145
|
1 #include <omp.h>
|
|
2 #include <stdlib.h>
|
|
3
|
|
4 struct S { unsigned long long int s, t; };
|
|
5
|
|
6 void
|
|
7 rbar (struct S *p, struct S *o)
|
|
8 {
|
|
9 p->s = 1;
|
|
10 if (o->t != 5)
|
|
11 abort ();
|
|
12 p->t = 9;
|
|
13 }
|
|
14
|
|
15 static inline void
|
|
16 rbaz (struct S *o, struct S *i)
|
|
17 {
|
|
18 if (o->t != 5 || i->t != 9)
|
|
19 abort ();
|
|
20 o->s *= i->s;
|
|
21 }
|
|
22
|
|
23 #pragma omp declare reduction (+: struct S : omp_out.s += omp_in.s) \
|
|
24 initializer (omp_priv = { 0, 3 })
|
|
25 #pragma omp declare reduction (*: struct S : rbaz (&omp_out, &omp_in)) \
|
|
26 initializer (rbar (&omp_priv, &omp_orig))
|
|
27
|
|
28 struct S g = { 0, 7 };
|
|
29 struct S h = { 1, 5 };
|
|
30
|
|
31 int
|
|
32 foo (int z, int *a, int *b)
|
|
33 {
|
|
34 int x = 0;
|
|
35 #pragma omp taskloop reduction (+:x) in_reduction (+:b[0])
|
|
36 for (int i = z; i < z + 8; i++)
|
|
37 {
|
|
38 x += a[i];
|
|
39 *b += a[i] * 2;
|
|
40 }
|
|
41 return x;
|
|
42 }
|
|
43
|
|
44 unsigned long long int
|
|
45 bar (int z, int *a, unsigned long long int *b, int *s)
|
|
46 {
|
|
47 unsigned long long int x = 1;
|
|
48 #pragma omp taskloop reduction (*:x) in_reduction (*:b[0]) \
|
|
49 in_reduction (+:s[0])
|
|
50 for (int i = z; i < z + 8; i++)
|
|
51 {
|
|
52 #pragma omp task in_reduction (*:x)
|
|
53 x *= a[i];
|
|
54 #pragma omp task in_reduction (*:b[0])
|
|
55 *b *= (3 - a[i]);
|
|
56 s[0]++;
|
|
57 }
|
|
58 return x;
|
|
59 }
|
|
60
|
|
61 void
|
|
62 baz (int i, int *a, int *c)
|
|
63 {
|
|
64 #pragma omp task in_reduction (*:h) in_reduction (+:g)
|
|
65 {
|
|
66 g.s += 7 * a[i];
|
|
67 h.s *= (3 - c[i]);
|
|
68 if ((g.t != 7 && g.t != 3) || (h.t != 5 && h.t != 9))
|
|
69 abort ();
|
|
70 }
|
|
71 }
|
|
72
|
|
73 int
|
|
74 main ()
|
|
75 {
|
|
76 int i, j = 0, a[64], b = 0, c[64], f = 0;
|
|
77 unsigned long long int d = 1, e = 1;
|
|
78 volatile int one = 1;
|
|
79 int r = 0, s = 0, t;
|
|
80 struct S m = { 0, 7 };
|
|
81 struct S n = { 1, 5 };
|
|
82 for (i = 0; i < 64; i++)
|
|
83 {
|
|
84 a[i] = 2 * i;
|
|
85 c[i] = 1 + ((i % 3) != 1);
|
|
86 }
|
|
87 #pragma omp parallel reduction (task, +:b) shared(t) reduction(+:r, s)
|
|
88 {
|
|
89 int z, q1, q2, q3;
|
|
90 #pragma omp master
|
|
91 t = omp_get_num_threads ();
|
|
92 #pragma omp for schedule(static) reduction (task, +: f) reduction (+: j)
|
|
93 for (z = 0; z < 64; z += 8)
|
|
94 {
|
|
95 f++;
|
|
96 j += foo (z, a, &b);
|
|
97 j += foo (z, a, &f);
|
|
98 }
|
|
99 if (j != 63 * 64 * 2 || f != 63 * 64 * 2 + 8)
|
|
100 abort ();
|
|
101 r++;
|
|
102 #pragma omp taskgroup task_reduction (+: s)
|
|
103 {
|
|
104 #pragma omp for schedule(static, 1) reduction(task, *: d) reduction (*: e)
|
|
105 for (z = 0; z < 64; z += 8)
|
|
106 e *= bar (z, c, &d, &s);
|
|
107 }
|
|
108 if (e != (1ULL << 43) || d != (1ULL << 21))
|
|
109 abort ();
|
|
110 #pragma omp for schedule(monotonic: dynamic, 1) reduction (task, +: g, m) \
|
|
111 reduction (task, *: h, n) collapse(3)
|
|
112 for (q1 = 0; q1 < one; q1++)
|
|
113 for (q2 = 0; q2 < 64; q2 += 8)
|
|
114 for (q3 = 0; q3 < one; ++q3)
|
|
115 #pragma omp taskloop in_reduction (+: g, m) in_reduction (*: h, n) \
|
|
116 nogroup
|
|
117 for (i = q2; i < q2 + 8; ++i)
|
|
118 {
|
|
119 g.s += 3 * a[i];
|
|
120 h.s *= (3 - c[i]);
|
|
121 m.s += 4 * a[i];
|
|
122 n.s *= c[i];
|
|
123 if ((g.t != 7 && g.t != 3) || (h.t != 5 && h.t != 9)
|
|
124 || (m.t != 7 && m.t != 3) || (n.t != 5 && n.t != 9))
|
|
125 abort ();
|
|
126 baz (i, a, c);
|
|
127 }
|
|
128 if (n.s != (1ULL << 43) || n.t != 5)
|
|
129 abort ();
|
|
130 if (g.s != 63 * 64 * 10 || g.t != 7)
|
|
131 abort ();
|
|
132 if (h.s != (1ULL << 42) || h.t != 5)
|
|
133 abort ();
|
|
134 if (m.s != 63 * 64 * 4 || m.t != 7)
|
|
135 abort ();
|
|
136 }
|
|
137 if (b != 63 * 64 * 2)
|
|
138 abort ();
|
|
139 if (r != t || s != 64)
|
|
140 abort ();
|
|
141 return 0;
|
|
142 }
|