LATX: optimize scalar pixel updates and VROUNDPS truncation - #442

Closed

luzeng87 wants to merge 12 commits into

lat-opensource:masterfrom

luzeng87:gb403-hdr-pr

Commits on Aug 28, 2026

LATX, AVX: defer redundant VEX.128 upper-half clears
luzeng87
committed
b245da0
LATX: optimize AVX scalar physics hot paths
luzeng87
committed
c743dd7
LATX: fold AVX sum-of-three vector patterns
luzeng87
committed
cae81e5

Commits on Aug 31, 2026

LATX, opt: streamline AVX audio hot paths
luzeng87
committed
86ae4ba
LATX, opt: forward scalar results into FMA
luzeng87
committed
388020b
LATX: preserve flags across VCOMIS branch patterns
luzeng87
committed
6fd7043
LATX: preserve VPSHUFD source before permutation
luzeng87
committed
e5154f0
LATX: classify VCOMIS flags in LBT fallback
luzeng87
committed
657a295
LATX: keep mixed XMM/YMM TBs out of SHBR
luzeng87
committed
cb121f8
LATX, opt: fuse scalar HDR pixel update pattern
luzeng87
committed
4a5d2c7
LATX, AVX: avoid FCSR serialization for VROUNDPS SAE truncation
luzeng87
committed
cd674b6

Commits on Sep 3, 2026

LATX, AVX: restore YMM mask and blend writeback
luzeng87
committed
27764e2