1 | // RUN: %libomp-compile-and-run |
2 | |
3 | // The test checks schedule(simd:runtime) |
4 | // in combination with omp_set_schedule() |
5 | #include <stdio.h> |
6 | #include <stdlib.h> |
7 | #include <omp.h> |
8 | |
9 | #if defined(WIN32) || defined(_WIN32) |
10 | #include <windows.h> |
11 | #define delay() Sleep(1); |
12 | #define seten(a,b,c) _putenv_s((a),(b)) |
13 | #else |
14 | #include <unistd.h> |
15 | #define delay() usleep(10); |
16 | #define seten(a,b,c) setenv((a),(b),(c)) |
17 | #endif |
18 | |
19 | #define SIMD_LEN 4 |
20 | int err = 0; |
21 | |
22 | // --------------------------------------------------------------------------- |
23 | // Various definitions copied from OpenMP RTL. |
24 | enum sched { |
25 | kmp_sch_static_balanced_chunked = 45, |
26 | kmp_sch_guided_simd = 46, |
27 | kmp_sch_runtime_simd = 47, |
28 | }; |
29 | typedef unsigned u32; |
30 | typedef long long i64; |
31 | typedef unsigned long long u64; |
32 | typedef struct { |
33 | int reserved_1; |
34 | int flags; |
35 | int reserved_2; |
36 | int reserved_3; |
37 | char *psource; |
38 | } id; |
39 | |
40 | #ifdef __cplusplus |
41 | extern "C" { |
42 | #endif |
43 | int __kmpc_global_thread_num(id*); |
44 | void __kmpc_barrier(id*, int gtid); |
45 | void __kmpc_dispatch_init_4(id*, int, enum sched, int, int, int, int); |
46 | void __kmpc_dispatch_init_8(id*, int, enum sched, i64, i64, i64, i64); |
47 | int __kmpc_dispatch_next_4(id*, int, void*, void*, void*, void*); |
48 | int __kmpc_dispatch_next_8(id*, int, void*, void*, void*, void*); |
49 | #ifdef __cplusplus |
50 | } // extern "C" |
51 | #endif |
52 | // End of definitions copied from OpenMP RTL. |
53 | // --------------------------------------------------------------------------- |
54 | static id loc = {.reserved_1: 0, .flags: 2, .reserved_2: 0, .reserved_3: 0, .psource: ";file;func;0;0;;" }; |
55 | |
56 | // --------------------------------------------------------------------------- |
57 | void |
58 | run_loop( |
59 | int loop_lb, // Loop lower bound. |
60 | int loop_ub, // Loop upper bound. |
61 | int loop_st, // Loop stride. |
62 | int lchunk |
63 | ) { |
64 | static int volatile loop_sync = 0; |
65 | int lb; // Chunk lower bound. |
66 | int ub; // Chunk upper bound. |
67 | int st; // Chunk stride. |
68 | int rc; |
69 | int nthreads = omp_get_num_threads(); |
70 | int tid = omp_get_thread_num(); |
71 | int gtid = __kmpc_global_thread_num(&loc); |
72 | int last; |
73 | int tc = (loop_ub - loop_lb) / loop_st + 1; |
74 | int ch; |
75 | int no_chunk = 0; |
76 | if (lchunk == 0) { |
77 | no_chunk = 1; |
78 | lchunk = 1; |
79 | } |
80 | ch = lchunk * SIMD_LEN; |
81 | #if _DEBUG > 1 |
82 | printf("run_loop gtid %d tid %d (lb=%d, ub=%d, st=%d, ch=%d)\n" , |
83 | gtid, tid, (int)loop_lb, (int)loop_ub, (int)loop_st, lchunk); |
84 | #endif |
85 | // Don't test degenerate cases that should have been discovered by codegen. |
86 | if (loop_st == 0) |
87 | return; |
88 | if (loop_st > 0 ? loop_lb > loop_ub : loop_lb < loop_ub) |
89 | return; |
90 | __kmpc_dispatch_init_4(&loc, gtid, kmp_sch_runtime_simd, |
91 | loop_lb, loop_ub, loop_st, SIMD_LEN); |
92 | { |
93 | // Let the master thread handle the chunks alone. |
94 | int chunk; // No of current chunk. |
95 | int last_ub; // Upper bound of the last processed chunk. |
96 | u64 cur; // Number of interations in current chunk. |
97 | u64 max; // Max allowed iterations for current chunk. |
98 | int undersized = 0; |
99 | last_ub = loop_ub; |
100 | chunk = 0; |
101 | max = (loop_ub - loop_lb) / loop_st + 1; |
102 | // The first chunk can consume all iterations. |
103 | while (__kmpc_dispatch_next_4(&loc, gtid, &last, &lb, &ub, &st)) { |
104 | ++ chunk; |
105 | #if _DEBUG |
106 | printf(format: "th %d: chunk=%d, lb=%d, ub=%d ch %d\n" , |
107 | tid, chunk, (int)lb, (int)ub, (int)(ub-lb+1)); |
108 | #endif |
109 | // Check if previous chunk (it is not the final chunk) is undersized. |
110 | if (undersized) |
111 | printf(format: "Error with chunk %d, th %d, err %d\n" , chunk, tid, ++err); |
112 | if (loop_st > 0) { |
113 | if (!(ub <= loop_ub)) |
114 | printf(format: "Error with ub %d, %d, ch %d, err %d\n" , |
