| 1 | //                               -*- Mode: C -*- | 
|---|
| 2 | // | 
|---|
| 3 | // Cforall Version 1.0.0 Copyright (C) 2021 University of Waterloo | 
|---|
| 4 | // | 
|---|
| 5 | // PRNG.c -- | 
|---|
| 6 | // | 
|---|
| 7 | // Author           : Peter A. Buhr | 
|---|
| 8 | // Created On       : Wed Dec 29 09:38:12 2021 | 
|---|
| 9 | // Last Modified By : Peter A. Buhr | 
|---|
| 10 | // Last Modified On : Sun Apr 23 22:02:09 2023 | 
|---|
| 11 | // Update Count     : 420 | 
|---|
| 12 | // | 
|---|
| 13 |  | 
|---|
| 14 | #include <fstream.hfa>                                                                  // sout | 
|---|
| 15 | #include <stdlib.hfa>                                                                   // PRNG | 
|---|
| 16 | #include <clock.hfa> | 
|---|
| 17 | #include <limits.hfa>                                                                   // MAX | 
|---|
| 18 | #include <math.hfa>                                                                             // sqrt | 
|---|
| 19 | #include <malloc.h>                                                                             // malloc_stats | 
|---|
| 20 | #include <locale.h>                                                                             // setlocale | 
|---|
| 21 | #include <thread.hfa> | 
|---|
| 22 | #include <mutex_stmt.hfa> | 
|---|
| 23 |  | 
|---|
| 24 | #define xstr(s) str(s) | 
|---|
| 25 | #define str(s) #s | 
|---|
| 26 |  | 
|---|
| 27 | #if defined( __x86_64__ ) || defined( __aarch64__ )             // 64-bit architecture | 
|---|
| 28 | #define PRNG PRNG64 | 
|---|
| 29 | #else                                                                                                   // 32-bit architecture | 
|---|
| 30 | #define PRNG PRNG32 | 
|---|
| 31 | #endif // __x86_64__ | 
|---|
| 32 |  | 
|---|
| 33 | //#define TIME | 
|---|
| 34 |  | 
|---|
| 35 | #ifdef TIME                                                                                             // use -O2 -nodebug | 
|---|
| 36 | #define STARTTIME start = timeHiRes() | 
|---|
| 37 | #define ENDTIME( extra ) sout | wd(0,1, (timeHiRes() - start)`ms / 1000.) | extra "seconds" | 
|---|
| 38 | enum { BUCKETS = 100_000, TRIALS = 1_000_000_000 }; | 
|---|
| 39 | #else | 
|---|
| 40 | #define STARTTIME | 
|---|
| 41 | #define ENDTIME( extra ) | 
|---|
| 42 | enum { BUCKETS = 100_000, TRIALS = 100_000_000 }; | 
|---|
| 43 | #endif // TIME | 
|---|
| 44 |  | 
|---|
| 45 | static void avgstd( size_t trials, size_t buckets[] ) { | 
|---|
| 46 | size_t min = MAX, max = 0; | 
|---|
| 47 | double sum = 0.0, diff; | 
|---|
| 48 | for ( i; BUCKETS ) { | 
|---|
| 49 | if ( buckets[i] < min ) min = buckets[i]; | 
|---|
| 50 | if ( buckets[i] > max ) max = buckets[i]; | 
|---|
| 51 | sum += buckets[i]; | 
|---|
| 52 | } // for | 
|---|
| 53 |  | 
|---|
| 54 | double avg = sum / BUCKETS;                                                     // average | 
|---|
| 55 | sum = 0.0; | 
|---|
| 56 | for ( i; BUCKETS ) {                                                            // sum squared differences from average | 
|---|
| 57 | diff = buckets[i] - avg; | 
|---|
| 58 | sum += diff * diff; | 
|---|
| 59 | } // for | 
|---|
| 60 | double std = sqrt( sum / BUCKETS ); | 
|---|
