| 1 | /* Copyright (c) 2006, 2011, Oracle and/or its affiliates. All rights reserved. |
| 2 | |
| 3 | This program is free software; you can redistribute it and/or modify |
| 4 | it under the terms of the GNU General Public License as published by |
| 5 | the Free Software Foundation; version 2 of the License. |
| 6 | |
| 7 | This program is distributed in the hope that it will be useful, |
| 8 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 9 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 10 | GNU General Public License for more details. |
| 11 | |
| 12 | You should have received a copy of the GNU General Public License |
| 13 | along with this program; if not, write to the Free Software |
| 14 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ |
| 15 | |
| 16 | #include "thr_template.c" |
| 17 | |
| 18 | volatile uint32 b32; |
| 19 | volatile int32 c32; |
| 20 | |
| 21 | /* add and sub a random number in a loop. Must get 0 at the end */ |
| 22 | pthread_handler_t test_atomic_add(void *arg) |
| 23 | { |
| 24 | int m= (*(int *)arg)/2; |
| 25 | int32 x; |
| 26 | for (x= ((int)(intptr)(&m)); m ; m--) |
| 27 | { |
| 28 | x= (x*m+0x87654321) & INT_MAX32; |
| 29 | my_atomic_add32(&bad, x); |
| 30 | my_atomic_add32(&bad, -x); |
| 31 | } |
| 32 | pthread_mutex_lock(&mutex); |
| 33 | if (!--running_threads) pthread_cond_signal(&cond); |
| 34 | pthread_mutex_unlock(&mutex); |
| 35 | return 0; |
| 36 | } |
| 37 | |
| 38 | volatile int64 a64; |
| 39 | /* add and sub a random number in a loop. Must get 0 at the end */ |
| 40 | pthread_handler_t test_atomic_add64(void *arg) |
| 41 | { |
| 42 | int m= (*(int *)arg)/2; |
| 43 | int64 x; |
| 44 | for (x= ((int64)(intptr)(&m)); m ; m--) |
| 45 | { |
| 46 | x= (x*m+0xfdecba987654321LL) & INT_MAX64; |
| 47 | my_atomic_add64(&a64, x); |
| 48 | my_atomic_add64(&a64, -x); |
| 49 | } |
| 50 | pthread_mutex_lock(&mutex); |
| 51 | if (!--running_threads) |
| 52 | { |
| 53 | bad= (a64 != 0); |
| 54 | pthread_cond_signal(&cond); |
| 55 | } |
| 56 | pthread_mutex_unlock(&mutex); |
| 57 | return 0; |
| 58 | } |
| 59 | |
| 60 | |
| 61 | /* |
| 62 | 1. generate thread number 0..N-1 from b32 |
| 63 | 2. add it to bad |
| 64 | 3. swap thread numbers in c32 |
| 65 | 4. (optionally) one more swap to avoid 0 as a result |
| 66 | 5. subtract result from bad |
| 67 | must get 0 in bad at the end |
| 68 | */ |
| 69 | pthread_handler_t test_atomic_fas(void *arg) |
| 70 | { |
| 71 | int m= *(int *)arg; |
| 72 | int32 x; |
| 73 | |
| 74 | x= my_atomic_add32(&b32, 1); |
| 75 | |
| 76 | my_atomic_add32(&bad, x); |
| 77 | |
| 78 | for (; m ; m--) |
| 79 | x= my_atomic_fas32(&c32, x); |
| 80 | |
| 81 | if (!x) |
| 82 | x= my_atomic_fas32(&c32, x); |
| 83 | |
| 84 | my_atomic_add32(&bad, -x); |
| 85 | |
| 86 | pthread_mutex_lock(&mutex); |
| 87 | if (!--running_threads) pthread_cond_signal(&cond); |
| 88 | pthread_mutex_unlock(&mutex); |
| 89 | return 0; |
| 90 | } |
| 91 | |
| 92 | /* |
| 93 | same as test_atomic_add, but my_atomic_add32 is emulated with |
| 94 | my_atomic_cas32 - notice that the slowdown is proportional to the |
| 95 | number of CPUs |
| 96 | */ |
| 97 | pthread_handler_t test_atomic_cas(void *arg) |
| 98 | { |
| 99 | int m= (*(int *)arg)/2, ok= 0; |
| 100 | int32 x, y; |
| 101 | for (x= ((int)(intptr)(&m)); m ; m--) |
| 102 | { |
| 103 | y= my_atomic_load32(&bad); |
| 104 | x= (x*m+0x87654321) & INT_MAX32; |
| 105 | do { |
| 106 | ok= my_atomic_cas32(&bad, &y, (uint32)y+x); |
| 107 | } while (!ok) ; |
| 108 | do { |
| 109 | ok= my_atomic_cas32(&bad, &y, y-x); |
| 110 | } while (!ok) ; |
| 111 | } |
| 112 | pthread_mutex_lock(&mutex); |
| 113 | if (!--running_threads) pthread_cond_signal(&cond); |
| 114 | pthread_mutex_unlock(&mutex); |
| 115 | return 0; |
| 116 | } |
| 117 | |
| 118 | |
| 119 | void do_tests() |
| 120 | { |
| 121 | plan(5); |
| 122 | |
| 123 | b32= c32= 0; |
| 124 | test_concurrently("my_atomic_add32" , test_atomic_add, THREADS, CYCLES); |
| 125 | b32= c32= 0; |
| 126 | test_concurrently("my_atomic_fas32" , test_atomic_fas, THREADS, CYCLES); |
| 127 | b32= c32= 0; |
| 128 | test_concurrently("my_atomic_cas32" , test_atomic_cas, THREADS, CYCLES); |
| 129 | |
| 130 | { |
| 131 | /* |
| 132 | If b is not volatile, the wrong assembly code is generated on OSX Lion |
| 133 | as the variable is optimized away as a constant. |
| 134 | See Bug#62533 / Bug#13030056. |
| 135 | Another workaround is to specify architecture explicitly using e.g. |
| 136 | CFLAGS/CXXFLAGS= "-m64". |
| 137 | */ |
| 138 | volatile int64 b=0x1000200030004000LL; |
| 139 | a64=0; |
| 140 | my_atomic_add64(&a64, b); |
| 141 | ok(a64==b, "add64" ); |
| 142 | } |
| 143 | a64=0; |
| 144 | test_concurrently("my_atomic_add64" , test_atomic_add64, THREADS, CYCLES); |
| 145 | } |
| 146 | |