Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
6977 Discussions

inconsistent results from mkl_dcscmv and mkl_dcsrmv

Kuo-ling_Huang
Beginner
‎06-30-2014 12:15 PM
304 Views

Hello,

I tried to calculate the (sparse) matrix-vector product using mkl_dcscmv and mkl_dcsrmv. However, sometimes they gave me different result. Here is the example I was using:

Let 

M : = [ 1  1  0  0 ], x := [ 1 ], and  sols := [ 1 ]  
         [ 1  0  1  0 ]         [ 1 ]                      [ 1 ]
         [ 1  0 -1  0 ]         [ 1 ]                      [ 1 ]
         [ 0  1  0  0 ]         [ 1 ]                      [ 1 ]
         [ 0  0  1  0 ]                                     [ 1 ].

I tried to calculate

  1. test 1: sols := M x - sols,
  2. test 2: sols := Mx, and
  3. test 3: sols := Mx + sols

 by using either mkl_dcscmv or mkl_dcsrmv. Here are the generated result:

=== Test 1: sols := M x - sols ===
(Mx - sols) using mkl_dcscmv (expected: [1 1 -1 0 0]^T)
sols[0] = 1.000000E+00.
sols[1] = 1.000000E+00.
sols[2] = -1.000000E+00.
sols[3] = 0.000000E+00.
sols[4] = 2.000000E+00.
(Mx - sols) using mkl_dcsrmv (expected: [1 1 -1 0 0]^T)
sols[0] = 1.000000E+00.
sols[1] = 1.000000E+00.
sols[2] = -1.000000E+00.
sols[3] = 0.000000E+00.
sols[4] = 0.000000E+00.

=== Test 2: sols := M x + 0 * sols ===
result 2.1: (Mx) using mkl_dcscmv (expected: [2 2 0 1 1]^T)
sols[0] = 2.000000E+00.
sols[1] = 2.000000E+00.
sols[2] = 0.000000E+00.
sols[3] = 1.000000E+00.
sols[4] = 2.000000E+00.
result 2.2: (Mx) using mkl_dcsrmv (expected: [2 2 0 1 1]^T)
sols[0] = 2.000000E+00.
sols[1] = 2.000000E+00.
sols[2] = 0.000000E+00.
sols[3] = 1.000000E+00.
sols[4] = 1.000000E+00.

=== Test 3: sols := M x + 1 * sols ===
result 3.1: (Mx + sol) using mkl_dcscmv (expected: [3 3 1 2 2]^T)
sols[0] = 3.000000E+00.
sols[1] = 3.000000E+00.
sols[2] = 1.000000E+00.
sols[3] = 2.000000E+00.
sols[4] = 2.000000E+00.
result: 3.2: (Mx + sol) using mkl_dcsrmv (expected: [3 3 1 2 2]^T)
sols[0] = 3.000000E+00.
sols[1] = 3.000000E+00.
sols[2] = 1.000000E+00.
sols[3] = 2.000000E+00.
sols[4] = 2.000000E+00.

===============================

My computational environment:

OS: Red Hat Enterprise Linux Server release 6.4 (Santiago)

Compiler and linker: Intel Composer-XE version 2013.2.146

GNU libc version: 2.12

===============================

Following is the complete code I was using for the test:

//===================

#include <stdlib.h>

#include <stdio.h>
#include <mkl.h>

int main (void)
{
#define NUM_ROWS 5
#define NUM_COLS 4
#define NUM_ENTS 8
    
    MKL_INT num_rows = NUM_ROWS;
    MKL_INT num_cols = NUM_COLS;
    MKL_INT num_ents = NUM_ENTS;
    MKL_INT i;
    
    double minus_one = -1.0;
    double one = +1.0;
    double zero = 0.0;
    char   notran = 'N';
    char   matdescra[4] = {'G', 'L', 'N', 'C'}; 
    
    /* M : = [ 1  1  0  0 ] x := [ 1 ] sols := [ 1 ]  
     *       [ 1  0  1  0 ]      [ 1 ]         [ 1 ]
     *       [ 1  0 -1  0 ]      [ 1 ]         [ 1 ]
     *       [ 0  1  0  0 ]      [ 1 ]         [ 1 ]
     *       [ 0  0  1  0 ]                    [ 1 ]
     * 
     */
    MKL_INT    M_bgn[NUM_COLS + 1] = {0, 3, 5, 8, 8};
    MKL_INT    M_idx[NUM_ENTS]     = {0, 1, 2, 0, 3, 1,  2, 4};
    double     M_val[NUM_ENTS]     = {1, 1, 1, 1, 1, 1, -1, 1};
   
    MKL_INT    MT_bgn[NUM_ROWS + 1] = {0, 2, 4, 6, 7, 8};
    MKL_INT    MT_idx[NUM_ENTS]     = {0, 1, 0, 0, 0, 2,  1, 2};
    double     MT_val[NUM_ENTS]     = {1, 1, 1, 1, 1, -1, 1, 1};
    
    double     sols[NUM_ROWS];
    double     x[NUM_COLS];
    
    /* initialize the solution and x */
#define INIT_SOL                                \
    for (i = 0; i < num_rows; ++i)              \
    {                                           \
        sols = 1.0;                          \
    }                                           \
    for (i = 0; i < num_cols; ++i)              \
    {                                           \
        x = 1.0;                             \
    }
    
#define PRINT_SOL(MSG)                          \
    printf("%s\n", MSG);                        \
    for (i = 0; i < num_rows; ++i)              \
    {                                           \
        printf("sols[%d] = %E.\n", i, sols); \
    }    
    
    /* test 1: compute sols := M x - sols
     *                       = [2 2 0 1 1]^T - [1 1 1 1 1]^T
     *                       = [1 1 -1 0 0]^T.
     */
    printf("\n=== Test 1: sols := M x - sols ===\n");
    INIT_SOL;
    /* test 1.1: using mkl_dcscmv */
    mkl_dcscmv(&notran, &num_rows, &num_cols, &one, matdescra,
        M_val, M_idx, M_bgn, M_bgn + 1, x, &minus_one, sols);     
    PRINT_SOL("(Mx - sols) using mkl_dcscmv (expected: [1 1 -1 0 0]^T)");
 
    INIT_SOL;    
    /* test 1.2: using mkl_dcsrmv */
    mkl_dcsrmv(&notran, &num_rows, &num_cols, &one, matdescra,
        MT_val, MT_idx, MT_bgn, MT_bgn + 1, x, &minus_one, sols);  
    PRINT_SOL("(Mx - sols) using mkl_dcsrmv (expected: [1 1 -1 0 0]^T)");
     
    /* test 2: compute sols := M x + 0 * sols
     *                       = [2 2 0 1 1]^T - 0 * [1 1 1 1 1]^T
     *                       = [2 2 0 1 1]^T.
     */
    printf("\n=== Test 2: sols := M x + 0 * sols ===\n");
    INIT_SOL;    
    /* test 2.1: using mkl_dcscmv */
    mkl_dcscmv(&notran, &num_rows, &num_cols, &one, matdescra, 
        M_val, M_idx, M_bgn, M_bgn + 1, x, &zero, sols);     
    PRINT_SOL("result 2.1: (Mx) using mkl_dcscmv (expected: [2 2 0 1 1]^T)");
    
    INIT_SOL;  
    /* test 2.2: using mkl_dcsrmv */
    mkl_dcsrmv(&notran, &num_rows, &num_cols, &one, matdescra,
        MT_val, MT_idx, MT_bgn, MT_bgn + 1, x, &zero, sols);  
    PRINT_SOL("result 2.2: (Mx) using mkl_dcsrmv (expected: [2 2 0 1 1]^T)");    
         
    /* test 3: compute sols := M x + 1 * sols
     *                       = [2 2 0 1 1]^T + 1 * [1 1 1 1 1]^T
     *                       = [3 3 1 2 2]^T.
     */ 
    printf("\n=== Test 3: sols := M x + 1 * sols ===\n");
    INIT_SOL;    
    /* test 3.1: using mkl_dcscmv */
    mkl_dcscmv(&notran, &num_rows, &num_cols, &one, matdescra,
        M_val, M_idx, M_bgn, M_bgn + 1, x, &one, sols);     
    PRINT_SOL("result 3.1: (Mx + sol) using mkl_dcscmv (expected: [3 3 1 2 2]^T)");
  
    INIT_SOL;  
    /* test 3.2: using mkl_dcsrmv */    
    mkl_dcsrmv(&notran, &num_rows, &num_cols, &one, matdescra,
        MT_val, MT_idx, MT_bgn, MT_bgn + 1, x, &one, sols);  
    PRINT_SOL("result: 3.2: (Mx + sol) using mkl_dcsrmv (expected: [3 3 1 2 2]^T)");    
 
    return 0;
}

//===================

0 Kudos

Link Copied

1 Reply
Ying_H_Intel
Employee
‎06-30-2014 07:09 PM
304 Views

Hello, 

I have only Composer XE 2013.5 (MKL 11.0.5)  installed, just try the code, it seems run fine.  It may be a bug in composer XE 2013.2 ( MKL 11.0.2) .  See https://software.intel.com/en-us/articles/intel-mkl-110-bug-fixes

Please upgrade the MKL version if possible ( the latest one : MKL 11.1. 3 ). 

Best Regards,

Ying 

$source /opt/intel/composer_xe_2013.5.192/mkl/bin/mklvars.sh intel64
$ gcc dcsrmv.c -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lm -lpthread
$ ./a.out

=== Test 1: sols := M x - sols ===
(Mx - sols) using mkl_dcscmv (expected: [1 1 -1 0 0]^T)
sols[0] = 1.000000E+00.
sols[1] = 1.000000E+00.
sols[2] = -1.000000E+00.
sols[3] = 0.000000E+00.
sols[4] = 0.000000E+00.
(Mx - sols) using mkl_dcsrmv (expected: [1 1 -1 0 0]^T)
sols[0] = 1.000000E+00.
sols[1] = 1.000000E+00.
sols[2] = -1.000000E+00.
sols[3] = 0.000000E+00.
sols[4] = 0.000000E+00.

=== Test 2: sols := M x + 0 * sols ===
result 2.1: (Mx) using mkl_dcscmv (expected: [2 2 0 1 1]^T)
sols[0] = 2.000000E+00.
sols[1] = 2.000000E+00.
sols[2] = 0.000000E+00.
sols[3] = 1.000000E+00.
sols[4] = 1.000000E+00.
result 2.2: (Mx) using mkl_dcsrmv (expected: [2 2 0 1 1]^T)
sols[0] = 2.000000E+00.
sols[1] = 2.000000E+00.
sols[2] = 0.000000E+00.
sols[3] = 1.000000E+00.
sols[4] = 1.000000E+00.

=== Test 3: sols := M x + 1 * sols ===
result 3.1: (Mx + sol) using mkl_dcscmv (expected: [3 3 1 2 2]^T)
sols[0] = 3.000000E+00.
sols[1] = 3.000000E+00.
sols[2] = 1.000000E+00.
sols[3] = 2.000000E+00.
sols[4] = 2.000000E+00.
result: 3.2: (Mx + sol) using mkl_dcsrmv (expected: [3 3 1 2 2]^T)
sols[0] = 3.000000E+00.
sols[1] = 3.000000E+00.
sols[2] = 1.000000E+00.
sols[3] = 2.000000E+00.
sols[4] = 2.000000E+00.

 

0 Kudos
Copy link
Reply

Community support is provided during standard business hours (Monday to Friday 7AM - 5PM PST). Other contact methods are available here.

Intel does not verify all solutions, including but not limited to any file transfers that may appear in this community. Accordingly, Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

For more complete information about compiler optimizations, see our Optimization Notice.