Commit | Line | Data |
---|---|---|
53cf6719 | 1 | #ifndef lint |
2d19cd70 | 2 | static char *sccsid = "@(#)prof.c 4.3 (Berkeley) %G%"; |
53cf6719 SL |
3 | #endif |
4 | /* | |
5 | * prof | |
6 | */ | |
7 | #include <stdio.h> | |
8 | #include <sys/types.h> | |
9 | #include <sys/stat.h> | |
10 | #include <a.out.h> | |
11 | ||
12 | typedef short UNIT; /* unit of profiling */ | |
2d19cd70 | 13 | #define PCFUDGE 11 |
53cf6719 SL |
14 | #define A_OUTNAME "a.out" |
15 | #define MON_OUTNAME "mon.out" | |
16 | #define MON_SUMNAME "mon.sum" | |
17 | ||
18 | /* | |
19 | * The symbol table; | |
20 | * for each external in the specified file we gather | |
21 | * its address, the number of calls and compute its share of cpu time. | |
22 | */ | |
23 | struct nl { | |
24 | char *name; | |
25 | unsigned value; | |
26 | float time; | |
27 | long ncall; | |
28 | } *nl; | |
29 | int nname; | |
30 | struct nl *np; | |
31 | struct nl *npe; | |
32 | ||
33 | /* | |
34 | * The header on the mon.out file. | |
35 | * Mon.out consists of one of these headers, an array of ncount | |
36 | * cnt structures (as below) and then an array of samples | |
37 | * representing the discretized program counter values. | |
38 | */ | |
39 | struct hdr { | |
40 | UNIT *lowpc, *highpc; | |
41 | int ncount; | |
42 | } h; | |
43 | ||
44 | /* | |
45 | * Each counter has an address and a number of calls. | |
46 | */ | |
47 | struct cnt { | |
48 | unsigned cvalue; | |
49 | long cncall; | |
50 | } *cbuf; | |
51 | ||
52 | /* | |
53 | * Each discretized pc sample has | |
54 | * a count of the number of samples in its range | |
55 | */ | |
56 | unsigned UNIT *samples; | |
57 | ||
58 | FILE *pfile, *nfile; | |
59 | ||
60 | unsigned lowpc, highpc; /* range profiled */ | |
61 | double ransca, ranoff; /* scaling for blowing up plots */ | |
62 | unsigned sampbytes; /* number of bytes of samples */ | |
63 | int nsamples; /* number of samples */ | |
64 | double totime; /* total time for all routines */ | |
65 | double maxtime; /* maximum time of any routine (for plot) */ | |
66 | double scale; /* scale factor converting samples to pc | |
67 | values: each sample covers scale bytes */ | |
68 | char *strtab; /* string table in core */ | |
69 | off_t ssiz; /* size of the string table */ | |
70 | struct exec xbuf; /* exec header of a.out */ | |
71 | ||
72 | int aflg; | |
73 | int nflg; | |
74 | int vflg; | |
75 | int lflg; | |
76 | int zflg; | |
77 | int sflag; | |
78 | ||
79 | char *namfil; | |
80 | ||
81 | int timcmp(), valcmp(), cntcmp(); | |
82 | ||
83 | main(argc, argv) | |
84 | char **argv; | |
85 | { | |
86 | int lowpct, highpct; | |
87 | ||
88 | /* | |
89 | * Use highpct and lowpc as percentages, temporarily | |
90 | * for graphing options involving blow-up | |
91 | */ | |
92 | lowpct = -1; | |
93 | highpct = -1; | |
94 | argv++; | |
95 | while ( *argv != 0 && **argv == '-' ) { | |
96 | *argv += 1; | |
97 | if (**argv == 'l') | |
98 | lflg++; | |
99 | else if (**argv == 'a') | |
100 | aflg++; | |
101 | else if (**argv == 'n') | |
102 | nflg++; | |
103 | else if (**argv == 'z') | |
104 | zflg++; | |
105 | else if (**argv == 'v') | |
106 | vflg++; | |
107 | else if ( **argv == 's' ) | |
108 | sflag++; | |
109 | else if (**argv >= '0' && **argv <= '9') { | |
110 | int i = atoi(*argv); | |
111 | if (lowpct == -1) | |
112 | lowpct = i; | |
113 | else | |
114 | highpct = i; | |
115 | } | |
116 | argv++; | |
117 | } | |
118 | if ( *argv != 0 ) { | |
119 | namfil = *argv; | |
