1 extern
2 #ifdef __cplusplus
3 "C"
4 #endif
5 void abort (void);
6
7 void
fn1(double * x,double * y,int z)8 fn1 (double *x, double *y, int z)
9 {
10 int i;
11 for (i = 0; i < z; i++)
12 {
13 x[i] = i & 31;
14 y[i] = (i & 63) - 30;
15 }
16 }
17
18 #pragma omp declare target
19 int tgtv = 6;
20 int
tgt(void)21 tgt (void)
22 {
23 #pragma omp atomic update
24 tgtv++;
25 return 0;
26 }
27 #pragma omp end declare target
28
29 double
fn2(int x,int y,int z)30 fn2 (int x, int y, int z)
31 {
32 double b[1024], c[1024], s = 0;
33 int i, j;
34 fn1 (b, c, x);
35 #pragma omp target data map(to: b)
36 {
37 #pragma omp target map(tofrom: c, s)
38 #pragma omp teams num_teams(y) thread_limit(z) reduction(+:s) firstprivate(x)
39 #pragma omp distribute dist_schedule(static, 4) collapse(1)
40 for (j=0; j < x; j += y)
41 #pragma omp parallel for reduction(+:s)
42 for (i = j; i < j + y; i++)
43 tgt (), s += b[i] * c[i];
44 #pragma omp target update from(b, tgtv)
45 }
46 return s;
47 }
48
49 double
fn3(int x)50 fn3 (int x)
51 {
52 double b[1024], c[1024], s = 0;
53 int i;
54 fn1 (b, c, x);
55 #pragma omp target map(to: b, c) map(tofrom:s)
56 #pragma omp parallel for reduction(+:s)
57 for (i = 0; i < x; i++)
58 tgt (), s += b[i] * c[i];
59 return s;
60 }
61
62 double
fn4(int x,double * p)63 fn4 (int x, double *p)
64 {
65 double b[1024], c[1024], d[1024], s = 0;
66 int i;
67 fn1 (b, c, x);
68 fn1 (d + x, p + x, x);
69 #pragma omp target map(to: b, c[0:x], d[x:x]) map(to:p[x:64 + (x & 31)]) \
70 map(tofrom: s)
71 #pragma omp parallel for reduction(+:s)
72 for (i = 0; i < x; i++)
73 s += b[i] * c[i] + d[x + i] + p[x + i];
74 return s;
75 }
76
77 int
main()78 main ()
79 {
80 double a = fn2 (128, 4, 6);
81 int b = tgtv;
82 double c = fn3 (61);
83 #pragma omp target update from(tgtv)
84 int d = tgtv;
85 double e[1024];
86 double f = fn4 (64, e);
87 if (a != 13888.0 || b != 6 + 128 || c != 4062.0 || d != 6 + 128 + 61
88 || f != 8032.0)
89 abort ();
90 return 0;
91 }
92