vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

backend : avoid double-ask callback calls

Browse source
This commit is contained in:
Georgi Gerganov 2024-01-16 11:02:51 +02:00
parent 0c96c72150
commit 012ecec506
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 6 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

14
ggml-backend.c
View file

@ -1337,24 +1337,22 @@ static void sched_compute_splits(ggml_backend_sched_t sched) {
for (int j0 = 0; j0 < split->graph.n_nodes; j0++) {
struct ggml_tensor * t = split->graph.nodes[j0];
// check if the user needs data from this node
bool need = sched->callback_eval(t, true, sched->callback_eval_user_data);
int j1 = j0;
// determine the range [j0, j1] of nodes that can be computed together
while (j1 < split->graph.n_nodes - 1) {
// check if the user needs data from this node
if (sched->callback_eval(t, true, sched->callback_eval_user_data)) {
break;
}
while (!need && j1 < split->graph.n_nodes - 1) {
t = split->graph.nodes[++j1];
need = sched->callback_eval(t, true, sched->callback_eval_user_data);
}
struct ggml_cgraph gv = ggml_graph_view(&split->graph, j0, j1 + 1);
ggml_backend_graph_compute(split_backend, &gv);
if (sched->callback_eval(t, true, sched->callback_eval_user_data) && // ask
!sched->callback_eval(t, false, sched->callback_eval_user_data)) { // eval
if (need && !sched->callback_eval(t, false, sched->callback_eval_user_data)) {
break;
}