int main(int argc, char *argv[])
{
int
datatype,
m_input, n_input,
m, n,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t nb_alg, n_threads;
char *colors = "brkgmcbrkgmcbrkgmc";
char *ticks = "o+*xso+*xso+*xso+*xs";
char m_dim_desc[14];
char n_dim_desc[14];
char m_dim_tag[10];
char n_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, B, C, C_ref;
FLA_Init( );
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d%d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%u", &n_threads );
fprintf( stdout, "%c %u\n", '%', n_threads );
fprintf( stdout, "\nclear all;\n\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
if ( n_input > 0 ) {
sprintf( n_dim_desc, "n = %d", n_input );
sprintf( n_dim_tag, "n%dc", n_input);
}
else if( n_input < -1 ) {
sprintf( n_dim_desc, "n = p/%d", -n_input );
sprintf( n_dim_tag, "n%dp", -n_input );
}
else if( n_input == -1 ) {
sprintf( n_dim_desc, "n = p" );
sprintf( n_dim_tag, "n%dp", 1 );
}
//datatype = FLA_COMPLEX;
datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
//FLASH_Queue_set_verbose_output( TRUE );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if( m < 0 ) m = p / abs(m_input);
if( n < 0 ) n = p / abs(n_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ ){
// If multiplying A on the left, A is m x m; ...on the right, A is n x n.
if ( pc_str[param_combo][0] == 'l' )
FLASH_Obj_create( datatype, m, m, 1, &nb_alg, &A );
else
FLASH_Obj_create( datatype, n, n, 1, &nb_alg, &A );
FLASH_Obj_create( datatype, m, n, 1, &nb_alg, &B );
FLASH_Obj_create( datatype, m, n, 1, &nb_alg, &C );
FLASH_Obj_create( datatype, m, n, 1, &nb_alg, &C_ref );
FLASH_Random_matrix( A );
FLASH_Random_matrix( B );
FLASH_Random_matrix( C );
FLASH_Copy( C, C_ref );
fprintf( stdout, "data_hemm_%s( %d, 1:5 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_Hemm( param_combo, FLA_ALG_REFERENCE, n_repeats, m, n,
A, B, C, C_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
time_Hemm( param_combo, FLA_ALG_FRONT, n_repeats, m, n,
A, B, C, C_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLASH_Obj_free( &A );
FLASH_Obj_free( &B );
FLASH_Obj_free( &C );
FLASH_Obj_free( &C_ref );
}
fprintf( stdout, "\n" );
}
/*
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_hemm_%s( :,1 ), data_hemm_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_hemm_%s( :,1 ), data_hemm_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_hemm\\_%s', 'fla\\_hemm\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME hemm front-end performance (%s, %s)' );\n",
m_dim_desc, n_dim_desc );
fprintf( stdout, "print -depsc hemm_front_%s_%s.eps\n", m_dim_tag, n_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
*/
FLA_Finalize( );
return 0;
}
int main( int argc, char *argv[] )
{
int
i, j,
n_threads,
n_repeats,
n_trials,
increment,
begin,
sorting,
caching,
work_stealing,
data_affinity;
dim_t
size,
nb_alg;
FLA_Datatype
datatype = FLA_DOUBLE;
FLA_Obj
A, x, b, b_norm,
AH, pH, bH;
double
b_norm_value,
dtime,
*dtimes,
*flops;
#ifndef FLA_ENABLE_WINDOWS_BUILD
char
output_file_m[100];
FILE
*fpp;
#endif
fprintf( stdout, "%c Enter number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c Enter blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c Enter problem size parameters: first, inc, num: ", '%' );
scanf( "%d%d%d", &begin, &increment, &n_trials );
fprintf( stdout, "%c %d %d %d\n", '%', begin, increment, n_trials );
fprintf( stdout, "%c Enter number of threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "%c Enter SuperMatrix parameters: sorting, caching, work stealing, data affinity: ", '%' );
scanf( "%d%d%d%d", &sorting, &caching, &work_stealing, &data_affinity );
fprintf( stdout, "%c %s %s %s %s\n\n", '%', ( sorting ? "TRUE" : "FALSE" ), ( caching ? "TRUE" : "FALSE" ), ( work_stealing ? "TRUE" : "FALSE" ), ( data_affinity ? ( data_affinity == 1 ? "FLASH_QUEUE_AFFINITY_2D_BLOCK_CYCLIC" : "FLASH_QUEUE_AFFINITY_OTHER" ) : "FLASH_QUEUE_AFFINITY_NONE" ) );
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, "%s_%u = [\n", OUTPUT_FILE, nb_alg );
#else
