Debian Codegraph @

libgomp/testsuite/libgomp.c++/scan-16.C

raw
1 // { dg-require-effective-target size32plus }
2 // { dg-additional-options "-O2 -fopenmp -fdump-tree-vect-details" }
3 // { dg-additional-options "-msse2" { target sse2_runtime } }
4 // { dg-additional-options "-mavx" { target avx_runtime } }
5 // { dg-final { scan-tree-dump-times "vectorized \[2-6] loops" 2 "vect" { target sse2_runtime } } }
6
7 extern "C" void abort ();
8
9 struct S {
10 inline S ();
11 inline ~S ();
12 inline S (const S &);
13 inline S & operator= (const S &);
14 int s;
15 };
16
17 S::S () : s (0)
18 {
19 }
20
21 S::~S ()
22 {
23 }
24
25 S::S (const S &x)
26 {
27 s = x.s;
28 }
29
30 S &
31 S::operator= (const S &x)
32 {
33 s = x.s;
34 return *this;
35 }
36
37 static inline void
38 ini (S &x)
39 {
40 x.s = 0;
41 }
42
43 S r, a[1024], b[1024];
44
45 #pragma omp declare reduction (+: S: omp_out.s += omp_in.s)
46 #pragma omp declare reduction (plus: S: omp_out.s += omp_in.s) initializer (ini (omp_priv))
47
48 __attribute__((noipa)) void
49 foo (S *a, S *b, S &r)
50 {
51 #pragma omp for simd simdlen (1) reduction (inscan, +:r)
52 for (int i = 0; i < 1024; i++)
53 {
54 b[i] = r;
55 #pragma omp scan exclusive(r)
56 r.s += a[i].s;
57 }
58 }
59
60 __attribute__((noipa)) S
61 bar (void)
62 {
63 S s;
64 #pragma omp parallel
65 #pragma omp for simd if (0) reduction (inscan, plus:s)
66 for (int i = 0; i < 1024; i++)
67 {
68 b[i] = s;
69 #pragma omp scan exclusive(s)
70 s.s += 2 * a[i].s;
71 }
72 return s;
73 }
74
75 __attribute__((noipa)) void
76 baz (S *a, S *b, S &r)
77 {
78 #pragma omp parallel for simd reduction (inscan, +:r)
79 for (int i = 0; i < 1024; i++)
80 {
81 b[i] = r;
82 #pragma omp scan exclusive(r)
83 r.s += a[i].s;
84 }
85 }
86
87 __attribute__((noipa)) S
88 qux (void)
89 {
90 S s;
91 #pragma omp parallel for simd reduction (inscan, plus:s)
92 for (int i = 0; i < 1024; i++)
93 {
94 b[i] = s;
95 #pragma omp scan exclusive(s)
96 s.s += 2 * a[i].s;
97 }
98 return s;
99 }
100
101 int
102 main ()
103 {
104 S s;
105 for (int i = 0; i < 1024; ++i)
106 {
107 a[i].s = i;
108 b[i].s = -1;
109 asm ("" : "+g" (i));
110 }
111 #pragma omp parallel
112 foo (a, b, r);
113 if (r.s != 1024 * 1023 / 2)
114 abort ();
115 for (int i = 0; i < 1024; ++i)
116 {
117 if (b[i].s != s.s)
118 abort ();
119 else
120 b[i].s = 25;
121 s.s += i;
122 }
123 if (bar ().s != 1024 * 1023)
124 abort ();
125 s.s = 0;
126 for (int i = 0; i < 1024; ++i)
127 {
128 if (b[i].s != s.s)
129 abort ();
130 s.s += 2 * i;
131 }
132 r.s = 0;
133 baz (a, b, r);
134 if (r.s != 1024 * 1023 / 2)
135 abort ();
136 s.s = 0;
137 for (int i = 0; i < 1024; ++i)
138 {
139 if (b[i].s != s.s)
140 abort ();
141 else
142 b[i].s = 25;
143 s.s += i;
144 }
145 if (qux ().s != 1024 * 1023)
146 abort ();
147 s.s = 0;
148 for (int i = 0; i < 1024; ++i)
149 {
150 if (b[i].s != s.s)
151 abort ();
152 s.s += 2 * i;
153 }
154 }