add GGML_USE_NUMA_MIGRATE feature to optimize cross NUMA op computation#13649

Closed

wenlujon wants to merge 8 commits intoggml-org:masterfrom

wenlujon:numa_opt

Commits on May 20, 2025

add GGML_USE_NUMA_MIGRATE feature to optimize cross NUMA op computation
boltliu85
committed
fix the threads number larger than numa node numbers
boltliu85
committed

Commits on May 21, 2025

fix the buffer allocate size for NUMA page migration
boltliu85
committed
fix the cgraph null issue when running with llama-bench
boltliu85
committed
use ggml_barrier instead of ggml_barrier_numa_aware for src1 data barrier as the barrier only happens in few cores
boltliu85
committed

Commits on May 22, 2025

remove debug code for ggml barrier
boltliu85
committed

Commits on Jun 3, 2025

opt the src1 tensor data with local numa data, 33% uplift for tg128
boltliu85
committed

Commits on Jun 12, 2025

fix the perf regression with numa_node_of_cpu() system call and the n_barrier_passed_last indexing issue
boltliu85
committed