upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-23 11:10:55 -05:00
Code Issues Packages Projects Releases Wiki Activity

ggml: fix backend split scheduler race condition

Browse Source 
splits without input were running concurrently with other splits, while potentially reusing memory the other split is accessing
This commit is contained in:
Ruben Ortlam
2026-07-23 14:07:43 +02:00
parent da296d6e72
commit 16a2011d40
1 changed files with 17 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
ggml/src/ggml-backend.cpp
View File

@@ -1546,11 +1546,23 @@ static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t s
std::vector<int32_t> ids;
std::vector<ggml_bitset_t> used_ids;
int prev_backend_id = -1;
for (int split_id = 0; split_id < sched->n_splits; split_id++) {
struct ggml_backend_sched_split * split = &splits[split_id];
int split_backend_id = split->backend_id;
ggml_backend_t split_backend = sched->backends[split_backend_id];
// ensure the previous split's async work has completed before we start
// this split, the allocator may have reused buffer regions across splits
if (prev_backend_id >= 0 && prev_backend_id != split_backend_id) {
if (sched->events[prev_backend_id][sched->cur_copy] != NULL) {
ggml_backend_event_synchronize(sched->events[prev_backend_id][sched->cur_copy]);
} else {
ggml_backend_synchronize(sched->backends[prev_backend_id]);
}
}
// copy the input tensors to the split backend
for (int input_id = 0; input_id < split->n_inputs; input_id++) {
ggml_backend_t input_backend = ggml_backend_sched_get_tensor_backend(sched, split->inputs[input_id]);
@@ -1713,12 +1725,12 @@ static enum ggml_status ggml_backend_sched_compute_splits(ggml_backend_sched_t s
}
}
// record the event of this copy
if (split->n_inputs > 0) {
if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
ggml_backend_event_record(sched->events[split_backend_id][sched->cur_copy], split_backend);
}
// record the event of this split
if (sched->events[split_backend_id][sched->cur_copy] != NULL) {
ggml_backend_event_record(sched->events[split_backend_id][sched->cur_copy], split_backend);
}
prev_backend_id = split_backend_id;
}
return GGML_STATUS_SUCCESS;