120 | argv++; | |
121 | } else { | |
122 | namfil = A_OUTNAME; | |
123 | } | |
124 | if (lowpct >= 100) | |
125 | lowpct = 0; | |
126 | if (highpct <= lowpct || highpct > 100) | |
127 | highpct = 100; | |
128 | ransca = 100./(highpct-lowpct); | |
129 | ranoff = 2040. + 40.8*lowpc*ransca; | |
130 | /* | |
131 | * get information about a.out file. | |
132 | */ | |
133 | getnfile(); | |
134 | /* | |
135 | * get information about mon.out file(s). | |
136 | */ | |
137 | if ( *argv == 0 ) { | |
138 | getpfile( MON_OUTNAME ); | |
139 | } else { | |
140 | do { | |
141 | getpfile( *argv ); | |
142 | argv++; | |
143 | } while ( *argv != 0 ); | |
144 | } | |
145 | asgnsamples(); /* assign samples to procedures */ | |
146 | #ifdef plot | |
147 | if (vflg) | |
148 | plotprof(); /* a plotted or ... */ | |
149 | else | |
150 | #endif | |
151 | printprof(); /* a printed profile */ | |
152 | if ( sflag != 0 ) { | |
153 | putprof(); | |
154 | } | |
155 | done(); | |
156 | } | |
157 | ||
158 | printprof() | |
159 | { | |
160 | double time, actime; | |
161 | ||
162 | actime = 0; | |
163 | printf(" %%time cumsecs #call ms/call name\n"); | |
164 | if (!lflg) | |
165 | qsort(nl, nname, sizeof(struct nl), timcmp); | |
166 | for (np = nl; np<npe-1; np++) { | |
167 | if (zflg == 0 && np->time == 0 && np->ncall == 0) | |
168 | continue; | |
169 | time = np->time/totime; | |
170 | actime += np->time; | |
171 | printf("%6.1f%9.2f", 100*time, actime/60); | |
172 | if (np->ncall != 0) | |
173 | printf("%7ld %8.2f", | |
174 | np->ncall, np->time/(np->ncall*.06)); | |
175 | else | |
176 | printf("%7.7s %8.8s", "", ""); | |
177 | printf(" %s\n", np->name); | |
178 | } | |
179 | } | |
180 | ||
181 | /* | |
182 | * Set up string and symbol tables from a.out. | |
183 | * On return symbol table is sorted by value. | |
184 | */ | |
185 | getnfile() | |
186 | { | |
187 | ||
188 | nfile = fopen(namfil,"r"); | |
189 | if (nfile == NULL) { | |
190 | perror(namfil); | |
191 | done(); | |
192 | } | |
193 | fread(&xbuf, 1, sizeof(xbuf), nfile); | |
194 | if (N_BADMAG(xbuf)) { | |
195 | fprintf(stderr, "%s: bad format\n", namfil); | |
196 | done(); | |
197 | } | |
198 | getstrtab(); | |
199 | getsymtab(); | |
200 | qsort(nl, nname, sizeof(struct nl), valcmp); | |
201 | } | |
202 | ||
203 | getstrtab() | |
204 | { | |
205 | ||
206 | fseek(nfile, N_SYMOFF(xbuf) + xbuf.a_syms, 0); | |
207 | if (fread(&ssiz, sizeof (ssiz), 1, nfile) == 0) { | |
208 | fprintf(stderr, "%s: no string table (old format?)\n", namfil); | |
209 | done(); | |
210 | } | |
211 | strtab = (char *)calloc(ssiz, 1); | |
212 | if (strtab == NULL) { | |
213 | fprintf(stderr, "%s: no room for %d bytes of string table", | |
214 | namfil, ssiz); | |
215 | done(); | |
216 | } | |
217 | if (fread(strtab+sizeof(ssiz), ssiz-sizeof(ssiz), 1, nfile) != 1) { | |
218 | fprintf(stderr, "%s: error reading string table\n", namfil); | |
219 | done(); | |
220 | } | |
221 | } | |
222 | ||
223 | /* | |
224 | * Read in symbol table | |
225 | */ | |
226 | getsymtab() | |
227 | { | |
228 | register int i; | |
229 | ||
230 | /* pass1 - count symbols */ | |
231 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
232 | nname = 0; | |
233 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
234 | struct nlist nbuf; | |
235 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
236 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
237 | continue; | |
238 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
