extern "C" int init_enviroment() { magma_init(); magma_print_environment(); magma_queue_create(dev, &queue); return 0; }
/* //////////////////////////////////////////////////////////////////////////// -- testing any solver */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_sopts zopts; magma_queue_t queue=NULL; magma_queue_create( 0, &queue ); magma_s_matrix Z={Magma_CSR}; int i=1; TESTING_CHECK( magma_sparse_opts( argc, argv, &zopts, &i, queue )); printf("matrixinfo = [\n"); printf("%% size (n) || nonzeros (nnz) || nnz/n\n"); printf("%%=============================================================%%\n"); while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_sm_5stencil( laplace_size, &Z, queue )); } else { // file-matrix test TESTING_CHECK( magma_s_csr_mtx( &Z, argv[i], queue )); } printf(" %10lld %10lld %10lld\n", (long long) Z.num_rows, (long long) Z.nnz, (long long) (Z.nnz/Z.num_rows) ); magma_smfree(&Z, queue ); i++; } printf("%%=============================================================%%\n"); printf("];\n"); magma_queue_destroy( queue ); TESTING_CHECK( magma_finalize() ); return info; }
/* //////////////////////////////////////////////////////////////////////////// -- testing any solver */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_copts zopts; magma_queue_t queue=NULL; magma_queue_create( 0, &queue ); real_Double_t res; magma_c_matrix A={Magma_CSR}, A2={Magma_CSR}, A3={Magma_CSR}, A4={Magma_CSR}, A5={Magma_CSR}; int i=1; TESTING_CHECK( magma_cparse_opts( argc, argv, &zopts, &i, queue )); while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_cm_5stencil( laplace_size, &A, queue )); } else { // file-matrix test TESTING_CHECK( magma_c_csr_mtx( &A, argv[i], queue )); } printf("%% matrix info: %lld-by-%lld with %lld nonzeros\n", (long long) A.num_rows, (long long) A.num_cols, (long long) A.nnz ); // filename for temporary matrix storage const char *filename = "testmatrix.mtx"; // write to file TESTING_CHECK( magma_cwrite_csrtomtx( A, filename, queue )); // read from file TESTING_CHECK( magma_c_csr_mtx( &A2, filename, queue )); // delete temporary matrix unlink( filename ); //visualize printf("A2:\n"); TESTING_CHECK( magma_cprint_matrix( A2, queue )); //visualize TESTING_CHECK( magma_cmconvert(A2, &A4, Magma_CSR, Magma_CSRL, queue )); printf("A4:\n"); TESTING_CHECK( magma_cprint_matrix( A4, queue )); TESTING_CHECK( magma_cmconvert(A4, &A5, Magma_CSR, Magma_ELL, queue )); printf("A5:\n"); TESTING_CHECK( magma_cprint_matrix( A5, queue )); // pass it to another application and back magma_int_t m, n; magma_index_t *row, *col; magmaFloatComplex *val=NULL; TESTING_CHECK( magma_ccsrget( A2, &m, &n, &row, &col, &val, queue )); TESTING_CHECK( magma_ccsrset( m, n, row, col, val, &A3, queue )); TESTING_CHECK( magma_cmdiff( A, A2, &res, queue )); printf("%% ||A-B||_F = %8.2e\n", res); if ( res < .000001 ) printf("%% tester IO: ok\n"); else printf("%% tester IO: failed\n"); TESTING_CHECK( magma_cmdiff( A, A3, &res, queue )); printf("%% ||A-B||_F = %8.2e\n", res); if ( res < .000001 ) printf("%% tester matrix interface: ok\n"); else printf("%% tester matrix interface: failed\n"); magma_cmfree(&A, queue ); magma_cmfree(&A2, queue ); magma_cmfree(&A4, queue ); magma_cmfree(&A5, queue ); i++; } magma_queue_destroy( queue ); TESTING_CHECK( magma_finalize() ); return info; }
