static_backward.yaml

# This file is to support those static ops different the dynamic.

- backward_op : amax_grad
  forward: amax (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1) -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray axis={}, bool keepdim=false, bool reduce_all=false)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : amax_grad

- backward_op : amin_grad
  forward: amin (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1) -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray axis={}, bool keepdim=false, bool reduce_all=false)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : amin_grad

- backward_op : assign_grad
  forward : assign (Tensor x) -> Tensor(out)
  args : (Tensor out_grad)
  output : Tensor(x_grad)
  composite: assign_grad(out_grad, x_grad)
  invoke : assign(out_grad)

- backward_op : conv2d_transpose_double_grad
  forward : conv2d_transpose_grad(Tensor x, Tensor filter, Tensor bias, Tensor grad_out, int[] strides, int[] paddings, int[] output_padding, IntArray output_size, str padding_algorithm, int groups, int[] dilations, str data_format) -> Tensor(grad_x), Tensor(grad_filter)
  args : (Tensor x, Tensor filter, Tensor grad_out, Tensor grad_x_grad, Tensor grad_filter_grad, int[] strides, int[] paddings, int[] output_padding, IntArray output_size, str padding_algorithm, int groups, int[] dilations, str data_format)
  output : Tensor(x_grad), Tensor(filter_grad), Tensor(grad_out_grad)
  infer_meta :
    func : Conv2dTransposeDoubleGradInferMeta
  kernel :
    func : conv2d_transpose_double_grad
    data_type : x

- backward_op : conv2d_transpose_grad
  forward : conv2d_transpose(Tensor x, Tensor filter, Tensor bias, int[] strides={1, 1}, int[] paddings={0, 0}, int[] output_padding={}, IntArray output_size={}, str padding_algorithm="EXPLICIT", int groups=1, int[] dilations={1, 1}, str data_format="NCHW") -> Tensor(out)
  args : (Tensor x, Tensor filter, Tensor bias, Tensor out_grad, int[] strides, int[] paddings, int[] output_padding, IntArray output_size, str padding_algorithm, int groups, int[] dilations, str data_format)
  output : Tensor(x_grad), Tensor(filter_grad)
  infer_meta :
    func : Conv2dTransposeGradInferMeta
    param : [x, filter, out_grad, strides, paddings, output_padding, output_size, padding_algorithm, groups, dilations, data_format]
  kernel :
    func : conv2d_transpose_grad
    param : [x, filter, out_grad, strides, paddings, output_padding, output_size, padding_algorithm, groups, dilations, data_format]
    data_type : x
  optional : bias
  backward : conv2d_transpose_double_grad

- backward_op : deformable_conv_grad
  forward : deformable_conv (Tensor x, Tensor offset, Tensor filter, Tensor mask, int[] strides={1, 1}, int[] paddings={0, 0}, int[] dilations={1, 1}, int deformable_groups=1, int groups=1, int im2col_step=64) -> Tensor(out)
  args : (Tensor x, Tensor offset, Tensor filter, Tensor mask, Tensor out_grad, int[] strides, int[] paddings, int[] dilations, int deformable_groups, int groups, int im2col_step)
  output : Tensor(x_grad), Tensor(offset_grad), Tensor(filter_grad), Tensor(mask_grad)
  infer_meta :
    func : DeformableConvGradInferMeta
  kernel :
    func : deformable_conv_grad
    data_type : x

- backward_op : depthwise_conv2d_transpose_grad
  forward : depthwise_conv2d_transpose(Tensor x, Tensor filter, Tensor bias, int[] strides={1, 1}, int[] paddings={0, 0}, int[] output_padding={}, IntArray output_size={}, str padding_algorithm="EXPLICIT", int groups=1, int[] dilations={1, 1}, str data_format="NCHW") -> Tensor(out)
  args : (Tensor x, Tensor filter, Tensor bias, Tensor out_grad, int[] strides, int[] paddings, int[] output_padding, IntArray output_size, str padding_algorithm, int groups, int[] dilations, str data_format)
  output : Tensor(x_grad), Tensor(filter_grad)
  infer_meta :
    func : Conv2dTransposeGradInferMeta
    param : [x, filter, out_grad, strides, paddings, output_padding, output_size, padding_algorithm, groups, dilations, data_format]
  kernel :
    func : depthwise_conv2d_transpose_grad
    param : [x, filter, out_grad, strides, paddings, output_padding, output_size, padding_algorithm, groups, dilations, data_format]
    data_type : x
  optional : bias

