Zuxy Meng f68e1afc1b avcodec/loongarch/h264dsp: Fix edge cases for bi-weight on loongarch ...

S16 saturating sum must be computed dot-product-first. The LSX/LASX code
accumulated the offset into the dot product with wrapping vmaddwev/wod,
so large-magnitude sums wrapped instead of saturating.

Accumulate the dot product from zero (wrapping is exact: the dot product
cannot overflow S16 for 8-bit with log2_denom < 7), then add the offset
with a saturating vsadd.

Also tweaked the checkasm test itself to cover only valid width and height
combo.  This fixes bi-weight checkasm test for all currently available
asm implementations.

Signed-off-by: Zuxy Meng <zuxy.meng@gmail.com>

2026-10-01 17:49:31 -07:00

..

cabac.h

…

h264_cabac.c

…

h264_deblock_lasx.c

…

h264_intrapred_init_loongarch.c

…

h264_intrapred_loongarch.h

…

h264chroma_init_loongarch.c

…

h264chroma_loongarch.h

…

h264chroma.S

…

h264dsp_init_loongarch.c

…

h264dsp_lasx.c

…

h264dsp_loongarch.h

…

h264dsp.S

…

h264idct_loongarch.c

…

h264idct.S

…

h264intrapred.S

…

h264qpel_init_loongarch.c

…

h264qpel_lasx.c

…

h264qpel_loongarch.h

…

h264qpel_lsx.c

…

h264qpel.S

…

hevc_add_res.S

…

hevc_idct_lsx.c

…

hevc_idct.S

…

hevc_lpf_sao_lsx.c

…

hevc_mc_bi_lsx.c

…

hevc_mc_uni_lsx.c

…

hevc_mc_uniw_lsx.c

…

hevc_mc.S

…

hevcdsp_init_loongarch.c

…

hevcdsp_lasx.h

…

hevcdsp_lsx.c

…

hevcdsp_lsx.h

…

hpeldsp_init_loongarch.c

…

hpeldsp_lasx.c

…

hpeldsp_lasx.h

…

idctdsp_init_loongarch.c

…

idctdsp_lasx.c

…

idctdsp_loongarch.h

…

loongson_asm.S

…

Makefile

…

simple_idct_lasx.c

…

vc1dsp_init_loongarch.c

…

vc1dsp_lasx.c

…

vc1dsp_loongarch.h

…

videodsp_init.c

…

vp8_lpf_lsx.c

…

vp8_mc_lsx.c

…

vp8dsp_init_loongarch.c

…

vp8dsp_loongarch.h

…

vp9_idct_lsx.c

…

vp9_intra_lsx.c

…

vp9_lpf_lsx.c

…

vp9_mc_lsx.c

…

vp9dsp_init_loongarch.c

…

vp9dsp_loongarch.h

…