llama.cpp
ggml-cpu: optimize ggml_vec_dot_bf16 for Power9
#18837

Merged

ggml-cpu: optimize ggml_vec_dot_bf16 for Power9 #18837

taronaeo merged 7 commits into ggml-org:master from shalinib-ibm:vec_dot_bf16_opt

shalinib-ibm requested a review from

ggerganov 68 days ago

shalinib-ibm force pushed from 92cac28b to 7cbfb64a 68 days ago

shalinib-ibm force pushed from 7cbfb64a to e9cde100 68 days ago

ggml-cpu: optimize ggml_vec_dot_bf16 for Power9

cfd1ed22

shalinib-ibm force pushed from e9cde100 to cfd1ed22 68 days ago

github-actions added ggml

taronaeo approved these changes on 2026-01-14

Update ggml/src/ggml-cpu/simd-mappings.h

0ca6fafa

Update simd-mappings.h

7b935c46

Update vec.cpp

d13f4c3d

Handle endian-ness during vec_mergeh/l while converting BF16 to FP32

5abb3897

taronaeo approved these changes on 2026-01-15

Update ggml/src/ggml-cpu/simd-mappings.h

40a9fed2

Merge branch 'ggml-org:master' into vec_dot_bf16_opt

9d235ca0

taronaeo merged 8cc0ba95 into master 67 days ago

Reviewers

taronaeo

ggerganov

Assignees

No one assigned

Labels

ggml

Milestone

No milestone