source: tests/PRNG.cfa@ 3a513d89

ADT
Last change on this file since 3a513d89 was 8a2f7f1, checked in by Peter A. Buhr <pabuhr@…>, 2 years ago

fix PRNG test to work correctly on ARM
  • Property mode set to 100644
File size: 7.2 KB
RevLine 
[5910fc0]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
[8a2f7f1]10// Last Modified On : Sat Apr 15 16:44:31 2023
11// Update Count : 419
[5910fc0]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
[8a2f7f1]21#include <thread.hfa>
[470618c0]22#include <mutex_stmt.hfa>
[5910fc0]23
[8a2f7f1]24#define xstr(s) str(s)
25#define str(s) #s
26
27#if defined( __x86_64__ ) || defined( __aarch64__ ) // 64-bit architecture
[d2ad151]28#define PRNG PRNG64
29#else // 32-bit architecture
30#define PRNG PRNG32
31#endif // __x86_64__
32
[8a2f7f1]33//#define TIME
34
[5910fc0]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"
38enum { BUCKETS = 100_000, TRIALS = 1_000_000_000 };
39#else
40#define STARTTIME
41#define ENDTIME( extra )
42enum { BUCKETS = 100_000, TRIALS = 100_000_000 };
43#endif // TIME
44
[8a2f7f1]45static void avgstd( size_t trials, size_t buckets[] ) {
46 size_t min = MAX, max = 0;
[5910fc0]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 );
[8a2f7f1]61 mutex( sout ) sout | "trials" | trials | "buckets" | BUCKETS
[470618c0]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)) | "%";
[5910fc0]64} // avgstd
65
66
[ae7a085c]67size_t seed = 1009;
[5910fc0]68
69thread T1 {};
70void main( T1 & ) {
[8a2f7f1]71 size_t * buckets = calloc( BUCKETS ); // too big for task stack
72 for ( TRIALS / 50 ) {
[5910fc0]73 buckets[rand() % BUCKETS] += 1; // concurrent
74 } // for
[8a2f7f1]75 avgstd( TRIALS / 50, buckets );
[5910fc0]76 free( buckets );
77} // main
78
79thread T2 {};
80void main( T2 & ) {
81 PRNG prng;
82 if ( seed != 0 ) set_seed( prng, seed );
[8a2f7f1]83 size_t * buckets = calloc( BUCKETS ); // too big for task stack
[5910fc0]84 for ( TRIALS ) {
85 buckets[prng( prng ) % BUCKETS] += 1; // concurrent
86 } // for
[8a2f7f1]87 avgstd( TRIALS, buckets );
[5910fc0]88 free( buckets );
89} // main
90
91thread T3 {};
92void main( T3 & th ) {
[8a2f7f1]93 size_t * buckets = calloc( BUCKETS ); // too big for task stack
94 for ( TRIALS / 5 ) {
[5910fc0]95 buckets[prng() % BUCKETS] += 1; // concurrent
96 } // for
[8a2f7f1]97 avgstd( TRIALS / 5, buckets );
[5910fc0]98 free( buckets );
99} // main
100
101thread T4 {};
102void main( T4 & th ) {
[8a2f7f1]103 size_t * buckets = calloc( BUCKETS ); // too big for task stack
[5910fc0]104 for ( TRIALS ) {
[8a2f7f1]105 buckets[prng( th ) % BUCKETS] += 1; // concurrent
[5910fc0]106 } // for
[8a2f7f1]107 avgstd( TRIALS, buckets );
[5910fc0]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.
113static void dummy( thread$ & th ) __attribute__(( noinline ));
114static void dummy( thread$ & th ) {
[8a2f7f1]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 ) {
[5910fc0]117 buckets[prng( th ) % BUCKETS] += 1; // sequential
118 } // for
[8a2f7f1]119 avgstd( TRIALS, buckets );
[5910fc0]120 free( buckets );
121} // dummy
122
[ae7a085c]123
[5910fc0]124int main() {
125 // causes leaked storage message
[ae7a085c]126 // setlocale( LC_NUMERIC, getenv( "LANG" ) ); // print digit separator
127 // locale_t loc = newlocale( LC_NUMERIC_MASK, getenv( "LANG" ), (locale_t)0p );
128 // if ( loc == (locale_t)0p ) abort( "newlocale" );
129 // uselocale( loc );
[5910fc0]130
131 enum { TASKS = 4 };
132 Time start;
[8a2f7f1]133
[5910fc0]134#ifdef TIME // too slow for test and generates non-repeatable results
135#if 1
[8a2f7f1]136 sout | "glib rand" | nl | nl;
137
138 size_t rseed;
[5910fc0]139 if ( seed != 0 ) rseed = seed;
140 else rseed = rdtscl();
141 srand( rseed );
142
143 sout | sepDisable;
[8a2f7f1]144 sout | nl | wd(26, "rand()" ) | wd(12, "rand(5)") | wd(12, "rand(0,5)" );
[5910fc0]145 for ( 20 ) {
[ae7a085c]146 sout | wd(26, rand()) | nonl;
147 sout | wd(12, rand() % 5) | nonl;
