Ejemplo n.º 1
0
/* ////////////////////////////////////////////////////////////////////////////
   -- Testing sgetrf
*/
int main( int argc, char** argv)
{
    TESTING_INIT();

    real_Double_t   gflops, gpu_perf, gpu_time, cpu_perf=0, cpu_time=0;
    float          error;
    float *h_A, *h_R;
    magmaFloat_ptr d_A;
    magma_int_t     *ipiv;
    magma_int_t M, N, n2, lda, ldda, info, min_mn;
    magma_int_t ione     = 1;
    magma_int_t ISEED[4] = {0,0,0,1};
    magma_int_t status = 0;

    magma_opts opts;
    parse_opts( argc, argv, &opts );

    float tol = opts.tolerance * lapackf77_slamch("E");
    
    printf("    M     N   CPU GFlop/s (ms)    GPU GFlop/s (ms)  Copy time (ms)  ||PA-LU||/(||A||*N)\n");
    printf("=======================================================================================\n");
    for( int itest = 0; itest < opts.ntest; ++itest ) {
        for( int iter = 0; iter < opts.niter; ++iter ) {
            M = opts.msize[itest];
            N = opts.nsize[itest];
            min_mn = min(M, N);
            lda    = M;
            n2     = lda*N;
            ldda   = ((M+31)/32)*32;
            gflops = FLOPS_SGETRF( M, N ) / 1e9;
            
            if ( N > 512 ) {
                printf( "%5d %5d   skipping because sgetf2 does not support N > 512\n", (int) M, (int) N );
                continue;
            }
            
            TESTING_MALLOC_CPU( ipiv, magma_int_t,        min_mn );
            TESTING_MALLOC_CPU( h_A,  float, n2     );
            TESTING_MALLOC_PIN( h_R,  float, n2     );
            TESTING_MALLOC_DEV( d_A,  float, ldda*N );
            
            /* Initialize the matrix */
            lapackf77_slarnv( &ione, ISEED, &n2, h_A );
            lapackf77_slacpy( MagmaUpperLowerStr, &M, &N, h_A, &lda, h_R, &lda );

            real_Double_t set_time = magma_wtime();
            magma_ssetmatrix( M, N, h_R, lda, d_A, ldda );
            set_time =  magma_wtime() - set_time;

            /* =====================================================================
               Performs operation using LAPACK
               =================================================================== */
            if ( opts.lapack ) {
                cpu_time = magma_wtime();
                lapackf77_sgetrf(&M, &N, h_A, &lda, ipiv, &info);
                cpu_time = magma_wtime() - cpu_time;
                cpu_perf = gflops / cpu_time;
                if (info != 0)
                    printf("lapackf77_sgetrf returned error %d: %s.\n",
                           (int) info, magma_strerror( info ));
            }
            
            /* ====================================================================
               Performs operation using MAGMA
               =================================================================== */
            gpu_time = magma_wtime();
            magma_sgetf2_gpu( M, N, d_A, ldda, ipiv, &info);
            gpu_time = magma_wtime() - gpu_time;
            gpu_perf = gflops / gpu_time;
            if (info != 0)
                printf("magma_sgetf2_gpu returned error %d: %s.\n",
                       (int) info, magma_strerror( info ));
            
            real_Double_t get_time = magma_wtime();
            magma_sgetmatrix( M, N, d_A, ldda, h_A, lda );
            get_time =  magma_wtime() - get_time;

            /* =====================================================================
               Check the factorization
               =================================================================== */
            if ( opts.lapack ) {
                printf("%5d %5d   %7.2f (%7.2f)   %7.2f (%7.2f)   %7.2f",
                       (int) M, (int) N, cpu_perf, cpu_time*1000., gpu_perf, gpu_time*1000.,
                       set_time*1000.+get_time*1000.);
            }
            else {
                printf("%5d %5d     ---   (  ---  )   %7.2f (%7.2f)   %7.2f",
                       (int) M, (int) N, gpu_perf, gpu_time*1000., set_time*1000.+get_time*1000. );
            }
            if ( opts.check ) {
                magma_sgetmatrix( M, N, d_A, ldda, h_A, lda );
                error = get_LU_error( M, N, h_R, lda, h_A, ipiv );
                printf("   %8.2e   %s\n", error, (error < tol ? "ok" : "failed") );
                status += ! (error < tol);
            }
            else {
                printf("     ---  \n");
            }
            
