Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[static op generation] tile #53317

Closed
wants to merge 10 commits into from
Closed

[static op generation] tile #53317

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
588d66e
[static op generation] tile
gouzil Apr 25, 2023
73ee06c
[fluid] rm tile_op.cc
gouzil Apr 25, 2023
1b82bb1
RollBACK tile_sig
gouzil Apr 25, 2023
8b15495
RollBACK tile_sig.cc
gouzil Apr 25, 2023
a9a6c3c
RollBACK tile_sig.cc
gouzil Apr 25, 2023
007f3b6
Merge branch 'develop' into mv_tile_yml
gouzil Apr 25, 2023
cd3240f
[phi] fix backward composite error
gouzil Apr 25, 2023
a1b78b9
[phi][api] add manual_signature in ops.yaml
gouzil Apr 26, 2023
4117f52
fix manual_signature
gouzil Apr 26, 2023
a24890d
[phi] fix manual_signature
gouzil Apr 26, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
238 changes: 0 additions & 238 deletions paddle/fluid/operators/tile_op.cc
View file
Open in desktop

This file was deleted.

19 changes: 19 additions & 0 deletions paddle/phi/api/yaml/backward.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1907,6 +1907,25 @@
func : thresholded_relu_grad
inplace : (out_grad -> x_grad)

- backward_op : tile_double_grad
forward : tile_grad (Tensor x, Tensor grad_out, IntArray repeat_times) -> Tensor(grad_x)
args : (Tensor grad_x_grad, IntArray repeat_times)
output : Tensor(grad_out_grad)
invoke : tile(grad_x_grad, repeat_times)

- backward_op : tile_grad
forward : tile (Tensor x, IntArray repeat_times) -> Tensor(out)
args : (Tensor x, Tensor out_grad, IntArray repeat_times)
output : Tensor(x_grad)
infer_meta :
func : UnchangedInferMeta
param : [x]
kernel :
func : tile_grad
no_need_buffer : x
composite : tile_grad(x, out_grad, repeat_times, x_grad)
backward : tile_double_grad

- backward_op : topk_grad
forward : topk (Tensor x, Scalar k, int axis = -1, bool largest = true, bool sorted = true) -> Tensor(out), Tensor(indices)
args : (Tensor x, Tensor indices, Tensor out_grad, Scalar k, int axis, bool largest, bool sorted)
Expand Down
19 changes: 0 additions & 19 deletions paddle/phi/api/yaml/legacy_backward.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1014,25 +1014,6 @@
data_type : out_grad
optional : reserve_space

- backward_op : tile_double_grad
forward : tile_grad (Tensor x, Tensor grad_out, IntArray repeat_times) -> Tensor(grad_x)
args : (Tensor grad_x_grad, IntArray repeat_times)
output : Tensor(grad_out_grad)
invoke : tile(grad_x_grad, repeat_times)

- backward_op : tile_grad
forward : tile (Tensor x, IntArray repeat_times) -> Tensor(out)
args : (Tensor x, Tensor out_grad, IntArray repeat_times)
output : Tensor(x_grad)
infer_meta :
func : UnchangedInferMeta
param : [x]
kernel :
func : tile_grad
no_need_buffer : x
composite : tile_grad(x, outgrad, repeat_times, x_grad)
backward : tile_double_grad

- backward_op : transpose_double_grad
forward : transpose_grad (Tensor grad_out, int[] perm) -> Tensor(grad_x)
args : (Tensor grad_x_grad, int[] perm)
Expand Down
9 changes: 0 additions & 9 deletions paddle/phi/api/yaml/legacy_ops.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1157,15 +1157,6 @@
backward : sync_batch_norm_grad
inplace : (mean -> mean_out), (variance -> variance_out)

- op : tile
args : (Tensor x, IntArray repeat_times = {})
output : Tensor(out)
infer_meta :
func : TileInferMeta
kernel :
func : tile
backward : tile_grad

- op : transpose
args : (Tensor x, int[] perm)
output : Tensor
Expand Down
1 change: 1 addition & 0 deletions paddle/phi/api/yaml/op_compat.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2274,6 +2274,7 @@
data_type : int
tensor_name : RepeatTimes
tensors_name : repeat_times_tensor
manual_signature : [tile, tile_grad]

- op : topk (top_k_v2)
backward : topk_grad (top_k_v2_grad)
Expand Down
10 changes: 10 additions & 0 deletions paddle/phi/api/yaml/ops.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1997,6 +1997,16 @@
func : thresholded_relu
backward : thresholded_relu_grad

- op : tile
args : (Tensor x, IntArray repeat_times = {})
output : Tensor(out)
infer_meta :
func : TileInferMeta
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

在op_compat.yaml中增加

manual_signature : [tile]

表示tile的signature保持手写，但是tle_grad和tile_double_grad不需要保持手写，所以需要删除paddle/phi/ops/compat/tile_sig.cc中grad的相关代码，保留tile的代码

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

应该写 manual_signature : [tile, tile_grad]吧, tle_grad 也需要保持手写

Paddle/paddle/phi/ops/compat/tile_sig.cc

Line 34 in 0fac328

KernelSignature TileGradOpArgumentMapping(const ArgumentMappingContext& ctx) {

kernel :
func : tile
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

增加data_type : x

data_type : x
backward : tile_grad

- op : topk
args : (Tensor x, Scalar(int) k = 1, int axis = -1, bool largest = true, bool sorted = true)
output : Tensor(out), Tensor(indices)
Expand Down