| 1 | //
|
|---|
| 2 | // Cforall Version 1.0.0 Copyright (C) 2021 University of Waterloo
|
|---|
| 3 | //
|
|---|
| 4 | // PRNG.c -- high-perforamnce pseudo-random numbers
|
|---|
| 5 | //
|
|---|
| 6 | // The contents of this file are covered under the licence agreement in the
|
|---|
| 7 | // file "LICENCE" distributed with Cforall.
|
|---|
| 8 | //
|
|---|
| 9 | // Author : Peter A. Buhr
|
|---|
| 10 | // Created On : Wed Dec 29 09:38:12 2021
|
|---|
| 11 | // Last Modified By : Peter A. Buhr
|
|---|
| 12 | // Last Modified On : Thu May 25 15:39:52 2023
|
|---|
| 13 | // Update Count : 422
|
|---|
| 14 | //
|
|---|
| 15 |
|
|---|
| 16 | #include <fstream.hfa> // sout
|
|---|
| 17 | #include <stdlib.hfa> // PRNG
|
|---|
| 18 | #include <clock.hfa>
|
|---|
| 19 | #include <limits.hfa> // MAX
|
|---|
| 20 | #include <math.hfa> // sqrt
|
|---|
| 21 | #include <malloc.h> // malloc_stats
|
|---|
| 22 | #include <locale.h> // setlocale
|
|---|
| 23 | #include <thread.hfa>
|
|---|
| 24 | #include <mutex_stmt.hfa>
|
|---|
| 25 |
|
|---|
| 26 | #define xstr(s) str(s)
|
|---|
| 27 | #define str(s) #s
|
|---|
| 28 |
|
|---|
| 29 | #if defined( __x86_64__ ) || defined( __aarch64__ ) // 64-bit architecture
|
|---|
| 30 | #define PRNG PRNG64
|
|---|
| 31 | #else // 32-bit architecture
|
|---|
| 32 | #define PRNG PRNG32
|
|---|
| 33 | #endif // __x86_64__
|
|---|
| 34 |
|
|---|
| 35 | //#define TIME
|
|---|
| 36 |
|
|---|
| 37 | #ifdef TIME // use -O2 -nodebug
|
|---|
| 38 | #define STARTTIME start = timeHiRes()
|
|---|
| 39 | #define ENDTIME( extra ) sout | wd(0,1, (timeHiRes() - start)`ms / 1000.) | extra "seconds"
|
|---|
| 40 | enum { BUCKETS = 100_000, TRIALS = 1_000_000_000 };
|
|---|
| 41 | #else
|
|---|
| 42 | #define STARTTIME
|
|---|
| 43 | #define ENDTIME( extra )
|
|---|
| 44 | enum { BUCKETS = 100_000, TRIALS = 100_000_000 };
|
|---|
| 45 | #endif // TIME
|
|---|
| 46 |
|
|---|
| 47 | static void avgstd( size_t trials, size_t buckets[] ) {
|
|---|
| 48 | size_t min = MAX, max = 0;
|
|---|
| 49 | double sum = 0.0, diff;
|
|---|
| 50 | for ( i; BUCKETS ) {
|
|---|
| 51 | if ( buckets[i] < min ) min = buckets[i];
|
|---|
| 52 | if ( buckets[i] > max ) max = buckets[i];
|
|---|
| 53 | sum += buckets[i];
|
|---|
| 54 | } // for
|
|---|
| 55 |
|
|---|
| 56 | double avg = sum / BUCKETS; // average
|
|---|
| 57 | sum = 0.0;
|
|---|
| 58 | for ( i; BUCKETS ) { // sum squared differences from average
|
|---|
| 59 | diff = buckets[i] - avg;
|
|---|
| 60 | sum += diff * diff;
|
|---|
| 61 | } // for
|
|---|
| 62 | double std = sqrt( sum / BUCKETS );
|
|---|
| 63 | mutex( sout ) sout | "trials" | trials | "buckets" | BUCKETS
|
|---|
| 64 | | "min" | min | "max" | max
|
|---|
| 65 | | "avg" | wd(0,1, avg) | "std" | wd(0,1, std) | "rstd" | wd(0,1, (avg == 0 ? 0.0 : std / avg * 100)) | "%";
|
|---|
| 66 | } // avgstd
|
