1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
|
/* AMD GCN does not use 32-lane vectors.
{ dg-skip-if "unsuitable dimensions" { openacc_amdgcn_accel_selected } { "*" } { "" } } */
/* { dg-additional-options "-fopenacc-dim=32" } */
#include <stdio.h>
#include <openacc.h>
#include <gomp-constants.h>
int check (const int *ary, int size, int gp, int wp, int vp)
{
int exit = 0;
int ix;
int gangs[32], workers[32], vectors[32];
for (ix = 0; ix < 32; ix++)
gangs[ix] = workers[ix] = vectors[ix] = 0;
for (ix = 0; ix < size; ix++)
{
vectors[ary[ix] & 0xff]++;
workers[(ary[ix] >> 8) & 0xff]++;
gangs[(ary[ix] >> 16) & 0xff]++;
}
for (ix = 0; ix < 32; ix++)
{
if (gp)
{
int expect = gangs[0];
if (gangs[ix] != expect)
{
exit = 1;
printf ("gang %d not used %d times\n", ix, expect);
}
}
else if (ix && gangs[ix])
{
exit = 1;
printf ("gang %d unexpectedly used\n", ix);
}
if (wp)
{
int expect = workers[0];
if (workers[ix] != expect)
{
exit = 1;
printf ("worker %d not used %d times\n", ix, expect);
}
}
else if (ix && workers[ix])
{
exit = 1;
printf ("worker %d unexpectedly used\n", ix);
}
if (vp)
{
int expect = vectors[0];
if (vectors[ix] != expect)
{
exit = 1;
printf ("vector %d not used %d times\n", ix, expect);
}
}
else if (ix && vectors[ix])
{
exit = 1;
printf ("vector %d unexpectedly used\n", ix);
}
}
return exit;
}
#pragma acc routine seq
static int __attribute__((noinline)) place ()
{
int r = 0;
int g = 0, w = 0, v = 0;
g = __builtin_goacc_parlevel_id (GOMP_DIM_GANG);
w = __builtin_goacc_parlevel_id (GOMP_DIM_WORKER);
v = __builtin_goacc_parlevel_id (GOMP_DIM_VECTOR);
r = (g << 16) | (w << 8) | v;
return r;
}
static void clear (int *ary, int size)
{
int ix;
for (ix = 0; ix < size; ix++)
ary[ix] = -1;
}
int vector_1 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop gang
for (int jx = 0; jx < 1; jx++)
#pragma acc loop auto
for (int ix = 0; ix < size; ix++)
ary[ix] = place ();
}
return check (ary, size, 0, 1, 1);
}
int vector_2 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop worker
for (int jx = 0; jx < size / 64; jx++)
#pragma acc loop auto
for (int ix = 0; ix < 64; ix++)
ary[ix + jx * 64] = place ();
}
return check (ary, size, 0, 1, 1);
}
int worker_1 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop gang
for (int kx = 0; kx < 1; kx++)
#pragma acc loop auto
for (int jx = 0; jx < size / 64; jx++)
#pragma acc loop vector
for (int ix = 0; ix < 64; ix++)
ary[ix + jx * 64] = place ();
}
return check (ary, size, 0, 1, 1);
}
int gang_1 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_gangs (32) num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop auto
for (int jx = 0; jx < size / 64; jx++)
#pragma acc loop worker
for (int ix = 0; ix < 64; ix++)
ary[ix + jx * 64] = place ();
}
return check (ary, size, 1, 1, 0);
}
int gang_2 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_gangs (32) num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop auto
for (int kx = 0; kx < size / (32 * 32); kx++)
#pragma acc loop auto
for (int jx = 0; jx < 32; jx++)
#pragma acc loop auto
for (int ix = 0; ix < 32; ix++)
ary[ix + jx * 32 + kx * 32 * 32] = place ();
}
return check (ary, size, 1, 1, 1);
}
int gang_3 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel num_workers (32) vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop auto
for (int jx = 0; jx < size / 64; jx++)
#pragma acc loop auto
for (int ix = 0; ix < 64; ix++)
ary[ix + jx * 64] = place ();
}
return check (ary, size, 1, 1, 1);
}
int gang_4 (int *ary, int size)
{
clear (ary, size);
#pragma acc parallel vector_length(32) copy(ary[0:size]) firstprivate (size)
{
#pragma acc loop auto
for (int jx = 0; jx < size; jx++)
ary[jx] = place ();
}
return check (ary, size, 1, 0, 1);
}
#define N (32*32*32*2)
int main ()
{
int ondev = 0;
#pragma acc parallel copy(ondev)
{
ondev = acc_on_device (acc_device_not_host);
}
if (!ondev)
return 0;
int ary[N];
if (vector_1 (ary, N))
return 1;
if (vector_2 (ary, N))
return 1;
if (worker_1 (ary, N))
return 1;
if (gang_1 (ary, N))
return 1;
if (gang_2 (ary, N))
return 1;
if (gang_3 (ary, N))
return 1;
if (gang_4 (ary, N))
return 1;
return 0;
}
|
