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

inductor: reduce compile time for cpu backend by reducing weight conversion (#104402)

Browse Source 
Before this PR, we always add ```to_mkldnn``` before doing weight packing, it is redundant, we can directly convert a dense tensor to block tensor.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/104402
Approved by: https://github.com/jgong5, https://github.com/jansel, https://github.com/eellison
This commit is contained in:
XiaobingSuper
2023-07-05 22:10:37 -04:00
committed by PyTorch MergeBot
parent 434fcffa21
commit 6bfd507c15
4 changed files with 15 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/test_mkldnn_fusion.py
View File

@ -366,7 +366,7 @@ class TestMkldnnFusion(JitTestCase):
if prepack_weight:
packed_weight = torch.ops.mkldnn._reorder_convolution_transpose_weight(
mod.conv_transpose.weight.to_mkldnn(),
mod.conv_transpose.weight,
mod.conv_transpose.padding,
mod.conv_transpose.output_padding,
mod.conv_transpose.stride,