OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[Intel GPU][PT2E] bugfix: use zero-point to decide conv src zp mask (#149473)

Browse Source 
# Motivation
The PR fix a bug that wrongly decides the zero-point mask setting. Specifically, it deems zero-point is always not zeros due to scale is used for judgement. Fortunately, the bug only affects the performance. The accuracy is not affected.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/149473
Approved by: https://github.com/EikanWang, https://github.com/guangyey
This commit is contained in:
ZhiweiYan-96 2025-03-19 02:22:00 +00:00 committed by PyTorch MergeBot
parent 496bbf38be
commit d67c1a027e
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
aten/src/ATen/native/mkldnn/xpu/detail/QConv.cpp
View File

@ -135,7 +135,7 @@ at::Tensor quantized_convolution(
mask_weight = (2 ^ 0) | (2 ^ 1); // 2^0 (group) | 2^1 (output channel)
dnnl::primitive_attr pattr;
bool src_need_zp = (act_scale != 0);
bool src_need_zp = (act_zero_point != 0);
bool dst_need_zp = (output_zero_point != 0);
// create usr_md for tensors, and md for conv primitive