| 61 | mutex( sout ) sout | "trials"  | trials | "buckets" | BUCKETS | 
|---|
| 62 | | "min" | min | "max" | max | 
|---|
| 63 | | "avg" | wd(0,1, avg) | "std" | wd(0,1, std) | "rstd" | wd(0,1, (avg == 0 ? 0.0 : std / avg * 100)) | "%"; | 
|---|
| 64 | } // avgstd | 
|---|
| 65 |  | 
|---|
| 66 |  | 
|---|
| 67 | size_t seed = 1009; | 
|---|
| 68 |  | 
|---|
| 69 | thread T1 {}; | 
|---|
| 70 | void main( T1 & ) { | 
|---|
| 71 | size_t * buckets = calloc( BUCKETS );                           // too big for task stack | 
|---|
| 72 | for ( TRIALS / 50 ) { | 
|---|
| 73 | buckets[rand() % BUCKETS] += 1;                                 // concurrent | 
|---|
| 74 | } // for | 
|---|
| 75 | avgstd( TRIALS / 50, buckets ); | 
|---|
| 76 | free( buckets ); | 
|---|
| 77 | } // main | 
|---|
| 78 |  | 
|---|
| 79 | thread T2 {}; | 
|---|
| 80 | void main( T2 & ) { | 
|---|
| 81 | PRNG prng; | 
|---|
| 82 | if ( seed != 0 ) set_seed( prng, seed ); | 
|---|
| 83 | size_t * buckets = calloc( BUCKETS );                           // too big for task stack | 
|---|
| 84 | for ( TRIALS ) { | 
|---|
| 85 | buckets[prng( prng ) % BUCKETS] += 1;                   // concurrent | 
|---|
| 86 | } // for | 
|---|
| 87 | avgstd( TRIALS, buckets ); | 
|---|
| 88 | free( buckets ); | 
|---|
| 89 | } // main | 
|---|
| 90 |  | 
|---|
| 91 | thread T3 {}; | 
|---|
| 92 | void main( T3 & th ) { | 
|---|
| 93 | size_t * buckets = calloc( BUCKETS );                           // too big for task stack | 
|---|
| 94 | for ( TRIALS / 5 ) { | 
|---|
| 95 | buckets[prng() % BUCKETS] += 1;                                 // concurrent | 
|---|
| 96 | } // for | 
|---|
| 97 | avgstd( TRIALS / 5, buckets ); | 
|---|
| 98 | free( buckets ); | 
|---|
| 99 | } // main | 
|---|
| 100 |  | 
|---|
| 101 | thread T4 {}; | 
|---|
| 102 | void main( T4 & th ) { | 
|---|
| 103 | size_t * buckets = calloc( BUCKETS );                           // too big for task stack | 
|---|
| 104 | for ( TRIALS ) { | 
|---|
| 105 | buckets[prng( th ) % BUCKETS] += 1;                             // concurrent | 
|---|
| 106 | } // for | 
|---|
| 107 | avgstd( TRIALS, buckets ); | 
|---|
| 108 | free( buckets ); | 
|---|
| 109 | } // main | 
|---|
| 110 |  | 
|---|
| 111 | // Compiler bug requires hiding declaration of th from the bucket access, otherwise the compiler thinks th is aliased | 
|---|
| 112 | // and continually reloads it from memory, which doubles the cost. | 
|---|
| 113 | static void dummy( thread$ & th ) __attribute__(( noinline )); | 
|---|
| 114 | static void dummy( thread$ & th ) { | 
|---|
| 115 | size_t * buckets = (size_t *)calloc( BUCKETS, sizeof(size_t) ); // too big for task stack | 
|---|
| 116 | for ( size_t i = 0; i < TRIALS; i += 1 ) { | 
|---|
| 117 | buckets[prng( th ) % BUCKETS] += 1;                             // sequential | 
|---|
| 118 | } // for | 
|---|
| 119 | avgstd( TRIALS, buckets ); | 
|---|
| 120 | free( buckets ); | 
|---|
| 121 | } // dummy | 
|---|
| 122 |  | 
|---|
| 123 |  | 
|---|
| 124 | int main() { | 
|---|
| 125 | // setlocale( LC_NUMERIC, getenv( "LANG" ) );           // causes leaked storage message | 
|---|
| 126 |  | 
|---|
| 127 | // only works on the current pthread thread | 
|---|
| 128 | // locale_t loc = newlocale( LC_NUMERIC_MASK, getenv( "LANG" ), (locale_t)0p ); | 
|---|
