1 | /************************************************************************* |
---|
2 | * * |
---|
3 | * Open Dynamics Engine, Copyright (C) 2001,2002 Russell L. Smith. * |
---|
4 | * All rights reserved. Email: russ@q12.org Web: www.q12.org * |
---|
5 | * * |
---|
6 | * This library is free software; you can redistribute it and/or * |
---|
7 | * modify it under the terms of EITHER: * |
---|
8 | * (1) The GNU Lesser General Public License as published by the Free * |
---|
9 | * Software Foundation; either version 2.1 of the License, or (at * |
---|
10 | * your option) any later version. The text of the GNU Lesser * |
---|
11 | * General Public License is included with this library in the * |
---|
12 | * file LICENSE.TXT. * |
---|
13 | * (2) The BSD-style license that is included with this library in * |
---|
14 | * the file LICENSE-BSD.TXT. * |
---|
15 | * * |
---|
16 | * This library is distributed in the hope that it will be useful, * |
---|
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of * |
---|
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the files * |
---|
19 | * LICENSE.TXT and LICENSE-BSD.TXT for more details. * |
---|
20 | * * |
---|
21 | *************************************************************************/ |
---|
22 | |
---|
23 | /* |
---|
24 | |
---|
25 | TODO |
---|
26 | ---- |
---|
27 | |
---|
28 | * gettimeofday() and the pentium time stamp counter return the real time, |
---|
29 | not the process time. fix this somehow! |
---|
30 | |
---|
31 | */ |
---|
32 | |
---|
33 | #include <ode/common.h> |
---|
34 | #include <ode/timer.h> |
---|
35 | |
---|
36 | // misc defines |
---|
37 | #define ALLOCA dALLOCA16 |
---|
38 | |
---|
39 | //**************************************************************************** |
---|
40 | // implementation for windows based on the multimedia performance counter. |
---|
41 | |
---|
42 | #ifdef WIN32 |
---|
43 | |
---|
44 | #include "windows.h" |
---|
45 | |
---|
46 | static inline void getClockCount (unsigned long cc[2]) |
---|
47 | { |
---|
48 | LARGE_INTEGER a; |
---|
49 | QueryPerformanceCounter (&a); |
---|
50 | cc[0] = a.LowPart; |
---|
51 | cc[1] = a.HighPart; |
---|
52 | } |
---|
53 | |
---|
54 | |
---|
55 | static inline void serialize() |
---|
56 | { |
---|
57 | } |
---|
58 | |
---|
59 | |
---|
60 | static inline double loadClockCount (unsigned long cc[2]) |
---|
61 | { |
---|
62 | LARGE_INTEGER a; |
---|
63 | a.LowPart = cc[0]; |
---|
64 | a.HighPart = cc[1]; |
---|
65 | return double(a.QuadPart); |
---|
66 | } |
---|
67 | |
---|
68 | |
---|
69 | double dTimerResolution() |
---|
70 | { |
---|
71 | return 1.0/dTimerTicksPerSecond(); |
---|
72 | } |
---|
73 | |
---|
74 | |
---|
75 | double dTimerTicksPerSecond() |
---|
76 | { |
---|
77 | static int query=0; |
---|
78 | static double hz=0.0; |
---|
79 | if (!query) { |
---|
80 | LARGE_INTEGER a; |
---|
81 | QueryPerformanceFrequency (&a); |
---|
82 | hz = double(a.QuadPart); |
---|
83 | query = 1; |
---|
84 | } |
---|
85 | return hz; |
---|
86 | } |
---|
87 | |
---|
88 | #endif |
---|
89 | |
---|
90 | //**************************************************************************** |
---|
91 | // implementation based on the pentium time stamp counter. the timer functions |
---|
92 | // can be serializing or non-serializing. serializing will ensure that all |
---|
93 | // instructions have executed and data has been written back before the cpu |
---|
