/* //////////////////////////////////////////////////////////////////////////// -- Testing cpotrf */ int main( int argc, char** argv) { TESTING_CUDA_INIT(); magma_timestr_t start, end; float flops, gpu_perf, cpu_perf; cuFloatComplex *h_A, *h_R; cuFloatComplex *d_A; magma_int_t N = 0, n2, lda, ldda; magma_int_t size[10] = {1024,2048,3072,4032,5184,6016,7040,8064,9088,10112}; magma_int_t i, info; const char *uplo = MagmaUpperStr; cuFloatComplex c_neg_one = MAGMA_C_NEG_ONE; magma_int_t ione = 1; magma_int_t ISEED[4] = {0,0,0,1}; float work[1], matnorm; if (argc != 1){ for(i = 1; i<argc; i++){ if (strcmp("-N", argv[i])==0) N = atoi(argv[++i]); } if (N>0) size[0] = size[9] = N; else exit(1); } else { printf("\nUsage: \n"); printf(" testing_cpotri_gpu -N %d\n\n", 1024); } /* Allocate host memory for the matrix */ n2 = size[9] * size[9]; ldda = ((size[9]+31)/32) * 32; TESTING_MALLOC( h_A, cuFloatComplex, n2); TESTING_HOSTALLOC( h_R, cuFloatComplex, n2); TESTING_DEVALLOC( d_A, cuFloatComplex, ldda*size[9] ); printf(" N CPU GFlop/s GPU GFlop/s ||R||_F / ||A||_F\n"); printf("========================================================\n"); for(i=0; i<10; i++){ N = size[i]; lda = N; n2 = lda*N; flops = FLOPS_CPOTRI( (float)N ) / 1000000; ldda = ((N+31)/32)*32; /* Initialize the matrix */ lapackf77_clarnv( &ione, ISEED, &n2, h_A ); /* Symmetrize and increase the diagonal */ { magma_int_t i, j; for(i=0; i<N; i++) { MAGMA_C_SET2REAL( h_A[i*lda+i], ( MAGMA_C_REAL(h_A[i*lda+i]) + 1.*N ) ); for(j=0; j<i; j++) h_A[i*lda+j] = cuConjf(h_A[j*lda+i]); } } lapackf77_clacpy( MagmaUpperLowerStr, &N, &N, h_A, &lda, h_R, &lda ); /* ==================================================================== Performs operation using MAGMA =================================================================== */ //cublasSetMatrix( N, N, sizeof(cuFloatComplex), h_A, lda, d_A, ldda); //magma_cpotrf_gpu(uplo[0], N, d_A, ldda, &info); /* factorize matrix */ magma_csetmatrix( N, N, h_A, lda, d_A, ldda ); magma_cpotrf_gpu(uplo[0], N, d_A, ldda, &info); // check for exact singularity //magma_cgetmatrix( N, N, d_A, ldda, h_R, lda ); //h_R[ 10 + 10*lda ] = MAGMA_C_MAKE( 0.0, 0.0 ); //magma_csetmatrix( N, N, h_R, lda, d_A, ldda ); start = get_current_time(); magma_cpotri_gpu(uplo[0], N, d_A, ldda, &info); end = get_current_time(); if (info != 0) printf("magma_cpotri_gpu returned error %d\n", (int) info); gpu_perf = flops / GetTimerValue(start, end); /* ===================================================================== Performs operation using LAPACK =================================================================== */ lapackf77_cpotrf(uplo, &N, h_A, &lda, &info); start = get_current_time(); lapackf77_cpotri(uplo, &N, h_A, &lda, &info); end = get_current_time(); if (info != 0) printf("lapackf77_cpotri returned error %d\n", (int) info); cpu_perf = flops / GetTimerValue(start, end); /* ===================================================================== Check the result compared to LAPACK =================================================================== */ magma_cgetmatrix( N, N, d_A, ldda, h_R, lda ); matnorm = lapackf77_clange("f", &N, &N, h_A, &lda, work); blasf77_caxpy(&n2, &c_neg_one, h_A, &ione, h_R, &ione); printf("%5d %6.2f %6.2f %e\n", (int) size[i], cpu_perf, gpu_perf, lapackf77_clange("f", &N, &N, h_R, &lda, work) / matnorm); if (argc != 1) break; } /* Memory clean up */ TESTING_FREE( h_A ); TESTING_HOSTFREE( h_R ); TESTING_DEVFREE( d_A ); /* Shutdown */ TESTING_CUDA_FINALIZE(); }
