Method Compute

tensorflow/core/kernels/quantized_conv_ops.cc:482–580

  }

  void Compute(OpKernelContext* context) override {
    // Input tensor is of the following dimensions:
    // [ batch, in_rows, in_cols, in_depth ]
    const Tensor& input = context->input(0);

    // Input filter is of the following dimensions:
    // [ filter_rows, filter_cols, in_depth, out_depth]
    const Tensor& filter = context->input(1);

    // For 2D convolution, there should be 4 dimensions.
    OP_REQUIRES(context, input.dims() == 4,
                errors::InvalidArgument("input must be 4-dimensional",
                                        input.shape().DebugString()));
    OP_REQUIRES(context, filter.dims() == 4,
                errors::InvalidArgument("filter must be 4-dimensional: ",
                                        filter.shape().DebugString()));

    const float min_input = context->input(2).flat<float>()(0);
    const float max_input = context->input(3).flat<float>()(0);
    const float min_filter = context->input(4).flat<float>()(0);
    const float max_filter = context->input(5).flat<float>()(0);
    const int32 offset_input =
        FloatToQuantizedUnclamped<T1>(0.0f, min_input, max_input);
    const int32 offset_filter =
        FloatToQuantizedUnclamped<T2>(0.0f, min_filter, max_filter);
    const int32 offset_output = 0;
    const int32 mult_output = 1;
    const int32 shift_output = 0;

    // The last dimension for input is in_depth. It must be the same as the
    // filter's in_depth.
    const int64 in_depth = input.dim_size(3);
    OP_REQUIRES(context, in_depth == filter.dim_size(2),
                errors::InvalidArgument(
                    "input and filter must have the same depth: ", in_depth,
                    " vs ", filter.dim_size(2)));

    // The last dimension for filter is out_depth.
    const int64 out_depth = filter.dim_size(3);

    // The second dimension for input is rows/height.
    // The first dimension for filter is rows/height.
    const int64 input_rows = input.dim_size(1);
    const int64 filter_rows = filter.dim_size(0);

    // The third dimension for input is columns/width.
    // The second dimension for filter is columns/width.
    const int64 input_cols = input.dim_size(2);
    const int64 filter_cols = filter.dim_size(1);

    // The first dimension for input is batch.
    const int64 batch = input.dim_size(0);

    // For now we take the stride from the second dimension only (we
    // assume row = col stride, and do not support striding on the
    // batch or depth dimension).
    const int stride = strides_[1];