sprintf( output_file_m, "%s/%s_output.m", OUTPUT_PATH, OUTPUT_FILE );
fpp = fopen( output_file_m, "a" );
fprintf( fpp, "%%\n" );
fprintf( fpp, "%% | Matrix Size | FLASH |\n" );
fprintf( fpp, "%% | n x n | GFlops |\n" );
fprintf( fpp, "%% -----------------------------\n" );
fprintf( fpp, "%s_%u = [\n", OUTPUT_FILE, nb_alg );
#endif
FLA_Init();
dtimes = ( double * ) FLA_malloc( n_repeats * sizeof( double ) );
flops = ( double * ) FLA_malloc( n_trials * sizeof( double ) );
FLASH_Queue_set_num_threads( n_threads );
FLASH_Queue_set_sorting( sorting );
FLASH_Queue_set_caching( caching );
FLASH_Queue_set_work_stealing( work_stealing );
FLASH_Queue_set_data_affinity( data_affinity );
for ( i = 0; i < n_trials; i++ )
{
size = begin + i * increment;
FLA_Obj_create( datatype, size, size, 0, 0, &A );
FLA_Obj_create( datatype, size, 1, 0, 0, &x );
FLA_Obj_create( datatype, size, 1, 0, 0, &b );
FLA_Obj_create( datatype, 1, 1, 0, 0, &b_norm );
for ( j = 0; j < n_repeats; j++ )
{
FLA_Random_matrix( A );
FLA_Random_matrix( b );
FLASH_Obj_create_hier_copy_of_flat( A, 1, &nb_alg, &AH );
FLASH_Obj_create( FLA_INT, size, 1, 1, &nb_alg, &pH );
FLASH_Obj_create_hier_copy_of_flat( b, 1, &nb_alg, &bH );
dtime = FLA_Clock();
FLASH_LU_piv( AH, pH );
dtime = FLA_Clock() - dtime;
dtimes[j] = dtime;
FLASH_Apply_pivots( FLA_LEFT, FLA_NO_TRANSPOSE, pH, bH );
FLASH_Trsv( FLA_LOWER_TRIANGULAR, FLA_NO_TRANSPOSE, FLA_UNIT_DIAG,
AH, bH );
FLASH_Trsv( FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE, FLA_NONUNIT_DIAG,
AH, bH );
FLASH_Obj_free( &AH );
FLASH_Obj_free( &pH );
FLASH_Obj_flatten( bH, x );
FLASH_Obj_free( &bH );
}
dtime = dtimes[0];
for ( j = 1; j < n_repeats; j++ )
dtime = min( dtime, dtimes[j] );
flops[i] = 2.0 / 3.0 * size * size * size / dtime / 1e9;
FLA_Gemv_external( FLA_NO_TRANSPOSE, FLA_ONE,
A, x, FLA_MINUS_ONE, b );
FLA_Nrm2_external( b, b_norm );
FLA_Obj_extract_real_scalar( b_norm, &b_norm_value );
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, " %d %6.3f %le\n", size, flops[i], b_norm_value );
#else
fprintf( fpp, " %d %6.3f\n", size, flops[i] );
fprintf( stdout, "Time: %e | GFlops: %6.3f\n", dtime, flops[i] );
fprintf( stdout, "Matrix size: %u x %u | nb_alg: %u\n",
size, size, nb_alg );
fprintf( stdout, "Norm of difference: %le\n\n", b_norm_value );
#endif
FLA_Obj_free( &A );
FLA_Obj_free( &x );
FLA_Obj_free( &b );
FLA_Obj_free( &b_norm );
}
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, "];\n\n" );
#else
fprintf( fpp, "];\n" );
fflush( fpp );
fclose( fpp );
#endif
FLA_free( dtimes );
FLA_free( flops );
FLA_Finalize();
return 0;
}
int main( int argc, char *argv[] )
{
int
i, j,
n_threads,
n_repeats,
n_trials,
increment,
begin,
sorting,
caching,
work_stealing,
data_affinity;
dim_t
size,
nb_alg;
FLA_Datatype
datatype = FLA_DOUBLE;
FLA_Inv
inv = FLA_NO_INVERSE;
FLA_Uplo
uplo = FLA_LOWER_TRIANGULAR;
FLA_Obj
A, B, x, b, b_norm,
AH, BH;
double
length,
b_norm_value = 0.0,
dtime,
*dtimes,
*flops;
#ifndef FLA_ENABLE_WINDOWS_BUILD
char
output_file_m[100];
FILE
*fpp;
#endif
fprintf( stdout, "%c Enter number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c Enter blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c Enter problem size parameters: first, inc, num: ", '%' );
scanf( "%d%d%d", &begin, &increment, &n_trials );
fprintf( stdout, "%c %d %d %d\n", '%', begin, increment, n_trials );
fprintf( stdout, "%c Enter number of threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "%c Enter SuperMatrix parameters: sorting, caching, work stealing, data affinity: ", '%' );
scanf( "%d%d%d%d", &sorting, &caching, &work_stealing, &data_affinity );
fprintf( stdout, "%c %s %s %s %s\n\n", '%', ( sorting ? "TRUE" : "FALSE" ), ( caching ? "TRUE" : "FALSE" ), ( work_stealing ? "TRUE" : "FALSE" ), ( data_affinity ? ( data_affinity == 1 ? "FLASH_QUEUE_AFFINITY_2D_BLOCK_CYCLIC" : "FLASH_QUEUE_AFFINITY_OTHER" ) : "FLASH_QUEUE_AFFINITY_NONE" ) );
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, "%s_%u = [\n", OUTPUT_FILE, nb_alg );
#else
sprintf( output_file_m, "%s/%s_output.m", OUTPUT_PATH, OUTPUT_FILE );
