为LAPACKE格式化带状矩阵

时间:2016-03-31 22:44:07

标签: c matrix lapack intel-mkl lapacke

我正在尝试使用英特尔MKL中名为LAPACKE的LAPACK的C接口来解决一般的带状矩阵。我试图调用的函数是*gbsv,其中*表示格式。不幸的是,我发现非常很难找到关于如何使用C接口格式化带状矩阵的工作示例。如果有人可以为所有C用户提供一个工作示例,我向您保证会有所帮助。

fortran布局作为示例here给出,但我不确定如何将其格式化以输入LAPACKE。我还应该注意,在我的问题中,我必须动态构建带状矩阵。所以我有每个i节点有5个系数A,B,C,D,E,它们必须被放入带状矩阵形式,然后传递给LAPACKE。

1 个答案:

答案 0 :(得分:2)

函数LAPACKE_dgbsv()的原型如下:

lapack_int LAPACKE_dgbsv( int matrix_layout, lapack_int n, lapack_int kl,
                      lapack_int ku, lapack_int nrhs, double* ab,
                      lapack_int ldab, lapack_int* ipiv, double* b,
                      lapack_int ldb )

与Lapack函数dgbsv()的主要区别在于参数matrix_layout,可以是LAPACK_ROW_MAJOR(C排序)或LAPACK_COL_MAJOR(Fortran排序)。如果LAPACK_ROW_MAJORLAPACKE_dgbsv将转置矩阵,请调用dgbsv(),然后将矩阵转置回C排序。

其他参数的含义与函数dgbsv()的含义相同。如果使用LAPACK_ROW_MAJOR,那么ldab的正确dgbsv()将由LAPACKE_dgbsv()计算,参数ldab可以设置为n。但是,就像dgbsv()一样,必须为矩阵ab分配额外空间以存储分解的细节。

以下示例利用LAPACKE_dgbsv()通过居中有限差分来解决一维平台扩散问题。考虑零温度边界条件,并使用正弦波之一作为检查正确性的源项。以下程序由gcc main3.c -o main3 -llapacke -llapack -lblas -Wall编译:

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <time.h>

#include <lapacke.h>

int main(void){

    srand (time(NULL));

    //size of the matrix
    int n=10;
    // number of right-hand size
    int nrhs=4;

    int ku=2;
    int kl=2;
    // ldab is larger than the number of bands, 
    // to store the details of factorization
    int ldab = 2*kl+ku+1;

    //memory initialization
    double *a=malloc(n*ldab*sizeof(double));
    if(a==NULL){fprintf(stderr,"malloc failed\n");exit(1);}

    double *b=malloc(n*nrhs*sizeof(double));
    if(b==NULL){fprintf(stderr,"malloc failed\n");exit(1);}

    int *ipiv=malloc(n*sizeof(int));
    if(ipiv==NULL){fprintf(stderr,"malloc failed\n");exit(1);}

    int i,j;

    double fact=1*((n+1.)*(n+1.));
    //matrix initialization : the different bands
    // are stored in rows kl <= j< 2kl+ku+1
    for(i=0;i<n;i++){
        a[(0+kl)*n+i]=0;
        a[(1+kl)*n+i]=-1*fact;
        a[(2+kl)*n+i]=2*fact;
        a[(3+kl)*n+i]=-1*fact;
        a[(4+kl)*n+i]=0;

        //initialize source terms 
        for(j=0;j<nrhs;j++){
            b[i*nrhs+j]=sin(M_PI*(i+1)/(n+1.));
        }
    }
    printf("end ini \n");

    int ierr;


    // ROW_MAJOR is C order, Lapacke will compute ldab by himself.
    ierr=LAPACKE_dgbsv(LAPACK_ROW_MAJOR, n, kl,ku,nrhs, a,n, ipiv,  b,nrhs );


    if(ierr<0){LAPACKE_xerbla( "LAPACKE_dgbsv", ierr );}

    printf("output of LAPACKE_dgbsv\n");
    for(i=0;i<n;i++){
        for(j=0;j<nrhs;j++){
            printf("%g ",b[i*nrhs+j]);
        }
        printf("\n");
    }

    //checking correctness
    double norm=0;
    double diffnorm=0;
    for(i=0;i<n;i++){
        for(j=0;j<nrhs;j++){
            norm+=b[i*nrhs+j]*b[i*nrhs+j];
            diffnorm+=(b[i*nrhs+j]-1./(M_PI*M_PI)*sin(M_PI*(i+1)/(n+1.)))*(b[i*nrhs+j]-1./(M_PI*M_PI)*sin(M_PI*(i+1)/(n+1.)));
        }
    }
    printf("analical solution is 1/(PI*PI)*sin(x)\n");
    printf("relative difference is %g\n",sqrt(diffnorm/norm));


    free(a);
    free(b);
    free(ipiv);

    return 0;
}