(Performance) Optimized x86 and generic q1_0(_g128) dot - #10

Closed

pl752 wants to merge 7 commits into

PrismML-Eng:masterfrom

pl752:perf/q1_0_g128_no_nofma

Commits on Apr 7, 2026

Implemented optimized q1_0 dot for x86 and generic
pl752
committed
195593b
Removed redundant helper definition
pl752
committed
e29cd48

Commits on Apr 10, 2026

Removed two redundant instructions from AVX q1_0 dot
pl752
committed
8587b5c

Commits on Apr 11, 2026

Fixed inconsistency with fp16 conversion for generic q1_0 dot and deduplicated generic fallback
pl752
committed
0c4fb41
Style cleanup around AVX q1_0 dot
pl752
committed
7f82cf0

Commits on Apr 14, 2026

Replaced explicitly unrolled blocks with inner for loop for q1_0
pl752
committed
67f8d32
Replaced scalar ARM q1_0 impl with new generic one
pl752
committed
715f62a