fpp = fopen( output_file_m, "a" );
fprintf( fpp, "%%\n" );
fprintf( fpp, "%% | Matrix Size | FLASH |\n" );
fprintf( fpp, "%% | n x n | GFlops |\n" );
fprintf( fpp, "%% -----------------------------\n" );
fprintf( fpp, "%s_%u = [\n", OUTPUT_FILE, nb_alg );
#endif
FLA_Init();
dtimes = ( double * ) FLA_malloc( n_repeats * sizeof( double ) );
flops = ( double * ) FLA_malloc( n_trials * sizeof( double ) );
FLASH_Queue_set_num_threads( n_threads );
FLASH_Queue_set_sorting( sorting );
FLASH_Queue_set_caching( caching );
FLASH_Queue_set_work_stealing( work_stealing );
FLASH_Queue_set_data_affinity( data_affinity );
for ( i = 0; i < n_trials; i++ )
{
size = begin + i * increment;
FLA_Obj_create( datatype, size, size, 0, 0, &A );
FLA_Obj_create( datatype, size, size, 0, 0, &B );
FLA_Obj_create( datatype, size, 1, 0, 0, &x );
FLA_Obj_create( datatype, size, 1, 0, 0, &b );
FLA_Obj_create( datatype, 1, 1, 0, 0, &b_norm );
for ( j = 0; j < n_repeats; j++ )
{
FLA_Random_matrix( A );
FLA_Random_matrix( B );
FLA_Random_matrix( x );
FLA_Random_matrix( b );
FLA_Symmetrize( uplo, A );
FLA_Symmetrize( uplo, B );
length = ( double ) FLA_Obj_length( B );
FLA_Add_to_diag( &length, B );
FLA_Symv_external( uplo, FLA_ONE, B, x, FLA_ZERO, b );
FLASH_Obj_create_hier_copy_of_flat( A, 1, &nb_alg, &AH );
FLASH_Obj_create_hier_copy_of_flat( B, 1, &nb_alg, &BH );
FLASH_Chol( uplo, BH );
dtime = FLA_Clock();
FLASH_Eig_gest( inv, uplo, AH, BH );
dtime = FLA_Clock() - dtime;
dtimes[j] = dtime;
FLASH_Obj_free( &AH );
FLASH_Obj_free( &BH );
}
dtime = dtimes[0];
for ( j = 1; j < n_repeats; j++ )
dtime = min( dtime, dtimes[j] );
flops[i] = 1.0 * size * size * size / dtime / 1e9;
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, " %d %6.3f %le\n", size, flops[i], b_norm_value );
#else
fprintf( fpp, " %d %6.3f\n", size, flops[i] );
fprintf( stdout, "Time: %e | GFlops: %6.3f\n", dtime, flops[i] );
fprintf( stdout, "Matrix size: %u x %u | nb_alg: %u\n",
size, size, nb_alg );
fprintf( stdout, "Norm of difference: %le\n\n", b_norm_value );
#endif
FLA_Obj_free( &A );
FLA_Obj_free( &B );
FLA_Obj_free( &x );
FLA_Obj_free( &b );
FLA_Obj_free( &b_norm );
}
#ifdef FLA_ENABLE_WINDOWS_BUILD
fprintf( stdout, "];\n\n" );
#else
fprintf( fpp, "];\n" );
fflush( fpp );
fclose( fpp );
#endif
FLA_free( dtimes );
FLA_free( flops );
FLA_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int
m_input,
m,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t b_flash;
dim_t n_threads;
FLA_Datatype datatype;
FLA_Uplo uplo;
FLA_Inv inv;
char *colors = "brkgmcbrkg";
char *ticks = "o+*xso+*xs";
char m_dim_desc[14];
char m_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, B, norm;
FLA_Init();
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &b_flash );
fprintf( stdout, "%c %u\n", '%', b_flash );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m (-1 means bind to problem size): ", '%' );
scanf( "%d", &m_input );
fprintf( stdout, "%c %d\n", '%', m_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
//datatype = FLA_FLOAT;
//datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
if( m < 0 ) m = p / abs(m_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ ){
if ( pc_str[param_combo][0] == 'i' ) inv = FLA_INVERSE;
else inv = FLA_NO_INVERSE;
if ( pc_str[param_combo][1] == 'l' ) uplo = FLA_LOWER_TRIANGULAR;
else uplo = FLA_UPPER_TRIANGULAR;
FLASH_Obj_create( datatype, m, m, 1, &b_flash, &A );
FLASH_Obj_create( datatype, m, m, 1, &b_flash, &B );
FLA_Obj_create( FLA_Obj_datatype_proj_to_real( A ), 1, 1, 0, 0, &norm );
FLASH_Random_spd_matrix( uplo, A );
FLASH_Hermitianize( uplo, A );
FLASH_Random_spd_matrix( uplo, B );
FLASH_Chol( uplo, B );
fprintf( stdout, "data_eig_gest_%s( %d, 1:3 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_Eig_gest( param_combo, FLA_ALG_FRONT, n_repeats, m,
inv, uplo, A, B, norm, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLASH_Obj_free( &A );
FLASH_Obj_free( &B );
FLA_Obj_free( &norm );
}
fprintf( stdout, "\n" );
}
/*
