happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Keep barrier at the end of loop to synchronise threads reducing cache-line contention (cache HITM) 

This improves throughput for cases where threads have to wait due to lack work and causing process
to spend many cycles in a spin loop. This enables to update dynamic chunk counter with static stride
partitioning which further helps to eliminate shared counter.

* remove one barrier in sgemm()

* static stride partitioning
This commit is contained in:
Uttam Pawar 2025-11-17 23:47:00 +00:00
parent 0de8878c96
commit e516cd0056
1 changed files with 1 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-cpu/llamafile/sgemm.cpp
View File

@ -446,10 +446,7 @@ class tinyBLAS {
ggml_threadpool_chunk_set(params->threadpool, params->nth); ggml_threadpool_chunk_set(params->threadpool, params->nth);
} }
ggml_barrier(params->threadpool); for (int64_t job = params->ith; job < nb_job; job += params->nth) {
int64_t job = params->ith;
while (job < nb_job) {
const int64_t ii = (job % ytiles) * RM * BM; const int64_t ii = (job % ytiles) * RM * BM;
const int64_t jb = job / ytiles; const int64_t jb = job / ytiles;
const int64_t jr0 = BLOC_POS(jb , jj_BN, SIZE_BN); const int64_t jr0 = BLOC_POS(jb , jj_BN, SIZE_BN);
@ -472,7 +469,6 @@ class tinyBLAS {
GGML_ASSERT(jj == jj2); GGML_ASSERT(jj == jj2);
} }
job = ggml_threadpool_chunk_add(params->threadpool, 1);
} }
ggml_barrier(params->threadpool); ggml_barrier(params->threadpool);