[Openmp-dev] orph_test_omp_parallel_copyin segfault

[Jack Howarth via Openmp-dev]

Jonathan,
      I am having trouble reproducing it outside of the
OpenMP3.1_Validation Makefile framework. However, when executing
'make clean' and 'make ctest' using clang 3.8svn built against Xcode
7.2 on OS X 10.11.2, I have determined that ...

CFLAGS = -fopenmp -lm -Os
CFLAGS = -fopenmp -lm -O3
CFLAGS = -fopenmp -lm -O2

all will trigger the segfault in the resulting
orph_test_omp_parallel_copyin test case whereas...

CFLAGS = -fopenmp -lm -O1
CFLAGS = -fopenmp -lm -O0
CFLAGS = -fopenmp -lm -Os -g
CFLAGS = -fopenmp -lm -O3 -g
CFLAGS = -fopenmp -lm -O2 -g
CFLAGS = -fopenmp -lm -O1 -g
CFLAGS = -fopenmp -lm -O0 -g

all suppress the problem. See
https://llvm.org/bugs/show_bug.cgi?id=25820#c6 for a backtrace of the
segfaulting orph_test_omp_parallel_copyin test case built at -fopenmp
-O3.

         Jack
ps This problem doesn't exist when using clang 3.7.0 built under Xcode
7.2 on OS X 10.11.2 so it is a new regression in clang 3.8svn.

On Tue, Dec 15, 2015 at 1:10 PM, Peyton, Jonathan L
<jonathan.l.peyton at intel.com> wrote:
> Jack,
>
> I wasn't able to replicate it at first go.  Can you tell me the OSX version, and perhaps the exact revision numbers for clang/llvm?
>
> -- Johnny
>
> -----Original Message-----
> From: Jack Howarth [mailto:howarth.mailing.lists at gmail.com]
> Sent: Friday, December 11, 2015 8:14 PM
> To: Peyton, Jonathan L
> Cc: openmp-dev (openmp-dev at lists.llvm.org)
> Subject: Re: [Openmp-dev] orph_test_omp_parallel_copyin segfault
>
> Jonathan,
>       The attached header omp_testsuite.h and orph_test_omp_parallel_copyin.c source file are required to reproduce the crash at any optimization level using -fopenmp.
>                  Jack
>
> On Fri, Dec 11, 2015 at 6:12 PM, Peyton, Jonathan L <jonathan.l.peyton at intel.com> wrote:
>> Are you using trunk versions of both clang and libomp?  I only use the check-libomp target now in CMake and don't see any problems with the trunk version of libomp on Darwin.  Can you send me the source file for orph_test_omp_parallel_copyin.c and I'll take a look at it.
>>
>> -- Johnny
>>
>> -----Original Message-----
>> From: Openmp-dev [mailto:openmp-dev-bounces at lists.llvm.org] On Behalf
>> Of Jack Howarth via Openmp-dev
>> Sent: Friday, December 11, 2015 4:17 PM
>> To: openmp-dev (openmp-dev at lists.llvm.org)
>> Subject: [Openmp-dev] orph_test_omp_parallel_copyin segfault
>>
>> Is anyone else seeing segfaults in the orph_test_omp_parallel_copyin
>> test case from the OpenMP3.1_Validation test suite? On
>> x86_64-apple-darwin15, todays trunk produces,,,
>>
>> Jacks-Mac-Pro:c howarth$ clang-3.8 -fopenmp -O0 -g
>> orph_test_omp_parallel_copyin.c -o orph_test_omp_parallel_copyin
>> Jacks-Mac-Pro:c howarth$ lldb ./orph_test_omp_parallel_copyin
>> (lldb) target create "./orph_test_omp_parallel_copyin"
>> Current executable set to './orph_test_omp_parallel_copyin' (x86_64).
>> (lldb) r
>> Process 38235 launched: './orph_test_omp_parallel_copyin' (x86_64) ######## OpenMP Validation Suite V 3.0a ######
>> ## Repetitions:  20                       ####
>> ## Loop Count :   1000                    ####
>> ##############################################
>> Testing omp parallel copyin
>>
>> Process 38235 stopped
>> * thread #1: tid = 0x17a2d4, 0x00007fff9ae780a5 libsystem_c.dylib`flockfile + 4, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x68)
>>     frame #0: 0x00007fff9ae780a5 libsystem_c.dylib`flockfile + 4
>> libsystem_c.dylib`flockfile:
>> ->  0x7fff9ae780a5 <+4>:  movq   0x68(%rdi), %rdi
>>     0x7fff9ae780a9 <+8>:  addq   $0x8, %rdi
>>     0x7fff9ae780ad <+12>: popq   %rbp
>>     0x7fff9ae780ae <+13>: jmp    0x7fff9aec1452            ; symbol
>> stub for: pthread_mutex_lock
>>
>> which is the only regression observed in the ctest run of the test suite.
>> _______________________________________________
>> Openmp-dev mailing list
>> Openmp-dev at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/openmp-dev