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_eig_gest_%s( :,1 ), data_eig_gest_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_eig_gest_%s( :,1 ), data_eig_gest_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_eig_gest\\_%s', 'fla\\_eig_gest\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME eig_gest front-end performance (%s)' );\n", m_dim_desc );
fprintf( stdout, "print -depsc eig_gest_front_%s.eps\n", m_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
*/
FLA_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
precision,
nb_alg,
n_threads,
m_input, n_input,
m, n,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
int one = 1;
char *colors = "brkgmcbrkgmcbrkgmc";
char *ticks = "o+*xso+*xso+*xso+*xs";
char m_dim_desc[14];
char n_dim_desc[14];
char m_dim_tag[10];
char n_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, x, y, y_ref;
FLA_Init( );
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%d", &nb_alg );
fprintf( stdout, "%c %d\n", '%', nb_alg );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d%d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\nclear all;\n\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
if ( n_input > 0 ) {
sprintf( n_dim_desc, "n = %d", n_input );
sprintf( n_dim_tag, "n%dc", n_input);
}
else if( n_input < -1 ) {
sprintf( n_dim_desc, "n = p/%d", -n_input );
sprintf( n_dim_tag, "n%dp", -n_input );
}
else if( n_input == -1 ) {
sprintf( n_dim_desc, "n = p" );
sprintf( n_dim_tag, "n%dp", 1 );
}
//precision = FLA_SINGLE_PRECISION;
precision = FLA_DOUBLE_PRECISION;
FLASH_Queue_set_num_threads( n_threads );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if( m < 0 ) m = p / abs(m_input);
if( n < 0 ) n = p / abs(n_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ ){
// Determine datatype based on trans argument.
if ( pc_str[param_combo][0] == 'c' )
{
if ( precision == FLA_SINGLE_PRECISION )
datatype = FLA_COMPLEX;
else
datatype = FLA_DOUBLE_COMPLEX;
}
else
{
if ( precision == FLA_SINGLE_PRECISION )
datatype = FLA_FLOAT;
else
datatype = FLA_DOUBLE;
}
// If transposing A, switch dimensions.
if ( pc_str[param_combo][0] == 'n' )
FLASH_Obj_create( datatype, m, n, 1, &nb_alg, &A );
else
FLASH_Obj_create( datatype, n, m, 1, &nb_alg, &A );
FLASH_Obj_create_ext( datatype, n, 1, 1, &nb_alg, &one, &x );
FLASH_Obj_create_ext( datatype, m, 1, 1, &nb_alg, &one, &y );
FLASH_Obj_create_ext( datatype, m, 1, 1, &nb_alg, &one, &y_ref );
FLASH_Random_matrix( A );
FLASH_Random_matrix( x );
FLASH_Random_matrix( y );
FLASH_Copy( y, y_ref );
fprintf( stdout, "data_gemv_%s( %d, 1:5 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_Gemv( param_combo, FLA_ALG_REFERENCE, n_repeats, m, n,
A, x, y, y_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
time_Gemv( param_combo, FLA_ALG_FRONT, n_repeats, m, n,
A, x, y, y_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLASH_Obj_free( &A );
FLASH_Obj_free( &x );
FLASH_Obj_free( &y );
FLASH_Obj_free( &y_ref );
}
fprintf( stdout, "\n" );
}
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_gemv_%s( :,1 ), data_gemv_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_gemv_%s( :,1 ), data_gemv_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_gemv\\_%s', 'fla\\_gemv\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME gemv front-end performance (%s, %s)' );\n",
m_dim_desc, n_dim_desc );
fprintf( stdout, "print -depsc gemv_front_%s_%s.eps\n", m_dim_tag, n_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
FLA_Finalize( );
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
n_threads,
m_input,
m,
n_input,
n,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t
b_flash,
b_alg;
char *colors = "brkgmcbrkgmcbrkgmc";
char *ticks = "o+*xso+*xso+*xso+*xs";
char m_dim_desc[14];
char m_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj A, TW, b, x;
FLA_Obj A_flat, b_flat, x_flat;
FLA_Init( );
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &b_flash );
fprintf( stdout, "%c %u\n", '%', b_flash );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d %d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
//datatype = FLA_FLOAT;
datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
//datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
//FLASH_Queue_set_verbose_output( TRUE );
//FLA_Check_error_level_set( FLA_NO_ERROR_CHECKING );
//FLASH_Queue_disable();
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if ( m < 0 ) m = p * abs(m_input);
if ( n < 0 ) n = p * abs(n_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ )
{
FLA_Obj_create( datatype, m, n, 0, 0, &A_flat );
FLA_Obj_create( datatype, n, 1, 0, 0, &x_flat );
FLA_Obj_create( datatype, m, 1, 0, 0, &b_flat );
FLA_Random_matrix( A_flat );
FLA_Random_matrix( b_flat );
FLASH_QR_UT_create_hier_matrices( A_flat, 1, &b_flash, &A, &TW );
FLASH_Obj_create_hier_copy_of_flat( b_flat, 1, &b_flash, &b );
FLASH_Obj_create_hier_copy_of_flat( x_flat, 1, &b_flash, &x );
fprintf( stdout, "data_qrut_%s( %d, 1:5 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_QR_UT( param_combo, FLA_ALG_FRONT, n_repeats, m, n,
A, TW, b, x, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLA_Obj_free( &A_flat );
FLA_Obj_free( &b_flat );
FLA_Obj_free( &x_flat );
FLASH_Obj_free( &A );
FLASH_Obj_free( &TW );
FLASH_Obj_free( &b );
FLASH_Obj_free( &x );
}
}
/*
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_qrut_%s( :,1 ), data_qrut_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_qrut_%s( :,1 ), data_qrut_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_qrut\\_%s', 'fla\\_qrut\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME qrut front-end performance (%s)' );\n",
m_dim_desc );
fprintf( stdout, "print -depsc qrut_front_%s.eps\n", m_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
*/
FLA_Finalize( );
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
n_threads,
m_input,
m,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t
nb_alg;
char *colors = "brkgmcbrkg";
char *ticks = "o+*xso+*xs";
char m_dim_desc[14];
char m_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, A_ref;
FLA_Init();
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m (-1 means bind to problem size): ", '%' );
scanf( "%d", &m_input );
fprintf( stdout, "%c %d\n", '%', m_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\nclear all;\n\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
//datatype = FLA_FLOAT;
datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
//datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
if( m < 0 ) m = p / f2c_abs(m_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ ){
FLASH_Obj_create( datatype, m, m, 1, &nb_alg, &A );
FLASH_Obj_create( datatype, m, m, 1, &nb_alg, &A_ref );
if ( pc_str[param_combo][0] == 'l' )
FLASH_Random_spd_matrix( FLA_LOWER_TRIANGULAR, A );
else
FLASH_Random_spd_matrix( FLA_UPPER_TRIANGULAR, A );
FLASH_Copy( A, A_ref );
fprintf( stdout, "data_chol_%s( %d, 1:5 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_Chol( param_combo, FLA_ALG_REFERENCE, n_repeats, m,
A, A_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
time_Chol( param_combo, FLA_ALG_FRONT, n_repeats, m,
A, A_ref, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLASH_Obj_free( &A );
FLASH_Obj_free( &A_ref );
}
fprintf( stdout, "\n" );
}
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_chol_%s( :,1 ), data_chol_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_chol_%s( :,1 ), data_chol_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_chol\\_%s', 'fla\\_chol\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME chol front-end performance (%s)' );\n", m_dim_desc );
fprintf( stdout, "print -depsc chol_front_%s.eps\n", m_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
FLA_Finalize( );
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
n_threads,
m_input,
m,
n_input,
n,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t
n_panels,
nb_flash,
nb_alg;
double
dtime,
gflops,
diff;
FLA_Obj A, ATW, R, RTW, b, x;
FLA_Obj A_flat, b_flat, x_flat;
FLA_Init( );
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter algorithmic blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &nb_flash );