|---|
| 67 |
|
|---|
| 68 |
|
|---|
| 69 | size_t seed = 1009;
|
|---|
| 70 |
|
|---|
| 71 | thread T1 {};
|
|---|
| 72 | void main( T1 & ) {
|
|---|
| 73 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 74 | for ( TRIALS / 50 ) {
|
|---|
| 75 | buckets[rand() % BUCKETS] += 1; // concurrent
|
|---|
| 76 | } // for
|
|---|
| 77 | avgstd( TRIALS / 50, buckets );
|
|---|
| 78 | free( buckets );
|
|---|
| 79 | } // main
|
|---|
| 80 |
|
|---|
| 81 | thread T2 {};
|
|---|
| 82 | void main( T2 & ) {
|
|---|
| 83 | PRNG prng;
|
|---|
| 84 | if ( seed != 0 ) set_seed( prng, seed );
|
|---|
| 85 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 86 | for ( TRIALS ) {
|
|---|
| 87 | buckets[prng( prng ) % BUCKETS] += 1; // concurrent
|
|---|
| 88 | } // for
|
|---|
| 89 | avgstd( TRIALS, buckets );
|
|---|
| 90 | free( buckets );
|
|---|
| 91 | } // main
|
|---|
| 92 |
|
|---|
| 93 | thread T3 {};
|
|---|
| 94 | void main( T3 & th ) {
|
|---|
| 95 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 96 | for ( TRIALS / 5 ) {
|
|---|
| 97 | buckets[prng() % BUCKETS] += 1; // concurrent
|
|---|
| 98 | } // for
|
|---|
| 99 | avgstd( TRIALS / 5, buckets );
|
|---|
| 100 | free( buckets );
|
|---|
| 101 | } // main
|
|---|
| 102 |
|
|---|
| 103 | thread T4 {};
|
|---|
| 104 | void main( T4 & th ) {
|
|---|
| 105 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 106 | for ( TRIALS ) {
|
|---|
| 107 | buckets[prng( th ) % BUCKETS] += 1; // concurrent
|
|---|
| 108 | } // for
|
|---|
| 109 | avgstd( TRIALS, buckets );
|
|---|
| 110 | free( buckets );
|
|---|
| 111 | } // main
|
|---|
| 112 |
|
|---|
| 113 | // Compiler bug requires hiding declaration of th from the bucket access, otherwise the compiler thinks th is aliased
|
|---|
| 114 | // and continually reloads it from memory, which doubles the cost.
|
|---|
| 115 | static void dummy( thread$ & th ) __attribute__(( noinline ));
|
|---|
| 116 | static void dummy( thread$ & th ) {
|
|---|
| 117 | size_t * buckets = (size_t *)calloc( BUCKETS, sizeof(size_t) ); // too big for task stack
|
|---|
| 118 | for ( size_t i = 0; i < TRIALS; i += 1 ) {
|
|---|
| 119 | buckets[prng( th ) % BUCKETS] += 1; // sequential
|
|---|
| 120 | } // for
|
|---|
| 121 | avgstd( TRIALS, buckets );
|
|---|
| 122 | free( buckets );
|
|---|
| 123 | } // dummy
|
|---|
| 124 |
|
|---|
| 125 |
|
|---|
| 126 | int main() {
|
|---|
| 127 | // setlocale( LC_NUMERIC, getenv( "LANG" ) ); // causes leaked storage message
|
|---|
| 128 |
|
|---|
| 129 | // only works on the current pthread thread
|
|---|
| 130 | // locale_t loc = newlocale( LC_NUMERIC_MASK, getenv( "LANG" ), (locale_t)0p );
|
|---|
| 131 | // if ( loc == (locale_t)0p ) abort( "newlocale" );
|
|---|
| 132 | // uselocale( loc );
|
|---|
| 133 |
|
|---|
| 134 | enum { TASKS = 4 };
|
|---|
| 135 | Time start;
|
|---|
| 136 |
|
|---|