            TESTING_FREE_CPU( ipiv );
            TESTING_FREE_CPU( h_A );
            TESTING_FREE_PIN( h_R );
            TESTING_FREE_DEV( d_A );
            fflush( stdout );
        }
        if ( opts.niter > 1 ) {
            printf( "\n" );
        }
    }

    TESTING_FINALIZE();
    return status;
}
Ejemplo n.º 2
0
/* ////////////////////////////////////////////////////////////////////////////
   -- Testing sgetrf
*/
int main( int argc, char** argv)
{
    TESTING_INIT();

    real_Double_t   gflops, gpu_perf, gpu_time, cpu_perf=0, cpu_time=0;
    float          error;
    float *h_A, *h_R;
    float *d_A;
    magma_int_t     *ipiv;
    magma_int_t M, N, n2, lda, ldda, info, min_mn;
    magma_int_t ione     = 1;
    magma_int_t ISEED[4] = {0,0,0,1};

    magma_opts opts;
    parse_opts( argc, argv, &opts );
    
    printf("  M     N     CPU GFlop/s (ms)    GPU GFlop/s (ms)    ||PA-LU||/(||A||*N)\n");
    printf("=========================================================================\n");
    for( int i = 0; i < opts.ntest; ++i ) {
        for( int iter = 0; iter < opts.niter; ++iter ) {
            M = opts.msize[i];
            N = opts.nsize[i];
            min_mn = min(M, N);
            lda    = M;
            n2     = lda*N;
            ldda   = ((M+31)/32)*32;
            gflops = FLOPS_SGETRF( M, N ) / 1e9;
            
            TESTING_MALLOC(    ipiv, magma_int_t,     min_mn );
            TESTING_MALLOC(    h_A,  float, n2     );
            TESTING_HOSTALLOC( h_R,  float, n2     );
            TESTING_DEVALLOC(  d_A,  float, ldda*N );
            
            /* Initialize the matrix */
            lapackf77_slarnv( &ione, ISEED, &n2, h_A );
            lapackf77_slacpy( MagmaUpperLowerStr, &M, &N, h_A, &lda, h_R, &lda );
            magma_ssetmatrix( M, N, h_R, lda, d_A, ldda );
            
            /* =====================================================================
               Performs operation using LAPACK
               =================================================================== */
            if ( opts.lapack ) {
                cpu_time = magma_wtime();
                lapackf77_sgetrf(&M, &N, h_A, &lda, ipiv, &info);
                cpu_time = magma_wtime() - cpu_time;
                cpu_perf = gflops / cpu_time;
                if (info != 0)
                    printf("lapackf77_sgetrf returned error %d: %s.\n",
                           (int) info, magma_strerror( info ));
            }
            
            /* ====================================================================
               Performs operation using MAGMA
               =================================================================== */
            gpu_time = magma_wtime();
            magma_sgetf2_gpu( M, N, d_A, ldda, ipiv, &info);
            gpu_time = magma_wtime() - gpu_time;
            gpu_perf = gflops / gpu_time;
            if (info != 0)
                printf("magma_sgetf2_gpu returned error %d: %s.\n",
                       (int) info, magma_strerror( info ));
            
            /* =====================================================================
               Check the factorization
               =================================================================== */
            if ( opts.lapack ) {
                printf("%5d %5d   %7.2f (%7.2f)   %7.2f (%7.2f)",
                       (int) M, (int) N, cpu_perf, cpu_time*1000., gpu_perf, gpu_time*1000. );
            }
            else {
                printf("%5d %5d     ---   (  ---  )   %7.2f (%7.2f)",
                       (int) M, (int) N, gpu_perf, gpu_time*1000. );
            }
            if ( opts.check ) {
                magma_sgetmatrix( M, N, d_A, ldda, h_A, lda );
                error = get_LU_error( M, N, h_R, lda, h_A, ipiv );
                printf("   %8.2e\n", error );
            }
            else {
                printf("     ---  \n");
            }
            
            TESTING_FREE( ipiv );
            TESTING_FREE( h_A );
            TESTING_HOSTFREE( h_R );
            TESTING_DEVFREE( d_A );
        }
        if ( opts.niter > 1 ) {
            printf( "\n" );
        }
    }

    TESTING_FINALIZE();
    return 0;
}