llama.cpp
opencl: perf optimization for mamba2 ssm_scan by folding 4 dim rows into one workgroup and add more coverage
#27775
Open
Open
opencl: perf optimization for mamba2 ssm_scan by folding 4 dim rows into one workgroup and add more coverage #27775
wanghqc
wants to merge
2 commits into
ggml-org:master
from
qualcomm:hq/opencl-ssm-scan-row-fold-r0822
opencl: fold 4 dim rows into one mamba2 ssm_scan workgroup
f485dd91
opencl: emit rollback snapshots from ssm_scan instead of falling back…
c904a135
Reviewers
No reviews
Assignees
No one assigned
Labels
ggml
OpenCL
Milestone
No milestone
Login to write a write a comment.
Login via GitHub