void libblis_test_ger_impl( iface_t iface, obj_t* alpha, obj_t* x, obj_t* y, obj_t* a ) { switch ( iface ) { case BLIS_TEST_SEQ_FRONT_END: bli_ger( alpha, x, y, a ); break; default: libblis_test_printf_error( "Invalid interface type.\n" ); } }
int main( int argc, char** argv ) { obj_t a, x, y; obj_t a_save; obj_t alpha; dim_t m, n; dim_t p; dim_t p_begin, p_end, p_inc; int m_input, n_input; num_t dt_a, dt_x, dt_y; num_t dt_alpha; int r, n_repeats; double dtime; double dtime_save; double gflops; bli_init(); n_repeats = 3; #ifndef PRINT p_begin = 40; p_end = 2000; p_inc = 40; m_input = -1; n_input = -1; #else p_begin = 16; p_end = 16; p_inc = 1; m_input = 15; n_input = 15; #endif dt_alpha = dt_x = dt_y = dt_a = BLIS_DOUBLE; for ( p = p_begin; p <= p_end; p += p_inc ) { if ( m_input < 0 ) m = p * ( dim_t )abs(m_input); else m = ( dim_t ) m_input; if ( n_input < 0 ) n = p * ( dim_t )abs(n_input); else n = ( dim_t ) n_input; bli_obj_create( dt_alpha, 1, 1, 0, 0, &alpha ); bli_obj_create( dt_x, m, 1, 0, 0, &x ); bli_obj_create( dt_y, n, 1, 0, 0, &y ); bli_obj_create( dt_a, m, n, 0, 0, &a ); bli_obj_create( dt_a, m, n, 0, 0, &a_save ); bli_randm( &x ); bli_randm( &y ); bli_randm( &a ); bli_setsc( (2.0/1.0), 0.0, &alpha ); bli_copym( &a, &a_save ); dtime_save = DBL_MAX; for ( r = 0; r < n_repeats; ++r ) { bli_copym( &a_save, &a ); dtime = bli_clock(); #ifdef PRINT bli_printm( "x", &x, "%4.1f", "" ); bli_printm( "y", &y, "%4.1f", "" ); bli_printm( "a", &a, "%4.1f", "" ); #endif #ifdef BLIS bli_ger( &alpha, &x, &y, &a ); #else f77_int mm = bli_obj_length( a ); f77_int nn = bli_obj_width( a ); f77_int incx = bli_obj_vector_inc( x ); f77_int incy = bli_obj_vector_inc( y ); f77_int lda = bli_obj_col_stride( a ); double* alphap = bli_obj_buffer( alpha ); double* xp = bli_obj_buffer( x ); double* yp = bli_obj_buffer( y ); double* ap = bli_obj_buffer( a ); dger_( &mm, &nn, alphap, xp, &incx, yp, &incy, ap, &lda ); #endif #ifdef PRINT bli_printm( "a after", &a, "%4.1f", "" ); exit(1); #endif dtime_save = bli_clock_min_diff( dtime_save, dtime ); } gflops = ( 2.0 * m * n ) / ( dtime_save * 1.0e9 ); #ifdef BLIS printf( "data_ger_blis" ); #else printf( "data_ger_%s", BLAS ); #endif printf( "( %2lu, 1:4 ) = [ %4lu %4lu %10.3e %6.3f ];\n", ( unsigned long )(p - p_begin + 1)/p_inc + 1, ( unsigned long )m, ( unsigned long )n, dtime_save, gflops ); bli_obj_free( &alpha ); bli_obj_free( &x ); bli_obj_free( &y ); bli_obj_free( &a ); bli_obj_free( &a_save ); } bli_finalize(); return 0; }