Commit | Line | Data |
---|---|---|
53cf6719 SL |
1 | #ifndef lint |
2 | static char *sccsid = "@(#)prof.c 4.1 (Berkeley) %G%"; | |
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 */ | |
13 | #define A_OUTNAME "a.out" | |
14 | #define MON_OUTNAME "mon.out" | |
15 | #define MON_SUMNAME "mon.sum" | |
16 | ||
17 | /* | |
18 | * The symbol table; | |
19 | * for each external in the specified file we gather | |
20 | * its address, the number of calls and compute its share of cpu time. | |
21 | */ | |
22 | struct nl { | |
23 | char *name; | |
24 | unsigned value; | |
25 | float time; | |
26 | long ncall; | |
27 | } *nl; | |
28 | int nname; | |
29 | struct nl *np; | |
30 | struct nl *npe; | |
31 | ||
32 | /* | |
33 | * The header on the mon.out file. | |
34 | * Mon.out consists of one of these headers, an array of ncount | |
35 | * cnt structures (as below) and then an array of samples | |
36 | * representing the discretized program counter values. | |
37 | */ | |
38 | struct hdr { | |
39 | UNIT *lowpc, *highpc; | |
40 | int ncount; | |
41 | } h; | |
42 | ||
43 | /* | |
44 | * Each counter has an address and a number of calls. | |
45 | */ | |
46 | struct cnt { | |
47 | unsigned cvalue; | |
48 | long cncall; | |
49 | } *cbuf; | |
50 | ||
51 | /* | |
52 | * Each discretized pc sample has | |
53 | * a count of the number of samples in its range | |
54 | */ | |
55 | unsigned UNIT *samples; | |
56 | ||
57 | FILE *pfile, *nfile; | |
58 | ||
59 | unsigned lowpc, highpc; /* range profiled */ | |
60 | double ransca, ranoff; /* scaling for blowing up plots */ | |
61 | unsigned sampbytes; /* number of bytes of samples */ | |
62 | int nsamples; /* number of samples */ | |
63 | double totime; /* total time for all routines */ | |
64 | double maxtime; /* maximum time of any routine (for plot) */ | |
65 | double scale; /* scale factor converting samples to pc | |
66 | values: each sample covers scale bytes */ | |
67 | char *strtab; /* string table in core */ | |
68 | off_t ssiz; /* size of the string table */ | |
69 | struct exec xbuf; /* exec header of a.out */ | |
70 | ||
71 | int aflg; | |
72 | int nflg; | |
73 | int vflg; | |
74 | int lflg; | |
75 | int zflg; | |
76 | int sflag; | |
77 | ||
78 | char *namfil; | |
79 | ||
80 | int timcmp(), valcmp(), cntcmp(); | |
81 | ||
82 | main(argc, argv) | |
83 | char **argv; | |
84 | { | |
85 | int lowpct, highpct; | |
86 | ||
87 | /* | |
88 | * Use highpct and lowpc as percentages, temporarily | |
89 | * for graphing options involving blow-up | |
90 | */ | |
91 | lowpct = -1; | |
92 | highpct = -1; | |
93 | argv++; | |
94 | while ( *argv != 0 && **argv == '-' ) { | |
95 | *argv += 1; | |
96 | if (**argv == 'l') | |
97 | lflg++; | |
98 | else if (**argv == 'a') | |
99 | aflg++; | |
100 | else if (**argv == 'n') | |
101 | nflg++; | |
102 | else if (**argv == 'z') | |
103 | zflg++; | |
104 | else if (**argv == 'v') | |
105 | vflg++; | |
106 | else if ( **argv == 's' ) | |
107 | sflag++; | |
108 | else if (**argv >= '0' && **argv <= '9') { | |
109 | int i = atoi(*argv); | |
110 | if (lowpct == -1) | |
111 | lowpct = i; | |
112 | else | |
113 | highpct = i; | |
114 | } | |
115 | argv++; | |
116 | } | |
117 | if ( *argv != 0 ) { | |
118 | namfil = *argv; | |
119 | argv++; | |
120 | } else { | |
121 | namfil = A_OUTNAME; | |
122 | } | |
123 | if (lowpct >= 100) | |
124 | lowpct = 0; | |
