llama.cpp
mmq.cu: tune mmq/rocblas switching for RDNA
#18537
Merged

mmq.cu: tune mmq/rocblas switching for RDNA #18537

Beinsezii
jiachengjason Patch perf regression for mmq kernels in ROCm
51e3d385
jiachengjason add n_experts branch like the cdna path
69a9a68b
Beinsezii mmq.cu: tune mmq/wmma switching for RDNA
a435c772
Beinsezii Beinsezii requested a review from JohannesGaessler JohannesGaessler 26 days ago
Beinsezii
Beinsezii commented on 2026-01-02
github-actions github-actions added Nvidia GPU
github-actions github-actions added ggml
Beinsezii
Beinsezii Beinsezii changed the title mmq.cu: tune mmq/wmma switching for RDNA mmq.cu: tune mmq/rocblas switching for RDNA 26 days ago
Beinsezii
JohannesGaessler
JohannesGaessler commented on 2026-01-02
IMbackK
IMbackK
JohannesGaessler
IMbackK
IMbackK
Beinsezii
Beinsezii
JohannesGaessler
Beinsezii
IMbackK
IMbackK
IMbackK approved these changes on 2026-01-02
Beinsezii
IMbackK
JohannesGaessler
Beinsezii
Beinsezii mmq.cu: move amd wmma mmq/wmma switching behind IS_RDNA3
3326fa23
Beinsezii
JohannesGaessler
JohannesGaessler commented on 2026-01-02
JohannesGaessler
Beinsezii Update ggml/src/ggml-cuda/mmq.cu
3fef966d
JohannesGaessler
JohannesGaessler
JohannesGaessler approved these changes on 2026-01-06
JohannesGaessler JohannesGaessler merged 96892952 into master 22 days ago
elfarolab
JohannesGaessler
elfarolab
Beinsezii Beinsezii deleted the beinsezii/rocm_mmq_tune branch 22 days ago
jiachengjason
jiachengjason
JohannesGaessler

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone