1000*1000 matrices A, B, c,c=a*b

Efficiency can be compared with MATLAB

int m;
int i, j, k;
double r;
double *A, *B, *C;
m = 1000;
A = new double[m*m];
B = new double[m*m];
C = new double[m*m];
//置初值
for (i=0; i<m*m; i++) {
A[i] = 1;
B[i] = 1;
C[i] = 0;
}
int bf; // blocking factor
int jj, kk, im;
int minj, mink;
bf = 48; // 可以修改
for (jj=0; jj<m; jj+=bf)
for (kk=0; kk<m; kk+=bf)
for ( i=0; i<m; ++i) {
minj = (jj+bf)<m ? (jj+bf):m;
for (j=jj; j<minj; ++j) {
r = 0;
im = i*m;
mink = (kk+bf)<m ? (kk+bf):m;
for (k=kk; k<mink; ++k) {
r += A[im+k]*B[k*m+j];
}
C[im+j] += r;
}
}
delete[] A;
delete[] B;
delete[] C;

This program is 1 time times slower than Matlab, I have to search the Web to a FORTRAN program (Block DGEMM), its speed is 0.5 times times slower than Matlab, do not know who can write a comparable with MATLAB program?