148 sout | wd(12, rand() % (5 - 0 + 1) + 0);
[5910fc0]149 } // for
150 sout | sepEnable;
151 sout | "seed" | rseed;
152
153 sout | nl | "Sequential";
154 STARTTIME;
155 {
[8a2f7f1]156 size_t * buckets = calloc( BUCKETS ); // too big for task stack
157 for ( i; TRIALS / 5 ) {
[5910fc0]158 buckets[rand() % BUCKETS] += 1; // sequential
159 } // for
[8a2f7f1]160 avgstd( TRIALS / 5, buckets );
[5910fc0]161 free( buckets );
162 }
[8a2f7f1]163 ENDTIME( " x 5 " );
[5910fc0]164
165 sout | nl | "Concurrent";
166 STARTTIME;
167 {
168 processor p[TASKS - 1]; // already 1 processor
169 {
170 T1 t[TASKS];
171 } // wait for threads to complete
172 }
[8a2f7f1]173 ENDTIME( " x 50 " );
[5910fc0]174#endif // 0
175#endif // TIME
[8a2f7f1]176
177 sout | nl | "CFA " xstr(PRNG_NAME);
178
[5910fc0]179#if 1
180 PRNG prng;
[d2ad151]181
[5910fc0]182 if ( seed != 0 ) set_seed( prng, seed );
183
184 sout | sepDisable;
[ae7a085c]185 sout | nl | wd(26, "PRNG()" ) | wd(12, "PRNG(5)") | wd(12, "PRNG(0,5)" );
[5910fc0]186 for ( 20 ) {
[ae7a085c]187 sout | wd(26, prng( prng )) | nonl; // cascading => side-effect functions called in arbitary order
188 sout | wd(12, prng( prng, 5 )) | nonl;
189 sout | wd(12, prng( prng, 0, 5 ));
[5910fc0]190 } // for
191 sout | sepEnable;
192 sout | "seed" | get_seed( prng );
193
194 sout | nl | "Sequential";
195 STARTTIME;
196 {
[8a2f7f1]197 size_t * buckets = calloc( BUCKETS ); // too big for task stack
[5910fc0]198 for ( TRIALS ) {
199 buckets[prng( prng ) % BUCKETS] += 1; // sequential
200 } // for
[8a2f7f1]201 avgstd( TRIALS, buckets );
[5910fc0]202 free( buckets );
203 }
204 ENDTIME();
205
206 sout | nl | "Concurrent";
207 STARTTIME;
208 {
209 processor p[TASKS - 1]; // already 1 processor
210 {
211 T2 t[TASKS];
212 } // wait for threads to complete
213 }
214 ENDTIME();
215#endif // 0
216#if 1
217 if ( seed != 0 ) set_seed( seed );
218
219 sout | sepDisable;
[ae7a085c]220 sout | nl | wd(26, "prng()" ) | wd(12, "prng(5)") | wd(12, "prng(0,5)" );
[5910fc0]221 for ( 20 ) {
[ae7a085c]222 sout | wd(26, prng()) | nonl; // cascading => side-effect functions called in arbitary order
223 sout | wd(12, prng( 5 )) | nonl;
224 sout | wd(12, prng( 0, 5 ));
[5910fc0]225 } // for
226 sout | sepEnable;
227 sout | "seed" | get_seed( prng );
228
229 sout | nl | "Sequential";
230 STARTTIME;
231 {
[8a2f7f1]232 size_t * buckets = calloc( BUCKETS ); // too big for task stack
233 for ( TRIALS / 5 ) {
[5910fc0]234 buckets[prng() % BUCKETS] += 1;
235 } // for
[8a2f7f1]236 avgstd( TRIALS / 5, buckets );
[5910fc0]237 free( buckets );
238 }
[8a2f7f1]239 ENDTIME( " x 5 " );
[5910fc0]240
241 sout | nl | "Concurrent";
242 STARTTIME;
243 {
244 processor p[TASKS - 1]; // already 1 processor
245 {
246 T3 t[TASKS];
247 } // wait for threads to complete
248 }
[8a2f7f1]249 ENDTIME( " x 5 " );
[5910fc0]250#endif // 0
251#if 1
252 if ( seed != 0 ) set_seed( seed );
253 thread$ & th = *active_thread();
254
255 sout | sepDisable;
[ae7a085c]256 sout | nl | wd(26, "prng(t)" ) | wd(12, "prng(t,5)") | wd(12, "prng(t,0,5)" );
[5910fc0]257 for ( 20 ) {
[ae7a085c]258 sout | wd(26, prng( th )) | nonl; // cascading => side-effect functions called in arbitary order
259 sout | wd(12, prng( th, 5 )) | nonl;
260 sout | wd(12, prng( th, 0, 5 ));
[5910fc0]261 } // for
262 sout | sepEnable;
263 sout | "seed" | get_seed( prng );
264
265 sout | nl | "Sequential";
266 STARTTIME;
267 {
268 dummy( th );
269 }
270 ENDTIME();
271
272 sout | nl | "Concurrent";
273 STARTTIME;
274 {
275 processor p[TASKS - 1]; // already 1 processor
276 {
277 T4 t[TASKS];
278 } // wait for threads to complete
279 }
280 ENDTIME();
281#endif // 0
282// malloc_stats();
[ae7a085c]283 // freelocale( loc );
[5910fc0]284} // main
285
286
287// Local Variables: //
288// compile-command: "cfa -DTIME -O2 -nodebug PRNG.cfa" //
289// End: //
Note: See TracBrowser for help on using the repository browser.