111
|
1 /* { dg-do run } */
|
|
2 /* { dg-require-effective-target sync_long_long_runtime } */
|
|
3 /* { dg-require-effective-target pthread_h } */
|
|
4 /* { dg-require-effective-target pthread } */
|
|
5 /* { dg-options "-pthread -std=gnu99" } */
|
|
6 /* { dg-additional-options "-march=pentium" { target { { i?86-*-* x86_64-*-* } && ia32 } } } */
|
|
7
|
|
8 /* test of long long atomic ops performed in parallel in 3 pthreads
|
|
9 david.gilbert@linaro.org */
|
|
10
|
|
11 #include <pthread.h>
|
|
12 #include <unistd.h>
|
|
13 #ifdef _WIN32
|
|
14 #include <windows.h>
|
|
15 #endif
|
|
16
|
|
17 /*#define DEBUGIT 1 */
|
|
18
|
|
19 #ifdef DEBUGIT
|
|
20 #include <stdio.h>
|
|
21
|
|
22 #define DOABORT(x,...) {\
|
|
23 fprintf (stderr, x, __VA_ARGS__); fflush (stderr); abort ();\
|
|
24 }
|
|
25
|
|
26 #else
|
|
27
|
|
28 #define DOABORT(x,...) abort ();
|
|
29
|
|
30 #endif
|
|
31
|
|
32 /* Passed to each thread to describe which bits it is going to work on. */
|
|
33 struct threadwork {
|
|
34 unsigned long long count; /* incremented each time the worker loops. */
|
|
35 unsigned int thread; /* ID */
|
|
36 unsigned int addlsb; /* 8 bit */
|
|
37 unsigned int logic1lsb; /* 5 bit */
|
|
38 unsigned int logic2lsb; /* 8 bit */
|
|
39 };
|
|
40
|
|
41 /* The shared word where all the atomic work is done. */
|
|
42 static volatile long long workspace;
|
|
43
|
|
44 /* A shared word to tell the workers to quit when non-0. */
|
|
45 static long long doquit;
|
|
46
|
|
47 extern void abort (void);
|
|
48
|
|
49 /* Note this test doesn't test the return values much. */
|
|
50 void*
|
|
51 worker (void* data)
|
|
52 {
|
|
53 struct threadwork *tw = (struct threadwork*)data;
|
|
54 long long add1bit = 1ll << tw->addlsb;
|
|
55 long long logic1bit = 1ll << tw->logic1lsb;
|
|
56 long long logic2bit = 1ll << tw->logic2lsb;
|
|
57
|
|
58 /* Clear the bits we use. */
|
|
59 __sync_and_and_fetch (&workspace, ~(0xffll * add1bit));
|
|
60 __sync_fetch_and_and (&workspace, ~(0x1fll * logic1bit));
|
|
61 __sync_fetch_and_and (&workspace, ~(0xffll * logic2bit));
|
|
62
|
|
63 do
|
|
64 {
|
|
65 long long tmp1, tmp2, tmp3;
|
|
66 /* OK, lets try and do some stuff to the workspace - by the end
|
|
67 of the main loop our area should be the same as it is now - i.e. 0. */
|
|
68
|
|
69 /* Push the arithmetic section up to 128 - one of the threads will
|
|
70 case this to carry across the 32bit boundary. */
|
|
71 for (tmp2 = 0; tmp2 < 64; tmp2++)
|
|
72 {
|
|
73 /* Add 2 using the two different adds. */
|
|
74 tmp1 = __sync_add_and_fetch (&workspace, add1bit);
|
|
75 tmp3 = __sync_fetch_and_add (&workspace, add1bit);
|
|
76
|
|
77 /* The value should be the intermediate add value in both cases. */
|
|
78 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
|
|
79 DOABORT ("Mismatch of add intermediates on thread %d "
|
|
80 "workspace=0x%llx tmp1=0x%llx "
|
|
81 "tmp2=0x%llx tmp3=0x%llx\n",
|
|
82 tw->thread, workspace, tmp1, tmp2, tmp3);
|
|
83 }
|
|
84
|
|
85 /* Set the logic bits. */
|
|
86 tmp2=__sync_or_and_fetch (&workspace,
|
|
87 0x1fll * logic1bit | 0xffll * logic2bit);
|
|
88
|
|
89 /* Check the logic bits are set and the arithmetic value is correct. */
|
|
90 if ((tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit
|
|
91 | 0xffll * add1bit))
|
|
92 != (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit))
|
|
93 DOABORT ("Midloop check failed on thread %d "
|
|
94 "workspace=0x%llx tmp2=0x%llx "
|
|
95 "masktmp2=0x%llx expected=0x%llx\n",
|
|
96 tw->thread, workspace, tmp2,
|
|
97 tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit |
|
|
98 0xffll * add1bit),
|
|
99 (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit));
|
|
100
|
|
101 /* Pull the arithmetic set back down to 0 - again this should cause a
