int main( int argc, char** argv)
{
real_Double_t gflops, gpu_perf, cpu_perf, gpu_time, cpu_time;
double *hA, *hR;
magmaDouble_ptr dA;
magma_int_t N = 0, n2, lda, ldda;
magma_int_t size[10] =
{ 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 8160, 8192 };
magma_int_t i, info;
double mz_one = MAGMA_D_NEG_ONE;
magma_int_t ione = 1;
magma_int_t ISEED[4] = {0,0,0,1};
double work[1], matnorm, diffnorm;
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_dpotrf2_gpu -N %d\n\n", 1024);
}
/* Initialize */
magma_queue_t queue1, queue2;
magma_device_t device;
int num = 0;
magma_err_t err;
magma_init();
err = magma_get_devices( &device, 2, &num );
if ( err != 0 or num < 1 ) {
fprintf( stderr, "magma_get_devices failed: %d\n", err );
exit(-1);
}
err = magma_queue_create( device, &queue1 );
if ( err != 0 ) {
fprintf( stderr, "magma_queue_create failed: %d\n", err );
exit(-1);
}
err = magma_queue_create( device, &queue2 );
if ( err != 0 ) {
fprintf( stderr, "magma_queue_create failed: %d\n", err );
exit(-1);
}
magma_queue_t queues[2] = {queue1, queue2};
/* Allocate memory for the largest matrix */
N = size[9];
n2 = N * N;
ldda = ((N+31)/32) * 32;
TESTING_MALLOC_CPU( hA, double, n2 );
TESTING_MALLOC_PIN( hR, double, n2 );
TESTING_MALLOC_DEV( dA, double, ldda*N );
printf("\n\n");
printf(" N CPU GFlop/s (sec) GPU GFlop/s (sec) ||R_magma-R_lapack||_F / ||R_lapack||_F\n");
printf("========================================================================================\n");
for(i=0; i<10; i++){
N = size[i];
lda = N;
n2 = lda*N;
ldda = ((N+31)/32)*32;
gflops = FLOPS( (double)N ) * 1e-9;
/* Initialize the matrix */
lapackf77_dlarnv( &ione, ISEED, &n2, hA );
/* Symmetrize and increase the diagonal */
for( int i = 0; i < N; ++i ) {
MAGMA_D_SET2REAL( hA(i,i), MAGMA_D_REAL(hA(i,i)) + N );
for( int j = 0; j < i; ++j ) {
hA(i, j) = MAGMA_D_CNJG( hA(j,i) );
}
}
lapackf77_dlacpy( MagmaFullStr, &N, &N, hA, &lda, hR, &lda );
/* Warm up to measure the performance */
magma_dsetmatrix( N, N, hA, 0, lda, dA, 0, ldda, queue1);
clFinish(queue1);
magma_dpotrf2_gpu( MagmaLower, N, dA, 0, ldda, &info, queues );
/* ====================================================================
Performs operation using MAGMA
=================================================================== */
magma_dsetmatrix( N, N, hA, 0, lda, dA, 0, ldda, queue1 );
clFinish(queue1);
gpu_time = magma_wtime();
magma_dpotrf2_gpu( MagmaLower, N, dA, 0, ldda, &info, queues );
gpu_time = magma_wtime() - gpu_time;
if (info != 0)
printf( "magma_dpotrf2 had error %d.\n", info );
gpu_perf = gflops / gpu_time;
/* =====================================================================
Performs operation using LAPACK
=================================================================== */
cpu_time = magma_wtime();
lapackf77_dpotrf( MagmaLowerStr, &N, hA, &lda, &info );
cpu_time = magma_wtime() - cpu_time;
if (info != 0)
printf( "lapackf77_dpotrf had error %d.\n", info );
cpu_perf = gflops / cpu_time;
/* =====================================================================
Check the result compared to LAPACK
|R_magma - R_lapack| / |R_lapack|
=================================================================== */
magma_dgetmatrix( N, N, dA, 0, ldda, hR, 0, lda, queue1 );
matnorm = lapackf77_dlange("f", &N, &N, hA, &lda, work);
blasf77_daxpy(&n2, &mz_one, hA, &ione, hR, &ione);
diffnorm = lapackf77_dlange("f", &N, &N, hR, &lda, work);
printf( "%5d %6.2f (%6.2f) %6.2f (%6.2f) %e\n",
N, cpu_perf, cpu_time, gpu_perf, gpu_time, diffnorm / matnorm );
if (argc != 1)
break;
}
/* clean up */
TESTING_FREE_CPU( hA );
TESTING_FREE_PIN( hR );
TESTING_FREE_DEV( dA );
magma_queue_destroy( queue1 );
magma_queue_destroy( queue2 );
magma_finalize();
}
/* ////////////////////////////////////////////////////////////////////////////
-- Testing dpotrf
*/
int main( int argc, char** argv)
{
TESTING_INIT();
real_Double_t gflops, gpu_perf, gpu_time, cpu_perf, cpu_time;
double *h_A, *h_R;
magmaDouble_ptr d_A;
magma_int_t N, n2, lda, ldda, info;
double c_neg_one = MAGMA_D_NEG_ONE;
magma_int_t ione = 1;
magma_int_t ISEED[4] = {0,0,0,1};
double work[1], error;
magma_int_t status = 0;
magma_opts opts;
parse_opts( argc, argv, &opts );
opts.lapack |= opts.check; // check (-c) implies lapack (-l)
double tol = opts.tolerance * lapackf77_dlamch("E");
printf("uplo = %s, version = %d\n", lapack_uplo_const(opts.uplo), opts.version );
printf(" N CPU GFlop/s (sec) GPU GFlop/s (sec) ||R_magma - R_lapack||_F / ||R_lapack||_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;
ldda = ((N+31)/32)*32;
gflops = FLOPS_DPOTRF( N ) / 1e9;
TESTING_MALLOC_CPU( h_A, double, n2 );
TESTING_MALLOC_PIN( h_R, double, n2 );
TESTING_MALLOC_DEV( d_A, double, ldda*N );
/* Initialize the matrix */
lapackf77_dlarnv( &ione, ISEED, &n2, h_A );
magma_dmake_hpd( N, h_A, lda );
lapackf77_dlacpy( MagmaUpperLowerStr, &N, &N, h_A, &lda, h_R, &lda );
magma_dsetmatrix( N, N, h_A, lda, d_A, 0, ldda, opts.queue );
/* ====================================================================
Performs operation using MAGMA
=================================================================== */
gpu_time = magma_wtime();
if ( opts.version == 1 ) {
magma_dpotrf_gpu( opts.uplo, N, d_A, 0, ldda, opts.queue, &info );
}
else if ( opts.version == 2 ) {
magma_dpotrf2_gpu( opts.uplo, N, d_A, 0, ldda, opts.queues2, &info );
}
else {
printf( "Unknown version %d\n", opts.version );
exit(1);
}
gpu_time = magma_wtime() - gpu_time;
gpu_perf = gflops / gpu_time;
if (info != 0)
printf("magma_dpotrf_gpu returned error %d: %s.\n",
(int) info, magma_strerror( info ));
if ( opts.lapack ) {
/* =====================================================================
Performs operation using LAPACK
=================================================================== */
cpu_time = magma_wtime();
lapackf77_dpotrf( 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_dpotrf returned error %d: %s.\n",
(int) info, magma_strerror( info ));
/* =====================================================================
Check the result compared to LAPACK
=================================================================== */
magma_dgetmatrix( N, N, d_A, 0, ldda, h_R, lda, opts.queue );
error = lapackf77_dlange("f", &N, &N, h_A, &lda, work);
blasf77_daxpy(&n2, &c_neg_one, h_A, &ione, h_R, &ione);
error = lapackf77_dlange("f", &N, &N, h_R, &lda, work) / error;
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 );
TESTING_FREE_DEV( d_A );
fflush( stdout );
}
if ( opts.niter > 1 ) {
printf( "\n" );
}
}
TESTING_FINALIZE();
return status;
}