abort openmp application

Dear All,

At MacOS X 10.7 Lion sudden aborts of our Huygens application was encountered by various customers. After investigation of the the log reports it appeared to happen at OpenMP pragmas and especially at rapid close-reopens of the OpenMP team. This abort behavior could also be fond with the following small test application where the OpenMP pragmas are used from within a loop, simulating in his way the rapid close-reopens of OpenMP:

#include <stdlib.h>

#include <stdio.h>

#include <omp.h>

#define NR_RUNS 1000

#define NR_SUMS 20

int main (int argc, char *argv[]) {

int runIndex, sumIndex;

int sum;

for (runIndex = 0; runIndex < NR_RUNS; runIndex++) {

sum = 0;

#pragma omp parallel for reduction(+:sum) shared(runIndex) private(sumIndex)

for (sumIndex = 1; sumIndex < NR_SUMS; sumIndex++) {

sum += sumIndex;

printf("Run: %d; thread: %d; index: %d; sum: %d\n",

runIndex, omp_get_thread_num(), sumIndex, sum);

fflush(stdout);

}

printf ("Final sum: %d\n", sum);

fflush(stdout);

}

return 0;

}

This application was compiled with the default GCC version for Lion: i686-apple-darwin11-llvm-gcc-4.2 (GCC) 4.2.1.

A typical log of such a crash is:

Parent Process: bash [850]

Report Version: 9

Crashed Thread: 0 Dispatch queue: com.apple.main-thread

Exception Type: EXC_CRASH (SIGABRT)

Exception Codes: 0x0000000000000000, 0x0000000000000000

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread

Thread 1:

Thread 2:

Thread 3:

Thread 0 crashed with X86 Thread State (64-bit):

rax: 0x0000000000000000 rbx: 0x00007fff71f77960 rcx: 0x00007fff6c15eb28 rdx: 0x0000000000000000

rdi: 0x00007fefe04000f8 rsi: 0x0000000000000300 rbp: 0x00007fff6c15ebe0 rsp: 0x00007fff6c15eb28

r8: 0x0000000000000000 r9: 0x0000000000000060 r10: 0x0000000000000000 r11: 0x0000000000000202

r12: 0x0000000000000000 r13: 0x0000000000000300 r14: 0x00007fefe0400118 r15: 0x00007fefe0400110

rip: 0x00007fff83362bca rfl: 0x0000000000000202 cr2: 0x000000010bc3a000

Logical CPU: 0

Binary Images:

External Modification Summary:

Calls made by other processes targeting this process:

Calls made by this process:

Calls made by all processes on this machine:

VM Region Summary:

ReadOnly portion of Libraries: Total=50.0M resident=28.2M(56%) swapped_out_or_unallocated=21.8M(44%)

Writable regions: Total=21.9M written=96K(0%) resident=180K(1%) swapped_out=0K(0%) unallocated=21.8M(99%)

From the crash log it appears that the abort took place at the moment there is waited for the worker threads to finish. This typically always is the case. However, not all runs of this program will lead to an abort.

These abort crashes where not found at previous MacOS X versions (Leopard, Tiger), and also not at Linux and Windows.

Could you please indicate what could be the problem here?

Many thanks in advance,

Kind regards,

Hans Blom,

Scientific Volume Imaging