239 | continue; | |
240 | nname++; | |
241 | } | |
242 | if (nname == 0) { | |
243 | fprintf(stderr, "%s: no symbols\n", namfil); | |
244 | done(); | |
245 | } | |
246 | nl = (struct nl *)calloc((nname+1), sizeof (struct nl)); | |
247 | if (nl == 0) { | |
248 | fprintf(stderr, "prof: No room for %d bytes of symbol table\n", | |
249 | (nname+1) * sizeof (struct nlist)); | |
250 | done(); | |
251 | } | |
252 | ||
253 | /* pass2 - read symbols */ | |
254 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
255 | npe = nl; | |
256 | nname = 0; | |
257 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
258 | struct nlist nbuf; | |
259 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
260 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
261 | continue; | |
262 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
263 | continue; | |
264 | npe->value = nbuf.n_value/sizeof(UNIT); | |
265 | npe->name = strtab+nbuf.n_un.n_strx; | |
266 | npe++; | |
267 | nname++; | |
268 | } | |
269 | npe->value = -1; | |
270 | npe++; | |
271 | } | |
272 | ||
273 | /* | |
274 | * information from a mon.out file is in two parts: | |
275 | * the counters of how many times each procedure was called, | |
276 | * if it was called at all; | |
277 | * and an array of sampling hits within pc ranges. | |
278 | * the counters must be dealt with on a file-by-file basis, | |
279 | * since which procedures are represented may vary. | |
280 | * the samples ranges are fixed, but must be summed across | |
281 | * files, and then distributed among procedures, because | |
282 | * of the wierd way the plotting is done. | |
283 | */ | |
284 | getpfile(filename) | |
285 | char *filename; | |
286 | { | |
287 | ||
288 | openpfile(filename); | |
289 | readcntrs(); | |
290 | asgncntrs(); /* assign counts to procedures */ | |
291 | readsamples(); | |
292 | closepfile(); | |
293 | } | |
294 | ||
295 | openpfile(filename) | |
296 | char *filename; | |
297 | { | |
298 | struct stat stb; | |
299 | ||
300 | if((pfile = fopen(filename, "r")) == NULL) { | |
301 | perror(filename); | |
302 | done(); | |
303 | } | |
304 | fstat(fileno(pfile), &stb); | |
305 | fread(&h, sizeof(struct hdr), 1, pfile); | |
306 | lowpc = h.lowpc - (UNIT *)0; | |
307 | highpc = h.highpc - (UNIT *)0; | |
308 | sampbytes = | |
309 | stb.st_size - sizeof(struct hdr) - h.ncount*sizeof(struct cnt); | |
310 | nsamples = sampbytes / sizeof (unsigned UNIT); | |
311 | } | |
312 | ||
313 | closepfile() | |
314 | { | |
315 | ||
316 | fclose(pfile); | |
317 | free(cbuf); | |
318 | } | |
319 | ||
320 | readcntrs() | |
321 | { | |
322 | struct cnt *kp; | |
323 | ||
324 | cbuf = (struct cnt *)calloc((h.ncount+1), sizeof (struct cnt)); | |
325 | if (cbuf == 0) { | |
326 | fprintf(stderr, "prof: No room for %d bytes of count buffer\n", | |
327 | (h.ncount+1) * sizeof (struct cnt)); | |
328 | exit(1); | |
329 | } | |
330 | fread(cbuf, sizeof(struct cnt), h.ncount, pfile); | |
331 | /* eliminate zero counters and scale counter pc values */ | |
332 | if (h.ncount) { | |
333 | kp = &cbuf[h.ncount - 1]; | |
334 | for (;;) { | |
335 | if (kp->cvalue==0) { | |
336 | h.ncount=kp-cbuf; | |
337 | ++kp; | |
338 | break; | |
339 | } | |
340 | if (kp == cbuf) { | |
341 | h.ncount = 0; | |
342 | break; | |
343 | } | |
344 | --kp; | |
345 | } | |
346 | for (; --kp>=cbuf; ) | |
347 | kp->cvalue /= sizeof(UNIT); | |
348 | } | |
349 | /* sort counters */ | |
350 | qsort(cbuf, h.ncount, sizeof(struct cnt), cntcmp); | |
