Page 1 of 1

testing_*blas core

Posted: Thu Feb 14, 2013 12:03 am
by marksieklucki
Hi,

I've managed to successfully compile and run many of the provided MAGMA tests.

However, I'm having an issue with the testing_*blas tests. Specifically, the test appears to run okay but them ends up core dumping.

From a quick glance at the source, it looks to me like the actual tests are completing okay, but that we fail somewhere around the cleanup section where we are making all the calls to magma_free.

An excerpt of the output around the point of interest:

Code: Select all

testing ztrsm
ztrsm( L, N ) diff 0
ztrsm( L, N ) diff 0
ztrsm( L, C ) diff 0
ztrsm( L, C ) diff 0
ztrsm( L, T ) diff 0
ztrsm( L, T ) diff 0
ztrsm( U, N ) diff 0
ztrsm( U, N ) diff 0
ztrsm( U, C ) diff 0
ztrsm( U, C ) diff 0
ztrsm( U, T ) diff 0
ztrsm( U, T ) diff 0
ztrsm( L, N ) diff 0
ztrsm( L, N ) diff 0
ztrsm( L, C ) diff 0
ztrsm( L, C ) diff 0
ztrsm( L, T ) diff 0
ztrsm( L, T ) diff 0
ztrsm( U, N ) diff 0
ztrsm( U, N ) diff 0
ztrsm( U, C ) diff 0
ztrsm( U, C ) diff 0
ztrsm( U, T ) diff 0
ztrsm( U, T ) diff 0
*** glibc detected *** ./testing_zblas: free(): invalid next size (fast): 0x0000000005892680 ***
======= Backtrace: =========
/lib64/libc.so.6[0x3236075916]
/lib64/libc.so.6[0x3236078443]
./testing_zblas[0x408e79]
./testing_zblas[0x40800d]
/lib64/libc.so.6(__libc_start_main+0xfd)[0x323601ecdd]
./testing_zblas[0x401549]
For this test, I am running MAGMA 1.3.0, statically linked, with no source changes.

Can anyone confirm this is an issue? Or is it just me doing something wrong?

Re: testing_*blas core

Posted: Tue Feb 26, 2013 8:01 pm
by Stan Tomov
Hello, I can confirm there is an issue. There is a bug related to the error checking, and in particular, the size of a pivot array piv in

Code: Select all

err = magma_malloc_cpu( (void**) &piv, N*sizeof(magma_int_t) );  assert( err == 0 );
should have been Ak, as in

Code: Select all

err = magma_malloc_cpu( (void**) &piv, Ak*sizeof(magma_int_t) );  assert( err == 0 );
We will check if there are other problems, but at least this change fixed the problem that I managed to reproduce.