CUDA: faster IQ2_K, IQ2_KS, IQ2_K_R4#716

Merged

ikawrakow merged 6 commits intomainfrom

ik/cuda_iq2k_use_bperm1

Aug 22, 2025

Commits on Aug 21, 2025

Use bperm trick for iq2_ks gemm -> 7% gain
Iwan Kawrakow
committed
Use bperm trick for iq2_k gemm -> ~5% gain
Iwan Kawrakow
committed
Use bperm trick for iq2_k_r4 gemm -> ~3% gain
Iwan Kawrakow
committed
Use bperm trick for iq2_ks gemv -> ~7% gain
Iwan Kawrakow
committed
Use bperm trick for iq2_k gemv -> ~3% gain
Iwan Kawrakow
committed
Use bperm trick for iq2_k_r4 gemv -> ~7% gain
Iwan Kawrakow
committed