LLVM OpenMP 22.0.0git
loop_dispatch.c
Go to the documentation of this file.
1// clang-format off
2// RUN: %libomp-compile-and-run | %sort-threads | FileCheck %s
3// REQUIRES: ompt
4// XFAIL: gcc
5// GCC doesn't call runtime for static schedule
6// clang-format on
7
8#include "callback.h"
9
10#define WORK_SIZE 64
11
12int main() {
13 int i;
14 int wait_s = 0;
15
16#pragma omp parallel num_threads(4)
17 {
18 int wait_id = 0;
19 int team_size = omp_get_num_threads();
20#pragma omp for schedule(static, WORK_SIZE / 4)
21 for (i = 0; i < WORK_SIZE; i++) {
22 }
23
24#pragma omp for schedule(dynamic)
25 for (i = 0; i < WORK_SIZE; i++) {
26 if (wait_id == 0) {
27 // Wait until every thread has at least one iteration assigned
28 OMPT_SIGNAL(wait_s);
29 OMPT_WAIT(wait_s, team_size);
30 wait_id++;
31 }
32 }
33
34#pragma omp for schedule(guided)
35 for (i = 0; i < WORK_SIZE; i++) {
36 if (wait_id == 1) {
37 // Wait until every thread has at least one iteration assigned
38 OMPT_SIGNAL(wait_s);
39 OMPT_WAIT(wait_s, 2 * team_size);
40 wait_id++;
41 }
42 }
43 }
44
45 return 0;
46}
47
48// clang-format off
49// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_parallel_begin'
50// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_implicit_task'
51// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_work'
52// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_dispatch'
53
54// CHECK: 0: NULL_POINTER=[[NULL:.*$]]
55// CHECK: {{^}}[[THREAD_ID0:[0-9]+]]: ompt_event_parallel_begin:
56// CHECK-SAME: parallel_id=[[PARALLEL_ID:[0-f]+]]
57
58// Each thread should have at least one ws-loop-chunk-begin event for each
59// for loop.
60
61// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_implicit_task_begin:
62// CHECK-SAME: task_id=[[TASK_ID0:[0-9]+]]
63// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_loop_static_begin:
64// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
65// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_ws_loop_chunk_begin:
66// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
67// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
68// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_loop_dynamic_begin:
69// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
70// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_ws_loop_chunk_begin:
71// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
72// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=1
73// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_loop_guided_begin:
74// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
75// CHECK: {{^}}[[THREAD_ID0]]: ompt_event_ws_loop_chunk_begin:
76// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID0]]
77// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations={{[1-9][0-9]*}}
78
79// CHECK: {{^}}[[THREAD_ID1:[0-9]+]]: ompt_event_implicit_task_begin:
80// CHECK-SAME: task_id=[[TASK_ID1:[0-9]+]]
81// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_loop_static_begin:
82// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
83// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_ws_loop_chunk_begin:
84// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
85// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
86// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_loop_dynamic_begin:
87// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
88// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_ws_loop_chunk_begin:
89// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
90// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=1
91// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_loop_guided_begin:
92// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
93// CHECK: {{^}}[[THREAD_ID1]]: ompt_event_ws_loop_chunk_begin:
94// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID1]]
95// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations={{[1-9][0-9]*}}
96
97// CHECK: {{^}}[[THREAD_ID2:[0-9]+]]: ompt_event_implicit_task_begin:
98// CHECK-SAME: task_id=[[TASK_ID2:[0-9]+]]
99// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_loop_static_begin:
100// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
101// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_ws_loop_chunk_begin:
102// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
103// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
104// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_loop_dynamic_begin:
105// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
106// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_ws_loop_chunk_begin:
107// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
108// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=1
109// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_loop_guided_begin:
110// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
111// CHECK: {{^}}[[THREAD_ID2]]: ompt_event_ws_loop_chunk_begin:
112// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID2]]
113// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations={{[1-9][0-9]*}}
114
115// CHECK: {{^}}[[THREAD_ID3:[0-9]+]]: ompt_event_implicit_task_begin:
116// CHECK-SAME: task_id=[[TASK_ID3:[0-9]+]]
117// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_loop_static_begin:
118// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
119// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_ws_loop_chunk_begin:
120// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
121// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=16
122// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_loop_dynamic_begin:
123// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
124// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_ws_loop_chunk_begin:
125// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
126// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations=1
127// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_loop_guided_begin:
128// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
129// CHECK: {{^}}[[THREAD_ID3]]: ompt_event_ws_loop_chunk_begin:
130// CHECK-SAME: parallel_id=[[PARALLEL_ID]], task_id=[[TASK_ID3]]
131// CHECK-SAME: chunk_start={{[0-9]+}}, chunk_iterations={{[1-9][0-9]*}}
132// clang-format on
WORK_SIZE
#define WORK_SIZE
GCC lowering of distribute results in calls to omp_get_num_teams/omp_get_team_num rather than region ...
Definition distribute_dispatch.c:11
i
#define i
Definition kmp_stub.cpp:87
main
int main()
Definition loop_dispatch.c:12
OMPT_WAIT
#define OMPT_WAIT(s, v)
Definition ompt-signal.h:20
OMPT_SIGNAL
#define OMPT_SIGNAL(s)
Definition ompt-signal.h:13
omp_get_num_threads
int omp_get_num_threads()
Generated on for LLVM OpenMP by doxygen 1.14.0