132 lines · c
1// clang-format off2// RUN: %libomp-compile-and-run | FileCheck %s3// REQUIRES: ompt4// clang-format on5 6#include "callback.h"7#include <omp.h>8 9__attribute__ ((noinline)) // workaround for bug in icc10void print_task_info_at(int ancestor_level, int id)11{12#pragma omp critical13 {14 int task_type;15char buffer[2048];16ompt_data_t *parallel_data;17ompt_data_t *task_data;18int thread_num;19ompt_get_task_info(ancestor_level, &task_type, &task_data, NULL, ¶llel_data,20 &thread_num);21format_task_type(task_type, buffer);22printf("%" PRIu64 ": ancestor_level=%d id=%d task_type=%s=%d "23 "parallel_id=%" PRIx64 " task_id=%" PRIx64 " thread_num=%d\n",24 ompt_get_thread_data()->value, ancestor_level, id, buffer, task_type,25 parallel_data->value, task_data->value, thread_num);26}27}28;29 30__attribute__ ((noinline)) // workaround for bug in icc31void print_innermost_task_info(int id)32{33 print_task_info_at(0, id);34}35 36int main() {37 38#pragma omp parallel num_threads(2)39 {40 // sync threads before checking the output41#pragma omp barrier42 // region 043 if (omp_get_thread_num() == 1) {44 // executed by worker thread only45 // assert that thread_num is 146 print_innermost_task_info(1);47 48#pragma omp parallel num_threads(1)49 {50 // serialized region 151 // assert that thread_num is 052 print_innermost_task_info(2);53 54#pragma omp parallel num_threads(1)55 {56 // serialized region 257 // assert that thread_num is 058 print_innermost_task_info(3);59 60 // Check the value of thread_num while iterating over the hierarchy61 // of active tasks.62 print_task_info_at(0, 3);63 print_task_info_at(1, 2);64 print_task_info_at(2, 1);65 }66 }67 }68 }69 70 // clang-format off71 // Check if libomp supports the callbacks for this test.72 // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'73 // CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'74 75 76 // CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]77 // CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_initial_task_begin: parallel_id=[[PARALLEL_ID_0:[0-9]+]], task_id=[[TASK_ID_0:[0-9]+]], actual_parallelism=1, index=1, flags=178 79 // region 080 // CHECK: {{^}}[[MASTER_ID]]: ompt_event_parallel_begin: parent_task_id=[[TASK_ID_0]],81 // CHECK-SAME: parallel_id=[[PARALLEL_ID_1:[0-9]+]]82 // CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID_1]], task_id=[[TASK_ID_1:[0-9]+]]83 // CHECK-DAG: {{^}}[[WORKER_ID:[0-9]+]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID_1]], task_id=[[TASK_ID_2:[0-9]+]]84 // assert some info about implicit task executed by worker thread85 // thread_num is the most important86 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=0 id=187 // CHECK-SAME: parallel_id=[[PARALLEL_ID_1]] task_id=[[TASK_ID_2]]88 // CHECK-SAME: thread_num=189 90 // serialized region 191 // CHECK: {{^}}[[WORKER_ID]]: ompt_event_parallel_begin: parent_task_id=[[TASK_ID_2]],92 // CHECK-SAME: parallel_id=[[PARALLEL_ID_2:[0-9]+]]93 // CHECK-DAG: {{^}}[[WORKER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID_2]], task_id=[[TASK_ID_3:[0-9]+]]94 // assert some information about the implicit task of the serialized region 195 // pay attention that thread_num should take value 096 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=0 id=297 // CHECK-SAME: parallel_id=[[PARALLEL_ID_2]] task_id=[[TASK_ID_3]]98 // CHECK-SAME: thread_num=099 100 // serialized region 2101 // CHECK: {{^}}[[WORKER_ID]]: ompt_event_parallel_begin: parent_task_id=[[TASK_ID_3]],102 // CHECK-SAME: parallel_id=[[PARALLEL_ID_3:[0-9]+]]103 // CHECK-DAG: {{^}}[[WORKER_ID]]: ompt_event_implicit_task_begin: parallel_id=[[PARALLEL_ID_3]], task_id=[[TASK_ID_4:[0-9]+]]104 // assert some information about the implicit task of the serialized region 2105 // pay attention that thread_num should take value 0106 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=0 id=3107 // CHECK-SAME: parallel_id=[[PARALLEL_ID_3]] task_id=[[TASK_ID_4]]108 // CHECK-SAME: thread_num=0109 110 // Check the value of thread_num argument while iterating over the hierarchy111 // of active tasks. The expected is that thread_num takes the value checked112 // above in the test case (0, 0, 1 - respectively).113 114 // Thread is the master thread of the region 2, so thread_num should be 0.115 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=0 id=3116 // CHECK-SAME: parallel_id=[[PARALLEL_ID_3]] task_id=[[TASK_ID_4]]117 // CHECK-SAME: thread_num=0118 119 // Thread is the master thread of the region 1, so thread_num should be 0.120 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=1 id=2121 // CHECK-SAME: parallel_id=[[PARALLEL_ID_2]] task_id=[[TASK_ID_3]]122 // CHECK-SAME: thread_num=0123 124 // Thread is the worker thread of the region 0, so thread_num should be 1.125 // CHECK: {{^}}[[WORKER_ID]]: ancestor_level=2 id=1126 // CHECK-SAME: parallel_id=[[PARALLEL_ID_1]] task_id=[[TASK_ID_2]]127 // CHECK-SAME: thread_num=1128 // clang-format on129 130 return 0;131}132