1 // RUN: %libomp-compile-and-run 2 // RUN: %libomp-compile && env KMP_TASKLOOP_MIN_TASKS=1 %libomp-run 3 // REQUIRES: openmp-4.5 4 5 // These compilers don't support the taskloop construct 6 // UNSUPPORTED: gcc-4, gcc-5, icc-16 7 // GCC 6 has support for taskloops, but at least 6.3.0 is crashing on this test 8 // UNSUPPORTED: gcc-6 9 10 /* 11 * Test for taskloop 12 * Method: caculate how many times the iteration space is dispatched 13 * and judge if each dispatch has the requested grainsize 14 * It is possible for two adjacent chunks are executed by the same thread 15 */ 16 #include <stdio.h> 17 #include <omp.h> 18 #include <stdlib.h> 19 #include "omp_testsuite.h" 20 21 #define CFDMAX_SIZE 1120 22 23 int test_omp_taskloop_grainsize() 24 { 25 int result = 0; 26 int i, grainsize, count, tmp_count, num_off; 27 int *tmp, *tids, *tidsArray; 28 29 tidsArray = (int *)malloc(sizeof(int) * CFDMAX_SIZE); 30 tids = tidsArray; 31 32 for (grainsize = 1; grainsize < 48; ++grainsize) { 33 fprintf(stderr, "Grainsize %d\n", grainsize); 34 count = tmp_count = num_off = 0; 35 36 for (i = 0; i < CFDMAX_SIZE; ++i) { 37 tids[i] = -1; 38 } 39 40 #pragma omp parallel shared(tids) 41 { 42 #pragma omp master 43 #pragma omp taskloop grainsize(grainsize) 44 for (i = 0; i < CFDMAX_SIZE; i++) { 45 tids[i] = omp_get_thread_num(); 46 } 47 } 48 49 for (i = 0; i < CFDMAX_SIZE; ++i) { 50 if (tids[i] == -1) { 51 fprintf(stderr, " Iteration %d not touched!\n", i); 52 result++; 53 } 54 } 55 56 for (i = 0; i < CFDMAX_SIZE - 1; ++i) { 57 if (tids[i] != tids[i + 1]) { 58 count++; 59 } 60 } 61 62 tmp = (int *)malloc(sizeof(int) * (count + 1)); 63 tmp[0] = 1; 64 65 for (i = 0; i < CFDMAX_SIZE - 1; ++i) { 66 if (tmp_count > count) { 67 printf("--------------------\nTestinternal Error: List too " 68 "small!!!\n--------------------\n"); 69 break; 70 } 71 if (tids[i] != tids[i + 1]) { 72 tmp_count++; 73 tmp[tmp_count] = 1; 74 } else { 75 tmp[tmp_count]++; 76 } 77 } 78 79 // is grainsize statement working? 80 int num_tasks = CFDMAX_SIZE / grainsize; 81 int multiple1 = CFDMAX_SIZE / num_tasks; 82 int multiple2 = CFDMAX_SIZE / num_tasks + 1; 83 for (i = 0; i < count; i++) { 84 // it is possible for 2 adjacent chunks assigned to a same thread 85 if (tmp[i] % multiple1 != 0 && tmp[i] % multiple2 != 0) { 86 num_off++; 87 } 88 } 89 90 if (num_off > 1) { 91 fprintf(stderr, " The number of bad chunks is %d\n", num_off); 92 result++; 93 } else { 94 fprintf(stderr, " Everything ok\n"); 95 } 96 97 free(tmp); 98 } 99 free(tidsArray); 100 return (result==0); 101 } 102 103 int main() 104 { 105 int i; 106 int num_failed=0; 107 108 for (i = 0; i < REPETITIONS; i++) { 109 if (!test_omp_taskloop_grainsize()) { 110 num_failed++; 111 } 112 } 113 return num_failed; 114 } 115