PR #22912 Ggml/cuda snake fusion hardening

Ggml/cuda snake fusion hardening #22912

ServeurpersoCom merged 5 commits into ggml-org:master from ServeurpersoCom:ggml/cuda-snake-fusion-hardening

cuda: tighten snake fusion type checks for all operands (defensive, s…

e4f8ba35

cuda: reject snake fusion when ne[2] or ne[3] > 1 (mirror vulkan PR r…

742dda0e

ServeurpersoCom requested a review 148 days ago

am17an commented on 2026-05-10

cuda: merge type_ok and types_ok into a single types_ok (address am17…

4cd7e1de

github-actions added Nvidia GPU

github-actions added ggml

am17an approved these changes on 2026-05-11

JohannesGaessler approved these changes on 2026-05-11

cuda: filter ADD/SUB/MUL/DIV in supports_op to F32/F16

5d325172

test-backend-ops: extend snake_fuse to rank-4 with ne[2]/ne[3] > 1 cases

c7d1a687

ServeurpersoCom requested a review from

ggerganov 147 days ago

github-actions added testing

pwilkin approved these changes on 2026-05-11

ServeurpersoCom merged e9366607 into master 147 days ago

Reviewers

pwilkin

JohannesGaessler

am17an

ggerganov

Assignees

No one assigned

Labels

testing Nvidia GPU ggml

Milestone

No milestone