void NM_gemm(const double alpha, NumericsMatrix* A, NumericsMatrix* B,
const double beta, NumericsMatrix* C)
{
switch(A->storageType)
{
case NM_DENSE:
{
cblas_dgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, A->size0, B->size1, B->size1, alpha, A->matrix0, A->size0, B->matrix0, B->size0, beta, C->matrix0, A->size0);
NM_clearSparseBlock(C);
NM_clearSparseStorage(C);
break;
}
case NM_SPARSE_BLOCK:
{
prodNumericsMatrixNumericsMatrix(alpha, A, B, beta, C);
NM_clearDense(C);
NM_clearSparseStorage(C);
break;
}
case NM_SPARSE:
{
CSparseMatrix* result = cs_add(cs_multiply(NM_csc(A), NM_csc(B)),
NM_csc(C), alpha, beta);
NM_clearDense(C);
NM_clearSparseBlock(C);
NM_clearSparseStorage(C);
NM_sparse(C)->csc = result;
C->size0 = (int)C->matrix2->csc->m;
C->size1 = (int)C->matrix2->csc->n;
break;
}
}
}
int test_prodNumericsMatrixNumericsMatrix(NumericsMatrix** MM)
{
NumericsMatrix * M1 = MM[0];
NumericsMatrix * M2 = MM[1];
NumericsMatrix * M3 = MM[2];
NumericsMatrix * M4 = MM[3];
int info = -1;
printf("== Numerics tests: prodNumericsMatrixNumericsMatrix(NumericsMatrix,NumericsMatrix) == \n");
int i, j, k;
double alpha = 1.0, beta = 0.0;
double tol = 1e-12;
NumericsMatrix C;
C.storageType = 0;
C.size0 = M1->size0;
C.size1 = M1->size1;
C.matrix0 = (double *)malloc(C.size0 * C.size1 * sizeof(double));
C.matrix1 = NULL;
C.matrix2 = NULL;
C.internalData = NULL;
prodNumericsMatrixNumericsMatrix(alpha, M1, M1, beta, &C);
double * Cref = (double *)malloc(C.size0 * C.size1 * sizeof(double));
double sum;
for (i = 0; i < C.size0; i++)
{
for (j = 0; j < C.size1; j++)
{
sum = 0.0;
for (k = 0; k < M1->size1; k++)
{
sum = sum + M1->matrix0[i + k * M1->size0] * M1->matrix0[k + j * M1->size0];
}
Cref[i + j * C.size0] = sum;
}
}
double err = 0.0;
for (i = 0; i < C.size0; i++)
{
for (j = 0; j < C.size1; j++)
{
printf("Cref[%i+%i*%i]= %lf\t\t", i, j, C.size0, Cref[i + j * C.size0]);
printf("Cmatrix0[%i+%i*%i]= %lf\t", i, j, C.size0, C.matrix0[i + j * C.size0]);
err += (Cref[i + j * C.size0] - C.matrix0[i + j * C.size0]) * (Cref[i + j * C.size0] - C.matrix0[i + j * C.size0]);
printf("err = %lf\n", err);
}
}
if (err < tol)
{
info = 0;
}
if (info == 0)
printf("Step 0 ( C = alpha*A*B + beta*C, double* storage, square matrix ) ok ...\n");
else
{
printf("Step 0 ( C = alpha*A*B + beta*C, double* storage, square matrix) failed ...\n");
goto exit_1;
}
NumericsMatrix C2;
C2.storageType = 0;
C2.size0 = M1->size0;
C2.size1 = M3->size1;
C2.matrix0 = (double *)malloc(C2.size0 * C2.size1 * sizeof(double));
C2.matrix1 = NULL;
C2.matrix2 = NULL;
C2.internalData = NULL;
prodNumericsMatrixNumericsMatrix(alpha, M1, M3, beta, &C2);
double * C2ref = (double *)malloc(C2.size0 * C2.size1 * sizeof(double));
for (i = 0; i < C2.size0; i++)
{
for (j = 0; j < C2.size1; j++)
{
sum = 0.0;
for (k = 0; k < M1->size1; k++)
{
sum = sum + M1->matrix0[i + k * M1->size0] * M3->matrix0[k + j * M3->size0];
