llama.cpp
ggml-cpu: arm64: q4_K repack gemm and gemv implementations (i8mm)
#16739
Merged
Merged
ggml-cpu: arm64: q4_K repack gemm and gemv implementations (i8mm) #16739
ggerganov
merged
15 commits into
ggml-org:master
from
Alcpz:Alcpz/arm_q4_k_repack
Alcpz
changed the title 
ggerganov
dismissed these changes
on 2025-10-27
ggerganov
dismissed their stale review
102 days ago
https://github.com/ggml-org/llama.cpp/pull/16739#issuecomment-3472806066
Enabled q4_K_8x8_q8_K path on ARM
b3011aa3
wip: I8mm qs multiplication, pending bias
d00dbf4c
cpu : arm : REPACK gemm q4_K8x8 implementation
b9b0b362
Guard gemm with proper features, improved superblock scale and min calc
f956373e
cpu: arm: Implemented REPACK gemv for Q4_K
eb8449b0
Removed completed TODO
8df2511b
Fixed missing guards when selecting optimal repack type for Q4_K
a66f6695
Fixed macro guard for gemv
1d7738ea
Fixed wrong comment in GEMV
b36356fd
Fixed warning for unused variable
92f61ead
vdotq_s32 -> ggml_vdotq_s32
b0b5a27c
Clang-format issues
8a2fd934
ggerganov
approved these changes
on 2025-11-20
Apply suggestions from code review
a8a1d19d
Removed unnecessary GGML_UNUSED
486c027e
Fixed guards in q4_k gemm and gemv (repack)
96af4393
slaren
approved these changes
on 2025-11-24
Login to write a write a comment.