145
|
1 // { dg-require-effective-target size32plus }
|
|
2 // { dg-additional-options "-fopenmp-simd" }
|
|
3 // { dg-additional-options "-mavx" { target avx_runtime } }
|
|
4 // { dg-final { scan-tree-dump-times "vectorized \[1-3] loops" 2 "vect" { target i?86-*-* x86_64-*-* } } }
|
|
5
|
|
6 #include "../../gcc.dg/vect/tree-vect.h"
|
|
7
|
|
8 template <typename T>
|
|
9 struct S {
|
|
10 inline S ();
|
|
11 inline ~S ();
|
|
12 inline S (const S &);
|
|
13 inline S & operator= (const S &);
|
|
14 T s;
|
|
15 };
|
|
16
|
|
17 template <typename T>
|
|
18 S<T>::S () : s (0)
|
|
19 {
|
|
20 }
|
|
21
|
|
22 template <typename T>
|
|
23 S<T>::~S ()
|
|
24 {
|
|
25 }
|
|
26
|
|
27 template <typename T>
|
|
28 S<T>::S (const S &x)
|
|
29 {
|
|
30 s = x.s;
|
|
31 }
|
|
32
|
|
33 template <typename T>
|
|
34 S<T> &
|
|
35 S<T>::operator= (const S &x)
|
|
36 {
|
|
37 s = x.s;
|
|
38 return *this;
|
|
39 }
|
|
40
|
|
41 template <typename T>
|
|
42 static inline void
|
|
43 ini (S<T> &x)
|
|
44 {
|
|
45 x.s = 0;
|
|
46 }
|
|
47
|
|
48 S<int> r, a[1024], b[1024];
|
|
49
|
|
50 #pragma omp declare reduction (+: S<int>: omp_out.s += omp_in.s)
|
|
51 #pragma omp declare reduction (plus: S<int>: omp_out.s += omp_in.s) initializer (ini (omp_priv))
|
|
52
|
|
53 template <typename T>
|
|
54 __attribute__((noipa)) void
|
|
55 foo (S<T> *a, S<T> *b)
|
|
56 {
|
|
57 #pragma omp simd reduction (inscan, +:r)
|
|
58 for (int i = 0; i < 1024; i++)
|
|
59 {
|
|
60 b[i] = r;
|
|
61 #pragma omp scan exclusive(r)
|
|
62 r.s += a[i].s;
|
|
63 }
|
|
64 }
|
|
65
|
|
66 template <typename T>
|
|
67 __attribute__((noipa)) S<T>
|
|
68 bar (void)
|
|
69 {
|
|
70 S<T> s;
|
|
71 #pragma omp simd reduction (inscan, plus:s)
|
|
72 for (int i = 0; i < 1024; i++)
|
|
73 {
|
|
74 b[i] = s;
|
|
75 #pragma omp scan exclusive(s)
|
|
76 s.s += 2 * a[i].s;
|
|
77 }
|
|
78 return S<T> (s);
|
|
79 }
|
|
80
|
|
81 __attribute__((noipa)) void
|
|
82 baz (S<int> *a, S<int> *b)
|
|
83 {
|
|
84 #pragma omp simd reduction (inscan, +:r) simdlen(1)
|
|
85 for (int i = 0; i < 1024; i++)
|
|
86 {
|
|
87 b[i] = r;
|
|
88 #pragma omp scan exclusive(r)
|
|
89 r.s += a[i].s;
|
|
90 }
|
|
91 }
|
|
92
|
|
93 __attribute__((noipa)) S<int>
|
|
94 qux (void)
|
|
95 {
|
|
96 S<int> s;
|
|
97 #pragma omp simd if (0) reduction (inscan, plus:s)
|
|
98 for (int i = 0; i < 1024; i++)
|
|
99 {
|
|
100 b[i] = s;
|
|
101 #pragma omp scan exclusive(s)
|
|
102 s.s += 2 * a[i].s;
|
|
103 }
|
|
104 return S<int> (s);
|
|
105 }
|
|
106
|
|
107 int
|
|
108 main ()
|
|
109 {
|
|
110 S<int> s;
|
|
111 check_vect ();
|
|
112 for (int i = 0; i < 1024; ++i)
|
|
113 {
|
|
114 a[i].s = i;
|
|
115 b[i].s = -1;
|
|
116 asm ("" : "+g" (i));
|
|
117 }
|
|
118 foo (a, b);
|
|
119 if (r.s != 1024 * 1023 / 2)
|
|
120 abort ();
|
|
121 for (int i = 0; i < 1024; ++i)
|
|
122 {
|
|
123 if (b[i].s != s.s)
|
|
124 abort ();
|
|
125 else
|
|
126 b[i].s = 25;
|
|
127 s.s += i;
|
|
128 }
|
|
129 if (bar<int> ().s != 1024 * 1023)
|
|
130 abort ();
|
|
131 s.s = 0;
|
|
132 for (int i = 0; i < 1024; ++i)
|
|
133 {
|
|
134 if (b[i].s != s.s)
|
|
135 abort ();
|
|
136 s.s += 2 * i;
|
|
137 }
|
|
138 r.s = 0;
|
|
139 baz (a, b);
|
|
140 if (r.s != 1024 * 1023 / 2)
|
|
141 abort ();
|
|
142 s.s = 0;
|
|
143 for (int i = 0; i < 1024; ++i)
|
|
144 {
|
|
145 if (b[i].s != s.s)
|
|
146 abort ();
|
|
147 else
|
|
148 b[i].s = 25;
|
|
149 s.s += i;
|
|
150 }
|
|
151 if (qux ().s != 1024 * 1023)
|
|
152 abort ();
|
|
153 s.s = 0;
|
|
154 for (int i = 0; i < 1024; ++i)
|
|
155 {
|
|
156 if (b[i].s != s.s)
|
|
157 abort ();
|
|
158 s.s += 2 * i;
|
|
159 }
|
|
160 return 0;
|
|
161 }
|