fprintf( stdout, "%c %u\n", '%', nb_flash );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d %d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of QR subproblem panels: ", '%' );
scanf( "%u", &n_panels );
fprintf( stdout, "%c %u\n", '%', n_panels );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
//datatype = FLA_FLOAT;
datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
//datatype = FLA_DOUBLE_COMPLEX;
//FLASH_Queue_disable();
FLASH_Queue_set_num_threads( n_threads );
//FLASH_Queue_set_verbose_output( TRUE );
// FLA_Check_error_level_set( FLA_NO_ERROR_CHECKING );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if ( m < 0 ) m = p * f2c_abs(m_input);
if ( n < 0 ) n = p * f2c_abs(n_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ )
{
FLA_Obj_create( datatype, m, n, 0, 0, &A_flat );
FLA_Obj_create( datatype, n, 1, 0, 0, &x_flat );
FLA_Obj_create( datatype, m, 1, 0, 0, &b_flat );
FLA_Random_matrix( A_flat );
FLA_Random_matrix( b_flat );
FLASH_CAQR_UT_inc_create_hier_matrices( n_panels, A_flat, 1, &nb_flash, nb_alg,
&A, &ATW, &R, &RTW );
FLASH_Obj_create_hier_copy_of_flat( b_flat, 1, &nb_flash, &b );
FLASH_Obj_create_hier_copy_of_flat( x_flat, 1, &nb_flash, &x );
fprintf( stdout, "data_caqrutinc_%s( %d, 1:3 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_CAQR_UT_inc( param_combo, FLA_ALG_FRONT, n_repeats, m, n, n_panels,
A, ATW, R, RTW, b, x, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLA_Obj_free( &A_flat );
FLA_Obj_free( &b_flat );
FLA_Obj_free( &x_flat );
FLASH_Obj_free( &A );
FLASH_Obj_free( &ATW );
FLASH_Obj_free( &R );
FLASH_Obj_free( &RTW );
FLASH_Obj_free( &b );
FLASH_Obj_free( &x );
}
}
FLA_Finalize( );
return 0;
}
int main( int argc, char *argv[] )
{
int
datatype,
n_threads,
m_input,
m,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i, j,
n_param_combos = N_PARAM_COMBOS;
int sign;
dim_t b_flash;
char *colors = "brkgmcbrkg";
char *ticks = "o+*xso+*xs";
char m_dim_desc[14];
char m_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, A_flat, C, C_flat, scale, isgn, norm;
FLA_Init();
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c Enter sign (-1 or 1):", '%' );
scanf( "%d", &sign );
fprintf( stdout, "%c %d\n", '%', sign );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &b_flash );
fprintf( stdout, "%c %u\n", '%', b_flash );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m (-1 means bind to problem size): ", '%' );
scanf( "%d", &m_input );
fprintf( stdout, "%c %d\n", '%', m_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
if ( 0 < sign )
isgn = FLA_ONE;
else
isgn = FLA_MINUS_ONE;
//datatype = FLA_FLOAT;
datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
//datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
//FLASH_Queue_disable();
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
if( m < 0 ) m = p / abs(m_input);
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ ){
FLA_Obj_create( datatype, m, m, 0, 0, &A_flat );
FLA_Obj_create( datatype, m, m, 0, 0, &C_flat );
FLA_Obj_create( FLA_Obj_datatype_proj_to_real( A_flat ), 1, 1, 0, 0, &scale );
FLA_Obj_create( FLA_Obj_datatype_proj_to_real( A_flat ), 1, 1, 0, 0, &norm );
FLA_Random_tri_matrix( FLA_UPPER_TRIANGULAR, FLA_NONUNIT_DIAG, A_flat );
FLA_Triangularize( FLA_UPPER_TRIANGULAR, FLA_NONUNIT_DIAG, A_flat );
FLA_Norm1( A_flat, norm );
FLA_Shift_diag( FLA_NO_CONJUGATE, norm, A_flat );
FLA_Random_matrix( C_flat );
FLA_Hermitianize( FLA_UPPER_TRIANGULAR, C_flat );
FLASH_Obj_create_hier_copy_of_flat( A_flat, 1, &b_flash, &A );
FLASH_Obj_create_hier_copy_of_flat( C_flat, 1, &b_flash, &C );
fprintf( stdout, "data_lyap_%s( %d, 1:3 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
/*
time_Lyap( param_combo, FLA_ALG_REFERENCE, n_repeats, m, n,
isgn, A, B, C, C_ref, scale, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
*/
time_Lyap( param_combo, FLA_ALG_FRONT, n_repeats, m,
isgn, A, C, scale, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLASH_Obj_free( &A );
FLASH_Obj_free( &C );
FLA_Obj_free( &A_flat );
FLA_Obj_free( &C_flat );
FLA_Obj_free( &scale );
FLA_Obj_free( &norm );
}
fprintf( stdout, "\n" );
}
/*