|
|
102 carry across the 32bit boundary in one thread. */
|
|
103
|
|
104 for (tmp2 = 0; tmp2 < 64; tmp2++)
|
|
105 {
|
|
106 /* Subtract 2 using the two different subs. */
|
|
107 tmp1=__sync_sub_and_fetch (&workspace, add1bit);
|
|
108 tmp3=__sync_fetch_and_sub (&workspace, add1bit);
|
|
109
|
|
110 /* The value should be the intermediate sub value in both cases. */
|
|
111 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
|
|
112 DOABORT ("Mismatch of sub intermediates on thread %d "
|
|
113 "workspace=0x%llx tmp1=0x%llx "
|
|
114 "tmp2=0x%llx tmp3=0x%llx\n",
|
|
115 tw->thread, workspace, tmp1, tmp2, tmp3);
|
|
116 }
|
|
117
|
|
118
|
|
119 /* Clear the logic bits. */
|
|
120 __sync_fetch_and_xor (&workspace, 0x1fll * logic1bit);
|
|
121 tmp3=__sync_and_and_fetch (&workspace, ~(0xffll * logic2bit));
|
|
122
|
|
123 /* The logic bits and the arithmetic bits should be zero again. */
|
|
124 if (tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit))
|
|
125 DOABORT ("End of worker loop; bits none 0 on thread %d "
|
|
126 "workspace=0x%llx tmp3=0x%llx "
|
|
127 "mask=0x%llx maskedtmp3=0x%llx\n",
|
|
128 tw->thread, workspace, tmp3, (0x1fll * logic1bit |
|
|
129 0xffll * logic2bit | 0xffll * add1bit),
|
|
130 tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit));
|
|
131
|
|
132 __sync_add_and_fetch (&tw->count, 1);
|
|
133 }
|
|
134 while (!__sync_bool_compare_and_swap (&doquit, 1, 1));
|
|
135
|
|
136 pthread_exit (0);
|
|
137 }
|
|
138
|
|
139 int
|
|
140 main ()
|
|
141 {
|
|
142 /* We have 3 threads doing three sets of operations, an 8 bit
|
|
143 arithmetic field, a 5 bit logic field and an 8 bit logic
|
|
144 field (just to pack them all in).
|
|
145
|
|
146 6 5 4 4 3 2 1
|
|
147 3 6 8 0 2 4 6 8 0
|
|
148 |...,...|...,...|...,...|...,...|...,...|...,...|...,...|...,...
|
|
149 - T0 -- T1 -- T2 --T2 -- T0 -*- T2-- T1-- T1 -***- T0-
|
|
150 logic2 logic2 arith log2 arith log1 log1 arith log1
|
|
151
|
|
152 */
|
|
153 unsigned int t;
|
|
154 long long tmp;
|
|
155 int err;
|
|
156
|
|
157 struct threadwork tw[3]={
|
|
158 { 0ll, 0, 27, 0, 56 },
|
|
159 { 0ll, 1, 8,16, 48 },
|
|
160 { 0ll, 2, 40,21, 35 }
|
|
161 };
|
|
162
|
|
163 pthread_t threads[3];
|
|
164
|
|
165 __sync_lock_release (&doquit);
|
|
166
|
|
167 /* Get the work space into a known value - All 1's. */
|
|
168 __sync_lock_release (&workspace); /* Now all 0. */
|
|
169 tmp = __sync_val_compare_and_swap (&workspace, 0, -1ll);
|
|
170 if (tmp!=0)
|
|
171 DOABORT ("Initial __sync_val_compare_and_swap wasn't 0 workspace=0x%llx "
|
|
172 "tmp=0x%llx\n", workspace,tmp);
|
|
173
|
|
174 for (t = 0; t < 3; t++)
|
|
175 {
|
|
176 err=pthread_create (&threads[t], NULL , worker, &tw[t]);
|
|
177 if (err) DOABORT ("pthread_create failed on thread %d with error %d\n",
|
|
178 t, err);
|
|
179 };
|
|
180
|
|
181 #ifdef _WIN32
|
|
182 Sleep (5000);
|
|
183 #else
|
|
184 sleep (5);
|
|
185 #endif
|
|
186
|
|
187 /* Stop please. */
|
|
188 __sync_lock_test_and_set (&doquit, 1ll);
|
|
189
|
|
190 for (t = 0; t < 3; t++)
|
|
191 {
|
|
192 err=pthread_join (threads[t], NULL);
|
|
193 if (err)
|
|
194 DOABORT ("pthread_join failed on thread %d with error %d\n", t, err);
|
|
195 };
|
|
196
|
|
197 __sync_synchronize ();
|
|
198
|
|
199 /* OK, so all the workers have finished -
|
|
200 the workers should have zero'd their workspace, the unused areas
|
|
201 should still be 1. */
|
|
202 if (!__sync_bool_compare_and_swap (&workspace, 0x040000e0ll, 0))
|
|
203 DOABORT ("End of run workspace mismatch, got %llx\n", workspace);
|
|
204
|
|
205 /* All the workers should have done some work. */
|
|
206 for (t = 0; t < 3; t++)
|
|
207 {
|
|
208 if (tw[t].count == 0) DOABORT ("Worker %d gave 0 count\n", t);
|
|
209 };
|
|
210
|
|
211 return 0;
|
|
212 }
|
|
213
|