Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[RUNTIME] Fast path for single thread run to allow app level threading #7454

Merged
merged 2 commits into from
Feb 18, 2021
Merged

[RUNTIME] Fast path for single thread run to allow app level threading #7454

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0c7303d
Fast path for single thread run to allow app level threading
masahi Feb 13, 2021
18fdf41
add sync counter to avoid error in one of tests
masahi Feb 13, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 20 additions & 11 deletions src/runtime/thread_pool.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -363,21 +363,30 @@ TVM_REGISTER_GLOBAL("runtime.config_threadpool").set_body([](TVMArgs args, TVMRe
} // namespace tvm

int TVMBackendParallelLaunch(FTVMParallelLambda flambda, void* cdata, int num_task) {
int num_workers = tvm::runtime::threading::MaxConcurrency();
if (num_workers == 1) {
std::atomic<int32_t> sync_counter{0};
TVMParallelGroupEnv env;
env.num_task = 1;
env.sync_handle = &sync_counter;
(*flambda)(0, &env, cdata);
return 0;
} else {
#if !TVM_THREADPOOL_USE_OPENMP
int res = tvm::runtime::ThreadPool::ThreadLocal()->Launch(flambda, cdata, num_task, 1);
return res;
int res = tvm::runtime::ThreadPool::ThreadLocal()->Launch(flambda, cdata, num_task, 1);
return res;
#else
int num_workers = tvm::runtime::threading::MaxConcurrency();
if (num_task == 0) num_task = num_workers;
omp_set_num_threads(num_task);
if (num_task == 0) num_task = num_workers;
omp_set_num_threads(num_task);
#pragma omp parallel num_threads(num_task)
{
TVMParallelGroupEnv env;
env.num_task = num_task;
(*flambda)(omp_get_thread_num(), &env, cdata);
}
return 0;
{
TVMParallelGroupEnv env;
env.num_task = num_task;
(*flambda)(omp_get_thread_num(), &env, cdata);
}
return 0;
#endif
}
}

int TVMBackendParallelBarrier(int task_id, TVMParallelGroupEnv* penv) {
Expand Down