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_lyap_%s( :,1 ), data_lyap_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_lyap_%s( :,1 ), data_lyap_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_lyap\\_%s', 'fla\\_lyap\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME lyap front-end performance (%s)' );\n", m_dim_desc );
fprintf( stdout, "print -depsc lyap_front_%s.eps\n", m_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
*/
FLA_Finalize( );
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
n_threads,
m_input, n_input,
m, n, min_m_n,
p_first, p_last, p_inc,
pp,
pivot_combo,
n_repeats,
i,
n_pivot_combos = N_PIVOT_COMBOS;
dim_t
nb_alg,
nb_flash;
char *colors = "brkgmcbrkg";
char *ticks = "o+*xso+*xs";
char m_dim_desc[14];
char n_dim_desc[14];
char m_dim_tag[10];
char n_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj C, p, x, b, norm;
FLA_Init();
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter algorithmic blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c enter FLASH blocksize: ", '%' );
scanf( "%u", &nb_flash );
fprintf( stdout, "%c %u\n", '%', nb_flash );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d %d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\nclear all;\n\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
if ( n_input > 0 ) {
sprintf( n_dim_desc, "n = %d", n_input );
sprintf( n_dim_tag, "n%dc", n_input);
}
else if( n_input < -1 ) {
sprintf( n_dim_desc, "n = p/%d", -n_input );
sprintf( n_dim_tag, "n%dp", -n_input );
}
else if( n_input == -1 ) {
sprintf( n_dim_desc, "n = p" );
sprintf( n_dim_tag, "n%dp", 1 );
}
//datatype = FLA_FLOAT;
datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
//datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
for ( pp = p_first, i = 1; pp <= p_last; pp += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if( m < 0 ) m = pp / abs(m_input);
if( n < 0 ) n = pp / abs(n_input);
min_m_n = min( m, n );
for ( pivot_combo = 0; pivot_combo < n_pivot_combos; pivot_combo++ ){
FLA_Obj_create( datatype, m, n, 0, 0, &C );
FLA_Obj_create( FLA_INT, min_m_n, 1, 0, 0, &p );
FLA_Obj_create( datatype, m, 1, 0, 0, &x );
FLA_Obj_create( datatype, m, 1, 0, 0, &b );
FLA_Obj_create( FLA_Obj_datatype_proj_to_real( C ), 1, 1, 0, 0, &norm );
FLA_Random_matrix( C );
FLA_Random_matrix( b );
fprintf( stdout, "data_lu_%s( %d, 1:5 ) = [ %d ", pc_str[pivot_combo], i, pp );
fflush( stdout );
time_LU( pivot_combo, FLA_ALG_FRONT_OPT0, n_repeats, m, n, nb_alg, nb_flash,
C, p, x, b, norm, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
time_LU( pivot_combo, FLA_ALG_FRONT_OPT1, n_repeats, m, n, nb_alg, nb_flash,
C, p, x, b, norm, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLA_Obj_free( &C );
FLA_Obj_free( &p );
FLA_Obj_free( &x );
FLA_Obj_free( &b );
FLA_Obj_free( &norm );
}
}
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_pivot_combos; i++ ) {
fprintf( stdout, "plot( data_lu_%s( :,1 ), data_lu_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_lu_%s( :,1 ), data_lu_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_pivot_combos; i++ )
fprintf( stdout, "'ref\\_lu\\_%s', 'fla\\_lu\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME LU front-end performance (%s, %s)' );\n",
m_dim_desc, n_dim_desc );
fprintf( stdout, "print -depsc lu_front_%s_%s.eps\n", m_dim_tag, n_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
FLA_Finalize( );
return 0;
}
int main(int argc, char *argv[])
{
int
datatype,
n_blocks_m,
n_threads,
m_input, n_input,
m, n,
p_first, p_last, p_inc,
p,
n_repeats,
param_combo,
i,
n_param_combos = N_PARAM_COMBOS;
dim_t
nb_flash, nb_alg;
char *colors = "brkgmcbrkgmcbrkgmc";
char *ticks = "o+*xso+*xso+*xso+*xs";
char m_dim_desc[14];
char n_dim_desc[14];
char m_dim_tag[10];
char n_dim_tag[10];
double max_gflops=6.0;
double
dtime,
gflops,
diff;
FLA_Obj
A, A_flat_ref, A_flat, B, B_flat, D, D_flat, t, T, T_flat;
FLA_Init( );
fprintf( stdout, "%c number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c enter algorithmic blocksize: ", '%' );
scanf( "%u", &nb_alg );