125 | if (highpct <= lowpct || highpct > 100) | |
126 | highpct = 100; | |
127 | ransca = 100./(highpct-lowpct); | |
128 | ranoff = 2040. + 40.8*lowpc*ransca; | |
129 | /* | |
130 | * get information about a.out file. | |
131 | */ | |
132 | getnfile(); | |
133 | /* | |
134 | * get information about mon.out file(s). | |
135 | */ | |
136 | if ( *argv == 0 ) { | |
137 | getpfile( MON_OUTNAME ); | |
138 | } else { | |
139 | do { | |
140 | getpfile( *argv ); | |
141 | argv++; | |
142 | } while ( *argv != 0 ); | |
143 | } | |
144 | asgnsamples(); /* assign samples to procedures */ | |
145 | #ifdef plot | |
146 | if (vflg) | |
147 | plotprof(); /* a plotted or ... */ | |
148 | else | |
149 | #endif | |
150 | printprof(); /* a printed profile */ | |
151 | if ( sflag != 0 ) { | |
152 | putprof(); | |
153 | } | |
154 | done(); | |
155 | } | |
156 | ||
157 | printprof() | |
158 | { | |
159 | double time, actime; | |
160 | ||
161 | actime = 0; | |
162 | printf(" %%time cumsecs #call ms/call name\n"); | |
163 | if (!lflg) | |
164 | qsort(nl, nname, sizeof(struct nl), timcmp); | |
165 | for (np = nl; np<npe-1; np++) { | |
166 | if (zflg == 0 && np->time == 0 && np->ncall == 0) | |
167 | continue; | |
168 | time = np->time/totime; | |
169 | actime += np->time; | |
170 | printf("%6.1f%9.2f", 100*time, actime/60); | |
171 | if (np->ncall != 0) | |
172 | printf("%7ld %8.2f", | |
173 | np->ncall, np->time/(np->ncall*.06)); | |
174 | else | |
175 | printf("%7.7s %8.8s", "", ""); | |
176 | printf(" %s\n", np->name); | |
177 | } | |
178 | } | |
179 | ||
180 | /* | |
181 | * Set up string and symbol tables from a.out. | |
182 | * On return symbol table is sorted by value. | |
183 | */ | |
184 | getnfile() | |
185 | { | |
186 | ||
187 | nfile = fopen(namfil,"r"); | |
188 | if (nfile == NULL) { | |
189 | perror(namfil); | |
190 | done(); | |
191 | } | |
192 | fread(&xbuf, 1, sizeof(xbuf), nfile); | |
193 | if (N_BADMAG(xbuf)) { | |
194 | fprintf(stderr, "%s: bad format\n", namfil); | |
195 | done(); | |
196 | } | |
197 | getstrtab(); | |
198 | getsymtab(); | |
199 | qsort(nl, nname, sizeof(struct nl), valcmp); | |
200 | } | |
201 | ||
202 | getstrtab() | |
203 | { | |
204 | ||
205 | fseek(nfile, N_SYMOFF(xbuf) + xbuf.a_syms, 0); | |
206 | if (fread(&ssiz, sizeof (ssiz), 1, nfile) == 0) { | |
207 | fprintf(stderr, "%s: no string table (old format?)\n", namfil); | |
208 | done(); | |
209 | } | |
210 | strtab = (char *)calloc(ssiz, 1); | |
211 | if (strtab == NULL) { | |
212 | fprintf(stderr, "%s: no room for %d bytes of string table", | |
213 | namfil, ssiz); | |
214 | done(); | |
215 | } | |
216 | if (fread(strtab+sizeof(ssiz), ssiz-sizeof(ssiz), 1, nfile) != 1) { | |
217 | fprintf(stderr, "%s: error reading string table\n", namfil); | |
218 | done(); | |
219 | } | |
220 | } | |
221 | ||
222 | /* | |
223 | * Read in symbol table | |
224 | */ | |
225 | getsymtab() | |
226 | { | |
227 | register int i; | |
228 | ||
229 | /* pass1 - count symbols */ | |
230 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
231 | nname = 0; | |
232 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
233 | struct nlist nbuf; | |
234 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
235 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
236 | continue; | |
237 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
238 | continue; | |
