xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-30 03:27:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add NUMA node validation for CPU thread binding (#28555)

Browse Source 
Signed-off-by: Bradley <bradley.b.pitt@gmail.com>
This commit is contained in:
usberkeley 2025-11-13 15:03:52 +08:00 committed by GitHub
parent c33b87e777
commit 4ab34f6ef1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
csrc/cpu/utils.cpp
View File

@ -45,6 +45,16 @@ std::string init_cpu_threads_env(const std::string& cpu_ids) {
// Memory node binding // Memory node binding
if (numa_available() != -1) { if (numa_available() != -1) {
int mem_node_id = numa_node_of_cpu(omp_cpu_ids.front()); int mem_node_id = numa_node_of_cpu(omp_cpu_ids.front());
// Verify all CPUs are on the same NUMA node
for (size_t i = 1; i < omp_cpu_ids.size(); ++i) {
int node_id = numa_node_of_cpu(omp_cpu_ids[i]);
TORCH_CHECK(node_id == mem_node_id, "CPU ", omp_cpu_ids[i],
" is on NUMA node ", node_id, ", but CPU ",
omp_cpu_ids.front(), " is on NUMA node ", mem_node_id,
". All CPUs should be on the same NUMA node for optimal "
"performance. Memory will be bound to NUMA node ",
mem_node_id, ".");
}
bitmask* mask = numa_parse_nodestring(std::to_string(mem_node_id).c_str()); bitmask* mask = numa_parse_nodestring(std::to_string(mem_node_id).c_str());
bitmask* src_mask = numa_get_membind(); bitmask* src_mask = numa_get_membind();