- backward_op : embedding_grad
  forward : embedding (Tensor x, Tensor weight, int64_t padding_idx=-1) -> Tensor(out)
  args : (Tensor x, Tensor weight, Tensor out_grad, int64_t padding_idx=-1)
  output : Tensor(weight_grad)
  infer_meta :
    func : EmbeddingGradInferMeta
    param : [x,weght]
  kernel :
    func : embedding_grad {dense, dense, dense -> dense}
           embedding_sparse_grad {dense, dense, dense -> selected_rows}
           sparse_weight_embedding_grad {selected_rows, dense, dense -> dense}
           sparse_weight_embedding_sparse_grad {selected_rows, dense, dense -> selected_rows}
    data_type : out_grad
  no_need_buffer : weight

- backward_op : frobenius_norm_grad
  forward: frobenius_norm (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1) -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param : [x]
  kernel :
    func : frobenius_norm_grad
    param : [x, out, out_grad, axis, keepdim, reduce_all]

- backward_op : hardswish_grad
  forward : hardswish (Tensor x, float threshold = 6.0f, float scale = 6.0f, float offset = 3.0f) -> Tensor(out)
  args : (Tensor x, Tensor out_grad)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param : [x]
  kernel :
    func : hardswish_grad
    param : [x, out_grad]
  inplace : (out_grad -> x_grad)

- backward_op : max_grad
  forward: max (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1) -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray axis={}, bool keepdim=false, bool reduce_all=false)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : max_grad
  composite: max_grad(x, out, out_grad, axis, keepdim, reduce_all, x_grad)

- backward_op : min_grad
  forward: min (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, int out_dtype=-1) -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray axis={}, bool keepdim=false, bool reduce_all=false)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : min_grad

- backward_op : pool2d_double_grad
  forward : pool2d_grad(Tensor x, Tensor out, Tensor grad_out, IntArray kernel_size, int[] strides, int[] paddings, bool ceil_mode, bool exclusive, str data_format, str pooling_type, bool global_pooling, bool adaptive, str padding_algorithm) -> Tensor(grad_x)
  args : (Tensor grad_x_grad, IntArray kernel_size, int[] strides, int[] paddings, bool ceil_mode, bool exclusive, str data_format, str pooling_type, bool global_pooling, bool adaptive, str padding_algorithm)
  output : Tensor(grad_out_grad)
  infer_meta :
    func : Pool2DInferMeta
    param : [grad_x_grad, kernel_size, strides, paddings, ceil_mode, exclusive, data_format, pooling_type, global_pooling, adaptive, padding_algorithm]
  kernel :
    func : pool2d_double_grad
    param : [grad_x_grad, kernel_size, strides, paddings, ceil_mode, exclusive, data_format, pooling_type, global_pooling, adaptive, padding_algorithm]

- backward_op : pool2d_grad
  forward : pool2d(Tensor x, IntArray kernel_size, int[] strides = {1,1}, int[] paddings = {0,0}, bool ceil_mode = false, bool exclusive = true, str data_format = "NCHW", str pooling_type = "", bool global_pooling = false, bool adaptive = false, str padding_algorithm = "EXPLICIT") -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, IntArray kernel_size, int[] strides, int[] paddings, bool ceil_mode, bool exclusive, str data_format, str pooling_type, bool global_pooling, bool adaptive, str padding_algorithm)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : pool2d_grad
    param : [x, out, out_grad, kernel_size, strides, paddings, ceil_mode, exclusive, data_format, pooling_type, global_pooling, adaptive, padding_algorithm]
  backward : pool2d_double_grad

- backward_op : pool3d_grad
  forward : pool3d(Tensor x, int[] kernel_size, int[] strides = {1,1,1}, int[] paddings = {0,0,0}, bool ceil_mode = false, bool exclusive = true, str data_format = "NCDHW", str pooling_type = "", bool global_pooling = false, bool adaptive = false, str padding_algorithm = "EXPLICIT") -> Tensor(out)
  args : (Tensor x, Tensor out, Tensor out_grad, int[] ksize, int[] strides, int[] paddings, bool ceil_mode, bool exclusive, str data_format, str pooling_type, bool global_pooling, bool adaptive, str padding_algorithm)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param: [x]
  kernel :
    func : pool3d_grad
    param : [x, out, out_grad, kernel_size, strides, paddings, ceil_mode, exclusive, data_format, pooling_type, global_pooling, adaptive, padding_algorithm]

- backward_op : relu6_grad
  forward : relu6 (Tensor x, float threshold = 6.0f) -> Tensor(out)
  args : (Tensor out, Tensor out_grad)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param : [out]
  kernel :
    func : relu6_grad
  inplace : (out_grad -> x_grad)

