Description / Steps to reproduce the issue
I found my code became significantly slower back in last October and VTune revealed the majority of time spent is in pthread_spin_lock(). I didn't report right away because libwinpthread.dll was changing rapidly at that time. Now I found the problem is still there. My code use boost::asio for thread pooling and a lot of context switches are expected. I understand on windows, one should never use spinlock. Could this sudden perf degradation be caused by some change of low-level threading model in libwinptghread?
Expected behavior
spinlock should not take significant time.
Actual behavior
spinlock takes 99% of the time
Verification
Windows Version
MINGW64_NT-10.0-26200
Are you willing to submit a PR?
No response
Description / Steps to reproduce the issue
I found my code became significantly slower back in last October and VTune revealed the majority of time spent is in pthread_spin_lock(). I didn't report right away because libwinpthread.dll was changing rapidly at that time. Now I found the problem is still there. My code use boost::asio for thread pooling and a lot of context switches are expected. I understand on windows, one should never use spinlock. Could this sudden perf degradation be caused by some change of low-level threading model in libwinptghread?
Expected behavior
spinlock should not take significant time.
Actual behavior
spinlock takes 99% of the time
Verification
Windows Version
MINGW64_NT-10.0-26200
Are you willing to submit a PR?
No response