llama.cpp
vulkan: use flops instead of weight tensor size for submission heuristic
#25005
Merged
Merged
vulkan: use flops instead of weight tensor size for submission heuristic #25005
0cc4m
merged
3 commits into
master
from
0cc4m/vulkan-submission-threshold-flops
vulkan: extract flops calculation into function
492adff8
use flops instead of matmul src0 tensor size for submission threshold
cb2a4259
use unsigned ints
bf05250d
jeffbolznv
approved these changes
on 2026-06-25
pwilkin
approved these changes
on 2026-06-29
Login to write a write a comment.
Login via GitHub