94 | // time stamp counter is read. the CPUID instruction is used to serialize. |
---|
95 | |
---|
96 | #if defined(PENTIUM) && !defined(WIN32) |
---|
97 | |
---|
98 | // we need to know the clock rate so that the timing function can report |
---|
99 | // accurate times. this number only needs to be set accurately if we're |
---|
100 | // doing performance tests and care about real-world time numbers - otherwise, |
---|
101 | // just ignore this. i have not worked out how to determine this number |
---|
102 | // automatically yet. |
---|
103 | |
---|
104 | #define PENTIUM_HZ (500e6) |
---|
105 | |
---|
106 | static inline void getClockCount (unsigned long cc[2]) |
---|
107 | { |
---|
108 | #ifndef X86_64_SYSTEM |
---|
109 | asm volatile ( |
---|
110 | "rdtsc\n" |
---|
111 | "movl %%eax,(%%esi)\n" |
---|
112 | "movl %%edx,4(%%esi)\n" |
---|
113 | : : "S" (cc) : "%eax","%edx","cc","memory"); |
---|
114 | #else |
---|
115 | asm volatile ( |
---|
116 | "rdtsc\n" |
---|
117 | "movl %%eax,(%%rsi)\n" |
---|
118 | "movl %%edx,4(%%rsi)\n" |
---|
119 | : : "S" (cc) : "%eax","%edx","cc","memory"); |
---|
120 | #endif |
---|
121 | } |
---|
122 | |
---|
123 | |
---|
124 | static inline void serialize() |
---|
125 | { |
---|
126 | #ifndef X86_64_SYSTEM |
---|
127 | asm volatile ( |
---|
128 | "mov $0,%%eax\n" |
---|
129 | "push %%ebx\n" |
---|
130 | "cpuid\n" |
---|
131 | "pop %%ebx\n" |
---|
132 | : : : "%eax","%ecx","%edx","cc","memory"); |
---|
133 | #else |
---|
134 | asm volatile ( |
---|
135 | "mov $0,%%rax\n" |
---|
136 | "push %%rbx\n" |
---|
137 | "cpuid\n" |
---|
138 | "pop %%rbx\n" |
---|
139 | : : : "%rax","%rcx","%rdx","cc","memory"); |
---|
140 | #endif |
---|
141 | } |
---|
142 | |
---|
143 | |
---|
144 | static inline double loadClockCount (unsigned long a[2]) |
---|
145 | { |
---|
146 | double ret; |
---|
147 | #ifndef X86_64_SYSTEM |
---|
148 | asm volatile ("fildll %1; fstpl %0" : "=m" (ret) : "m" (a[0]) : |
---|
149 | "cc","memory"); |
---|
150 | #else |
---|
151 | asm volatile ("fildll %1; fstpl %0" : "=m" (ret) : "m" (a[0]) : |
---|
152 | "cc","memory"); |
---|
153 | #endif |
---|
154 | return ret; |
---|
155 | } |
---|
156 | |
---|
157 | |
---|
158 | double dTimerResolution() |
---|
159 | { |
---|
160 | return 1.0/PENTIUM_HZ; |
---|
161 | } |
---|
162 | |
---|
163 | |
---|
164 | double dTimerTicksPerSecond() |
---|
165 | { |
---|
166 | return PENTIUM_HZ; |
---|
167 | } |
---|
168 | |
---|
169 | #endif |
---|
170 | |
---|
171 | //**************************************************************************** |
---|
172 | // otherwise, do the implementation based on gettimeofday(). |
---|
173 | |
---|
174 | #if !defined(PENTIUM) && !defined(WIN32) |
---|
175 | |
---|
176 | #ifndef macintosh |
---|
177 | |
---|
178 | #include <sys/time.h> |
---|
179 | #include <unistd.h> |
---|
180 | |
---|
181 | |
---|
182 | static inline void getClockCount (unsigned long cc[2]) |
---|
183 | { |
---|
184 | struct timeval tv; |
---|
185 | gettimeofday (&tv,0); |
---|
186 | cc[0] = tv.tv_usec; |
---|
187 | cc[1] = tv.tv_sec; |
---|
188 | } |
---|
189 | |
---|
190 | #else // macintosh |
---|
191 | |
---|
192 | #include <MacTypes.h> |
---|
193 | #include <Timer.h> |
---|
194 | |
---|
195 | static inline void getClockCount (unsigned long cc[2]) |
---|
196 | { |
---|
197 | UnsignedWide ms; |
---|
198 | Microseconds (&ms); |
---|
199 | cc[1] = ms.lo / 1000000; |
---|
200 | cc[0] = ms.lo - ( cc[1] * 1000000 ); |
---|
201 | } |
---|
202 | |