239 | nname++; | |
240 | } | |
241 | if (nname == 0) { | |
242 | fprintf(stderr, "%s: no symbols\n", namfil); | |
243 | done(); | |
244 | } | |
245 | nl = (struct nl *)calloc((nname+1), sizeof (struct nl)); | |
246 | if (nl == 0) { | |
247 | fprintf(stderr, "prof: No room for %d bytes of symbol table\n", | |
248 | (nname+1) * sizeof (struct nlist)); | |
249 | done(); | |
250 | } | |
251 | ||
252 | /* pass2 - read symbols */ | |
253 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
254 | npe = nl; | |
255 | nname = 0; | |
256 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
257 | struct nlist nbuf; | |
258 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
259 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
260 | continue; | |
261 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
262 | continue; | |
263 | npe->value = nbuf.n_value/sizeof(UNIT); | |
264 | npe->name = strtab+nbuf.n_un.n_strx; | |
265 | npe++; | |
266 | nname++; | |
267 | } | |
268 | npe->value = -1; | |
269 | npe++; | |
270 | } | |
271 | ||
272 | /* | |
273 | * information from a mon.out file is in two parts: | |
274 | * the counters of how many times each procedure was called, | |
275 | * if it was called at all; | |
276 | * and an array of sampling hits within pc ranges. | |
277 | * the counters must be dealt with on a file-by-file basis, | |
278 | * since which procedures are represented may vary. | |
279 | * the samples ranges are fixed, but must be summed across | |
280 | * files, and then distributed among procedures, because | |
281 | * of the wierd way the plotting is done. | |
282 | */ | |
283 | getpfile(filename) | |
284 | char *filename; | |
285 | { | |
286 | ||
287 | openpfile(filename); | |
288 | readcntrs(); | |
289 | asgncntrs(); /* assign counts to procedures */ | |
290 | readsamples(); | |
291 | closepfile(); | |
292 | } | |
293 | ||
294 | openpfile(filename) | |
295 | char *filename; | |
296 | { | |
297 | struct stat stb; | |
298 | ||
299 | if((pfile = fopen(filename, "r")) == NULL) { | |
300 | perror(filename); | |
301 | done(); | |
302 | } | |
303 | fstat(fileno(pfile), &stb); | |
304 | fread(&h, sizeof(struct hdr), 1, pfile); | |
305 | lowpc = h.lowpc - (UNIT *)0; | |
306 | highpc = h.highpc - (UNIT *)0; | |
307 | sampbytes = | |
308 | stb.st_size - sizeof(struct hdr) - h.ncount*sizeof(struct cnt); | |
309 | nsamples = sampbytes / sizeof (unsigned UNIT); | |
310 | } | |
311 | ||
312 | closepfile() | |
313 | { | |
314 | ||
315 | fclose(pfile); | |
316 | free(cbuf); | |
317 | } | |
318 | ||
319 | readcntrs() | |
320 | { | |
321 | struct cnt *kp; | |
322 | ||
323 | cbuf = (struct cnt *)calloc((h.ncount+1), sizeof (struct cnt)); | |
324 | if (cbuf == 0) { | |
325 | fprintf(stderr, "prof: No room for %d bytes of count buffer\n", | |
326 | (h.ncount+1) * sizeof (struct cnt)); | |
327 | exit(1); | |
328 | } | |
329 | fread(cbuf, sizeof(struct cnt), h.ncount, pfile); | |
330 | /* eliminate zero counters and scale counter pc values */ | |
331 | if (h.ncount) { | |
332 | kp = &cbuf[h.ncount - 1]; | |
333 | for (;;) { | |
334 | if (kp->cvalue==0) { | |
335 | h.ncount=kp-cbuf; | |
336 | ++kp; | |
337 | break; | |
338 | } | |
339 | if (kp == cbuf) { | |
340 | h.ncount = 0; | |
341 | break; | |
342 | } | |
343 | --kp; | |
344 | } | |
345 | for (; --kp>=cbuf; ) | |
346 | kp->cvalue /= sizeof(UNIT); | |
347 | } | |
