frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix spelling of word - when (#160185)

Browse Source 
just found a typo while understanding the codebase while working on another PR

This fixes typo in word `when` in files

```
native/cpu/PaddingKernel.cpp
native/cpu/batch_norm_kernel.cpp
```

@eqy

Pull Request resolved: https://github.com/pytorch/pytorch/pull/160185
Approved by: https://github.com/yewentao256, https://github.com/ezyang
This commit is contained in:
Raman Kumar
2025-08-31 13:38:23 +00:00
committed by PyTorch MergeBot
parent 91f0bcf43f
commit 0d421ace32
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
aten/src/ATen/native/cpu/PaddingKernel.cpp
View File

@ -156,7 +156,7 @@ void cpu_padding(
int64_t offset_h = ndim >= 2 ? p.offsets[ndim - 2] : 0;
int64_t offset_w = p.offsets[ndim - 1];
// do vectorized copy whe output is overlapped with input on W,
// do vectorized copy when output is overlapped with input on W,
// only applies to positive padding
auto loop = [=](scalar_t* out, const scalar_t* in, bool positive_padding) {
if (positive_padding) {

2
aten/src/ATen/native/cpu/batch_norm_kernel.cpp
View File

@ -318,7 +318,7 @@ batch_norm_cpu_collect_stats_channels_last_impl(
//
// The optimal THRESHOLD to tile was found empirically.
// When C > THRESHOLD, C is large enough that the benefit from tiling and vectorization outweigh the synchronization overhead.
// Wehn C <= TILE_SIZE, the problem size is small enough (C <= TILE_SIZE && NHW <= max_threads) that it's better to launch single thread with vectorization than C threads without vectorization.
// When C <= TILE_SIZE, the problem size is small enough (C <= TILE_SIZE && NHW <= max_threads) that it's better to launch single thread with vectorization than C threads without vectorization.
//
// When num_threads == 1, always use Method 2 as there is no synchronization overhead.
//