frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 12:54:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add caveats for native_functions.yaml.

Stefan Krah
2019-04-25 20:32:46 +02:00
parent 48d73f7b09
commit 3ee45c52de
1 changed files with 11 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
TH-to-ATen-porting-guide.md

@ -152,3 +152,14 @@ General guidance: maybe someone has ported something similar before! You can use
* `#pragma omp`. This parallelizes CPU loops, with a huge impact on performance. Don't forget to preserve these when you move loops over!
## Caveats for [native_functions.yaml](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml)
* The argument order in [native_functions.yaml](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml) does not match the order in [NativeFunctions.h](https://github.com/pytorch/pytorch/blob/master/torch/include/ATen/NativeFunctions.h). Example:
1) Signature:
`adaptive_max_pool2d(Tensor self, int[2] output_size, *, Tensor(a!) out, Tensor(b!) indices) -> (Tensor(a!), Tensor(b!))`
2) Function prototype:
`std::tuple<Tensor &,Tensor &> adaptive_max_pool2d_out_cpu(Tensor & out, Tensor & indices, const Tensor & self, IntArrayRef output_size);`
* Argument names matter, the convention is to use `out` for output arguments. See: https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/README.md)