CUDA: faster IQ3_K, IQ3_KS, IQ3_K_R4#714

Merged

ikawrakow merged 6 commits intomainfrom

ik/cuda_iq3k_use_bperm1

Aug 21, 2025

Commits on Aug 21, 2025

Use bperm trick for iq3_ks - 5% PP performance gain
Iwan Kawrakow
committed
Use bperm trick for iq3_k -> 5% PP performance gain
Iwan Kawrakow
committed
Use bperm trick for iq3_k -> 8% PP performance gain
Iwan Kawrakow
committed
Use bperm trick for iq3_k_r4 gemv -> ~5% faster
Iwan Kawrakow
committed
Use bperm trick for iq3_k gemv -> ~3% faster
Iwan Kawrakow
committed
Use bperm trick for iq3_k gemv -> 4.5% gain
Iwan Kawrakow
committed