llama.cpp
CUDA: Accelerate MXFP4 table lookup using `__byte_perm`
#15451

Merged

CUDA: Accelerate MXFP4 table lookup using `__byte_perm` #15451

JohannesGaessler merged 3 commits into ggml-org:master from Qeeweew:byte_perm_opt

ggerganov requested a review from

JohannesGaessler 1 year ago

github-actions added Nvidia GPU

github-actions added ggml

JohannesGaessler commented on 2025-08-20

JohannesGaessler commented on 2025-08-21

CUDA: optimize get_int_from_table_16

43ecc1a7

CUDA: use v_perm_b32 to replace byte_perm on AMD GPUs

f011aac3

revise documentation

741934c9

JohannesGaessler force pushed from 7720366f to 741934c9 1 year ago

JohannesGaessler approved these changes on 2025-08-25

IMbackK approved these changes on 2025-08-25

JohannesGaessler merged 74f52f77 into master 1 year ago

Reviewers

IMbackK

JohannesGaessler

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone