topic Hi caden, in Intel® oneAPI Math Kernel Library

I get an error in MKL function vsldCorrExec, why?

caden_g_ — Mon, 19 Sep 2016 10:02:26 GMT

I want to get

array h[16]={16,15,14,13,

12,11,10,9,

8,7,6,5,

4,3,2,1,},

but I get an error:

line 41:"status = vsldCorrExec(task, f, f_stride, g, g_stride, h, h_stride);" status is not VSL_STATUS_OK,it is VSL_CC_ERROR_DECIMATION,so how to set the decimation with vslCorrSetDecimation?

If I delete "status = vslCorrSetStart(task, h_start);" in code, it will work normally, and

array h[16]={0,0,0,0,

0,16,15,14,

0,12,11,10,

0,8,7,6}.

What can I do?

#include <assert.h>
#include <mkl.h>
#include <iostream>
using namespace std;  

int main()
{
   //input
   const double f[16] = { 1 , 2 , 3 , 4,
                          5 , 6 , 7 , 8,
                          9 , 10, 11, 12,
                          13, 14, 15, 16,};
   //kernel, a simple example
   const double g[9] = { 0, 0, 0,
                         0, 1, 0,
                         0, 0, 0 };
   //output
   double h[16] = { 0 };

   VSLCorrTaskPtr task;

   MKL_INT f_shape[2] = { 4, 4 };
   MKL_INT g_shape[2] = { 3, 3 };

   MKL_INT h_shape[2] = { 4, 4 };
   MKL_INT h_start[2] = { 1, 1 };

   MKL_INT f_stride[2] = { f_shape[1], 1 };
   MKL_INT g_stride[2] = { g_shape[1], 1 };
   MKL_INT h_stride[2] = { h_shape[1], 1 };

   int status;

   status = vsldCorrNewTask(&task, VSL_CORR_MODE_DIRECT, 2, f_shape, g_shape, h_shape);
   assert(status == VSL_STATUS_OK);

   status = vslCorrSetStart(task, h_start);
   assert(status == VSL_STATUS_OK);

   //always get wrong,return -2303(VSL_CC_ERROR_DECIMATION) I can't find and solve the problem
   status = vsldCorrExec(task, f, f_stride, g, g_stride, h, h_stride);
   assert(status == VSL_STATUS_OK);

   status = vslCorrDeleteTask(&task);
   assert(status == VSL_STATUS_OK);

   //print the result 
   for (int r = 0; r < 4; r++)
   {
      cout << r << "(output): ";
      for (int c = 0; c < 4; c++)
      {
         cout << h[r * 4 + c] << " ";
      }
      cout << endl;
    }
   return 0;
}

Hi caden,

Zhen_Z_Intel — Tue, 20 Sep 2016 07:29:44 GMT

Hi caden,

It must belong to the interval R_n^min≤ start(n) ≤ R_n^max. And while you setStart, the R_n^min will be replaced by start(n). And the output(n) should be equals to start(n) + (kn-1) that 1 ≤ kn≤ xshape(n).

In your program, xshape[0]&xshape[1]=4,
=> the R_n^min=-3

If you set size of zshape=4*4, that means you could move from left-top side to right-bottom side by 2 steps, thus, R_n^max=-3+2=-1

int h_start[2]={-3,-3};

And the output should be same as without using SetStart function. If you set zshape to 3*3, that means, -3≤ start(n) ≤0, that R_n^max=0.

Best regards,
Fiona

Quote:Fiona Z. (Intel) wrote:

caden_g_ — Wed, 21 Sep 2016 01:40:48 GMT

Fiona Z. (Intel) wrote:

Hi caden,

It must belong to the interval R_n^min≤ start(n) ≤ R_n^max. And while you setStart, the R_n^min will be replaced by start(n). And the output(n) should be equals to start(n) + (kn-1) that 1 ≤ kn≤ xshape(n).

In your program, xshape[0]&xshape[1]=4,
=> the R_n^min=-3

If you set size of zshape=4*4, that means you could move from left-top side to right-bottom side by 2 steps, thus, R_n^max=-3+2=-1
int h_start[2]={-3,-3};
And the output should be same as without using SetStart function. If you set zshape to 3*3, that means, -3≤ start(n) ≤0, that R_n^max=0.

Best regards,
Fiona

thank you,Fiona.

According to your suggestion,I solve this problem.But if I write "int h_start[2]={-3,-3};",the result is wrong,too.

the array h[16] is {0,0,0,0,

0,16,15,14,

0,12,11,10,

0,8,7,6}.

Instead of

int h_start[2]={-2,-2};

then it will be right, the array h[16] is

{16,15,14,13,

12,11,10,9,

8,7,6,5,

4,3,2,1,},

Hi caden,

Zhen_Z_Intel — Wed, 21 Sep 2016 06:27:51 GMT

Hi caden,

I just thought you want to print left-top corner of result matrix. The scale of start(n) is between -3~-1 in this program, if you set start(n) out of range. There will be an error. Anyway, you could fix program and get result what you want.

Best regards,
Fiona