensorIdx];
    assert(input1_data != nullptr);
    assert(input2_data != nullptr);
    assert(output_data != nullptr);
  }
 
  core::OMRuntimeShape input1_shape(input1);
  core::OMRuntimeShape input2_shape(input2);
  core::OMRuntimeShape output_shape(output);
 
  core::BinaryArithmeticBroadcastParams params{};
  const bool need_broadcast = pal::processBroadcastShapes(input1_shape, input2_shape, &params);
 
  switch (input1->type())
  {
#ifndef DIS_FLOAT
    case circle::TensorType_FLOAT32:
    {
      status = execute::calculateActivationRange(
        circle::ActivationFunctionType::ActivationFunctionType_NONE, &params.float_activation_min,
        &params.float_activation_max);
      if (need_broadcast)
      {
        status = pal::BroadcastSquaredDifference4DSlow(
          params, input1_shape, core::utils::castInputData<float>(input1_data), input2_shape,
          core::utils::castInputData<float>(input2_data), output_shape,
          core::utils::castOutputData<float>(output_data));
      }
      else
      {
        status = pal::SquaredDifference(params, input1_shape.flatSize(),
                                        core::utils::castInputData<float>(input1_data),
                                        core::utils::castInputData<float>(input2_data),
                                        core::utils::castOutputData<float>(output_data));
      }
    }
    break;
#endif // DIS_FLOAT
#ifndef DIS_QUANT
    case circle::TensorType_INT8:
    {
      // TODO support CWQ
 
      status = execute::calculateActivationRange(
        circle::ActivationFunctionType::ActivationFunctionType_NONE, &params.float_activation_min,
        &params.float_activation_max);
 
      onert_micro::core::QuantizationParams in1_qparams = {
        (*input1->quantization()->scale())[0],
        static_cast<int32_t>((*input1->quantization()->zero_point())[0])};
      onert_micro::core::QuantizationParams in2_qparams = {
        (*input2->quantization()->scale())[0],
        static_cast<int32_t>((*input2->quantization()->zero_point())[0])};
      onert_micro::core::QuantizationParams out_qparams = {
        (*output->quantization()->scale())[0],
        static_cast<int32_t>((*output->quantization()->zero_point())[0])};
 
      if (need_broadcast)
      {
        status = pal::QuantizedBroadcastSquaredDifference4DSlow(
          params, input1_shape, in1_qparams, core::utils::castInputData<int8_t>(input1_data),
          input2_shape, in2_qparams, core::utils::castInputData<int8_t>(input2_data), output_shape,
          out_qparams, core::utils::castOutputData<int8_t>(output_data));
      }
      else
      {
        status = pal::QuantizedSquaredDifference(
          params, input1_shape.flatSize(), in1_qparams,
          core::utils::castInputData<int8_t>(input1_data), in2_qparams,
          core::utils::castInputData<int8_t>(input2_data), out_qparams,
          core::utils::castOutputData<int8_t>(output_data));
      }
    }
    break;
#endif // DIS_QUANT
    default:
    {
      status = UnsupportedType;
      assert(false && "Unsupported type.");
    }
  }
 
  return status;
}
 
} // namespace execute
} // namespace onert_micro