348 | /* sort counters */ | |
349 | qsort(cbuf, h.ncount, sizeof(struct cnt), cntcmp); | |
350 | } | |
351 | ||
352 | /* | |
353 | * Assign counters to the procedures to which they belong | |
354 | */ | |
355 | asgncntrs() | |
356 | { | |
357 | register int i; | |
358 | struct cnt *kp; | |
359 | ||
360 | kp = &cbuf[h.ncount-1]; | |
361 | np = npe; | |
362 | while (--np>=nl) { | |
363 | if (kp<cbuf || np->value > kp->cvalue) | |
364 | continue; | |
365 | /* skip ``static'' functions */ | |
366 | while (kp >= cbuf && kp->cvalue > np->value + 11) | |
367 | --kp; | |
368 | if (kp->cvalue >= np->value) { | |
369 | np->ncall += kp->cncall; | |
370 | --kp; | |
371 | } | |
372 | } | |
373 | } | |
374 | ||
375 | readsamples() | |
376 | { | |
377 | register i; | |
378 | unsigned UNIT sample; | |
379 | int totalt; | |
380 | ||
381 | if (samples == 0) { | |
382 | samples = (unsigned UNIT *) | |
383 | calloc(sampbytes, sizeof (unsigned UNIT)); | |
384 | if (samples == 0) { | |
385 | printf("prof: No room for %d sample pc's\n", | |
386 | sampbytes / sizeof (unsigned UNIT)); | |
387 | done(); | |
388 | } | |
389 | } | |
390 | for (i = 0; ; i++) { | |
391 | fread(&sample, sizeof (unsigned UNIT), 1, pfile); | |
392 | if (feof(pfile)) | |
393 | break; | |
394 | samples[i] += sample; | |
395 | totalt += sample; | |
396 | } | |
397 | if (i != nsamples) { | |
398 | fprintf(stderr, | |
399 | "prof: unexpected EOF after reading %d/%d samples\n", | |
400 | --i, nsamples); | |
401 | done(); | |
402 | } | |
403 | } | |
404 | ||
405 | /* | |
406 | * Assign samples to the procedures to which they belong. | |
407 | */ | |
408 | asgnsamples() | |
409 | { | |
410 | register j; | |
411 | unsigned UNIT ccnt; | |
412 | double time; | |
413 | unsigned pcl, pch; | |
414 | register int i; | |
415 | int overlap; | |
416 | ||
417 | /* read samples and assign to namelist symbols */ | |
418 | scale = highpc - lowpc; | |
419 | scale /= nsamples; | |
420 | for (i=0; i < nsamples; i++) { | |
421 | ccnt = samples[i]; | |
422 | if (ccnt == 0) | |
423 | continue; | |
424 | pcl = lowpc + scale*i; | |
425 | pch = lowpc + scale*(i+1); | |
426 | time = ccnt; | |
427 | totime += time; | |
428 | if(time > maxtime) | |
429 | maxtime = time; | |
430 | for (j=0; j<nname; j++) { | |
431 | if (pch < nl[j].value) | |
432 | break; | |
433 | if (pcl >= nl[j+1].value) | |
434 | continue; | |
435 | overlap=(min(pch,nl[j+1].value)-max(pcl,nl[j].value)); | |
436 | if (overlap>0) | |
437 | nl[j].time += overlap*time/scale; | |
438 | } | |
439 | } | |
440 | if (totime==0.0) { | |
441 | fprintf(stderr, "No time accumulated\n"); | |
442 | /* | |
443 | done(); | |
444 | */ | |
445 | totime=1.0; | |
446 | } | |
447 | } | |
448 | ||
449 | /* | |
450 | * dump what you have out to a mon.out style file. | |
451 | */ | |
452 | putprof() | |
453 | { | |
454 | FILE *sfile; | |
455 | struct nl *np; | |
456 | struct cnt kp; | |
457 | int i; | |
458 | ||
459 | sfile = fopen(MON_SUMNAME, "w"); | |
460 | if (sfile == NULL) { | |
461 | perror(MON_SUMNAME); | |
462 | done(); | |
463 | } | |
464 | /* | |
465 | * build a new header. | |
466 | * h.lowpc and h.highpc are already fine. | |
467 | * fix h.ncount to count non-zero calls, | |
468 | * and the one zero call which marks the end. | |
469 | */ | |
470 | h.ncount = 0; | |
471 | for (np = nl; np < npe-1 ; np++) | |
472 | if (np->ncall > 0) | |