/* //////////////////////////////////////////////////////////////////////////// -- testing sparse matrix vector product */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_queue_t queue=NULL; magma_queue_create( 0, &queue ); magma_s_matrix hA={Magma_CSR}, hA_SELLP={Magma_CSR}, dA={Magma_CSR}, dA_SELLP={Magma_CSR}; magma_s_matrix hx={Magma_CSR}, hy={Magma_CSR}, dx={Magma_CSR}, dy={Magma_CSR}, hrefvec={Magma_CSR}, hcheck={Magma_CSR}; hA_SELLP.blocksize = 8; hA_SELLP.alignment = 8; real_Double_t start, end, res; #ifdef MAGMA_WITH_MKL magma_int_t *pntre=NULL; #endif cusparseHandle_t cusparseHandle = NULL; cusparseMatDescr_t descr = NULL; float c_one = MAGMA_S_MAKE(1.0, 0.0); float c_zero = MAGMA_S_MAKE(0.0, 0.0); float accuracy = 1e-10; #define PRECISION_s #if defined(PRECISION_c) accuracy = 1e-4; #endif #if defined(PRECISION_s) accuracy = 1e-4; #endif magma_int_t i, j; for( i = 1; i < argc; ++i ) { if ( strcmp("--blocksize", argv[i]) == 0 ) { hA_SELLP.blocksize = atoi( argv[++i] ); } else if ( strcmp("--alignment", argv[i]) == 0 ) { hA_SELLP.alignment = atoi( argv[++i] ); } else break; } printf("\n# usage: ./run_sspmm" " [ --blocksize %lld --alignment %lld (for SELLP) ] matrices\n\n", (long long) hA_SELLP.blocksize, (long long) hA_SELLP.alignment ); while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_sm_5stencil( laplace_size, &hA, queue )); } else { // file-matrix test TESTING_CHECK( magma_s_csr_mtx( &hA, argv[i], queue )); } printf("%% matrix info: %lld-by-%lld with %lld nonzeros\n", (long long) hA.num_rows, (long long) hA.num_cols, (long long) hA.nnz ); real_Double_t FLOPS = 2.0*hA.nnz/1e9; // m - number of rows for the sparse matrix // n - number of vectors to be multiplied in the SpMM product magma_int_t m, n; m = hA.num_rows; n = 48; // init CPU vectors TESTING_CHECK( magma_svinit( &hx, Magma_CPU, m, n, c_one, queue )); TESTING_CHECK( magma_svinit( &hy, Magma_CPU, m, n, c_zero, queue )); // init DEV vectors TESTING_CHECK( magma_svinit( &dx, Magma_DEV, m, n, c_one, queue )); TESTING_CHECK( magma_svinit( &dy, Magma_DEV, m, n, c_zero, queue )); // calling MKL with CSR #ifdef MAGMA_WITH_MKL TESTING_CHECK( magma_imalloc_cpu( &pntre, m + 1 ) ); pntre[0] = 0; for (j=0; j < m; j++ ) { pntre[j] = hA.row[j+1]; } MKL_INT num_rows = hA.num_rows; MKL_INT num_cols = hA.num_cols; MKL_INT nnz = hA.nnz; MKL_INT num_vecs = n; MKL_INT *col; TESTING_CHECK( magma_malloc_cpu( (void**) &col, nnz * sizeof(MKL_INT) )); for( magma_int_t t=0; t < hA.nnz; ++t ) { col[ t ] = hA.col[ t ]; } MKL_INT *row; TESTING_CHECK( magma_malloc_cpu( (void**) &row, num_rows * sizeof(MKL_INT) )); for( magma_int_t t=0; t < hA.num_rows; ++t ) { row[ t ] = hA.col[ t ]; } // === Call MKL with consecutive SpMVs, using mkl_scsrmv === // warmp up mkl_scsrmv( "N", &num_rows, &num_cols, MKL_ADDR(&c_one), "GFNC", MKL_ADDR(hA.val), col, row, pntre, MKL_ADDR(hx.val), MKL_ADDR(&c_zero), MKL_ADDR(hy.val) ); start = magma_wtime(); for (j=0; j < 10; j++ ) { mkl_scsrmv( "N", &num_rows, &num_cols, MKL_ADDR(&c_one), "GFNC", MKL_ADDR(hA.val), col, row, pntre, MKL_ADDR(hx.val), MKL_ADDR(&c_zero), MKL_ADDR(hy.val) ); } end = magma_wtime(); printf( "\n > MKL SpMVs : %.2e seconds %.2e GFLOP/s (CSR).