llama.cpp
ggml-webgpu: compute pass batching and removing profiling overhead
#21873
Merged
Merged
ggml-webgpu: compute pass batching and removing profiling overhead #21873
ggerganov
merged
13 commits into
ggml-org:master
from
reeselevine:command-batching
Update register tiling matmul to use f32 accumulation
e94d13cf
fix profiling code
af4c1d51
Fix register tiling matmul for chrome, i'm blaming dawn
ac5267d4
Update batch tuning value for iOS
4edf91b8
compile fix
354cb5c0
Merge remote-tracking branch 'upstream/master' into reg-tile-accum-fix
bde59228
Fix use of new load function
0928d310
Move to a single query set for GPU profiling
0dfdf157
Move to batching compute passes when not profiling
55c05a9a
Refactor build_multi
6468f393
remove iOS throttling now that we're batching compute passes
aa2b3805
Merge remote-tracking branch 'upstream/master' into command-batching
6f6c6a70
Merge remote-tracking branch 'upstream/master' into command-batching
bd6e48a2
reeselevine
changed the title 
CISC
approved these changes
on 2026-04-15
Login to write a write a comment.
Login via GitHub