happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge 10482f55b7 into 2d2d9c2062

This commit is contained in:
Aaron Teo 2026-03-24 14:49:45 +06:00 committed by GitHub
parent 2d2d9c2062 10482f55b7
commit 67a2e7bb51
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 27 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
tools/llama-bench/llama-bench.cpp
View File

@ -1051,6 +1051,24 @@ static cmd_params parse_cmd_params(int argc, char ** argv) {
if (params.tensor_buft_overrides.empty()) {
params.tensor_buft_overrides = cmd_params_defaults.tensor_buft_overrides;
}
// check if user specified `--direct-io` without `--mmap`
// if so, we disable mmap to avoid the situation where they see no difference
if (params.use_mmap.empty() && !params.use_direct_io.empty()) {
params.use_mmap.push_back(false);
} else if (params.use_mmap.size() > 0 && params.use_direct_io.empty()) {
params.use_direct_io.push_back(false);
} else if (params.use_mmap.size() != params.use_direct_io.size()) {
// if both are specified, they must have the same number of values
fprintf(stderr, "error: --mmap and --direct-io must have the same number of values\n");
exit(1);
} else {
for (size_t i = 0; i < params.use_direct_io.size(); ++i) {
if (params.use_direct_io[i] == true && params.use_mmap[i] == true) {
fprintf(stderr, "error: --direct-io cannot be enabled with --mmap\n");
exit(1);
}
}
}
if (params.use_mmap.empty()) {
params.use_mmap = cmd_params_defaults.use_mmap;
}
@ -2085,11 +2103,12 @@ int main(int argc, char ** argv) {
fprintf(stderr, "warning: sanitizer enabled, performance may be affected\n");
#endif
// parse command line parameters first to exit early if there are problems
cmd_params params = parse_cmd_params(argc, argv);
// initialize backends
ggml_backend_load_all();
cmd_params params = parse_cmd_params(argc, argv);
auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
if (!cpu_dev) {
fprintf(stderr, "%s: error: CPU backend is not loaded\n", __func__);
@ -2138,6 +2157,12 @@ int main(int argc, char ** argv) {
auto params_count = params_instances.size();
for (const auto & inst : params_instances) {
params_idx++;
// skip test if use_mmap and use_direct_io are turned on
if (inst.use_mmap == true && inst.use_direct_io == true) {
continue;
}
if (params.progress) {
fprintf(stderr, "llama-bench: benchmark %d/%zu: starting\n", params_idx, params_count);
}