Repository navigation

cuda: prefetch the next PTQ1_0 mat-vec's weights into L2 from the last CTAs - #314

Open

sb32445 wants to merge 3 commits into

PrismML-Eng:prismfrom

sb32445:pr/ptq1-l2-prefetch

Open

cuda: prefetch the next PTQ1_0 mat-vec's weights into L2 from the last CTAs#314
sb32445 wants to merge 3 commits into
PrismML-Eng:prismfrom
sb32445:pr/ptq1-l2-prefetch

Commits on Oct 15, 2026
- cuda: prefetch the next PTQ1_0 mat-vec's weights into L2 from the last CTAs
  
  authored and
  bri-prism
  committed
  90d040b
- cuda: fix the L2 prefetch parameters and remove its switches
  
  authored and
  bri-prism
  committed
  462eb85
- cuda : cap the PTQ1_0 L2 prefetch at 8 MiB on Ada and 2 MiB elsewhere
  
  sb32445
  authored and
  bri-prism
  committed
  dce8d99