source: CIVL/examples/omp/DataRaceBench/micro-benchmarks/adi-tile-no.c@ af3b8e4

1.23 2.0 main test-branch
Last change on this file since af3b8e4 was 36a61f3, checked in by Ziqing Luo <ziqing@…>, 9 years ago

Commit DataRaceBench into CIVL examples

git-svn-id: ​svn://vsl.cis.udel.edu/civl/trunk@4225 fb995dde-84ed-4084-dfe6-e5aef3e2452c
  • Property mode set to 100644
File size: 7.9 KB
Line 
1/**
2 * adi.c: This file is part of the PolyBench/C 3.2 test suite.
3 * Alternating Direction Implicit solver with tiling and nested SIMD.
4 *
5 * Contact: Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
6 * Web address: http://polybench.sourceforge.net
7 */
8#include <stdio.h>
9#include <unistd.h>
10#include <string.h>
11#include <math.h>
12/* Include polybench common header. */
13#include <polybench.h>
14/* Include benchmark-specific header. */
15/* Default data type is double, default size is 10x1024x1024. */
16#include "adi.h"
17/* Array initialization. */
18
19static void init_array(int n,double X[500 + 0][500 + 0],double A[500 + 0][500 + 0],double B[500 + 0][500 + 0])
20{
21 int i;
22 int j;
23{
24 int c1;
25 int c3;
26 int c2;
27 int c4;
28 if (n >= 1) {
29#pragma omp parallel for private(c4, c2, c3)
30 for (c1 = 0; c1 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c1++) {
31 for (c2 = 0; c2 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c2++) {
32 for (c3 = 16 * c1; c3 <= ((16 * c1 + 15 < n + -1?16 * c1 + 15 : n + -1)); c3++) {
33#pragma ivdep
34#pragma vector always
35#pragma simd
36 for (c4 = 16 * c2; c4 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c4++) {
37 X[c3][c4] = (((double )c3) * (c4 + 1) + 1) / n;
38 A[c3][c4] = (((double )c3) * (c4 + 2) + 2) / n;
39 B[c3][c4] = (((double )c3) * (c4 + 3) + 3) / n;
40 }
41 }
42 }
43 }
44 }
45 }
46}
47/* DCE code. Must scan the entire live-out data.
48 Can be used also to check the correctness of the output. */
49
50static void print_array(int n,double X[500 + 0][500 + 0])
51{
52 int i;
53 int j;
54 for (i = 0; i < n; i++)
55 for (j = 0; j < n; j++) {
56 fprintf(stderr,"%0.2lf ",X[i][j]);
57 if ((i * 500 + j) % 20 == 0)
58 fprintf(stderr,"\n");
59 }
60 fprintf(stderr,"\n");
61}
62/* Main computational kernel. The whole function will be timed,
63 including the call and return. */
64
65static void kernel_adi(int tsteps,int n,double X[500 + 0][500 + 0],double A[500 + 0][500 + 0],double B[500 + 0][500 + 0])
66{
67 int t;
68 int i1;
69 int i2;
70
71#pragma scop
72{
73 int c0;
74 int c2;
75 int c8;
76 int c9;
77 int c15;
78 if (n >= 1 && tsteps >= 1) {
79 for (c0 = 0; c0 <= tsteps + -1; c0++) {
80 if (n >= 2) {
81#pragma omp parallel for private(c15, c9, c8)
82 for (c2 = 0; c2 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c2++) {
83 for (c8 = 0; c8 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c8++) {
84 for (c9 = (1 > 16 * c8?1 : 16 * c8); c9 <= ((16 * c8 + 15 < n + -1?16 * c8 + 15 : n + -1)); c9++) {
85#pragma ivdep
86#pragma vector always
87#pragma simd
88 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
89 B[c15][c9] = B[c15][c9] - A[c15][c9] * A[c15][c9] / B[c15][c9 - 1];
90 }
91 }
92 }
93 for (c8 = 0; c8 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c8++) {
94 for (c9 = (1 > 16 * c8?1 : 16 * c8); c9 <= ((16 * c8 + 15 < n + -1?16 * c8 + 15 : n + -1)); c9++) {
95#pragma ivdep
96#pragma vector always
97#pragma simd
98 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
99 X[c15][c9] = X[c15][c9] - X[c15][c9 - 1] * A[c15][c9] / B[c15][c9 - 1];
100 }
101 }
102 }
103 for (c8 = 0; c8 <= (((n + -3) * 16 < 0?((16 < 0?-((-(n + -3) + 16 + 1) / 16) : -((-(n + -3) + 16 - 1) / 16))) : (n + -3) / 16)); c8++) {
104 for (c9 = 16 * c8; c9 <= ((16 * c8 + 15 < n + -3?16 * c8 + 15 : n + -3)); c9++) {