\n", (end-start)/10, FLOPS*10/(end-start) ); // === Call MKL with blocked SpMVs, using mkl_scsrmm === char transa = 'n'; MKL_INT ldb = n, ldc=n; char matdescra[6] = {'g', 'l', 'n', 'c', 'x', 'x'}; // warm up mkl_scsrmm( &transa, &num_rows, &num_vecs, &num_cols, MKL_ADDR(&c_one), matdescra, MKL_ADDR(hA.val), col, row, pntre, MKL_ADDR(hx.val), &ldb, MKL_ADDR(&c_zero), MKL_ADDR(hy.val), &ldc ); start = magma_wtime(); for (j=0; j < 10; j++ ) { mkl_scsrmm( &transa, &num_rows, &num_vecs, &num_cols, MKL_ADDR(&c_one), matdescra, MKL_ADDR(hA.val), col, row, pntre, MKL_ADDR(hx.val), &ldb, MKL_ADDR(&c_zero), MKL_ADDR(hy.val), &ldc ); } end = magma_wtime(); printf( "\n > MKL SpMM : %.2e seconds %.2e GFLOP/s (CSR).\n", (end-start)/10, FLOPS*10.*n/(end-start) ); magma_free_cpu( row ); magma_free_cpu( col ); row = NULL; col = NULL; #endif // MAGMA_WITH_MKL // copy matrix to GPU TESTING_CHECK( magma_smtransfer( hA, &dA, Magma_CPU, Magma_DEV, queue )); // SpMV on GPU (CSR) start = magma_sync_wtime( queue ); for (j=0; j < 10; j++) { TESTING_CHECK( magma_s_spmv( c_one, dA, dx, c_zero, dy, queue )); } end = magma_sync_wtime( queue ); printf( " > MAGMA: %.2e seconds %.2e GFLOP/s (standard CSR).\n", (end-start)/10, FLOPS*10.*n/(end-start) ); TESTING_CHECK( magma_smtransfer( dy, &hrefvec , Magma_DEV, Magma_CPU, queue )); magma_smfree(&dA, queue ); // convert to SELLP and copy to GPU TESTING_CHECK( magma_smconvert( hA, &hA_SELLP, Magma_CSR, Magma_SELLP, queue )); TESTING_CHECK( magma_smtransfer( hA_SELLP, &dA_SELLP, Magma_CPU, Magma_DEV, queue )); magma_smfree(&hA_SELLP, queue ); magma_smfree( &dy, queue ); TESTING_CHECK( magma_svinit( &dy, Magma_DEV, dx.num_rows, dx.num_cols, c_zero, queue )); // SpMV on GPU (SELLP) start = magma_sync_wtime( queue ); for (j=0; j < 10; j++) { TESTING_CHECK( magma_s_spmv( c_one, dA_SELLP, dx, c_zero, dy, queue )); } end = magma_sync_wtime( queue ); printf( " > MAGMA: %.2e seconds %.2e GFLOP/s (SELLP).\n", (end-start)/10, FLOPS*10.*n/(end-start) ); TESTING_CHECK( magma_smtransfer( dy, &hcheck , Magma_DEV, Magma_CPU, queue )); res = 0.0; for(magma_int_t k=0; k < hA.num_rows; k++ ) { res=res + MAGMA_S_REAL(hcheck.val[k]) - MAGMA_S_REAL(hrefvec.val[k]); } printf("%% |x-y|_F = %8.2e\n", res); if ( res < accuracy ) printf("%% tester spmm SELL-P: ok\n"); else printf("%% tester spmm SELL-P: failed\n"); magma_smfree( &hcheck, queue ); magma_smfree(&dA_SELLP, queue ); // SpMV on GPU (CUSPARSE - CSR) // CUSPARSE context // magma_smfree( &dy, queue ); TESTING_CHECK( magma_svinit( &dy, Magma_DEV, dx.num_rows, dx.num_cols, c_zero, queue )); //#ifdef PRECISION_d start = magma_sync_wtime( queue ); TESTING_CHECK( cusparseCreate( &cusparseHandle )); TESTING_CHECK( cusparseSetStream( cusparseHandle, magma_queue_get_cuda_stream(queue) )); TESTING_CHECK( cusparseCreateMatDescr( &descr )); TESTING_CHECK( cusparseSetMatType( descr, CUSPARSE_MATRIX_TYPE_GENERAL )); TESTING_CHECK( cusparseSetMatIndexBase( descr, CUSPARSE_INDEX_BASE_ZERO )); float alpha = c_one; float beta = c_zero; // copy matrix to GPU TESTING_CHECK( magma_smtransfer( hA, &dA, Magma_CPU, Magma_DEV, queue) ); for (j=0; j < 10; j++) { cusparseScsrmm(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, dA.num_rows, n, dA.num_cols, dA.nnz, &alpha, descr, dA.dval, dA.drow, dA.dcol, dx.dval, dA.num_cols, &beta, dy.dval, dA.num_cols); } end = magma_sync_wtime( queue ); printf( " > CUSPARSE: %.2e seconds %.2e GFLOP/s (CSR).\n", (end-start)/10, FLOPS*10*n/(end-start) ); TESTING_CHECK( magma_smtransfer( dy, &hcheck , Magma_DEV, Magma_CPU, queue )); res = 0.0; for(magma_int_t k=0; k < hA.num_rows; k++ ) { res = res + MAGMA_S_REAL(hcheck.val[k]) - MAGMA_S_REAL(hrefvec.val[k]); } printf("%% |x-y|_F = %8.2e\n", res); if ( res < accuracy ) printf("%% tester spmm cuSPARSE: ok\n"); else printf("%% tester spmm cuSPARSE: failed\n"); magma_smfree( &hcheck, queue ); cusparseDestroyMatDescr( descr ); cusparseDestroy( cusparseHandle ); descr = NULL; cusparseHandle = NULL; //#endif printf("\n\n"); // free CPU memory magma_smfree( &hA, queue ); magma_smfree( &hx, queue ); magma_smfree( &hy, queue ); magma_smfree( &hrefvec, queue ); // free GPU memory magma_smfree( &dx, queue ); magma_smfree( &dy, queue ); magma_smfree( &dA, queue); #ifdef MAGMA_WITH_MKL magma_free_cpu( pntre ); #endif i++; } magma_queue_destroy( queue ); TESTING_CHECK( magma_finalize() ); return info; }
/* //////////////////////////////////////////////////////////////////////////// -- testing any solver */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_queue_t queue=NULL; magma_queue_create( 0, &queue ); magmaFloatComplex one = MAGMA_C_MAKE(1.0, 0.0); magmaFloatComplex zero = MAGMA_C_MAKE(0.0, 0.0); magma_c_matrix A={Magma_CSR}, B_d={Magma_CSR}; magma_c_matrix x={Magma_CSR}, b={Magma_CSR}; int i=1; while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_cm_5stencil( laplace_size, &A, queue )); } else { // file-matrix test TESTING_CHECK( magma_c_csr_mtx( &A, argv[i], queue )); } printf( "\n# matrix info: %lld-by-%lld with %lld nonzeros\n\n", (long long) A.num_rows, (long long) A.num_cols, (long long) A.nnz ); magma_int_t n = A.num_rows; TESTING_CHECK( magma_cmtransfer( A, &B_d, Magma_CPU, Magma_DEV, queue )); // vectors and initial guess TESTING_CHECK( magma_cvinit( &b, Magma_DEV, A.num_cols, 1, zero, queue )); TESTING_CHECK( magma_cvinit( &x, Magma_DEV, A.num_cols, 1, one, queue )); TESTING_CHECK( magma_cprint_vector( b, 90, 10, queue )); TESTING_CHECK( magma_cprint_matrix( A, queue )); printf("\n\n\n"); TESTING_CHECK( magma_cprint_matrix( B_d, queue )); float res; res = magma_scnrm2( n, b.dval, 1, queue ); printf("norm0: %f\n", res); TESTING_CHECK( magma_c_spmv( one, B_d, x, zero, b, queue )); // b = A x TESTING_CHECK( magma_cprint_vector( b, 0, 100, queue )); TESTING_CHECK( magma_cprint_vector( b, b.num_rows-10, 10, queue )); res = magma_scnrm2( n, b.dval, 1, queue ); printf("norm: %f\n", res); TESTING_CHECK( magma_cresidual( B_d, x, b, &res, queue )); printf("res: %f\n", res); magma_cmfree(&B_d, queue ); magma_cmfree(&A, queue ); magma_cmfree(&x, queue ); magma_cmfree(&b, queue ); i++; } magma_queue_destroy( queue ); magma_finalize(); return info; }
