diff options
| author | Henrik Gramner <gramner@twoorioles.com> | 2022-09-07 03:06:25 +0300 |
|---|---|---|
| committer | Henrik Gramner <henrik@gramner.com> | 2022-09-07 03:09:49 +0300 |
| commit | 58b15237eef865b2673fc3eecf5a07735db32545 (patch) | |
| tree | 0415ea8914888d1adff44db94fcad66334355d82 | |
| parent | 895fed08e1de3d0a90c9ecd588cf61d8eee8457b (diff) | |
x86: Fix an alignment issue in 8-bit AVX-512 loop restoration
We don't have a separate 8-bit AVX-512 5-tap Wiener filter so the 7-tap
function is used for chroma as well, and in some esoteric edge cases
chroma dst pointers may only have a 32-byte alignment despite having
a width larger than 32, so use an unaligned store as a workaround.
| -rw-r--r-- | src/x86/looprestoration_avx512.asm | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/src/x86/looprestoration_avx512.asm b/src/x86/looprestoration_avx512.asm index 5669ce6..1e57177 100644 --- a/src/x86/looprestoration_avx512.asm +++ b/src/x86/looprestoration_avx512.asm @@ -329,11 +329,11 @@ ALIGN function_align packuswb m2, m4 psrlw m2, 8 vpackuswb m2{k2}, m3, m5 - mova [dstq+r10], m2 - add r10, 64 - jl .hv_loop - mov t6, t5 - mov t5, t4 + movu [dstq+r10], m2 ; We don't have a separate 5-tap version so the 7-tap + add r10, 64 ; function is used for chroma as well, and in some + jl .hv_loop ; esoteric edge cases chroma dst pointers may only + mov t6, t5 ; have a 32-byte alignment despite having a width + mov t5, t4 ; larger than 32, so use an unaligned store here. mov t4, t3 mov t3, t2 mov t2, t1 @@ -379,7 +379,7 @@ ALIGN function_align packuswb m0, m2 psrlw m0, 8 vpackuswb m0{k2}, m1, m3 - mova [dstq+r10], m0 + movu [dstq+r10], m0 add r10, 64 jl .v_loop mov t6, t5 |