351 | } | |
352 | ||
353 | /* | |
354 | * Assign counters to the procedures to which they belong | |
355 | */ | |
356 | asgncntrs() | |
357 | { | |
358 | register int i; | |
359 | struct cnt *kp; | |
360 | ||
361 | kp = &cbuf[h.ncount-1]; | |
362 | np = npe; | |
363 | while (--np>=nl) { | |
364 | if (kp<cbuf || np->value > kp->cvalue) | |
365 | continue; | |
366 | /* skip ``static'' functions */ | |
2d19cd70 | 367 | while (kp >= cbuf && kp->cvalue > np->value + PCFUDGE) |
53cf6719 SL |
368 | --kp; |
369 | if (kp->cvalue >= np->value) { | |
370 | np->ncall += kp->cncall; | |
371 | --kp; | |
372 | } | |
373 | } | |
374 | } | |
375 | ||
376 | readsamples() | |
377 | { | |
378 | register i; | |
379 | unsigned UNIT sample; | |
380 | int totalt; | |
381 | ||
382 | if (samples == 0) { | |
383 | samples = (unsigned UNIT *) | |
384 | calloc(sampbytes, sizeof (unsigned UNIT)); | |
385 | if (samples == 0) { | |
386 | printf("prof: No room for %d sample pc's\n", | |
387 | sampbytes / sizeof (unsigned UNIT)); | |
388 | done(); | |
389 | } | |
390 | } | |
391 | for (i = 0; ; i++) { | |
392 | fread(&sample, sizeof (unsigned UNIT), 1, pfile); | |
393 | if (feof(pfile)) | |
394 | break; | |
395 | samples[i] += sample; | |
396 | totalt += sample; | |
397 | } | |
398 | if (i != nsamples) { | |
399 | fprintf(stderr, | |
400 | "prof: unexpected EOF after reading %d/%d samples\n", | |
401 | --i, nsamples); | |
402 | done(); | |
403 | } | |
404 | } | |
405 | ||
406 | /* | |
407 | * Assign samples to the procedures to which they belong. | |
408 | */ | |
409 | asgnsamples() | |
410 | { | |
411 | register j; | |
412 | unsigned UNIT ccnt; | |
413 | double time; | |
414 | unsigned pcl, pch; | |
415 | register int i; | |
416 | int overlap; | |
417 | ||
418 | /* read samples and assign to namelist symbols */ | |
419 | scale = highpc - lowpc; | |
420 | scale /= nsamples; | |
421 | for (i=0; i < nsamples; i++) { | |
422 | ccnt = samples[i]; | |
423 | if (ccnt == 0) | |
424 | continue; | |
425 | pcl = lowpc + scale*i; | |
426 | pch = lowpc + scale*(i+1); | |
427 | time = ccnt; | |
428 | totime += time; | |
429 | if(time > maxtime) | |
430 | maxtime = time; | |
431 | for (j=0; j<nname; j++) { | |
432 | if (pch < nl[j].value) | |
433 | break; | |
434 | if (pcl >= nl[j+1].value) | |
435 | continue; | |
436 | overlap=(min(pch,nl[j+1].value)-max(pcl,nl[j].value)); | |
437 | if (overlap>0) | |
438 | nl[j].time += overlap*time/scale; | |
439 | } | |
440 | } | |
441 | if (totime==0.0) { | |
442 | fprintf(stderr, "No time accumulated\n"); | |
443 | /* | |
444 | done(); | |
445 | */ | |
446 | totime=1.0; | |
447 | } | |
448 | } | |
449 | ||
450 | /* | |
451 | * dump what you have out to a mon.out style file. | |
452 | */ | |
453 | putprof() | |
454 | { | |
455 | FILE *sfile; | |
456 | struct nl *np; | |
457 | struct cnt kp; | |
458 | int i; | |
459 | ||
460 | sfile = fopen(MON_SUMNAME, "w"); | |
461 | if (sfile == NULL) { | |
462 | perror(MON_SUMNAME); | |
463 | done(); | |
464 | } | |
465 | /* | |
466 | * build a new header. | |
467 | * h.lowpc and h.highpc are already fine. | |
468 | * fix h.ncount to count non-zero calls, | |
469 | * and the one zero call which marks the end. | |
470 | */ | |
471 | h.ncount = 0; | |
472 | for (np = nl; np < npe-1 ; np++) | |
473 | if (np->ncall > 0) | |
474 | h.ncount++; | |
475 | h.ncount++; | |
476 | fwrite(&h, sizeof (struct hdr), 1, sfile); | |
477 | for (np = nl; np < npe-1; np++) { | |
