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

fixed hpx runtime initialization and finalization

Browse source
This commit is contained in:
ct-clmsn 2023-12-23 22:35:47 -05:00
parent e1eb3d14f1
commit 1cdfdb34e0
1 changed files with 16 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

23
llama.cpp
View file

@ -8771,13 +8771,6 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
std::vector<int64_t> hist_all(1 << 4, 0);
#if defined(GGML_USE_HPX)
{
std::string thread_arg = "--hpx:threads=" + std::to_string(nthread);
hpx::init_params params;
params.cfg = { thread_arg };
hpx::start(nullptr, 0, nullptr, params);
}
std::vector<hpx::future<void>> futures;
futures.reserve(nthread);
#else
@ -9352,6 +9345,7 @@ void llama_backend_init(bool numa) {
struct ggml_init_params params = { 0, NULL, false };
struct ggml_context * ctx = ggml_init(params);
ggml_free(ctx);
}
if (numa) {
@ -9361,12 +9355,27 @@ void llama_backend_init(bool numa) {
#ifdef GGML_USE_MPI
ggml_mpi_backend_init();
#endif
#ifdef GGML_USE_HPX
{
const auto nthread = std::thread::hardware_concurrency();
std::string thread_arg = "--hpx:threads=" + std::to_string(nthread);
hpx::init_params params;
params.cfg = { thread_arg };
hpx::start(nullptr, 0, nullptr, params);
}
#endif
}
void llama_backend_free(void) {
#ifdef GGML_USE_MPI
ggml_mpi_backend_free();
#endif
#ifdef GGML_USE_HPX
{
hpx::post([]() { hpx::finalize(); });
hpx::stop();
}
#endif
}
int64_t llama_time_us(void) {