topic zfeast_hrci initialization shows strange behaviour in Intel® oneAPI Math Kernel Library

zfeast_hrci initialization shows strange behaviour

Christian_M_3 — Sun, 27 Mar 2016 10:42:45 GMT

Hi all,

i am working my way through the example zexample_sparse_rci_c.c. I have it working now for my matrix, BUT found a strange behaviour for the input arrays Aq and Sq:

when declaring them like this:

MKL_Complex16 Aq[M0SQUARE]; (where MOSquare is an integer) everything works fine.

When declaring them like

MKL_Complex16 *Sq = new MKL_Complex16[M0SQUARE];

the first call to zfeast_hrci fails with:

Thread #1 [feastTrial] 5489 [core: 2] (Suspended : Signal : SIGSEGV:Segmentation fault)
   mkl_feast_zfeast_hrci() at 0x7ffff617fedf
   zfeast_hrci_() at 0x7ffff7959141
   testAddition() at feastTrial.cpp:251 0x4047da
   0x0

. Strangely enough: when declaring workc and work with new .... no problems occur.

What am I missing here?

I want to have all memory on the heap in order to cope with LARGE matrices.

Hi Christian,

Ying_H_Intel — Mon, 28 Mar 2016 07:47:55 GMT

Hi Christian,

how do you declare the M0SQUARE? Could you please provide us one test case?

I use the zexample_sparse_rci_c.c -> covert to CPP file. with MKL_Complex16 *Sq = new MKL_Complex16[M0SQUARE];

It seems work fine. I attached the cpp code for your reference.

Best Regards,

Ying

Hello Ying,

Christian_M_3 — Mon, 28 Mar 2016 10:00:54 GMT

Hello Ying,

MKL_INT M0;

   if (dimension < 500)
   {
       M0=dimension*0.5;
   }
   else
   {
       M0=dimension*0.1;
   }
   int workSize = M0*dimension;
   MKL_INT M0StartValue = M0;
   MKL_INT M0SQUARE = M0*M0;

The values in my testcase are dimension=495, M0= 247 and so on.

One more thing I forgot to mention:

I use the gcc compiler, (4.8.4) on ubuntu linux.

Thanks for your help.

Christian

OK. It seems to me that

Christian_M_3 — Mon, 28 Mar 2016 17:31:23 GMT

OK. It seems to me that cannot mix the declaration/allocation types with my version of gcc

(i.e. having

MKL_Complex16 Aq[M0SQUARE]; (where MOSquare is an integer) everything works fine.

and

MKL_Complex16 *Sq = new MKL_Complex16[M0SQUARE];

in the program

)

replacing all declarations with new [] helped.

BUT now comes the next problem:

When trying to delete those arrays, the program crashes when trying to delete [] Sq; claiming the pointer would be invalid.

Any ideas why this happens? All other pointers can be deleted properly.

Hi Christian,

Ying_H_Intel — Tue, 29 Mar 2016 08:36:40 GMT

Hi Christian,

is there any special compiler option in your CPP code? or is there any special code (for example, delete [] Sq in while loop) in your CPP code?

I add

printf( "Max of (conjugate transposed of X)*X-I %.15e \n", smax);

delete [] Sq;

return 0;
}

tried the command;

$source /opt/intel/compilers_and_libraries_2016.2.181/linux/mkl/bin/mklvars.sh intel64

$g++ zexample_sparse_rci_c.cpp -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lm

hu5@ivb01-ub:~/Merry$ ./a.out

FEAST ZFEAST_HRCI EXAMPLE PROGRAM
Sparse matrix size 10
Search interval [ 2.000000000000000e+00, 1.200000000000000e+01 ]
Intel MKL Extended Eigensolvers: complex double precision driver
Intel MKL Extended Eigensolvers: List of input parameters fpm(1:64)-- if different from default
Intel MKL Extended Eigensolvers: fpm(1)=1
Intel MKL Extended Eigensolvers: fpm(6)=1

It run fines.

yhu5@ivb01-ub:~/Merry$ g++ --version
g++ (Ubuntu/Linaro 4.8.1-10ubuntu9) 4.8.1.

Could you provide us a test main cpp file, so we can debug at our sides?

Regards,

Ying