CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-04 09:32:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-cpu : fix ggml_graph_compute_thread did not terminate on abort. (ggml/1065)

Browse Source 
some threads kept looping and failed to terminate properly after an abort during CPU execution.

Co-authored-by: issi <issi@gmail.com>
This commit is contained in:
issixx
2025-01-17 21:29:08 +09:00
committed by Georgi Gerganov
parent b636228c0a
commit d2e518e9b4
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
ggml/src/ggml-cpu/ggml-cpu.c
View File

@@ -1302,7 +1302,7 @@ struct ggml_threadpool {
// these are atomic as an annotation for thread-sanitizer
atomic_bool stop; // Used for stopping the threadpool altogether
atomic_bool pause; // Used for pausing the threadpool or individual threads
atomic_bool abort; // Used for aborting processing of a graph
atomic_int abort; // Used for aborting processing of a graph
struct ggml_compute_state * workers; // per thread state
int n_threads_max; // number of threads in the pool
@@ -13851,14 +13851,14 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
/*.threadpool=*/ tp,
};
for (int node_n = 0; node_n < cgraph->n_nodes && !tp->abort; node_n++) {
for (int node_n = 0; node_n < cgraph->n_nodes && atomic_load_explicit(&tp->abort, memory_order_relaxed) != node_n; node_n++) {
struct ggml_tensor * node = cgraph->nodes[node_n];
ggml_compute_forward(&params, node);
if (state->ith == 0 && cplan->abort_callback &&
cplan->abort_callback(cplan->abort_callback_data)) {
tp->abort = true;
atomic_store_explicit(&tp->abort, node_n + 1, memory_order_relaxed);
tp->ec = GGML_STATUS_ABORTED;
}
@@ -14031,7 +14031,7 @@ static struct ggml_threadpool * ggml_threadpool_new_impl(
threadpool->current_chunk = 0;
threadpool->stop = false;
threadpool->pause = tpp->paused;
threadpool->abort = false;
threadpool->abort = -1;
threadpool->workers = NULL;
threadpool->n_threads_max = tpp->n_threads;
threadpool->n_threads_cur = tpp->n_threads;
@@ -14110,7 +14110,7 @@ enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cpl
threadpool->cgraph = cgraph;
threadpool->cplan = cplan;
threadpool->current_chunk = 0;
threadpool->abort = false;
threadpool->abort = -1;
threadpool->ec = GGML_STATUS_SUCCESS;
}