| 129 | // if ( loc == (locale_t)0p ) abort( "newlocale" ); | 
|---|
| 130 | // uselocale( loc ); | 
|---|
| 131 |  | 
|---|
| 132 | enum { TASKS = 4 }; | 
|---|
| 133 | Time start; | 
|---|
| 134 |  | 
|---|
| 135 | #ifdef TIME                                                                                             // too slow for test and generates non-repeatable results | 
|---|
| 136 | #if 1 | 
|---|
| 137 | sout | "glib rand" | nl | nl; | 
|---|
| 138 |  | 
|---|
| 139 | size_t rseed; | 
|---|
| 140 | if ( seed != 0 ) rseed = seed; | 
|---|
| 141 | else rseed = rdtscl(); | 
|---|
| 142 | srand( rseed ); | 
|---|
| 143 |  | 
|---|
| 144 | sout | sepDisable; | 
|---|
| 145 | sout | nl | wd(26, "rand()" ) | wd(12, "rand(5)") | wd(12, "rand(0,5)" ); | 
|---|
| 146 | for ( 20 ) { | 
|---|
| 147 | sout | wd(26, rand()) | nonl; | 
|---|
| 148 | sout | wd(12, rand() % 5) | nonl; | 
|---|
| 149 | sout | wd(12, rand() % (5 - 0 + 1) + 0); | 
|---|
| 150 | } // for | 
|---|
| 151 | sout | sepEnable; | 
|---|
| 152 | sout | "seed" | rseed; | 
|---|
| 153 |  | 
|---|
| 154 | sout | nl | "Sequential"; | 
|---|
| 155 | STARTTIME; | 
|---|
| 156 | { | 
|---|
| 157 | size_t * buckets = calloc( BUCKETS );                   // too big for task stack | 
|---|
| 158 | for ( i; TRIALS / 5 ) { | 
|---|
| 159 | buckets[rand() % BUCKETS] += 1;                         // sequential | 
|---|
| 160 | } // for | 
|---|
| 161 | avgstd( TRIALS / 5, buckets ); | 
|---|
| 162 | free( buckets ); | 
|---|
| 163 | } | 
|---|
| 164 | ENDTIME( " x 5 " ); | 
|---|
| 165 |  | 
|---|
| 166 | sout | nl | "Concurrent"; | 
|---|
| 167 | STARTTIME; | 
|---|
| 168 | { | 
|---|
| 169 | processor p[TASKS - 1];                                                 // already 1 processor | 
|---|
| 170 | { | 
|---|
| 171 | T1 t[TASKS]; | 
|---|
| 172 | } // wait for threads to complete | 
|---|
| 173 | } | 
|---|
| 174 | ENDTIME( " x 50 " ); | 
|---|
| 175 | #endif // 0 | 
|---|
| 176 | #endif // TIME | 
|---|
| 177 |  | 
|---|
| 178 | sout | nl | "CFA " xstr(PRNG_NAME); | 
|---|
| 179 |  | 
|---|
| 180 | #if 1 | 
|---|
| 181 | PRNG prng; | 
|---|
| 182 |  | 
|---|
| 183 | if ( seed != 0 ) set_seed( prng, seed ); | 
|---|
| 184 |  | 
|---|
| 185 | sout | sepDisable; | 
|---|
| 186 | sout | nl | wd(26, "PRNG()" ) | wd(12, "PRNG(5)") | wd(12, "PRNG(0,5)" ); | 
|---|
| 187 | for ( 20 ) { | 
|---|
| 188 | sout | wd(26, prng( prng )) | nonl;                             // cascading => side-effect functions called in arbitary order | 
|---|
| 189 | sout | wd(12, prng( prng, 5 )) | nonl; | 
|---|
| 190 | sout | wd(12, prng( prng, 0, 5 )); | 
|---|
| 191 | } // for | 
|---|
| 192 | sout | sepEnable; | 
|---|
| 193 | sout | "seed" | get_seed( prng ); | 
|---|
| 194 |  | 
|---|
| 195 | sout | nl | "Sequential"; | 
|---|
| 196 | STARTTIME; | 
|---|
| 197 | { | 
|---|
| 198 | size_t * buckets = calloc( BUCKETS );                   // too big for task stack | 
|---|
| 199 | for ( TRIALS ) { | 
|---|
| 200 | buckets[prng( prng ) % BUCKETS] += 1;           // sequential | 
|---|
| 201 | } // for | 
|---|
| 202 | avgstd( TRIALS, buckets ); | 
|---|
| 203 | free( buckets ); | 
|---|
| 204 | } | 
|---|
| 205 | ENDTIME(); | 
|---|
| 206 |  | 
|---|
| 207 | sout | nl | "Concurrent"; | 
|---|
| 208 | STARTTIME; | 