105#pragma ivdep
106#pragma vector always
107#pragma simd
108 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
109 X[c15][n - c9 - 2] = (X[c15][n - 2 - c9] - X[c15][n - 2 - c9 - 1] * A[c15][n - c9 - 3]) / B[c15][n - 3 - c9];
110 }
111 }
112 }
113 }
114 }
115#pragma omp parallel for private(c15)
116 for (c2 = 0; c2 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c2++) {
117#pragma ivdep
118#pragma vector always
119#pragma simd
120 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
121 X[c15][n - 1] = X[c15][n - 1] / B[c15][n - 1];
122 }
123 }
124 if (n >= 2) {
125#pragma omp parallel for private(c15, c9, c8)
126 for (c2 = 0; c2 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c2++) {
127 for (c8 = 0; c8 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c8++) {
128 for (c9 = (1 > 16 * c8?1 : 16 * c8); c9 <= ((16 * c8 + 15 < n + -1?16 * c8 + 15 : n + -1)); c9++) {
129#pragma ivdep
130#pragma vector always
131#pragma simd
132 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
133 B[c9][c15] = B[c9][c15] - A[c9][c15] * A[c9][c15] / B[c9 - 1][c15];
134 }
135 }
136 }
137 for (c8 = 0; c8 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c8++) {
138 for (c9 = (1 > 16 * c8?1 : 16 * c8); c9 <= ((16 * c8 + 15 < n + -1?16 * c8 + 15 : n + -1)); c9++) {
139#pragma ivdep
140#pragma vector always
141#pragma simd
142 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
143 X[c9][c15] = X[c9][c15] - X[c9 - 1][c15] * A[c9][c15] / B[c9 - 1][c15];
144 }
145 }
146 }
147 for (c8 = 0; c8 <= (((n + -3) * 16 < 0?((16 < 0?-((-(n + -3) + 16 + 1) / 16) : -((-(n + -3) + 16 - 1) / 16))) : (n + -3) / 16)); c8++) {
148 for (c9 = 16 * c8; c9 <= ((16 * c8 + 15 < n + -3?16 * c8 + 15 : n + -3)); c9++) {
149#pragma ivdep
150#pragma vector always
151#pragma simd
152 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
153 X[n - 2 - c9][c15] = (X[n - 2 - c9][c15] - X[n - c9 - 3][c15] * A[n - 3 - c9][c15]) / B[n - 2 - c9][c15];
154 }
155 }
156 }
157 }
158 }
159#pragma omp parallel for private(c15)
160 for (c2 = 0; c2 <= (((n + -1) * 16 < 0?((16 < 0?-((-(n + -1) + 16 + 1) / 16) : -((-(n + -1) + 16 - 1) / 16))) : (n + -1) / 16)); c2++) {
161#pragma ivdep
162#pragma vector always
163#pragma simd
164 for (c15 = 16 * c2; c15 <= ((16 * c2 + 15 < n + -1?16 * c2 + 15 : n + -1)); c15++) {
165 X[n - 1][c15] = X[n - 1][c15] / B[n - 1][c15];
166 }
167 }
168 }
169 }
170 }
171
172#pragma endscop
173}
174
175int main(int argc,char **argv)
176{
177/* Retrieve problem size. */
178 int n = 500;
179 int tsteps = 10;
180/* Variable declaration/allocation. */
181 double (*X)[500 + 0][500 + 0];
182 X = ((double (*)[500 + 0][500 + 0])(polybench_alloc_data(((500 + 0) * (500 + 0)),(sizeof(double )))));
183 ;
184 double (*A)[500 + 0][500 + 0];
185 A = ((double (*)[500 + 0][500 + 0])(polybench_alloc_data(((500 + 0) * (500 + 0)),(sizeof(double )))));
186 ;
187 double (*B)[500 + 0][500 + 0];
188 B = ((double (*)[500 + 0][500 + 0])(polybench_alloc_data(((500 + 0) * (500 + 0)),(sizeof(double )))));
189 ;
190/* Initialize array(s). */
191 init_array(n, *X, *A, *B);
192/* Start timer. */
193 polybench_timer_start();
194 ;
195/* Run kernel. */
196 kernel_adi(tsteps,n, *X, *A, *B);
197/* Stop and print timer. */
198 polybench_timer_stop();
199 ;
200 polybench_timer_print();
201 ;
202/* Prevent dead-code elimination. All live-out data must be printed
203 by the function call in argument. */
204 if (argc > 42 && !strcmp(argv[0],""))
205 print_array(n, *X);
206/* Be clean. */
207 free(((void *)X));
208 ;
209 free(((void *)A));
210 ;
211 free(((void *)B));
212 ;
213 return 0;
214}
Note: See TracBrowser for help on using the repository browser.