Function Sub

tensorflow/lite/kernels/internal/reference/reference_ops.h:999–1032

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) {
  NdArrayDesc<4> desc1;
  NdArrayDesc<4> desc2;
  NdArrayDescsForElementwiseBroadcast(input1_shape, input2_shape, &desc1,
                                      &desc2);
  const RuntimeShape extended_output_shape =
      RuntimeShape::ExtendedShape(4, output_shape);

  // In Tensorflow, the dimensions are canonically named (batch_number, row,
  // col, channel), with extents (batches, height, width, depth), with the
  // trailing dimension changing most rapidly (channels has the smallest stride,
  // typically 1 element).
  //
  // In generated C code, we store arrays with the dimensions reversed. The
  // first dimension has smallest stride.
  //
  // We name our variables by their Tensorflow convention, but generate C code
  // nesting loops such that the innermost loop has the smallest stride for the
  // best cache behavior.
  for (int b = 0; b < extended_output_shape.Dims(0); ++b) {
    for (int y = 0; y < extended_output_shape.Dims(1); ++y) {
      for (int x = 0; x < extended_output_shape.Dims(2); ++x) {
        for (int c = 0; c < extended_output_shape.Dims(3); ++c) {
          output_data[Offset(extended_output_shape, b, y, x, c)] =
              input1_data[SubscriptToIndex(desc1, b, y, x, c)] -
              input2_data[SubscriptToIndex(desc2, b, y, x, c)];
        }
      }
    }
  }
}

inline void SubWithActivation(const ArithmeticParams& params,
                              const RuntimeShape& input1_shape,