- backward_op : rnn_grad
  forward : rnn (Tensor x, Tensor[] pre_state, Tensor[] weight_list, Tensor sequence_length, float dropout_prob=0.0, bool is_bidirec=false, int input_size=10, int hidden_size=100, int num_layers=1, str mode="RNN_TANH", int seed=0, bool is_test=false) -> Tensor(out), Tensor(dropout_state_out), Tensor[](state), Tensor(reserve)
  args : (Tensor x, Tensor[] pre_state, Tensor[] weight_list, Tensor sequence_length, Tensor out, Tensor dropout_state_out, Tensor reserve, Tensor out_grad, Tensor[] state_grad, float dropout_prob, bool is_bidirec, int input_size, int hidden_size, int num_layers, str mode, int seed, bool is_test)
  output : Tensor(x_grad), Tensor[](pre_state_grad){pre_state.size()}, Tensor[](weight_list_grad){weight_list.size()}
  infer_meta :
    func : RnnGradInferMeta
    param : [x, pre_state, weight_list]
  kernel :
    func : rnn_grad
    data_type: out_grad

- backward_op : softmax_grad
  forward : softmax (Tensor x, int axis=-1) -> Tensor(out)
  args : (Tensor out, Tensor out_grad, int axis)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param : [out]
  kernel :
    func : softmax_grad
  composite : softmax_grad(out, out_grad, axis, x_grad)

- backward_op : strided_slice_grad
  forward : strided_slice (Tensor x, int[] axes, IntArray starts={}, IntArray ends={}, IntArray strides={}, int[] infer_flags={}, int[] decrease_axis={}) -> Tensor(out)
  args : (Tensor x, Tensor out_grad, int[] axes, IntArray starts, IntArray ends, IntArray strides, int[] infer_flags, int[] decrease_axis)
  output : Tensor(x_grad)
  infer_meta :
    func : GeneralUnaryGradInferMeta
    param : [x]
  kernel :
    func : strided_slice_grad
    param : [x, axes, starts, ends, strides]
    data_type : out_grad
  no_need_buffer : x

- backward_op : sum_double_grad
  forward : sum_grad (Tensor x, Tensor grad_out, IntArray axis, bool keepdim, bool reduce_all=false) -> Tensor(grad_x)
  args : (Tensor grad_x_grad, IntArray axis={}, bool keepdim=false, bool reduce_all=false)
  output : Tensor(grad_out_grad)
  invoke : sum(grad_x_grad, axis, keepdim, reduce_all)

- backward_op : sum_grad
  forward : sum (Tensor x, IntArray axis={0}, bool keepdim=false, bool reduce_all=false, int in_dtype=-1, DataType out_dtype=DataType::UNDEFINED) -> Tensor(out)
  args : (Tensor x, Tensor out_grad, IntArray axis, bool keepdim, bool reduce_all=false)
  output : Tensor(x_grad)
  infer_meta :
    func : UnchangedInferMeta
    param : [x]
  kernel :
    func : sum_grad
  composite : sum_grad(x, out_grad, axis, keepdim, reduce_all, x_grad)
  no_need_buffer : x
  backward : sum_double_grad

- backward_op : swish_grad
  forward : swish (Tensor x, float beta = 1.0f) -> Tensor(out)
  args : (Tensor x, Tensor out_grad)
  output : Tensor(x_grad)
  infer_meta :
    func : GeneralUnaryGradInferMeta
    param : [x]
  kernel :
    func : swish_grad
  inplace : (out_grad -> x_grad)

- backward_op : tril_triu_grad
  forward : tril_triu (Tensor x, int diagonal = 0, bool lower = false)  -> Tensor(out)
  args : (Tensor out_grad, int diagonal, bool lower)
  output : Tensor(x_grad)
  infer_meta :
      func : UnchangedInferMeta
      param : [out_grad]
  kernel:
    func : tril_triu_grad

- backward_op: unpool_grad
  forward: unpool (Tensor x, Tensor indices, int[] ksize, str unpooling_type, int[] strides = {1,1}, int[] paddings ={0,0} ,IntArray output_size = {0,0}, str data_format="NCHW") -> Tensor(out)
  args: (Tensor x, Tensor indices, Tensor out, Tensor out_grad, int[] ksize, int[] strides, int[] paddings, IntArray output_size, str data_format)
  output: Tensor(x_grad)
  infer_meta:
    func: UnchangedInferMeta
    param : [x]
  kernel:
    func: unpool_grad
    data_type: x