vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-08 13:40:59 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add sched in device_flops but disabled by default

Browse source
This commit is contained in:
Lizonghang 2024-12-04 23:54:57 +04:00
parent 7521e532c4
commit b9538f048a
1 changed files with 33 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

33
common/profiler.cpp
View file

@ -164,6 +164,39 @@ static float device_flops(struct llama_model * model, enum ggml_type src0t, enum
ggml_backend_cpu_set_n_threads(backend, n_threads); ggml_backend_cpu_set_n_threads(backend, n_threads);
} }
#if 0
// use scheduler
std::vector<ggml_backend_buffer_type_t> backend_buft;
std::vector<ggml_backend_t> backends = {backend};
if (!ggml_backend_is_cpu(backend)) {
backends.push_back(ggml_backend_cpu_init());
}
for (ggml_backend_t bak : backends) {
if (ggml_backend_is_cpu(bak)) {
backend_buft.push_back(ggml_backend_cpu_buffer_type());
} else {
backend_buft.push_back(ggml_backend_get_default_buffer_type(bak));
}
}
ggml_backend_sched_t sched = ggml_backend_sched_new(backends.data(), backend_buft.data(), backends.size(), 128, false);
bool ok = ggml_backend_sched_reserve(sched, gf);
if (!ok) {
LOG_INF("%s: failed to allocate compute buffers\n", __func__);
ggml_free(ctx_cgraph);
ggml_gallocr_free(allocr);
ggml_free(ctx);
ggml_backend_buffer_free(buffer);
ggml_backend_free(backend);
return 0.0f;
}
ggml_backend_sched_reset(sched);
ggml_backend_sched_alloc_graph(sched, gf);
#endif
// warm-up // warm-up
// ggml_backend_graph_compute(backend, gf); // ggml_backend_graph_compute(backend, gf);