}
C2ref[i + j * C2.size0] = sum;
/* printf("C2ref(%i,%i)=%f\n", i,j,C2ref[i+j*C2.size0] ); */
}
}
err = 0.0;
for (i = 0; i < C2.size0; i++)
{
for (j = 0; j < C2.size1; j++)
{
err += (C2ref[i + j * C2.size0] - C2.matrix0[i + j * C2.size0]) * (C2ref[i + j * C2.size0] - C2.matrix0[i + j * C2.size0]);
}
}
if (err < tol)
{
info = 0;
}
if (info == 0)
printf("Step 1 ( C = alpha*A*B + beta*C, double* storage, non square) ok ...\n");
else
{
printf("Step 1 ( C = alpha*A*B + beta*C, double* storage, non square) failed ...\n");
goto exit_2;
}
NumericsMatrix C3;
C3.storageType = 1;
C3.size0 = M2->size0;
C3.size1 = M2->size1;
SparseBlockStructuredMatrix * SBM3 = (SparseBlockStructuredMatrix *)malloc(sizeof(SparseBlockStructuredMatrix));
C3.matrix1 = SBM3;
C3.matrix2 = NULL;
C3.internalData = NULL;
beta = 1.0;
i = 1;
// while (i > 0)
allocateMemoryForProdSBMSBM(M2->matrix1, M2->matrix1, SBM3);
/* printSBM(SBM3); */
prodNumericsMatrixNumericsMatrix(alpha, M2, M2, beta, &C3);
//freeSBM(SBM3);
printf("i= %i\n", i++);
/* Check if it is correct */
/* C3 and CRef must have the same values.*/
for (i = 0; i < C3.size0; i++)
{
for (j = 0; j < C3.size1; j++)
{
if (fabs(Cref[i + j * C3.size0] - getValueSBM(C3.matrix1, i, j)) > tol) info = 1;
/* printf("%i\t%i\n",i,j); */
/* printf("%lf\n",fabs(Cref[i+j*C3.size0]-getValueSBM(C3.matrix1,i,j) )); */
/* printf("%lf\n",Cref[i+j*C3.size0]); */
/* printf("%lf\n",getValueSBM(C3.matrix1,i,j)); */
if (info == 1) break;
}
if (info == 1) break ;
}
if (info == 0)
printf("Step 2 ( C = alpha*A*B + beta*C, sparse storage) ok ...\n");
else
{
printf("Step 2 ( C = alpha*A*B + beta*C, sparse storage) failed ...\n");
goto exit_3;
}
NumericsMatrix C4;
C4.storageType = 1;
C4.size0 = M2->size0;
C4.size1 = M4->size1;
SparseBlockStructuredMatrix * SBM4 = (SparseBlockStructuredMatrix *)malloc(sizeof(SparseBlockStructuredMatrix));
C4.matrix1 = SBM4;
C4.matrix2 = NULL;
C4.internalData = NULL;
allocateMemoryForProdSBMSBM(M2->matrix1, M4->matrix1, SBM4);
/* printSBM(SBM4); */
prodNumericsMatrixNumericsMatrix(alpha, M2, M4, beta, &C4);
/* Check if it is correct */
/* C4 and C2Ref must have the same values.*/
for (i = 0; i < C4.size0; i++)
{
for (j = 0; j < C4.size1; j++)
{
if (fabs(C2ref[i + j * C4.size0] - getValueSBM(C4.matrix1, i, j)) > tol) info = 1;
/* printf("%i\t%i\n",i,j); */
/* printf("%lf\n",fabs(C2ref[i+j*C4.size0]-getValueSBM(C4.matrix1,i,j) )); */
/* printf("%lf\n",C2ref[i+j*C4.size0]); */
/* printf("%lf\n",getValueSBM(C4.matrix1,i,j)); */
if (info == 1) break;
}
if (info == 1) break ;
}
if (info == 0)
printf("Step 3 ( C = alpha*A*B + beta*C, sparse storage) ok ...\n");
else
{
printf("Step 3 ( C = alpha*A*B + beta*C, sparse storage) failed ...\n");
}
freeNumericsMatrix(&C4);
exit_3:
freeNumericsMatrix(&C3);
exit_2:
free(C2.matrix0);
free(C2ref);
exit_1:
free(Cref);
free(C.matrix0);
return info;
}