115 | (int)ub, (int)loop_ub, chunk, ++err); |
116 | if (!(lb <= ub)) |
117 | printf(format: "Error with bounds %d, %d, %d, err %d\n" , |
118 | (int)lb, (int)ub, chunk, ++err); |
119 | } else { |
120 | if (!(ub >= loop_ub)) |
121 | printf(format: "Error with ub %d, %d, %d, err %d\n" , |
122 | (int)ub, (int)loop_ub, chunk, ++err); |
123 | if (!(lb >= ub)) |
124 | printf(format: "Error with bounds %d, %d, %d, err %d\n" , |
125 | (int)lb, (int)ub, chunk, ++err); |
126 | }; // if |
127 | // Stride should not change. |
128 | if (!(st == loop_st)) |
129 | printf(format: "Error with st %d, %d, ch %d, err %d\n" , |
130 | (int)st, (int)loop_st, chunk, ++err); |
131 | cur = ( ub - lb ) / loop_st + 1; |
132 | // Guided scheduling uses FP computations, so current chunk may |
133 | // be a bit bigger (+1) than allowed maximum. |
134 | if (!( cur <= max + 1)) |
135 | printf(format: "Error with iter %llu, %llu, err %d\n" , cur, max, ++err); |
136 | // Update maximum for the next chunk. |
137 | if (last) { |
138 | if (!no_chunk && cur > ch && nthreads > 1) |
139 | printf(format: "Error: too big last chunk %d (%d), tid %d, err %d\n" , |
140 | (int)cur, ch, tid, ++err); |
141 | } else { |
142 | if (cur % ch) |
143 | printf(format: "Error with chunk %d, %d, ch %d, tid %d, err %d\n" , |
144 | chunk, (int)cur, ch, tid, ++err); |
145 | } |
146 | if (cur < max) |
147 | max = cur; |
148 | last_ub = ub; |
149 | undersized = (cur < ch); |
150 | #if _DEBUG > 1 |
151 | if (last) |
152 | printf("under%d cur %d, ch %d, tid %d, ub %d, lb %d, st %d =======\n" , |
153 | undersized,cur,ch,tid,ub,lb,loop_st); |
154 | #endif |
155 | } // while |
156 | // Must have the right last iteration index. |
157 | if (loop_st > 0) { |
158 | if (!(last_ub <= loop_ub)) |
159 | printf(format: "Error with last1 %d, %d, ch %d, err %d\n" , |
160 | (int)last_ub, (int)loop_ub, chunk, ++err); |
161 | if (last && !(last_ub + loop_st > loop_ub)) |
162 | printf(format: "Error with last2 %d, %d, %d, ch %d, err %d\n" , |
163 | (int)last_ub, (int)loop_st, (int)loop_ub, chunk, ++err); |
164 | } else { |
165 | if (!(last_ub >= loop_ub)) |
166 | printf(format: "Error with last1 %d, %d, ch %d, err %d\n" , |
167 | (int)last_ub, (int)loop_ub, chunk, ++err); |
168 | if (last && !(last_ub + loop_st < loop_ub)) |
169 | printf(format: "Error with last2 %d, %d, %d, ch %d, err %d\n" , |
170 | (int)last_ub, (int)loop_st, (int)loop_ub, chunk, ++err); |
171 | } // if |
172 | } |
173 | __kmpc_barrier(&loc, gtid); |
174 | } // run_loop |
175 | |
176 | int main(int argc, char *argv[]) |
177 | { |
178 | int chunk = 0; |
179 | // static (no chunk) |
180 | omp_set_schedule(omp_sched_static,0); |
181 | #pragma omp parallel// num_threads(num_th) |
182 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
183 | |
184 | // auto (chunk should be ignorted) |
185 | omp_set_schedule(omp_sched_auto,0); |
186 | #pragma omp parallel// num_threads(num_th) |
187 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
188 | |
189 | // static,1 |
190 | chunk = 1; |
191 | omp_set_schedule(omp_sched_static,1); |
192 | #pragma omp parallel// num_threads(num_th) |
193 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
194 | |
195 | // dynamic,1 |
196 | omp_set_schedule(omp_sched_dynamic,1); |
197 | #pragma omp parallel// num_threads(num_th) |
198 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
199 | |
200 | // guided,1 |
201 | omp_set_schedule(omp_sched_guided,1); |
202 | #pragma omp parallel// num_threads(num_th) |
203 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
204 | |
205 | // dynamic,0 - use default chunk size 1 |
206 | omp_set_schedule(omp_sched_dynamic,0); |
207 | #pragma omp parallel// num_threads(num_th) |
208 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
209 | |
210 | // guided,0 - use default chunk size 1 |
211 | omp_set_schedule(omp_sched_guided,0); |
212 | #pragma omp parallel// num_threads(num_th) |
213 | run_loop(loop_lb: 0, loop_ub: 26, loop_st: 1, lchunk: chunk); |
214 | |
215 | if (err) { |
216 | printf(format: "failed, err = %d\n" , err); |
217 | return 1; |
218 | } else { |
219 | printf(format: "passed\n" ); |
220 | return 0; |
221 | } |
222 | } |
223 | |