1 /*
2
3 Copyright (C) 2014, The University of Texas at Austin
4
5 This file is part of libflame and is available under the 3-Clause
6 BSD license, which can be found in the LICENSE file at the top-level
7 directory, or at http://opensource.org/licenses/BSD-3-Clause
8
9 */
10
11 #include "FLAME.h"
12 #include "test_libflame.h"
13
14 #define NUM_PARAM_COMBOS 1
15 #define NUM_MATRIX_ARGS 1
16 #define FIRST_VARIANT 1
17 #define LAST_VARIANT 1
18
19 // Static variables.
20 static char* op_str = "LU factorization with incremental pivoting";
21 static char* flash_front_str = "FLASH_LU_incpiv";
22 //static char* fla_front_str = "";
23 //static char* fla_unb_var_str = "";
24 //static char* fla_opt_var_str = "";
25 //static char* fla_blk_var_str = "";
26 static char* pc_str[NUM_PARAM_COMBOS] = { "" };
27 static test_thresh_t thresh = { 1e-02, 1e-03, // warn, pass for s
28 1e-10, 1e-11, // warn, pass for d
29 1e-02, 1e-03, // warn, pass for c
30 1e-10, 1e-11 }; // warn, pass for z
31
32 // Local prototypes.
33 void libfla_test_lu_incpiv_experiment( test_params_t params,
34 unsigned int var,
35 char* sc_str,
36 FLA_Datatype datatype,
37 unsigned int p,
38 unsigned int pci,
39 unsigned int n_repeats,
40 signed int impl,
41 double* perf,
42 double* residual );
43 void libfla_test_lu_incpiv_impl( int impl,
44 FLA_Obj A,
45 FLA_Obj p,
46 FLA_Obj L );
47
48
libfla_test_lu_incpiv(FILE * output_stream,test_params_t params,test_op_t op)49 void libfla_test_lu_incpiv( FILE* output_stream, test_params_t params, test_op_t op )
50 {
51 libfla_test_output_info( "--- %s ---\n", op_str );
52 libfla_test_output_info( "\n" );
53
54 if ( op.flash_front == ENABLE )
55 {
56 //libfla_test_output_info( "%s() front-end...\n", flash_front_str );
57 //libfla_test_output_info( "\n" );
58 libfla_test_op_driver( flash_front_str, NULL,
59 FIRST_VARIANT, LAST_VARIANT,
60 NUM_PARAM_COMBOS, pc_str,
61 NUM_MATRIX_ARGS,
62 FLA_TEST_HIER_FRONT_END,
63 params, thresh, libfla_test_lu_incpiv_experiment );
64 }
65
66 }
67
68
69
libfla_test_lu_incpiv_experiment(test_params_t params,unsigned int var,char * sc_str,FLA_Datatype datatype,unsigned int p_cur,unsigned int pci,unsigned int n_repeats,signed int impl,double * perf,double * residual)70 void libfla_test_lu_incpiv_experiment( test_params_t params,
71 unsigned int var,
72 char* sc_str,
73 FLA_Datatype datatype,
74 unsigned int p_cur,
75 unsigned int pci,
76 unsigned int n_repeats,
77 signed int impl,
78 double* perf,
79 double* residual )
80 {
81 dim_t b_flash = params.b_flash;
82 dim_t b_alg_hier = params.b_alg_hier;
83 double time_min = 1e9;
84 double time;
85 unsigned int i;
86 unsigned int m, n;
87 signed int m_input = -1;
88 signed int n_input = -1;
89 FLA_Obj A, x, b, norm;
90 FLA_Obj A_save;
91 FLA_Obj A_test, p_test, L_test, x_test, b_test;
92
93 // Determine the dimensions.
94 if ( m_input < 0 ) m = p_cur / abs(m_input);
95 else m = p_cur;
96 if ( n_input < 0 ) n = p_cur / abs(n_input);
97 else n = p_cur;
98
99 // Create the matrices for the current operation.
100 libfla_test_obj_create( datatype, FLA_NO_TRANSPOSE, sc_str[0], m, n, &A );
101
102 // Initialize the test matrices.
103 FLA_Random_matrix( A );
104
105 // Save the original object contents in a temporary object.
106 FLA_Obj_create_copy_of( FLA_NO_TRANSPOSE, A, &A_save );
107
108 // Create vectors to form a linear system.
109 FLA_Obj_create( datatype, n, 1, 0, 0, &x );
110 FLA_Obj_create( datatype, n, 1, 0, 0, &b );
111
112 // Create a real scalar object to hold the norm of A.
113 FLA_Obj_create( FLA_Obj_datatype_proj_to_real( A ), 1, 1, 0, 0, &norm );
114
115 // Create a random right-hand side vector.
116 FLA_Random_matrix( b );
117
118 // Use hierarchical matrices since we're testing the FLASH front-end.
119 FLASH_LU_incpiv_create_hier_matrices( A, 1, &b_flash, b_alg_hier, &A_test, &p_test, &L_test );
120 FLASH_Obj_create_hier_copy_of_flat( b, 1, &b_flash, &b_test );
121 FLASH_Obj_create_hier_copy_of_flat( x, 1, &b_flash, &x_test );
122
123 // Repeat the experiment n_repeats times and record results.
124 for ( i = 0; i < n_repeats; ++i )
125 {
126 FLASH_Obj_hierarchify( A_save, A_test );
127
128 time = FLA_Clock();
129
130 libfla_test_lu_incpiv_impl( impl, A_test, p_test, L_test );
131
132 time = FLA_Clock() - time;
133 time_min = min( time_min, time );
134 }
135
136 // Perform a linear solve with the result.
137 FLASH_LU_incpiv_solve( A_test, p_test, L_test, b_test, x_test );
138 FLASH_Obj_flatten( x_test, x );
139
140 // Free the FLASH objects.
141 FLASH_Obj_free( &A_test );
142 FLASH_Obj_free( &p_test );
143 FLASH_Obj_free( &L_test );
144 FLASH_Obj_free( &b_test );
145 FLASH_Obj_free( &x_test );
146
147 // Compute the performance of the best experiment repeat.
148 *perf = 2.0 / 3.0 * m * m * m / time_min / FLOPS_PER_UNIT_PERF;
149 if ( FLA_Obj_is_complex( A ) ) *perf *= 4.0;
150
151 // Compute the residual.
152 FLA_Gemv_external( FLA_NO_TRANSPOSE,
153 FLA_ONE, A_save, x, FLA_MINUS_ONE, b );
154 FLA_Nrm2_external( b, norm );
155 FLA_Obj_extract_real_scalar( norm, residual );
156
157 // Free the supporting flat objects.
158 FLA_Obj_free( &x );
159 FLA_Obj_free( &b );
160 FLA_Obj_free( &norm );
161 FLA_Obj_free( &A_save );
162
163 // Free the flat test matrices.
164 FLA_Obj_free( &A );
165 }
166
167
168
libfla_test_lu_incpiv_impl(int impl,FLA_Obj A,FLA_Obj p,FLA_Obj L)169 void libfla_test_lu_incpiv_impl( int impl,
170 FLA_Obj A,
171 FLA_Obj p,
172 FLA_Obj L )
173 {
174 switch ( impl )
175 {
176 case FLA_TEST_HIER_FRONT_END:
177 FLASH_LU_incpiv( A, p, L );
178 break;
179
180 default:
181 libfla_test_output_error( "Invalid implementation type.\n" );
182 }
183 }
184
185