/* //////////////////////////////////////////////////////////////////////////// -- Testing cpotri */ int main( int argc, char** argv) { TESTING_INIT(); real_Double_t gflops, gpu_perf, gpu_time, cpu_perf, cpu_time; magmaFloatComplex *h_A, *h_R; magmaFloatComplex c_neg_one = MAGMA_C_NEG_ONE; magma_int_t N, n2, lda, info; magma_int_t ione = 1; magma_int_t ISEED[4] = {0,0,0,1}; float Anorm, error, work[1]; magma_int_t status = 0; magma_opts opts; opts.parse_opts( argc, argv ); opts.lapack |= opts.check; // check (-c) implies lapack (-l) float tol = opts.tolerance * lapackf77_slamch("E"); printf("%% uplo = %s\n", lapack_uplo_const(opts.uplo) ); printf("%% N CPU Gflop/s (sec) GPU Gflop/s (sec) ||R||_F / ||A||_F\n"); printf("%%================================================================\n"); for( int itest = 0; itest < opts.ntest; ++itest ) { for( int iter = 0; iter < opts.niter; ++iter ) { N = opts.nsize[itest]; lda = N; n2 = lda*N; gflops = FLOPS_CPOTRI( N ) / 1e9; TESTING_MALLOC_CPU( h_A, magmaFloatComplex, n2 ); TESTING_MALLOC_PIN( h_R, magmaFloatComplex, n2 ); /* ==================================================================== Initialize the matrix =================================================================== */ lapackf77_clarnv( &ione, ISEED, &n2, h_A ); magma_cmake_hpd( N, h_A, lda ); lapackf77_clacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda ); /* ==================================================================== Performs operation using MAGMA =================================================================== */ if ( opts.warmup ) { magma_cpotrf( opts.uplo, N, h_R, lda, &info ); magma_cpotri( opts.uplo, N, h_R, lda, &info ); lapackf77_clacpy( MagmaFullStr, &N, &N, h_A, &lda, h_R, &lda ); } /* factorize matrix */ magma_cpotrf( opts.uplo, N, h_R, lda, &info ); // check for exact singularity //h_R[ 10 + 10*lda ] = MAGMA_C_MAKE( 0.0, 0.0 ); gpu_time = magma_wtime(); magma_cpotri( opts.uplo, N, h_R, lda, &info ); gpu_time = magma_wtime() - gpu_time; gpu_perf = gflops / gpu_time; if (info != 0) { printf("magma_cpotri returned error %d: %s.\n", (int) info, magma_strerror( info )); } /* ===================================================================== Performs operation using LAPACK =================================================================== */ if ( opts.lapack ) { lapackf77_cpotrf( lapack_uplo_const(opts.uplo), &N, h_A, &lda, &info ); cpu_time = magma_wtime(); lapackf77_cpotri( lapack_uplo_const(opts.uplo), &N, h_A, &lda, &info ); cpu_time = magma_wtime() - cpu_time; cpu_perf = gflops / cpu_time; if (info != 0) { printf("lapackf77_cpotri returned error %d: %s.\n", (int) info, magma_strerror( info )); } /* ===================================================================== Check the result compared to LAPACK =================================================================== */ blasf77_caxpy(&n2, &c_neg_one, h_A, &ione, h_R, &ione); Anorm = lapackf77_clange("f", &N, &N, h_A, &lda, work); error = lapackf77_clange("f", &N, &N, h_R, &lda, work) / Anorm; printf("%5d %7.2f (%7.2f) %7.2f (%7.2f) %8.2e %s\n", (int) N, cpu_perf, cpu_time, gpu_perf, gpu_time, error, (error < tol ? "ok" : "failed") ); status += ! (error < tol); } else { printf("%5d --- ( --- ) %7.2f (%7.2f) ---\n", (int) N, gpu_perf, gpu_time ); } TESTING_FREE_CPU( h_A ); TESTING_FREE_PIN( h_R ); fflush( stdout ); } if ( opts.niter > 1 ) { printf( "\n" ); } } opts.cleanup(); TESTING_FINALIZE(); return status; }