chromium/third_party/tflite/src/tensorflow/lite/kernels/internal/reference/sub.h

/* Copyright 2020 The TensorFlow Authors. 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.
==============================================================================*/
#ifndef TENSORFLOW_LITE_KERNELS_INTERNAL_REFERENCE_SUB_H_
#define TENSORFLOW_LITE_KERNELS_INTERNAL_REFERENCE_SUB_H_

#include <stdint.h>

#include <algorithm>
#include <cstddef>
#include <limits>

#include "ruy/profiler/instrumentation.h"  // from @ruy
#include "tensorflow/lite/kernels/internal/common.h"
#include "tensorflow/lite/kernels/internal/compatibility.h"
#include "tensorflow/lite/kernels/internal/types.h"

namespace tflite {

namespace reference_ops {

template <class T>
struct SubImpl { … };

template <>
struct SubImpl<int32_t> { … };

template <typename T, typename F>
inline void BroadcastSubRecursiveDimensions(
    int dimension, const ArithmeticParams& params, const T* input1_data,
    const T* input2_data, T* output_data, size_t* input1_offset_p,
    size_t* input2_offset_p, size_t* output_offset,
    size_t* compressed_input1_stride, size_t* compressed_input2_stride,
    size_t* compressed_output_shape, F binary_func) { … }

// TODO: b/296510380 - we may be able to factor out this to common.h for all
// binary arithmetic ops (add, sub, mul).
template <typename T, typename F>
inline void BroadcastSubCommon(const ArithmeticParams& params,
                               const RuntimeShape& input1_shape,
                               const T* input1_data,
                               const RuntimeShape& input2_shape,
                               const T* input2_data,
                               const RuntimeShape& output_shape, T* output_data,
                               F binary_func) { … }

// TODO(b/151345304): We can implement BroadcastSub on buffers of arbitrary
// dimensionality if the runtime code does a single loop over one dimension
// that handles broadcasting as the base case. The code generator would then
// generate max(D1, D2) nested for loops.
template <typename T>
void BroadcastSubSlow(const ArithmeticParams& params,
                      const RuntimeShape& input1_shape, const T* input1_data,
                      const RuntimeShape& input2_shape, const T* input2_data,
                      const RuntimeShape& output_shape, T* output_data) { … }

inline void BroadcastSub16POTSlow(const ArithmeticParams& params,
                                  const RuntimeShape& input1_shape,
                                  const int16_t* input1_data,
                                  const RuntimeShape& input2_shape,
                                  const int16_t* input2_data,
                                  const RuntimeShape& output_shape,
                                  int16_t* output_data) { … }

template <typename T>
void BroadcastQuantSubSlow(const ArithmeticParams& params,
                           const RuntimeShape& input1_shape,
                           const T* input1_data,
                           const RuntimeShape& input2_shape,
                           const T* input2_data,
                           const RuntimeShape& output_shape, T* output_data) { … }

// Element-wise add that can often be used for inner loop of broadcast add as
// well as the non-broadcast add.
template <typename T>
inline void SubElementwise(int size, const ArithmeticParams& params,
                           const T* input1_data, const T* input2_data,
                           T* output_data) { … }

inline void Sub(const ArithmeticParams& params,
                const RuntimeShape& input1_shape, const uint8_t* input1_data,
                const RuntimeShape& input2_shape, const uint8_t* input2_data,
                const RuntimeShape& output_shape, uint8_t* output_data) { … }

inline void Sub(const ArithmeticParams& params,
                const RuntimeShape& input1_shape, const int8_t* input1_data,
                const RuntimeShape& input2_shape, const int8_t* input2_data,
                const RuntimeShape& output_shape, int8_t* output_data) { … }

inline void Sub(const ArithmeticParams& params,
                const RuntimeShape& input1_shape, const int16_t* input1_data,
                const RuntimeShape& input2_shape, const int16_t* input2_data,
                const RuntimeShape& output_shape, int16_t* output_data) { … }

template <typename T>
void Sub(const ArithmeticParams& params, const RuntimeShape& input1_shape,
         const T* input1_data, const RuntimeShape& input2_shape,
         const T* input2_data, const RuntimeShape& output_shape,
         T* output_data) { … }

inline void SetActivationMinMax(const ArithmeticParams& params,
                                int32_t* activation_min,
                                int32_t* activation_max) { … }

inline void SetActivationMinMax(const ArithmeticParams& params,
                                float* activation_min, float* activation_max) { … }

inline void SetActivationMinMax(const ArithmeticParams& params,
                                int64_t* activation_min,
                                int64_t* activation_max) { … }

template <typename T>
inline void SubWithActivation(
    const ArithmeticParams& params, const RuntimeShape& input1_shape,
    const T* input1_data, const RuntimeShape& input2_shape,
    const T* input2_data, const RuntimeShape& output_shape, T* output_data) { … }

}  // namespace reference_ops
}  // namespace tflite

#endif  // TENSORFLOW_LITE_KERNELS_INTERNAL_REFERENCE_SUB_H_