Permutation vector for dgetrf

Jason_Jones1 — Mon, 15 Mar 2010 21:09:03 GMT

I am having trouble with the output ipiv from dgetrf. My understanding is that it should return an array the length of the number of rows of my input matrix. Each member of this array should contain a different number. These numbers are the column in which a 1 would be placed for the corresponding column i.e. if:

ipiv =

This would correspond to a permutation matrix that looks like:

1 0 0 0 0

0 0 0 1 0

0 0 0 0 1

0 1 0 0 0

0 0 1 0 0

And that this permutation matrix P*X(my input matrix) = L*U. However, the ipiv outputs I get from dgetrf have many repetititions such as the output below and P*X != L*U:

ipiv = 4
ipiv = 2
ipiv = 4
ipiv = 4
ipiv = 5

Here is my example program:

#define M 5
#define N 5
int main()
{

int i,m, n, *ipiv, info;
double a[M*N] = { 3, 4, 6, 8, 2, 2, 9, 6, 1, 8, 12, 8, 4, 6, 2, 6, 6, 7, 1, 2, 9, 4, 3, 2, 8 };
ipiv = malloc(M*sizeof(int));
m = M;
n = N;
info = 0;

dgetrf(&m,&n,a,&m,ipiv,&info);
printf("info = %d\\n",info);

for (i=0;i{
printf("ipiv = %d\\n",ipiv);
}
return(0);
}

Thanks!

Permutation vector for dgetrf

TimP — Wed, 17 Mar 2010 17:56:52 GMT

If you wish to understand how this works, you will need to refer to the public netlib source code. I haven't looked for a while, but I believe your output indicates that the pivoting went like:
swap rows 1 and 4
continue processing
swap rows 3 and 4 (which was originally row 1)
continue...
The swap vector is intended to be used only by the companion BLAS functions.

Permutation vector for dgetrf

Jason_Jones1 — Wed, 17 Mar 2010 23:20:25 GMT

Yes, that works. Thank you!

topic Permutation vector for dgetrf in Intel® oneAPI Math Kernel Library

Permutation vector for dgetrf

Permutation vector for dgetrf

Permutation vector for dgetrf