| 1 | // RUN: %libomp-compile-and-run |
| 2 | // RUN: %libomp-compile && env KMP_TASKLOOP_MIN_TASKS=1 %libomp-run |
| 3 | |
| 4 | #include <stdio.h> |
| 5 | #include <omp.h> |
| 6 | #include "omp_my_sleep.h" |
| 7 | |
| 8 | #define N 4 |
| 9 | #define ST 3 |
| 10 | #define UB 118 |
| 11 | #define LB 0 |
| 12 | |
| 13 | // globals |
| 14 | int counter; |
| 15 | int task_count; |
| 16 | |
| 17 | // Compiler-generated code (emulation) |
| 18 | typedef struct ident { |
| 19 | void* dummy; |
| 20 | } ident_t; |
| 21 | |
| 22 | typedef struct shar { |
| 23 | int *pcounter; |
| 24 | int *pj; |
| 25 | int *ptask_count; |
| 26 | } *pshareds; |
| 27 | |
| 28 | typedef struct task { |
| 29 | pshareds shareds; |
| 30 | int(* routine)(int,struct task*); |
| 31 | int part_id; |
| 32 | unsigned long long lb; // library always uses ULONG |
| 33 | unsigned long long ub; |
| 34 | int st; |
| 35 | int last; |
| 36 | int i; |
| 37 | int j; |
| 38 | int th; |
| 39 | } *ptask, kmp_task_t; |
| 40 | |
| 41 | typedef int(* task_entry_t)( int, ptask ); |
| 42 | |
| 43 | void |
| 44 | __task_dup_entry(ptask task_dst, ptask task_src, int lastpriv) |
| 45 | { |
| 46 | // setup lastprivate flag |
| 47 | task_dst->last = lastpriv; |
| 48 | // could be constructor calls here... |
| 49 | } |
| 50 | |
| 51 | // OpenMP RTL interfaces |
| 52 | typedef unsigned long long kmp_uint64; |
| 53 | typedef long long kmp_int64; |
| 54 | |
| 55 | #ifdef __cplusplus |
| 56 | extern "C" { |
| 57 | #endif |
| 58 | void |
| 59 | __kmpc_taskloop_5(ident_t *loc, int gtid, kmp_task_t *task, int if_val, |
| 60 | kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st, |
| 61 | int nogroup, int sched, kmp_int64 grainsize, int modifier, |
| 62 | void *task_dup); |
| 63 | ptask |
| 64 | __kmpc_omp_task_alloc(ident_t *loc, int gtid, int flags, |
| 65 | size_t sizeof_kmp_task_t, size_t sizeof_shareds, |
| 66 | task_entry_t task_entry); |
| 67 | void __kmpc_atomic_fixed4_add(void *id_ref, int gtid, int * lhs, int rhs); |
| 68 | int __kmpc_global_thread_num(void *id_ref); |
| 69 | #ifdef __cplusplus |
| 70 | } |
| 71 | #endif |
| 72 | |
| 73 | // User's code |
| 74 | int task_entry(int gtid, ptask task) |
| 75 | { |
| 76 | pshareds pshar = task->shareds; |
| 77 | __kmpc_atomic_fixed4_add(NULL, gtid, lhs: pshar->ptask_count, rhs: 1); |
| 78 | |
| 79 | for (task->i = task->lb; task->i <= (int)task->ub; task->i += task->st) { |
| 80 | task->th = omp_get_thread_num(); |
| 81 | __kmpc_atomic_fixed4_add(NULL,gtid,lhs: pshar->pcounter,rhs: 1); |
| 82 | task->j = task->i; |
| 83 | } |
| 84 | my_sleep( 0.1 ); // sleep 100 ms in order to allow other threads to steal tasks |
| 85 | if (task->last) { |
| 86 | *(pshar->pj) = task->j; // lastprivate |
| 87 | } |
| 88 | return 0; |
| 89 | } |
| 90 | |
| 91 | void task_loop(int sched_type, int sched_val, int modifier) |
| 92 | { |
| 93 | int i, j, gtid = __kmpc_global_thread_num(NULL); |
| 94 | ptask task; |
| 95 | pshareds psh; |
| 96 | omp_set_dynamic(0); |
| 97 | counter = 0; |
| 98 | task_count = 0; |
| 99 | #pragma omp parallel num_threads(N) |
| 100 | { |
| 101 | #pragma omp master |
| 102 | { |
| 103 | int gtid = __kmpc_global_thread_num(NULL); |
| 104 | task = __kmpc_omp_task_alloc(NULL, gtid, flags: 1, sizeof_kmp_task_t: sizeof(struct task), |
| 105 | sizeof_shareds: sizeof(struct shar), task_entry: &task_entry); |
| 106 | psh = task->shareds; |
| 107 | psh->pcounter = &counter; |
| 108 | psh->ptask_count = &task_count; |
| 109 | psh->pj = &j; |
| 110 | task->lb = LB; |
| 111 | task->ub = UB; |
| 112 | task->st = ST; |
| 113 | |
| 114 | __kmpc_taskloop_5( |
| 115 | NULL, // location |
| 116 | gtid, // gtid |
| 117 | task, // task structure |
| 118 | if_val: 1, // if clause value |
| 119 | lb: &task->lb, // lower bound |
| 120 | ub: &task->ub, // upper bound |
| 121 | ST, // loop increment |
| 122 | nogroup: 0, // 1 if nogroup specified |
| 123 | sched: sched_type, // schedule type: 0-none, 1-grainsize, 2-num_tasks |
| 124 | grainsize: sched_val, // schedule value (ignored for type 0) |
| 125 | modifier, // strict modifier |
| 126 | task_dup: (void*)&__task_dup_entry // tasks duplication routine |
| 127 | ); |
| 128 | } // end master |
| 129 | } // end parallel |
| 130 | // check results |
| 131 | int tc; |
| 132 | if (ST == 1) { // most common case |
| 133 | tc = UB - LB + 1; |
| 134 | } else if (ST < 0) { |
| 135 | tc = (LB - UB) / (-ST) + 1; |
| 136 | } else { // ST > 0 |
| 137 | tc = (UB - LB) / ST + 1; |
| 138 | } |
| 139 | int count; |
| 140 | if (sched_type == 1) { |
| 141 | count = (sched_val > tc) ? 1 : (tc + sched_val - 1) / sched_val; |
| 142 | } else { |
| 143 | count = (sched_val > tc) ? tc : sched_val; |
| 144 | } |
| 145 | if (j != LB + (tc - 1) * ST) { |
| 146 | printf(format: "Error in lastprivate, %d != %d\n" , j, LB + (tc - 1) * ST); |
| 147 | exit(1); |
| 148 | } |
| 149 | if (counter != tc) { |
| 150 | printf(format: "Error, counter %d != %d\n" , counter, tc); |
| 151 | exit(1); |
| 152 | } |
| 153 | if (task_count != count) { |
| 154 | printf(format: "Error, task count %d != %d\n" , task_count, count); |
| 155 | exit(1); |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | int main(int argc, char *argv[]) { |
| 160 | task_loop(sched_type: 1, sched_val: 6, modifier: 1); // create 7 tasks |
| 161 | task_loop(sched_type: 2, sched_val: 6, modifier: 1); // create 6 tasks |
| 162 | task_loop(sched_type: 1, sched_val: 50, modifier: 1); // create 1 task |
| 163 | task_loop(sched_type: 2, sched_val: 50, modifier: 1); // create 40 tasks |
| 164 | |
| 165 | printf(format: "Test passed\n" ); |
| 166 | return 0; |
| 167 | } |
| 168 | |