2 /* { dg-require-effective-target sync_longlong } */
3 /* { dg-require-effective-target pthread_h } */
4 /* { dg-require-effective-target pthread } */
5 /* { dg-options "-pthread -std=gnu99" } */
7 /* test of long long atomic ops performed in parallel in 3 pthreads
8 david.gilbert@linaro.org */
13 /*#define DEBUGIT 1 */
18 #define DOABORT(x,...) {\
19 fprintf (stderr, x, __VA_ARGS__); fflush (stderr); abort ();\
24 #define DOABORT(x,...) abort ();
28 /* Passed to each thread to describe which bits it is going to work on. */
30 unsigned long long count; /* incremented each time the worker loops. */
31 unsigned int thread; /* ID */
32 unsigned int addlsb; /* 8 bit */
33 unsigned int logic1lsb; /* 5 bit */
34 unsigned int logic2lsb; /* 8 bit */
37 /* The shared word where all the atomic work is done. */
38 static volatile long long workspace;
40 /* A shared word to tell the workers to quit when non-0. */
41 static long long doquit;
43 extern void abort (void);
45 /* Note this test doesn't test the return values much. */
49 struct threadwork *tw = (struct threadwork*)data;
50 long long add1bit = 1ll << tw->addlsb;
51 long long logic1bit = 1ll << tw->logic1lsb;
52 long long logic2bit = 1ll << tw->logic2lsb;
54 /* Clear the bits we use. */
55 __sync_and_and_fetch (&workspace, ~(0xffll * add1bit));
56 __sync_fetch_and_and (&workspace, ~(0x1fll * logic1bit));
57 __sync_fetch_and_and (&workspace, ~(0xffll * logic2bit));
61 long long tmp1, tmp2, tmp3;
62 /* OK, lets try and do some stuff to the workspace - by the end
63 of the main loop our area should be the same as it is now - i.e. 0. */
65 /* Push the arithmetic section upto 128 - one of the threads will
66 case this to carry accross the 32bit boundary. */
67 for (tmp2 = 0; tmp2 < 64; tmp2++)
69 /* Add 2 using the two different adds. */
70 tmp1 = __sync_add_and_fetch (&workspace, add1bit);
71 tmp3 = __sync_fetch_and_add (&workspace, add1bit);
73 /* The value should be the intermediate add value in both cases. */
74 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
75 DOABORT ("Mismatch of add intermediates on thread %d "
76 "workspace=0x%llx tmp1=0x%llx "
77 "tmp2=0x%llx tmp3=0x%llx\n",
78 tw->thread, workspace, tmp1, tmp2, tmp3);
81 /* Set the logic bits. */
82 tmp2=__sync_or_and_fetch (&workspace,
83 0x1fll * logic1bit | 0xffll * logic2bit);
85 /* Check the logic bits are set and the arithmetic value is correct. */
86 if ((tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit
88 != (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit))
89 DOABORT ("Midloop check failed on thread %d "
90 "workspace=0x%llx tmp2=0x%llx "
91 "masktmp2=0x%llx expected=0x%llx\n",
92 tw->thread, workspace, tmp2,
93 tmp2 & (0x1fll * logic1bit | 0xffll * logic2bit |
95 (0x1fll * logic1bit | 0xffll * logic2bit | 0x80ll * add1bit));
97 /* Pull the arithmetic set back down to 0 - again this should cause a
98 carry across the 32bit boundary in one thread. */
100 for (tmp2 = 0; tmp2 < 64; tmp2++)
102 /* Subtract 2 using the two different subs. */
103 tmp1=__sync_sub_and_fetch (&workspace, add1bit);
104 tmp3=__sync_fetch_and_sub (&workspace, add1bit);
106 /* The value should be the intermediate sub value in both cases. */
107 if ((tmp1 & (add1bit * 0xff)) != (tmp3 & (add1bit * 0xff)))
108 DOABORT ("Mismatch of sub intermediates on thread %d "
109 "workspace=0x%llx tmp1=0x%llx "
110 "tmp2=0x%llx tmp3=0x%llx\n",
111 tw->thread, workspace, tmp1, tmp2, tmp3);
115 /* Clear the logic bits. */
116 __sync_fetch_and_xor (&workspace, 0x1fll * logic1bit);
117 tmp3=__sync_and_and_fetch (&workspace, ~(0xffll * logic2bit));
119 /* The logic bits and the arithmetic bits should be zero again. */
120 if (tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit))
121 DOABORT ("End of worker loop; bits none 0 on thread %d "
122 "workspace=0x%llx tmp3=0x%llx "
123 "mask=0x%llx maskedtmp3=0x%llx\n",
124 tw->thread, workspace, tmp3, (0x1fll * logic1bit |
125 0xffll * logic2bit | 0xffll * add1bit),
126 tmp3 & (0x1fll * logic1bit | 0xffll * logic2bit | 0xffll * add1bit));
128 __sync_add_and_fetch (&tw->count, 1);
130 while (!__sync_bool_compare_and_swap (&doquit, 1, 1));
138 /* We have 3 threads doing three sets of operations, an 8 bit
139 arithmetic field, a 5 bit logic field and an 8 bit logic
140 field (just to pack them all in).
144 |...,...|...,...|...,...|...,...|...,...|...,...|...,...|...,...
145 - T0 -- T1 -- T2 --T2 -- T0 -*- T2-- T1-- T1 -***- T0-
146 logic2 logic2 arith log2 arith log1 log1 arith log1
153 struct threadwork tw[3]={
154 { 0ll, 0, 27, 0, 56 },
155 { 0ll, 1, 8,16, 48 },
156 { 0ll, 2, 40,21, 35 }
159 pthread_t threads[3];
161 __sync_lock_release (&doquit);
163 /* Get the work space into a known value - All 1's. */
164 __sync_lock_release (&workspace); /* Now all 0. */
165 tmp = __sync_val_compare_and_swap (&workspace, 0, -1ll);
167 DOABORT ("Initial __sync_val_compare_and_swap wasn't 0 workspace=0x%llx "
168 "tmp=0x%llx\n", workspace,tmp);
170 for (t = 0; t < 3; t++)
172 err=pthread_create (&threads[t], NULL , worker, &tw[t]);
173 if (err) DOABORT ("pthread_create failed on thread %d with error %d\n",
180 __sync_lock_test_and_set (&doquit, 1ll);
182 for (t = 0; t < 3; t++)
184 err=pthread_join (threads[t], NULL);
186 DOABORT ("pthread_join failed on thread %d with error %d\n", t, err);
189 __sync_synchronize ();
191 /* OK, so all the workers have finished -
192 the workers should have zero'd their workspace, the unused areas
193 should still be 1. */
194 if (!__sync_bool_compare_and_swap (&workspace, 0x040000e0ll, 0))
195 DOABORT ("End of run workspace mismatch, got %llx\n", workspace);
197 /* All the workers should have done some work. */
198 for (t = 0; t < 3; t++)
200 if (tw[t].count == 0) DOABORT ("Worker %d gave 0 count\n", t);