🐛 Bug

derivatives.yaml
has a declaration of thnn_conv2d_forward().

However, when changing the declaration, e.g. by adding a new parameter ( ..., Scalar newParam, ...) the generated Declarations.yaml (/build/aten/src/ATen/Declarations.yaml) does not get updated.
Subsequently, "aten/src/ATen/TypeDefault.h" holds the wrong function definition, which leads to a compile error.

derivatives.yaml:

• name: thnn_conv2d_forward(Tensor self, Tensor weight, IntList kernel_size, Tensor bias, IntList stride, IntList padding, Scalar new Param)

Expected behavior

Change in the Declarations.yaml file (build folder)

• name: thnn_conv2d_forward
  method_prefix_derived: ''
  arguments:

  • dynamic_type: Tensor
    name: self
    type: const Tensor &
  • dynamic_type: Tensor
    name: weight
    type: const Tensor &
  • dynamic_type: IntList
    name: kernel_size
    size: 2
    type: IntList
  • dynamic_type: Tensor
    is_nullable: true
    name: bias
    type: const Tensor &
  • dynamic_type: IntList
    name: stride
    size: 2
    type: IntList
  • dynamic_type: IntList
    name: padding
    size: 2
    type: IntList
    - additional parameter is not listed here
    method_of:
  • Type
  • namespace
    mode: NN
    buffers: []
    returns:
  • dynamic_type: Tensor
    name: output
    type: Tensor
  • dynamic_type: Tensor
    name: finput
    type: Tensor
  • dynamic_type: Tensor
    name: fgrad_input
    type: Tensor
    inplace: false
    abstract: true
    device_guard: true
    with_gil: false
    deprecated: false

• OS (e.g., Linux): Ubuntu 16.04

• Build command you used (if compiling from source): python setup.py build develop

• Python version: 3.5.6

• CUDA/cuDNN version: 9.2