PR #18896 HIP: add mmf for CDNA

HIP: add mmf for CDNA #18896

JohannesGaessler merged 23 commits into ggml-org:master from zhang-hui-yulo:mmf_for_cdna

refactor mmf rows_per_block

fe24848d

speed up compile

be5ec5a6

pass cdna compile

a31225cc

fix cuda error

8c875b23

clean up mmf

7b43cbc0

f32 mmf

cd8a31ce

clean float mma

45024ee8

fix mmf error

7444a7a1

faster mmf

250ae9ae

github-actions added Nvidia GPU

github-actions added ggml

extend tile k

4d2ef3d4

fix compile error

47a7defe

Merge branch 'ggml-org:master' into mmf_for_cdna

8375cc65

Revert "extend tile k"

ee0443fb

zhang-hui-yulo marked this pull request as ready for review 194 days ago

zhang-hui-yulo requested a review from

JohannesGaessler 194 days ago

zhang-hui-yulo requested a review from

am17an 194 days ago

fix smem overflow

3bb26c7e

speed up compiling mmf

b06565ae

speed up compile for hip

8331ffd0

512 block for cdna

09527b88

config pad size

ee0c740b

JohannesGaessler commented on 2026-01-27

fix as comment

a66d4b1b

update select logic

1e6ef5e8

move some code to cuh

298b97ab

fix as comment

c3c69230

correct cdna3 config

5dabdd6e

JohannesGaessler approved these changes on 2026-01-28

JohannesGaessler merged f3dd7b8e into master 184 days ago

zhang-hui-yulo deleted the mmf_for_cdna branch 166 days ago

Reviewers

JohannesGaessler

am17an

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone

llama.cpp HIP: add mmf for CDNA #18896 Merged

HIP: add mmf for CDNA #18896

llama.cpp
HIP: add mmf for CDNA
#18896

Merged