yidawang commented on a change in pull request #4747: [ThreadPool] Solve ARM
BIG.LITTLE heterogeneous multicores
URL: https://github.com/apache/incubator-tvm/pull/4747#discussion_r373950272
##########
File path: src/runtime/threading_backend.cc
##########
@@ -133,34 +133,42 @@ class ThreadGroup::Impl {
#endif
}
if (exclude_worker0) { // master thread run task
-#if defined(__ANDROID__)
- SetFullCpuAffinity();
-#else
- // if we set TVM_BIND_MASTER_THREAD to be 1, we will bind master thread
- // to core 0.
- const char* bind_master_thread = getenv("TVM_BIND_MASTER_THREAD");
- if (bind_master_thread && atoi(bind_master_thread) == 1) {
- cpu_set_t cpuset;
- CPU_ZERO(&cpuset);
- if (reverse) {
- CPU_SET(sorted_order_[sorted_order_.size() - 1], &cpuset);
- } else {
- CPU_SET(sorted_order_[0], &cpuset);
- }
- pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset);
- }
- pthread_atfork(nullptr, nullptr, ThreadGroup::Impl::SetFullCpuAffinity);
-#endif
+ // Master thread will have free migration on needed cores.
Review comment:
I would suggest add one sentence in the comments: Typically, the OS will
schedule the master thread to run at core 0, which is idle otherwise, when
other works are running.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