int main( int argc, char** argv ) { magma_init(); magma_print_environment(); magma_int_t err; magma_int_t num = 0; magma_device_t dev; magma_queue_t queues; magma_queue_create( 0, &queues ); const double c_zero = MAGMA_D_ZERO; const double c_one = MAGMA_D_ONE; const double c_neg_one = MAGMA_D_NEG_ONE; double dummy[1]; magma_int_t M, N, MN,lda, ldb, ldc, ldda, info; double *h_A, *h_S, *h_U, *h_VT; double *d_Acu, *d_test; magmaDouble_ptr d_A, d_U, d_S, d_VT; magma_int_t ione = 1; magma_int_t ISEED[4] = {0, 0, 0, 1}; double tmp; double error, rwork[1]; magma_int_t status = 0; M = 3; N = 4; MN = M*N; ldda = magma_roundup(N, 32); lda = N; h_A = (double*)malloc(M*N*sizeof(double)); cudaMalloc((void**)& d_Acu, M*N*sizeof(double)); cudaMemset(d_Acu, 0, M*N*sizeof(double)); cudaPrintMatrix(d_Acu, M, N); magma_malloc((void**)&d_A, M*ldda*sizeof(double)); magma_malloc((void**)&d_test, M*ldda*sizeof(double)); cudaMemset(d_test, 0, M*ldda*sizeof(double)); //magma_malloc((void**)&d_A, M*ldda*sizeof(double)); //magma_malloc((void**)&d_S, N*ldda*sizeof(double)); //magma_malloc((void**)&d_U, lddbm*lddbn*sizeof(double)); //magma_malloc((void**)&d_VT, lddcm*lddcn*sizeof(double)); //printMatrix(d_A, M, K, lddan); //exit(0); //printf("\n\n ldda: %d, M: %d, N: %d \n\n", (int)lddan, (int)M, (int)N); // Initialize the matrix lapackf77_dlarnv(&ione, ISEED, &MN, h_A); //cudaMemcpy(d_Acu, h_A, M*N*sizeof(double), cudaMemcpyHostToDevice); magma_dsetmatrix(N, M, h_A, lda, d_A, ldda, queues); printMatrix(h_A, M, N, lda, queues); printf("========MTOC============\n"); magma_to_cuda(M, N, d_A, ldda, d_Acu); //printMatrix(d_test, M, N, lda, queues); printMatrix(d_A, M, N, ldda, queues); cudaPrintMatrix(d_Acu, M, N); printf("========CTOM============\n"); cuda_to_magma(M, N, d_Acu, d_test, ldda); cudaPrintMatrix(d_Acu, M, N); printMatrix(d_test, M, N, ldda, queues); //exit(0); printf("====================\n"); //cudaMemcpy(d_A, h_A, n2*sizeof(double), cudaMemcpyHostToDevice); //if(M >= N){ // printMatrix(h_U, M, N, ldda, queues); // printMatrix(h_S, 1, N, ldda, queues); // printMatrix(h_VT, N, N, ldda, queues); //}else{ // printMatrix(h_U, M, M, ldda, queues); // printMatrix(h_S, 1, M, ldda, queues); // printMatrix(h_VT, M, N, ldda, queues); //} magma_finalize(); return 0; }
/* //////////////////////////////////////////////////////////////////////////// -- testing any solver */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_copts zopts; magma_queue_t queue; magma_queue_create( 0, &queue ); magmaFloatComplex one = MAGMA_C_MAKE(1.0, 0.0); magmaFloatComplex zero = MAGMA_C_MAKE(0.0, 0.0); magma_c_matrix A={Magma_CSR}, B={Magma_CSR}, B_d={Magma_CSR}; magma_c_matrix x={Magma_CSR}, b={Magma_CSR}; int i=1; TESTING_CHECK( magma_cparse_opts( argc, argv, &zopts, &i, queue )); B.blocksize = zopts.blocksize; B.alignment = zopts.alignment; TESTING_CHECK( magma_csolverinfo_init( &zopts.solver_par, &zopts.precond_par, queue )); while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_cm_5stencil( laplace_size, &A, queue )); } else { // file-matrix test TESTING_CHECK( magma_c_csr_mtx( &A, argv[i], queue )); } // for the eigensolver case zopts.solver_par.ev_length = A.num_cols; TESTING_CHECK( magma_ceigensolverinfo_init( &zopts.solver_par, queue )); // scale matrix TESTING_CHECK( magma_cmscale( &A, zopts.scaling, queue )); // preconditioner if ( zopts.solver_par.solver != Magma_ITERREF ) { TESTING_CHECK( magma_c_precondsetup( A, b, &zopts.solver_par, &zopts.precond_par, queue ) ); } TESTING_CHECK( magma_cmconvert( A, &B, Magma_CSR, zopts.output_format, queue )); printf( "\n%% matrix info: %lld-by-%lld with %lld nonzeros\n\n", (long long) A.num_rows, (long long) A.num_cols, (long long) A.nnz ); printf("matrixinfo = [\n"); printf("%% size (m x n) || nonzeros (nnz) || nnz/m || stored nnz\n"); printf("%%============================================================================%%\n"); printf(" %8lld %8lld %10lld %4lld %10lld\n", (long long) B.num_rows, (long long) B.num_cols, (long long) B.true_nnz, (long long) (B.true_nnz/B.num_rows), (long long) B.nnz ); printf("%%============================================================================%%\n"); printf("];\n"); TESTING_CHECK( magma_cmtransfer( B, &B_d, Magma_CPU, Magma_DEV, queue )); // vectors and initial guess TESTING_CHECK( magma_cvinit( &b, Magma_DEV, A.num_rows, 1, one, queue )); //magma_cvinit( &x, Magma_DEV, A.num_cols, 1, one, queue ); //magma_c_spmv( one, B_d, x, zero, b, queue ); // b = A x //magma_cmfree(&x, queue ); TESTING_CHECK( magma_cvinit( &x, Magma_DEV, A.num_cols, 1, zero, queue )); info = magma_c_solver( B_d, b, &x, &zopts, queue ); if( info != 0 ) { printf("%%error: solver returned: %s (%lld).\n", magma_strerror( info ), (long long) info ); } printf("convergence = [\n"); magma_csolverinfo( &zopts.solver_par, &zopts.precond_par, queue ); printf("];\n\n"); zopts.solver_par.verbose = 0; printf("solverinfo = [\n"); magma_csolverinfo( &zopts.solver_par, &zopts.precond_par, queue ); printf("];\n\n"); printf("precondinfo = [\n"); printf("%% setup runtime\n"); printf(" %.6f %.6f\n", zopts.precond_par.setuptime, zopts.precond_par.runtime ); printf("];\n\n"); magma_cmfree(&B_d, queue ); magma_cmfree(&B, queue ); magma_cmfree(&A, queue ); magma_cmfree(&x, queue ); magma_cmfree(&b, queue ); i++; } magma_queue_destroy( queue ); TESTING_CHECK( magma_finalize() ); return info; }
/* //////////////////////////////////////////////////////////////////////////// -- testing any solver */ int main( int argc, char** argv ) { magma_int_t info = 0; TESTING_CHECK( magma_init() ); magma_print_environment(); magma_zopts zopts; magma_queue_t queue=NULL; magma_queue_create( 0, &queue ); real_Double_t res; magma_z_matrix Z={Magma_CSR}, Z2={Magma_CSR}, A={Magma_CSR}, A2={Magma_CSR}, AT={Magma_CSR}, AT2={Magma_CSR}, B={Magma_CSR}; int i=1; TESTING_CHECK( magma_zparse_opts( argc, argv, &zopts, &i, queue )); B.blocksize = zopts.blocksize; B.alignment = zopts.alignment; while( i < argc ) { if ( strcmp("LAPLACE2D", argv[i]) == 0 && i+1 < argc ) { // Laplace test i++; magma_int_t laplace_size = atoi( argv[i] ); TESTING_CHECK( magma_zm_5stencil( laplace_size, &Z, queue )); } else { // file-matrix test TESTING_CHECK( magma_z_csr_mtx( &Z, argv[i], queue )); } printf("%% matrix info: %lld-by-%lld with %lld nonzeros\n", (long long) Z.num_rows, (long long) Z.num_cols, (long long) Z.nnz ); // convert to be non-symmetric TESTING_CHECK( magma_zmconvert( Z, &A, Magma_CSR, Magma_CSRL, queue )); TESTING_CHECK( magma_zmconvert( Z, &B, Magma_CSR, Magma_CSRU, queue )); // transpose TESTING_CHECK( magma_zmtranspose( A, &AT, queue )); // quite some conversions //ELL TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_ELL, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_ELL, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //ELLPACKT TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_ELLPACKT, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_ELLPACKT, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //ELLRT AT2.blocksize = 8; AT2.alignment = 8; TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_ELLRT, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_ELLRT, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //SELLP AT2.blocksize = 8; AT2.alignment = 8; TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_SELLP, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_SELLP, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //ELLD TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_ELLD, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_ELLD, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //CSRCOO TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_CSRCOO, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_CSRCOO, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //CSRLIST TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_CSRLIST, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_CSRLIST, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); //CSRD TESTING_CHECK( magma_zmconvert( AT, &AT2, Magma_CSR, Magma_CSRD, queue )); magma_zmfree(&AT, queue ); TESTING_CHECK( magma_zmconvert( AT2, &AT, Magma_CSRD, Magma_CSR, queue )); magma_zmfree(&AT2, queue ); // transpose TESTING_CHECK( magma_zmtranspose( AT, &A2, queue )); TESTING_CHECK( magma_zmdiff( A, A2, &res, queue)); printf("%% ||A-A2||_F = %8.2e\n", res); if ( res < .000001 ) printf("%% conversion tester: ok\n"); else printf("%% conversion tester: failed\n"); TESTING_CHECK( magma_zmlumerge( A2, B, &Z2, queue )); TESTING_CHECK( magma_zmdiff( Z, Z2, &res, queue)); printf("%% ||Z-Z2||_F = %8.2e\n", res); if ( res < .000001 ) printf("%% LUmerge tester: ok\n"); else printf("%% LUmerge tester: failed\n"); magma_zmfree(&A, queue ); magma_zmfree(&A2, queue ); magma_zmfree(&AT, queue ); magma_zmfree(&AT2, queue ); magma_zmfree(&B, queue ); magma_zmfree(&Z2, queue ); magma_zmfree(&Z, queue ); i++; } magma_queue_destroy( queue ); TESTING_CHECK( magma_finalize() ); return info; }