---|
203 | #endif |
---|
204 | |
---|
205 | |
---|
206 | static inline void serialize() |
---|
207 | { |
---|
208 | } |
---|
209 | |
---|
210 | |
---|
211 | static inline double loadClockCount (unsigned long a[2]) |
---|
212 | { |
---|
213 | return a[1]*1.0e6 + a[0]; |
---|
214 | } |
---|
215 | |
---|
216 | |
---|
217 | double dTimerResolution() |
---|
218 | { |
---|
219 | unsigned long cc1[2],cc2[2]; |
---|
220 | getClockCount (cc1); |
---|
221 | do { |
---|
222 | getClockCount (cc2); |
---|
223 | } |
---|
224 | while (cc1[0]==cc2[0] && cc1[1]==cc2[1]); |
---|
225 | do { |
---|
226 | getClockCount (cc1); |
---|
227 | } |
---|
228 | while (cc1[0]==cc2[0] && cc1[1]==cc2[1]); |
---|
229 | double t1 = loadClockCount (cc1); |
---|
230 | double t2 = loadClockCount (cc2); |
---|
231 | return (t1-t2) / dTimerTicksPerSecond(); |
---|
232 | } |
---|
233 | |
---|
234 | |
---|
235 | double dTimerTicksPerSecond() |
---|
236 | { |
---|
237 | return 1000000; |
---|
238 | } |
---|
239 | |
---|
240 | #endif |
---|
241 | |
---|
242 | //**************************************************************************** |
---|
243 | // stop watches |
---|
244 | |
---|
245 | void dStopwatchReset (dStopwatch *s) |
---|
246 | { |
---|
247 | s->time = 0; |
---|
248 | s->cc[0] = 0; |
---|
249 | s->cc[1] = 0; |
---|
250 | } |
---|
251 | |
---|
252 | |
---|
253 | void dStopwatchStart (dStopwatch *s) |
---|
254 | { |
---|
255 | serialize(); |
---|
256 | getClockCount (s->cc); |
---|
257 | } |
---|
258 | |
---|
259 | |
---|
260 | void dStopwatchStop (dStopwatch *s) |
---|
261 | { |
---|
262 | unsigned long cc[2]; |
---|
263 | serialize(); |
---|
264 | getClockCount (cc); |
---|
265 | double t1 = loadClockCount (s->cc); |
---|
266 | double t2 = loadClockCount (cc); |
---|
267 | s->time += t2-t1; |
---|
268 | } |
---|
269 | |
---|
270 | |
---|
271 | double dStopwatchTime (dStopwatch *s) |
---|
272 | { |
---|
273 | return s->time / dTimerTicksPerSecond(); |
---|
274 | } |
---|
275 | |
---|
276 | //**************************************************************************** |
---|
277 | // code timers |
---|
278 | |
---|
279 | // maximum number of events to record |
---|
280 | #define MAXNUM 100 |
---|
281 | |
---|
282 | static int num = 0; // number of entries used in event array |
---|
283 | static struct { |
---|
284 | unsigned long cc[2]; // clock counts |
---|
285 | double total_t; // total clocks used in this slot. |
---|
286 | double total_p; // total percentage points used in this slot. |
---|
287 | int count; // number of times this slot has been updated. |
---|
288 | char *description; // pointer to static string |
---|
289 | } event[MAXNUM]; |
---|
290 | |
---|
291 | |
---|
292 | // make sure all slot totals and counts reset to 0 at start |
---|
293 | |
---|
294 | static void initSlots() |
---|
295 | { |
---|
296 | static int initialized=0; |
---|
297 | if (!initialized) { |
---|
298 | for (int i=0; i<MAXNUM; i++) { |
---|
299 | event[i].count = 0; |
---|
300 | event[i].total_t = 0; |
---|
301 | event[i].total_p = 0; |
---|
302 | } |
---|
303 | initialized = 1; |
---|
304 | } |
---|
305 | } |
---|
306 | |
---|
307 | |
---|
308 | void dTimerStart (const char *description) |
---|
309 | { |
---|
310 | initSlots(); |
---|
311 | event[0].description = const_cast<char*> (description); |
---|
312 | num = 1; |
---|
313 | serialize(); |
---|
314 | getClockCount (event[0].cc); |
---|
315 | } |
---|
316 | |
---|
317 | |
---|
