CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : better PERF prints + support "LLAMA_PERF=1 make"

Browse Source
This commit is contained in:
Georgi Gerganov
2023-04-23 18:15:39 +03:00
parent 53c8434398
commit e4422e299c
3 changed files with 9 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml.c
View File

@@ -11239,7 +11239,7 @@ void ggml_graph_print(const struct ggml_cgraph * cgraph) {
perf_total_per_op_us[node->op] += node->perf_time_us;
GGML_PRINT(" - %3d: [ %" PRId64 ", %" PRId64 ", %" PRId64 "] %16s %s (%3d) cpu = %7.3f / %7.3f ms, wall = %7.3f / %7.3f ms\n",
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 ", %5" PRId64 "] %16s %s (%3d) cpu = %7.3f / %7.3f ms, wall = %7.3f / %7.3f ms\n",
i,
node->ne[0], node->ne[1], node->ne[2],
GGML_OP_LABEL[node->op], node->is_param ? "x" : node->grad ? "g" : " ", node->perf_runs,
@@ -11253,7 +11253,7 @@ void ggml_graph_print(const struct ggml_cgraph * cgraph) {
for (int i = 0; i < cgraph->n_leafs; i++) {
struct ggml_tensor * node = cgraph->leafs[i];
GGML_PRINT(" - %3d: [ %" PRId64 ", %" PRId64 "] %8s\n",
GGML_PRINT(" - %3d: [ %5" PRId64 ", %5" PRId64 "] %8s\n",
i,
node->ne[0], node->ne[1],
GGML_OP_LABEL[node->op]);