PR #14366 vulkan: Add fusion support for RMS_NORM+MUL

vulkan: Add fusion support for RMS_NORM+MUL #14366

0cc4m merged 12 commits into ggml-org:master from jeffbolznv:rms_mul_fusion_2

jeffbolznv requested a review from

slaren 84 days ago

github-actions added testing

github-actions added Vulkan

github-actions added ggml

jeffbolznv marked this pull request as draft 84 days ago

jeffbolznv commented on 2025-06-24

vulkan: Add fusion support for RMS_NORM+MUL

8643fea3

extract some common fusion logic

18f2d000

fix -Winconsistent-missing-override

b84cb4ac

move ggml_can_fuse to a common function

5e13dcf7

jeffbolznv force pushed from c2245299 to 5e13dcf7 82 days ago

jeffbolznv commented on 2025-06-26

build fix

e6f3c067

C and C++ versions of can_fuse

da2fce0b

move use count to the graph to avoid data races and double increments…

8c50a9b4

slaren commented on 2025-06-27

use hash table lookup to find node index

9ddd4252

change use_counts to be indexed by hash table slot

16304359

minimize hash lookups

2dbac5a0

slaren approved these changes on 2025-06-28

ggerganov approved these changes on 2025-06-28

last node doesn't need single use.

6894905c

jeffbolznv marked this pull request as ready for review 80 days ago

jeffbolznv requested a review from

0cc4m 80 days ago

0cc4m approved these changes on 2025-06-28

remove redundant parameter

a75e9837

0cc4m merged bd9c981d into master 79 days ago

Reviewers

0cc4m

ggerganov

slaren

Assignees

No one assigned

Labels

testing Vulkan ggml

Milestone

No milestone

llama.cpp vulkan: Add fusion support for RMS_NORM+MUL #14366 Merged

vulkan: Add fusion support for RMS_NORM+MUL #14366

llama.cpp
vulkan: Add fusion support for RMS_NORM+MUL
#14366

Merged