318 | void dTimerNow (const char *description) |
---|
319 | { |
---|
320 | if (num < MAXNUM) { |
---|
321 | // do not serialize |
---|
322 | getClockCount (event[num].cc); |
---|
323 | event[num].description = const_cast<char*> (description); |
---|
324 | num++; |
---|
325 | } |
---|
326 | } |
---|
327 | |
---|
328 | |
---|
329 | void dTimerEnd() |
---|
330 | { |
---|
331 | if (num < MAXNUM) { |
---|
332 | serialize(); |
---|
333 | getClockCount (event[num].cc); |
---|
334 | event[num].description = "TOTAL"; |
---|
335 | num++; |
---|
336 | } |
---|
337 | } |
---|
338 | |
---|
339 | //**************************************************************************** |
---|
340 | // print report |
---|
341 | |
---|
342 | static void fprintDoubleWithPrefix (FILE *f, double a, char *fmt) |
---|
343 | { |
---|
344 | if (a >= 0.999999) { |
---|
345 | fprintf (f,fmt,a); |
---|
346 | return; |
---|
347 | } |
---|
348 | a *= 1000.0; |
---|
349 | if (a >= 0.999999) { |
---|
350 | fprintf (f,fmt,a); |
---|
351 | fprintf (f,"m"); |
---|
352 | return; |
---|
353 | } |
---|
354 | a *= 1000.0; |
---|
355 | if (a >= 0.999999) { |
---|
356 | fprintf (f,fmt,a); |
---|
357 | fprintf (f,"u"); |
---|
358 | return; |
---|
359 | } |
---|
360 | a *= 1000.0; |
---|
361 | fprintf (f,fmt,a); |
---|
362 | fprintf (f,"n"); |
---|
363 | } |
---|
364 | |
---|
365 | |
---|
366 | void dTimerReport (FILE *fout, int average) |
---|
367 | { |
---|
368 | int i; |
---|
369 | size_t maxl; |
---|
370 | double ccunit = 1.0/dTimerTicksPerSecond(); |
---|
371 | fprintf (fout,"\nTimer Report ("); |
---|
372 | fprintDoubleWithPrefix (fout,ccunit,"%.2f "); |
---|
373 | fprintf (fout,"s resolution)\n------------\n"); |
---|
374 | if (num < 1) return; |
---|
375 | |
---|
376 | // get maximum description length |
---|
377 | maxl = 0; |
---|
378 | for (i=0; i<num; i++) { |
---|
379 | size_t l = strlen (event[i].description); |
---|
380 | if (l > maxl) maxl = l; |
---|
381 | } |
---|
382 | |
---|
383 | // calculate total time |
---|
384 | double t1 = loadClockCount (event[0].cc); |
---|
385 | double t2 = loadClockCount (event[num-1].cc); |
---|
386 | double total = t2 - t1; |
---|
387 | if (total <= 0) total = 1; |
---|
388 | |
---|
389 | // compute time difference for all slots except the last one. update totals |
---|
390 | double *times = (double*) ALLOCA (num * sizeof(double)); |
---|
391 | for (i=0; i < (num-1); i++) { |
---|
392 | double t1 = loadClockCount (event[i].cc); |
---|
393 | double t2 = loadClockCount (event[i+1].cc); |
---|
394 | times[i] = t2 - t1; |
---|
395 | event[i].count++; |
---|
396 | event[i].total_t += times[i]; |
---|
397 | event[i].total_p += times[i]/total * 100.0; |
---|
398 | } |
---|
399 | |
---|
400 | // print report (with optional averages) |
---|
401 | for (i=0; i<num; i++) { |
---|
402 | double t,p; |
---|
403 | if (i < (num-1)) { |
---|
404 | t = times[i]; |
---|
405 | p = t/total * 100.0; |
---|
406 | } |
---|
407 | else { |
---|
408 | t = total; |
---|
409 | p = 100.0; |
---|
410 | } |
---|
411 | fprintf (fout,"%-*s %7.2fms %6.2f%%",maxl,event[i].description, |
---|
412 | t*ccunit * 1000.0, p); |
---|
413 | if (average && i < (num-1)) { |
---|
414 | fprintf (fout," (avg %7.2fms %6.2f%%)", |
---|
415 | (event[i].total_t / event[i].count)*ccunit * 1000.0, |
---|
416 | event[i].total_p / event[i].count); |
---|
417 | } |
---|
418 | fprintf (fout,"\n"); |
---|
419 | } |
---|
420 | fprintf (fout,"\n"); |
---|
421 | } |
---|