1// XFAIL: msvc
2// Fails on windows due to issue #129023
3// RUN: %libomp-compile-and-run
4#include <stdio.h>
5#include <stdlib.h>
6#include <string.h>
7#include "omp.h"
8
9#ifndef MAX_BOUND
10#define MAX_BOUND 64
11#endif
12#ifndef _MSC_VER
13#define NO_EFFICIENCY_CHECK
14#endif
15
16/* To ensure Correctness, only valid iterations are executed and are executed
17 only once. Stores the number of times an iteration is executed. */
18unsigned *execution_count = NULL;
19/* Stores the number of iterations executed by each thread. */
20unsigned *iterations_per_thread = NULL;
21
22unsigned *Alloc(unsigned bound1, unsigned bound2) {
23 return (unsigned *)(malloc(size: bound1 * bound2 * sizeof(unsigned)));
24}
25
26void ZeroOut(unsigned *p, unsigned bound1, unsigned bound2) {
27 memset(s: p, c: 0, n: bound1 * bound2 * sizeof(unsigned));
28}
29
30void Free(unsigned *p) { free(ptr: (void *)p); }
31
32unsigned *Index(unsigned *p, unsigned i, unsigned j, unsigned bound2) {
33 return &p[i * bound2 + j];
34}
35
36int test(unsigned upper_bound) {
37
38 unsigned total_iterations = upper_bound * (upper_bound - 1) / 2;
39 unsigned num_threads = omp_get_max_threads();
40 unsigned lower_per_chunk = total_iterations / num_threads;
41 unsigned upper_per_chunk =
42 lower_per_chunk + ((total_iterations % num_threads) ? 1 : 0);
43 int i, j;
44
45 omp_set_num_threads(num_threads);
46
47 ZeroOut(p: execution_count, bound1: upper_bound, bound2: upper_bound);
48 ZeroOut(p: iterations_per_thread, bound1: num_threads, bound2: 1);
49
50#ifdef VERBOSE
51 fprintf(stderr,
52 "INFO: Using %6d threads for %6d outer iterations with %6d [%6d:%6d] "
53 "chunks "
54 "loop type lower triangle <,< - ",
55 num_threads, upper_bound, total_iterations, lower_per_chunk,
56 upper_per_chunk);
57#endif
58
59#pragma omp parallel shared(iterations_per_thread, execution_count)
60 { /* begin of parallel */
61 /* Lower triangular execution_count matrix */
62#pragma omp for schedule(static) collapse(2)
63 for (i = 0; i < upper_bound; i++) {
64 for (j = 0; j < i; j++) {
65 (*Index(p: iterations_per_thread, i: omp_get_thread_num(), j: 0, bound2: 1))++;
66 (*Index(p: execution_count, i, j, bound2: upper_bound))++;
67 }
68 } /* end of for*/
69 } /* end of parallel */
70
71 /* check the execution_count array */
72 for (i = 0; i < upper_bound; i++) {
73 for (j = 0; j < i; j++) {
74 unsigned value = *Index(p: execution_count, i, j, bound2: upper_bound);
75 /* iteration with j<=i are valid, but should have been executed only once
76 */
77 if (value != 1) {
78 fprintf(stderr, format: "ERROR: valid iteration [%i,%i] executed %i times.\n",
79 i, j, value);
80 return 0;
81 }
82 }
83 for (j = i; j < upper_bound; j++) {
84 unsigned value = *Index(p: execution_count, i, j, bound2: upper_bound);
85 /* iteration with j>=i are invalid and should not have been executed
86 */
87 if (value > 0) {
88 fprintf(stderr, format: "ERROR: invalid iteration [%i,%i] executed %i times.\n",
89 i, j, value);
90 return 0;
91 }
92 }
93 }
94
95#ifndef NO_EFFICIENCY_CHECK
96 /* Ensure the number of iterations executed by each thread is within bounds */
97 for (i = 0; i < num_threads; i++) {
98 unsigned value = *Index(iterations_per_thread, i, 0, 1);
99 if (value < lower_per_chunk || value > upper_per_chunk) {
100 fprintf(stderr,
101 "ERROR: Inefficient Collapse thread %d of %d assigned %i "
102 "iterations; must be between %d and %d\n",
103 i, num_threads, value, lower_per_chunk, upper_per_chunk);
104 return 0;
105 }
106 }
107#endif
108#ifdef VERBOSE
109 fprintf(stderr, "PASSED\r\n");
110#endif
111 return 1;
112}
113
114int main() {
115
116 execution_count = Alloc(MAX_BOUND, MAX_BOUND);
117 iterations_per_thread = Alloc(bound1: omp_get_max_threads(), bound2: 1);
118
119 for (unsigned j = 0; j < MAX_BOUND; j++) {
120 if (!test(upper_bound: j))
121 return 1;
122 }
123 Free(p: execution_count);
124 Free(p: iterations_per_thread);
125 return 0;
126}
127

source code of openmp/runtime/test/worksharing/for/omp_for_collapse_LowerTriangularLess.c