OSSForks/opencv
1
0
Fork 0
mirror of https://github.com/zebrajr/opencv.git synced 2026-01-15 12:15:17 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Merge pull request #28211 from pratham-mcw:fix/neon-dotprod-expand-overflow

Browse Source 
NEON: fix dot product accumulation causing NORM test failures on Windows ARM64
This commit is contained in:
Alexander Smorkalov
2025-12-18 11:39:46 +03:00
committed by GitHub
parent 721bb7289d ddf2863aaa
commit 0af685708b
1 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
modules/core/include/opencv2/core/hal/intrin_neon.hpp
View File

@@ -888,9 +888,10 @@ inline v_uint32x4 v_dotprod_expand_fast(const v_uint8x16& a, const v_uint8x16& b
inline v_int32x4 v_dotprod_expand_fast(const v_int8x16& a, const v_int8x16& b)
{
int16x8_t prod = vmull_s8(vget_low_s8(a.val), vget_low_s8(b.val));
prod = vmlal_s8(prod, vget_high_s8(a.val), vget_high_s8(b.val));
return v_int32x4(vaddl_s16(vget_low_s16(prod), vget_high_s16(prod)));
int16x8_t p0 = vmull_s8(vget_low_s8(a.val), vget_low_s8(b.val));
int16x8_t p1 = vmull_s8(vget_high_s8(a.val), vget_high_s8(b.val));
int32x4_t s0 = vaddl_s16(vget_low_s16(p0), vget_low_s16(p1));
return v_int32x4(vaddq_s32(s0, vaddl_s16(vget_high_s16(p0), vget_high_s16(p1))));
}
inline v_int32x4 v_dotprod_expand_fast(const v_int8x16& a, const v_int8x16& b, const v_int32x4& c)
{