OpenMP strange behaviour

Question

Hello I have the following code, which I compile with gcc (>4.2) with -fopenmp flag:

int main(void)
{
#pragma omp parallel for
    int i; 
    for(i=0;i<4;i++) while(1);

    return 0;
}

I get a SIGSEGV on OSX Lion (ver 1.7.3, llvm-gcc 4.2.1) and CentOS 6.2 . What am I doing wrong here? Thanks

I get the same, Win7/cygwin, gcc 4.5.0. I ran it though gdb: it creates the threads and then I get `Program received signal SIGSEGV, Segmentation fault. 0x63602726 in omp_get_max_active_levels ()`. It runs fine without the `while(1)`. How is OpenMP treating that infinite loop? — Steve Blackwell, Feb 06 '12 at 16:37
You're using an undeclared variable. But that should generate a compilation error, not a segfault. But with `i` declared I also get a segfault, gcc-4.5.1, openSuSE 11.4. — Daniel Fischer, Feb 06 '12 at 16:41

score 2 · Answer 1 · edited May 23 '17 at 11:48

2

Not sure if this is relevant to the compiler version and configuration but while(true){} terminates

More precisely, if you write a loop which

makes no calls to library I/O functions, and
does not access or modify volatile objects, and
performs no synchronization operations (1.10) or atomic operations (Clause 29)

and does not terminate, you have undefined behaviour.

This may end up not applying to your situation, but as C++11 becomes more established, watch out.

edited May 23 '17 at 11:48

Community

1
1

answered Feb 06 '12 at 18:49

spraff

32,570
22
121
229

Irrelevant, this is C, and in C, the loop may only be assumed to terminate if the controlling expression is **not a constant expression**. Since `while(1);` is controlled by a constant expression, the loop must not terminate. – Daniel Fischer Feb 06 '12 at 19:56
Interesting, I didn't realise C had a not-constant-expression quality like that. Can you point to the standard, please? – spraff Feb 07 '12 at 08:32
Hello spraff, I don't see any undefined behavior in C => while(1); (point to the C standard please). It should be an unconditional jump (jmp) jumping at its own address. If I create 4 threads (with pthread_create) and I put in the start function a "while(1);" I get no SIGSEGV, and my 4 cores are at 100%. I wanted to achieve the same behavior with openmp, but it seems it's not that reliable as I thought. – sfa Feb 07 '12 at 09:35
2

I'm going by n1570 since I don't have the actual standard, in 6.8.5 (6): "An iteration statement whose controlling expression is not a constant expression,that performs no input/output operations, ...". It's possible that the 'not a constant expression' qualifier has been removed again, but since it has been added later to the paragraph, I don't expect that. – Daniel Fischer Feb 07 '12 at 10:10

score 1 · Answer 2 · answered Feb 06 '12 at 17:24

1

Very interesting.

I changed the code a little so

int main(void)
{
int i;
#pragma omp parallel 
  {
        while(1);
    }
    return 0;
}

and so

inline void func() {
    while (1) ;
}

int main(void)
{
int i;
#pragma omp parallel for 
    for(i=0;i<8;i++) {
        func();
    }
    return 0;
}

And they both worked OK.

answered Feb 06 '12 at 17:24

mikithskegg

806
6
10

I know, I also tried the latter. (also works without the inline). But what do you think is wrong with the code I wrote,(except that is missing a... int i). On my CentOS gomp_loop_static_start() gets called, with parameters like => chunk_size=140737488347560, istart=0x7fffffffe1b0, iend=0xca. The last one is a pointer which gets dereferenced.. hence getting the SIGSEGV. If I explicitly use schedule() to set chunk size to a value, the iend pointer is a valid address.. and I get no SIG 11. Also if I debug your latter code(with the inline fnc) gomp_loop_static_start() doesn't get called at all. – sfa Feb 06 '12 at 21:33
if you compile with optimizations enabled then you also get segfault for `func()` case. – jfs Feb 06 '12 at 23:14
@J.F.Sebastian Not here, with `func()`, it properly does nothing at 390+% CPU, from -O0 to -O3 and -Os. Segfaults with an unadorned `while(1);` on all optimisation levels. gcc is 4.5.1. Very strange. – Daniel Fischer Feb 07 '12 at 00:18
gcc 4.6.1: `-O0` no segfault, `-O1+` -- segfault. – jfs Feb 07 '12 at 00:57
@sfa, I really don't know what happens. Think one should look at the generated assembler code. Latter i will try to use Intel C++ compiler, now i have no time. – mikithskegg Feb 07 '12 at 11:06
I tried compilation with Intel C++ Composer 12.1 and everything WORKS properly !!! – mikithskegg Feb 09 '12 at 21:19
Thanks mikithskegg, the problem was in gcc. – sfa Feb 22 '12 at 11:49

score 1 · Accepted Answer · answered Feb 22 '12 at 11:46

1

There was a bug in the gcc regarding this issue, I reported it and they will provide a fix. Here is the link: GCC bug

answered Feb 22 '12 at 11:46

sfa

101
4

OpenMP strange behaviour

3 Answers3