[llvm-bugs] [Bug 32840] New: NONDETERMINIZM: It seems, ThreadSanitizer does not understand OpenMP's critical section in C++

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=32840

            Bug ID: 32840
           Summary: NONDETERMINIZM: It seems, ThreadSanitizer does not
                    understand OpenMP's critical section in C++
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: lebedev.ri at gmail.com
                CC: llvm-bugs at lists.llvm.org

Example one: (good)

#include <stdio.h>

int main() {
#ifdef _OPENMP
#pragma omp parallel num_threads(2)
#endif
#ifdef _OPENMP
#pragma omp critical(io)
#endif
  {
    printf("thread\n");
  }

  return 0;
}

$ clang-5.0 openmp-critical.c -fsanitize=thread
$ ./a.out 
thread

As you can see, TSan does not complain, and there were only one print.
Let's enable OpenMP.

$ clang-5.0 openmp-critical.c -fsanitize=thread -fopenmp
$ ./a.out 
thread
thread

As you can see, TSan does not complain, and there were two prints.
Now, let's check C++.

Example two (bad):

#include <cstdio>
#include <iostream>

int main() {
#ifdef _OPENMP
#pragma omp parallel num_threads(2)
#endif
#ifdef _OPENMP
#pragma omp critical(io)
#endif
  {
    std::cout << "thread\n";
  }

  return 0;
}

$ clang++-5.0 openmp-critical.cpp -fsanitize=thread 
$ ./a.out 
thread

Again all fine. Let's enable OpenMP.

$ clang++-5.0 openmp-critical.cpp -fsanitize=thread -fopenmp
$ ./a.out 
==================
WARNING: ThreadSanitizer: data race (pid=30729)
  Atomic read of size 1 at 0x7b6800000b00 by main thread:
    #0 pthread_mutex_lock <null> (a.out+0x432067)
    #1 <null> <null> (libomp.so.5+0x477a2)
    #2 __libc_start_main <null> (libc.so.6+0x202b0)

  Previous write of size 1 at 0x7b6800000b00 by thread T2:
    #0 pthread_mutex_init <null> (a.out+0x42ae0a)
    #1 <null> <null> (libomp.so.5+0x44c51)

  Location is heap block of size 1440 at 0x7b6800000600 allocated by main
thread:
    #0 malloc <null> (a.out+0x4296dd)
    #1 getenv <null> (libomp.so.5+0x18202)
    #2 __libc_start_main <null> (libc.so.6+0x202b0)

  Thread T2 (tid=30732, running) created by main thread at:
    #0 pthread_create <null> (a.out+0x42ab26)
    #1 <null> <null> (libomp.so.5+0x458bb)
    #2 __libc_start_main <null> (libc.so.6+0x202b0)

SUMMARY: ThreadSanitizer: data race (/tmp/a.out+0x432067) in pthread_mutex_lock
==================
thread
thread
ThreadSanitizer: reported 1 warnings

I'd expect that it should not complain. 

And now the weirdest part: it only happens if the system is not in idle, i.e.
if there are some other computational/compilation processes running!

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170428/79608590/attachment.html>