| 137 | #ifdef TIME // too slow for test and generates non-repeatable results
|
|---|
| 138 | #if 1
|
|---|
| 139 | sout | "glib rand" | nl | nl;
|
|---|
| 140 |
|
|---|
| 141 | size_t rseed;
|
|---|
| 142 | if ( seed != 0 ) rseed = seed;
|
|---|
| 143 | else rseed = rdtscl();
|
|---|
| 144 | srand( rseed );
|
|---|
| 145 |
|
|---|
| 146 | sout | sepDisable;
|
|---|
| 147 | sout | nl | wd(26, "rand()" ) | wd(12, "rand(5)") | wd(12, "rand(0,5)" );
|
|---|
| 148 | for ( 20 ) {
|
|---|
| 149 | sout | wd(26, rand()) | nonl;
|
|---|
| 150 | sout | wd(12, rand() % 5) | nonl;
|
|---|
| 151 | sout | wd(12, rand() % (5 - 0 + 1) + 0);
|
|---|
| 152 | } // for
|
|---|
| 153 | sout | sepEnable;
|
|---|
| 154 | sout | "seed" | rseed;
|
|---|
| 155 |
|
|---|
| 156 | sout | nl | "Sequential";
|
|---|
| 157 | STARTTIME;
|
|---|
| 158 | {
|
|---|
| 159 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 160 | for ( i; TRIALS / 5 ) {
|
|---|
| 161 | buckets[rand() % BUCKETS] += 1; // sequential
|
|---|
| 162 | } // for
|
|---|
| 163 | avgstd( TRIALS / 5, buckets );
|
|---|
| 164 | free( buckets );
|
|---|
| 165 | }
|
|---|
| 166 | ENDTIME( " x 5 " );
|
|---|
| 167 |
|
|---|
| 168 | sout | nl | "Concurrent";
|
|---|
| 169 | STARTTIME;
|
|---|
| 170 | {
|
|---|
| 171 | processor p[TASKS - 1]; // already 1 processor
|
|---|
| 172 | {
|
|---|
| 173 | T1 t[TASKS];
|
|---|
| 174 | } // wait for threads to complete
|
|---|
| 175 | }
|
|---|
| 176 | ENDTIME( " x 50 " );
|
|---|
| 177 | #endif // 0
|
|---|
| 178 | #endif // TIME
|
|---|
| 179 |
|
|---|
| 180 | sout | nl | "CFA " xstr(PRNG_NAME);
|
|---|
| 181 |
|
|---|
| 182 | #if 1
|
|---|
| 183 | PRNG prng;
|
|---|
| 184 |
|
|---|
| 185 | if ( seed != 0 ) set_seed( prng, seed );
|
|---|
| 186 |
|
|---|
| 187 | sout | sepDisable;
|
|---|
| 188 | sout | nl | wd(26, "PRNG()" ) | wd(12, "PRNG(5)") | wd(12, "PRNG(0,5)" );
|
|---|
| 189 | for ( 20 ) {
|
|---|
| 190 | sout | wd(26, prng( prng )) | nonl; // cascading => side-effect functions called in arbitary order
|
|---|
| 191 | sout | wd(12, prng( prng, 5 )) | nonl;
|
|---|
| 192 | sout | wd(12, prng( prng, 0, 5 ));
|
|---|
| 193 | } // for
|
|---|
| 194 | sout | sepEnable;
|
|---|
| 195 | sout | "seed" | get_seed( prng );
|
|---|
| 196 |
|
|---|
| 197 | sout | nl | "Sequential";
|
|---|
| 198 | STARTTIME;
|
|---|
| 199 | {
|
|---|
| 200 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 201 | for ( TRIALS ) {
|
|---|
| 202 | buckets[prng( prng ) % BUCKETS] += 1; // sequential
|
|---|
| 203 | } // for
|
|---|
| 204 | avgstd( TRIALS, buckets );
|
|---|
| 205 | free( buckets );
|
|---|
| 206 | }
|
|---|
| 207 | ENDTIME();
|
|---|
| 208 |
|
|---|
| 209 | sout | nl | "Concurrent";
|
|---|
| 210 | STARTTIME;
|
|---|
| 211 | {
|
|---|
| 212 | processor p[TASKS - 1]; // already 1 processor
|
|---|
| 213 | {
|
|---|
| 214 | T2 t[TASKS];
|
