ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-10 00:10:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

A better way to measure the cost of ggml_barrier (#238)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-03-01 17:12:58 +02:00
committed by GitHub
parent a79ab8f342
commit ef9a3d17b5
1 changed files with 16 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
ggml/src/ggml.c
View File

@@ -21459,14 +21459,26 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
/*.shared=*/ state->shared,
};
#if IK_PRINT_TIMING
int64_t t_start = ggml_time_us();
int64_t t_eval = 0;
#endif
for (int node_n = 0; node_n < cgraph->n_nodes; node_n++) {
struct ggml_tensor * node = cgraph->nodes[node_n];
if (ggml_is_noop(node)) continue;
#if IK_PRINT_TIMING
int64_t tim1 = ggml_time_us();
#endif
if (ggml_compute_forward(&params, node, node_n < cgraph->n_nodes-1 ? cgraph->nodes[node_n+1] : NULL)) {
++node_n;
}
#if IK_PRINT_TIMING
int64_t tim2 = ggml_time_us();
t_eval += tim2 - tim1;
#endif
if (state->ith == 0 && cplan->abort_callback && cplan->abort_callback(cplan->abort_callback_data)) {
state->shared->ec = GGML_STATUS_ABORTED;
@@ -21478,6 +21490,10 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
break;
}
}
#if IK_PRINT_TIMING
int64_t t_end = ggml_time_us();
if (state->ith == 0) printf("ggml_barrier(...): %d us\n", (int)(t_end - t_start - t_eval));
#endif
return 0;
}