On Fri, Jan 27, 2012 at 11:52 AM, Ian Lance Taylor <iant@xxxxxxxxxx> wrote:
> David Mebane <mebane@xxxxxxxxxx> writes:
>
>> The following code extest.cpp:
>>
>> extern "C" { int ilaenv_( int *ispec, const char *name, const char
>> *uplo, int *N, int *N2, int *N3, int *N4 ); }
>>
>> int main ()
>> {
>> int ispec = 1;
>> const char* name = 'DSYTRD';
>> const char uplo = 'U';
>> int N = 4;
>> int N2 = -1;
>> int N3 = -1;
>> int N4 = -1;
>> int NB;
>>
>> NB = ilaenv_( &ispec, name, &uplo, &N, &N2, &N3, &N4 );
>>
>> return 0;
>>
>> }
>>
>> compiled with:
>>
>> g++ -ggdb -L/.../lapack/ extest.cpp -llapack -lblas -lgfortran
>>
>> using g++ 4.4.1, when run under gdb, segfaults at the external
>> function call, giving:
>>
>> Program received signal SIGSEGV, Segmentation fault.
>> _wordcopy_bwd_aligned (dstp=140737353957304, srcp=-130065848,
>> len=2305843009196895680) at wordcopy.c:282
>>
>> the backtrace yields:
>>
>> #0 _wordcopy_bwd_aligned (dstp=140737353957304, srcp=-130065848,
>> len=2305843009196895680) at wordcopy.c:282
>> #1 0x00007ffff709ccaa in memmove (dest=0x7fffffffd1f0, src=<value
>> optimized out>, len=18446744073575165448) at memmove.c:99
>> #2 0x000000000040f7c1 in ilaenv_ ()
>> #3 0x00007fffffffda90 in ?? ()
>> #4 0x00007fffffffdbf0 in ?? ()
>> #5 0x0000000000000002 in ?? ()
>> #6 0x00007fffffffdbf0 in ?? ()
>> #7 0x00007fffffffd25f in ?? ()
>> #8 0x00007fffffffda60 in ?? ()
>> #9 0x0000000000000000 in ?? ()
>>
>> I have only a faint idea of what wordcopy.c does, but I'm guessing
>> that the length parameter here is perhaps a bit large. I thought I'd
>> post here in case there's a known issue with wordcopy.c.
>
> Forget wordcopy.c. It's just a subroutine of memmove, which is a
> standard function. The problem here is that memmove is being called
> with a very very large length. It's impossible for us to tell you
> anything about why that is happening, because you haven't shown us the
> source code for ilaenv_, which is the function that is calling memmove.
>
> Ian
So ilaenv_ is a LAPACK routine, written and compiled with
FORTRAN, the source code for which is here:
http://www.netlib.org/lapack/explore-html/d8/d6d/_s_r_c_2ilaenv_8f_source.html
A colleague reports that he cannot replicate the problem using gcc
4.6.1. I suppose this more or less solves the issue for me.
Thanks,
-David