|---|
| 215 | } // wait for threads to complete
|
|---|
| 216 | }
|
|---|
| 217 | ENDTIME();
|
|---|
| 218 | #endif // 0
|
|---|
| 219 | #if 1
|
|---|
| 220 | if ( seed != 0 ) set_seed( seed );
|
|---|
| 221 |
|
|---|
| 222 | sout | sepDisable;
|
|---|
| 223 | sout | nl | wd(26, "prng()" ) | wd(12, "prng(5)") | wd(12, "prng(0,5)" );
|
|---|
| 224 | for ( 20 ) {
|
|---|
| 225 | sout | wd(26, prng()) | nonl; // cascading => side-effect functions called in arbitary order
|
|---|
| 226 | sout | wd(12, prng( 5 )) | nonl;
|
|---|
| 227 | sout | wd(12, prng( 0, 5 ));
|
|---|
| 228 | } // for
|
|---|
| 229 | sout | sepEnable;
|
|---|
| 230 | sout | "seed" | get_seed( prng );
|
|---|
| 231 |
|
|---|
| 232 | sout | nl | "Sequential";
|
|---|
| 233 | STARTTIME;
|
|---|
| 234 | {
|
|---|
| 235 | size_t * buckets = calloc( BUCKETS ); // too big for task stack
|
|---|
| 236 | for ( TRIALS / 5 ) {
|
|---|
| 237 | buckets[prng() % BUCKETS] += 1;
|
|---|
| 238 | } // for
|
|---|
| 239 | avgstd( TRIALS / 5, buckets );
|
|---|
| 240 | free( buckets );
|
|---|
| 241 | }
|
|---|
| 242 | ENDTIME( " x 5 " );
|
|---|
| 243 |
|
|---|
| 244 | sout | nl | "Concurrent";
|
|---|
| 245 | STARTTIME;
|
|---|
| 246 | {
|
|---|
| 247 | processor p[TASKS - 1]; // already 1 processor
|
|---|
| 248 | {
|
|---|
| 249 | T3 t[TASKS];
|
|---|
| 250 | } // wait for threads to complete
|
|---|
| 251 | }
|
|---|
| 252 | ENDTIME( " x 5 " );
|
|---|
| 253 | #endif // 0
|
|---|
| 254 | #if 1
|
|---|
| 255 | if ( seed != 0 ) set_seed( seed );
|
|---|
| 256 | thread$ & th = *active_thread();
|
|---|
| 257 |
|
|---|
| 258 | sout | sepDisable;
|
|---|
| 259 | sout | nl | wd(26, "prng(t)" ) | wd(12, "prng(t,5)") | wd(12, "prng(t,0,5)" );
|
|---|
| 260 | for ( 20 ) {
|
|---|
| 261 | sout | wd(26, prng( th )) | nonl; // cascading => side-effect functions called in arbitary order
|
|---|
| 262 | sout | wd(12, prng( th, 5 )) | nonl;
|
|---|
| 263 | sout | wd(12, prng( th, 0, 5 ));
|
|---|
| 264 | } // for
|
|---|
| 265 | sout | sepEnable;
|
|---|
| 266 | sout | "seed" | get_seed( prng );
|
|---|
| 267 |
|
|---|
| 268 | sout | nl | "Sequential";
|
|---|
| 269 | STARTTIME;
|
|---|
| 270 | {
|
|---|
| 271 | dummy( th );
|
|---|
| 272 | }
|
|---|
| 273 | ENDTIME();
|
|---|
| 274 |
|
|---|
| 275 | sout | nl | "Concurrent";
|
|---|
| 276 | STARTTIME;
|
|---|
| 277 | {
|
|---|
| 278 | processor p[TASKS - 1]; // already 1 processor
|
|---|
| 279 | {
|
|---|
| 280 | T4 t[TASKS];
|
|---|
| 281 | } // wait for threads to complete
|
|---|
| 282 | }
|
|---|
| 283 | ENDTIME();
|
|---|
| 284 | #endif // 0
|
|---|
| 285 | // malloc_stats();
|
|---|
| 286 | // freelocale( loc );
|
|---|
| 287 | } // main
|
|---|
| 288 |
|
|---|
| 289 |
|
|---|
| 290 | // Local Variables: //
|
|---|
| 291 | // compile-command: "cfa -DTIME -O2 -nodebug PRNG.cfa" //
|
|---|
| 292 | // End: //
|
|---|