|---|
| 209 | { | 
|---|
| 210 | processor p[TASKS - 1];                                                 // already 1 processor | 
|---|
| 211 | { | 
|---|
| 212 | T2 t[TASKS]; | 
|---|
| 213 | } // wait for threads to complete | 
|---|
| 214 | } | 
|---|
| 215 | ENDTIME(); | 
|---|
| 216 | #endif // 0 | 
|---|
| 217 | #if 1 | 
|---|
| 218 | if ( seed != 0 ) set_seed( seed ); | 
|---|
| 219 |  | 
|---|
| 220 | sout | sepDisable; | 
|---|
| 221 | sout | nl | wd(26, "prng()" ) | wd(12, "prng(5)") | wd(12, "prng(0,5)" ); | 
|---|
| 222 | for ( 20 ) { | 
|---|
| 223 | sout | wd(26, prng()) | nonl;                                   // cascading => side-effect functions called in arbitary order | 
|---|
| 224 | sout | wd(12, prng( 5 )) | nonl; | 
|---|
| 225 | sout | wd(12, prng( 0, 5 )); | 
|---|
| 226 | } // for | 
|---|
| 227 | sout | sepEnable; | 
|---|
| 228 | sout | "seed" | get_seed( prng ); | 
|---|
| 229 |  | 
|---|
| 230 | sout | nl | "Sequential"; | 
|---|
| 231 | STARTTIME; | 
|---|
| 232 | { | 
|---|
| 233 | size_t * buckets = calloc( BUCKETS );                   // too big for task stack | 
|---|
| 234 | for ( TRIALS / 5 ) { | 
|---|
| 235 | buckets[prng() % BUCKETS] += 1; | 
|---|
| 236 | } // for | 
|---|
| 237 | avgstd( TRIALS / 5, buckets ); | 
|---|
| 238 | free( buckets ); | 
|---|
| 239 | } | 
|---|
| 240 | ENDTIME( " x 5 " ); | 
|---|
| 241 |  | 
|---|
| 242 | sout | nl | "Concurrent"; | 
|---|
| 243 | STARTTIME; | 
|---|
| 244 | { | 
|---|
| 245 | processor p[TASKS - 1];                                                 // already 1 processor | 
|---|
| 246 | { | 
|---|
| 247 | T3 t[TASKS]; | 
|---|
| 248 | } // wait for threads to complete | 
|---|
| 249 | } | 
|---|
| 250 | ENDTIME( " x 5 " ); | 
|---|
| 251 | #endif // 0 | 
|---|
| 252 | #if 1 | 
|---|
| 253 | if ( seed != 0 ) set_seed( seed ); | 
|---|
| 254 | thread$ & th = *active_thread(); | 
|---|
| 255 |  | 
|---|
| 256 | sout | sepDisable; | 
|---|
| 257 | sout | nl | wd(26, "prng(t)" ) | wd(12, "prng(t,5)") | wd(12, "prng(t,0,5)" ); | 
|---|
| 258 | for ( 20 ) { | 
|---|
| 259 | sout | wd(26, prng( th )) | nonl;                               // cascading => side-effect functions called in arbitary order | 
|---|
| 260 | sout | wd(12, prng( th, 5 )) | nonl; | 
|---|
| 261 | sout | wd(12, prng( th, 0, 5 )); | 
|---|
| 262 | } // for | 
|---|
| 263 | sout | sepEnable; | 
|---|
| 264 | sout | "seed" | get_seed( prng ); | 
|---|
| 265 |  | 
|---|
| 266 | sout | nl | "Sequential"; | 
|---|
| 267 | STARTTIME; | 
|---|
| 268 | { | 
|---|
| 269 | dummy( th ); | 
|---|
| 270 | } | 
|---|
| 271 | ENDTIME(); | 
|---|
| 272 |  | 
|---|
| 273 | sout | nl | "Concurrent"; | 
|---|
| 274 | STARTTIME; | 
|---|
| 275 | { | 
|---|
| 276 | processor p[TASKS - 1];                                                 // already 1 processor | 
|---|
| 277 | { | 
|---|
| 278 | T4 t[TASKS]; | 
|---|
| 279 | } // wait for threads to complete | 
|---|
| 280 | } | 
|---|
| 281 | ENDTIME(); | 
|---|
| 282 | #endif // 0 | 
|---|
| 283 | //      malloc_stats(); | 
|---|
| 284 | // freelocale( loc ); | 
|---|
| 285 | } // main | 
|---|
| 286 |  | 
|---|
| 287 |  | 
|---|
| 288 | // Local Variables: // | 
|---|
| 289 | // compile-command: "cfa -DTIME -O2 -nodebug PRNG.cfa" // | 
|---|
| 290 | // End: // | 
|---|