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

[QNNPACK] Add more unaligned attributes (#91208)

Browse Source 
Summary: Bypass "Runtime error: store to misaligned address [...] for type 'uint16_t' (aka 'unsigned short'), which requires 2 byte alignment" for q8conv.

Reviewed By: scramsby

Differential Revision: D42179009

Pull Request resolved: https://github.com/pytorch/pytorch/pull/91208
Approved by: https://github.com/kimishpatel
This commit is contained in:
Digant Desai 2022-12-21 03:01:11 +00:00 committed by PyTorch MergeBot
parent a274b5b99e
commit 37ea99cd25
1 changed files with 5 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
aten/src/ATen/native/quantized/cpu/qnnpack/src/q8conv/4x4c2-sse2.c
View File

@ -337,14 +337,15 @@ void pytorch_q8conv_ukernel_4x4c2__sse2(
(uint32_t)_mm_cvtsi128_si32(_mm_unpackhi_epi32(vout, vout));
*((uint32_t*)c3) = (uint32_t)_mm_cvtsi128_si32(_mm_srli_si128(vout, 12));
} else {
typedef PYTORCH_QNNP_UNALIGNED uint16_t unaligned_uint16_t;
if (nr >= 2) {
*((uint16_t*)c0) = (uint16_t)_mm_extract_epi16(vout, 0);
*((unaligned_uint16_t*)c0) = (uint16_t)_mm_extract_epi16(vout, 0);
c0 += 2;
*((uint16_t*)c1) = (uint16_t)_mm_extract_epi16(vout, 2);
*((unaligned_uint16_t*)c1) = (uint16_t)_mm_extract_epi16(vout, 2);
c1 += 2;
*((uint16_t*)c2) = (uint16_t)_mm_extract_epi16(vout, 4);
*((unaligned_uint16_t*)c2) = (uint16_t)_mm_extract_epi16(vout, 4);
c2 += 2;
*((uint16_t*)c3) = (uint16_t)_mm_extract_epi16(vout, 6);
*((unaligned_uint16_t*)c3) = (uint16_t)_mm_extract_epi16(vout, 6);
c3 += 2;
vout = _mm_srli_epi32(vout, 16);
nr -= 2;