478 | if (np->ncall > 0) { | |
479 | kp.cvalue = np->value * sizeof (unsigned UNIT); | |
480 | kp.cncall = np->ncall; | |
481 | fwrite(&kp, sizeof (struct cnt), 1, sfile); | |
482 | } | |
483 | } | |
484 | kp.cvalue = 0; | |
485 | kp.cncall = 0; | |
486 | fwrite(&kp, sizeof (struct cnt), 1, sfile); | |
487 | fwrite(samples, sizeof (unsigned UNIT), nsamples, sfile); | |
488 | fclose(sfile); | |
489 | } | |
490 | ||
491 | min(a, b) | |
492 | { | |
493 | if (a<b) | |
494 | return(a); | |
495 | return(b); | |
496 | } | |
497 | ||
498 | max(a, b) | |
499 | { | |
500 | if (a>b) | |
501 | return(a); | |
502 | return(b); | |
503 | } | |
504 | ||
505 | valcmp(p1, p2) | |
506 | struct nl *p1, *p2; | |
507 | { | |
508 | ||
509 | return(p1->value - p2->value); | |
510 | } | |
511 | ||
512 | timcmp(p1, p2) | |
513 | struct nl *p1, *p2; | |
514 | { | |
515 | float d; | |
516 | ||
517 | if (nflg && p2->ncall != p1->ncall) | |
518 | return (p2->ncall - p1->ncall); | |
519 | d = p2->time - p1->time; | |
520 | if (d > 0.0) | |
521 | return(1); | |
522 | if (d < 0.0) | |
523 | return(-1); | |
524 | return(strcmp(p1->name,p2->name)); | |
525 | } | |
526 | ||
527 | cntcmp(p1, p2) | |
528 | struct cnt *p1, *p2; | |
529 | { | |
530 | ||
531 | return(p1->cvalue - p2->cvalue); | |
532 | } | |
533 | ||
534 | done() | |
535 | { | |
536 | ||
537 | #ifdef plot | |
538 | if(vflg) { | |
539 | point(0, -2040); | |
540 | closepl(); | |
541 | } | |
542 | #endif | |
543 | exit(0); | |
544 | } | |
545 | ||
546 | #ifdef plot | |
547 | plotprof() | |
548 | { | |
549 | double time, lastx, lasty, lastsx; | |
550 | register i; | |
551 | ||
552 | openpl(); | |
553 | erase(); | |
554 | space(-2048, -2048, 2048, 2048); | |
555 | line(-2040, -2040, -2040, 2040); | |
556 | line(0, 2040, 0, -2040); | |
557 | for(i=0; i<11; i++) | |
558 | line(-2040, 2040-i*408, 0, 2040-i*408); | |
559 | lastx = 0.; | |
560 | lasty = ranoff; | |
561 | scale = (4080.*ransca)/(sampbytes/sizeof(UNIT)); | |
562 | lastsx = 0.0; | |
563 | for(i = 0; i < nsamples; i++) { | |
564 | unsigned UNIT ccnt; | |
565 | double tx, ty; | |
566 | ccnt = samples[i]; | |
567 | time = ccnt; | |
568 | tx = lastsx; | |
569 | ty = lasty; | |
570 | lastsx -= 2000.*time/totime; | |
571 | lasty -= scale; | |
572 | if(lasty >= -2040. && ty <= 2040.) { | |
573 | line((int)tx, (int)ty, (int)lastsx, (int)lasty); | |
574 | if (ccnt!=0 || lastx!=0.0) { | |
575 | tx = lastx; | |
576 | lastx = -time*2000./maxtime; | |
577 | ty += scale/2; | |
578 | line(0, (int)ty, (int)tx, (int)ty); | |
579 | } | |
580 | } | |
581 | } | |
582 | scale = (4080.*ransca)/(highpc-lowpc); | |
583 | lastx = 50.; | |
584 | for(np = nl; np<npe; np++) { | |
585 | if(np->value < lowpc) | |
586 | continue; | |
587 | if(np->value >= highpc) | |
588 | continue; | |
d28c78f5 BJ |
589 | if(zflg == 0 && np->time == 0 && np->ncall == 0) |
590 | continue; | |
53cf6719 SL |
591 | time = np->time/totime; |
592 | lasty = ranoff - (np->value - lowpc)*scale; | |
593 | if(lasty >= -2040. && lasty <= 2040.) { | |
594 | char bufl[BUFSIZ], *namp; | |
595 | register j; | |
596 | line(0, (int)lasty, 50, (int)lasty); | |
597 | line((int)(lastx-50),(int)lasty,(int)lastx,(int)lasty); | |
598 | move((int)(lastx+30), (int)(lasty+10)); | |
599 | sprintf(bufl, "%s", np->name + (np->name[0] == '_')); | |
600 | label(bufl); | |
601 | } | |
602 | lastx += 500.; | |
603 | if(lastx > 2000.) | |
604 | lastx = 50.; | |
605 | } | |
606 | } | |
607 | #endif |