fprintf( stdout, "%c %u\n", '%', nb_alg );
fprintf( stdout, "%c enter problem size first, last, inc: ", '%' );
scanf( "%d%d%d", &p_first, &p_last, &p_inc );
fprintf( stdout, "%c %d %d %d\n", '%', p_first, p_last, p_inc );
fprintf( stdout, "%c enter m n (-1 means bind to problem size): ", '%' );
scanf( "%d%d", &m_input, &n_input );
fprintf( stdout, "%c %d %d\n", '%', m_input, n_input );
fprintf( stdout, "%c enter the number of SuperMatrix threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n", '%', n_threads );
fprintf( stdout, "\nclear all;\n\n" );
if ( m_input > 0 ) {
sprintf( m_dim_desc, "m = %d", m_input );
sprintf( m_dim_tag, "m%dc", m_input);
}
else if( m_input < -1 ) {
sprintf( m_dim_desc, "m = p/%d", -m_input );
sprintf( m_dim_tag, "m%dp", -m_input );
}
else if( m_input == -1 ) {
sprintf( m_dim_desc, "m = p" );
sprintf( m_dim_tag, "m%dp", 1 );
}
if ( n_input > 0 ) {
sprintf( n_dim_desc, "n = %d", n_input );
sprintf( n_dim_tag, "n%dc", n_input);
}
else if( n_input < -1 ) {
sprintf( n_dim_desc, "n = p/%d", -n_input );
sprintf( n_dim_tag, "n%dp", -n_input );
}
else if( n_input == -1 ) {
sprintf( n_dim_desc, "n = p" );
sprintf( n_dim_tag, "n%dp", 1 );
}
//datatype = FLA_FLOAT;
//datatype = FLA_DOUBLE;
//datatype = FLA_COMPLEX;
datatype = FLA_DOUBLE_COMPLEX;
FLASH_Queue_set_num_threads( n_threads );
for ( p = p_first, i = 1; p <= p_last; p += p_inc, i += 1 )
{
m = m_input;
n = n_input;
if( m < 0 ) m = p / abs(m_input);
if( n < 0 ) n = p / abs(n_input);
nb_flash = n;
for ( param_combo = 0; param_combo < n_param_combos; param_combo++ )
{
FLA_Obj_create( datatype, m, nb_flash, &A_flat );
FLA_Obj_create( datatype, m, nb_flash, &A_flat_ref );
FLA_Obj_create( datatype, m, nb_flash, &T_flat );
FLA_Obj_create( datatype, nb_flash, 1, &t );
FLASH_Obj_create( datatype, m, nb_flash, 1, &nb_flash, &A );
n_blocks_m = FLA_Obj_length( A );
FLASH_Obj_create_ext( datatype, nb_alg * n_blocks_m, nb_flash, 1, &nb_alg, &nb_flash, &T );
FLA_Set( FLA_ZERO, T_flat );
FLASH_Set( FLA_ZERO, T );
FLASH_Random_matrix( A );
FLASH_Obj_flatten( A, A_flat );
FLA_Part_2x1( A, &B,
&D, 1, FLA_TOP );
FLA_Part_2x1( A_flat, &B_flat,
&D_flat, FLA_Obj_width( A_flat ), FLA_TOP );
FLA_Triangularize( FLA_UPPER_TRIANGULAR, FLA_NONUNIT_DIAG, *(FLASH_OBJ_PTR_AT(B)) );
FLA_Triangularize( FLA_UPPER_TRIANGULAR, FLA_NONUNIT_DIAG, B_flat );
fprintf( stdout, "data_qr2ut_%s( %d, 1:5 ) = [ %d ", pc_str[param_combo], i, p );
fflush( stdout );
time_QR2_UT( param_combo, FLA_ALG_REFERENCE, n_repeats, m, n,
A, A_flat_ref, B, B_flat, D, D_flat, A_flat, t, T, T_flat, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
time_QR2_UT( param_combo, FLA_ALG_FRONT, n_repeats, m, n,
A, A_flat_ref, B, B_flat, D, D_flat, A_flat, t, T, T_flat, &dtime, &diff, &gflops );
fprintf( stdout, "%6.3lf %6.2le ", gflops, diff );
fflush( stdout );
fprintf( stdout, " ]; \n" );
fflush( stdout );
FLA_Obj_free( &A_flat );
FLA_Obj_free( &A_flat_ref );
FLA_Obj_free( &T_flat );
FLA_Obj_free( &t );
FLASH_Obj_free( &A );
FLASH_Obj_free( &T );
}
fprintf( stdout, "\n" );
}
fprintf( stdout, "figure;\n" );
fprintf( stdout, "hold on;\n" );
for ( i = 0; i < n_param_combos; i++ ) {
fprintf( stdout, "plot( data_qr2ut_%s( :,1 ), data_qr2ut_%s( :, 2 ), '%c:%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
fprintf( stdout, "plot( data_qr2ut_%s( :,1 ), data_qr2ut_%s( :, 4 ), '%c-.%c' ); \n",
pc_str[i], pc_str[i], colors[ i ], ticks[ i ] );
}
fprintf( stdout, "legend( ... \n" );
for ( i = 0; i < n_param_combos; i++ )
fprintf( stdout, "'ref\\_qr2ut\\_%s', 'fla\\_qr2ut\\_%s', ... \n", pc_str[i], pc_str[i] );
fprintf( stdout, "'Location', 'SouthEast' ); \n" );
fprintf( stdout, "xlabel( 'problem size p' );\n" );
fprintf( stdout, "ylabel( 'GFLOPS/sec.' );\n" );
fprintf( stdout, "axis( [ 0 %d 0 %.2f ] ); \n", p_last, max_gflops );
fprintf( stdout, "title( 'FLAME qr2ut front-end performance (%s, %s)' );\n",
m_dim_desc, n_dim_desc );
fprintf( stdout, "print -depsc qr2ut_front_%s_%s.eps\n", m_dim_tag, n_dim_tag );
fprintf( stdout, "hold off;\n");
fflush( stdout );
FLA_Finalize( );
return 0;
}