473 | h.ncount++; | |
474 | h.ncount++; | |
475 | fwrite(&h, sizeof (struct hdr), 1, sfile); | |
476 | for (np = nl; np < npe-1; np++) { | |
477 | if (np->ncall > 0) { | |
478 | kp.cvalue = np->value * sizeof (unsigned UNIT); | |
479 | kp.cncall = np->ncall; | |
480 | fwrite(&kp, sizeof (struct cnt), 1, sfile); | |
481 | } | |
482 | } | |
483 | kp.cvalue = 0; | |
484 | kp.cncall = 0; | |
485 | fwrite(&kp, sizeof (struct cnt), 1, sfile); | |
486 | fwrite(samples, sizeof (unsigned UNIT), nsamples, sfile); | |
487 | fclose(sfile); | |
488 | } | |
489 | ||
490 | min(a, b) | |
491 | { | |
492 | if (a<b) | |
493 | return(a); | |
494 | return(b); | |
495 | } | |
496 | ||
497 | max(a, b) | |
498 | { | |
499 | if (a>b) | |
500 | return(a); | |
501 | return(b); | |
502 | } | |
503 | ||
504 | valcmp(p1, p2) | |
505 | struct nl *p1, *p2; | |
506 | { | |
507 | ||
508 | return(p1->value - p2->value); | |
509 | } | |
510 | ||
511 | timcmp(p1, p2) | |
512 | struct nl *p1, *p2; | |
513 | { | |
514 | float d; | |
515 | ||
516 | if (nflg && p2->ncall != p1->ncall) | |
517 | return (p2->ncall - p1->ncall); | |
518 | d = p2->time - p1->time; | |
519 | if (d > 0.0) | |
520 | return(1); | |
521 | if (d < 0.0) | |
522 | return(-1); | |
523 | return(strcmp(p1->name,p2->name)); | |
524 | } | |
525 | ||
526 | cntcmp(p1, p2) | |
527 | struct cnt *p1, *p2; | |
528 | { | |
529 | ||
530 | return(p1->cvalue - p2->cvalue); | |
531 | } | |
532 | ||
533 | done() | |
534 | { | |
535 | ||
536 | #ifdef plot | |
537 | if(vflg) { | |
538 | point(0, -2040); | |
539 | closepl(); | |
540 | } | |
541 | #endif | |
542 | exit(0); | |
543 | } | |
544 | ||
545 | #ifdef plot | |
546 | plotprof() | |
547 | { | |
548 | double time, lastx, lasty, lastsx; | |
549 | register i; | |
550 | ||
551 | openpl(); | |
552 | erase(); | |
553 | space(-2048, -2048, 2048, 2048); | |
554 | line(-2040, -2040, -2040, 2040); | |
555 | line(0, 2040, 0, -2040); | |
556 | for(i=0; i<11; i++) | |
557 | line(-2040, 2040-i*408, 0, 2040-i*408); | |
558 | lastx = 0.; | |
559 | lasty = ranoff; | |
560 | scale = (4080.*ransca)/(sampbytes/sizeof(UNIT)); | |
561 | lastsx = 0.0; | |
562 | for(i = 0; i < nsamples; i++) { | |
563 | unsigned UNIT ccnt; | |
564 | double tx, ty; | |
565 | ccnt = samples[i]; | |
566 | time = ccnt; | |
567 | tx = lastsx; | |
568 | ty = lasty; | |
569 | lastsx -= 2000.*time/totime; | |
570 | lasty -= scale; | |
571 | if(lasty >= -2040. && ty <= 2040.) { | |
572 | line((int)tx, (int)ty, (int)lastsx, (int)lasty); | |
573 | if (ccnt!=0 || lastx!=0.0) { | |
574 | tx = lastx; | |
575 | lastx = -time*2000./maxtime; | |
576 | ty += scale/2; | |
577 | line(0, (int)ty, (int)tx, (int)ty); | |
578 | } | |
579 | } | |
580 | } | |
581 | scale = (4080.*ransca)/(highpc-lowpc); | |
582 | lastx = 50.; | |
583 | for(np = nl; np<npe; np++) { | |
584 | if(np->value < lowpc) | |
585 | continue; | |
586 | if(np->value >= highpc) | |
587 | continue; | |
588 | time = np->time/totime; | |
589 | lasty = ranoff - (np->value - lowpc)*scale; | |
590 | if(lasty >= -2040. && lasty <= 2040.) { | |
591 | char bufl[BUFSIZ], *namp; | |
592 | register j; | |
593 | line(0, (int)lasty, 50, (int)lasty); | |
594 | line((int)(lastx-50),(int)lasty,(int)lastx,(int)lasty); | |
595 | move((int)(lastx+30), (int)(lasty+10)); | |
596 | sprintf(bufl, "%s", np->name + (np->name[0] == '_')); | |
597 | label(bufl); | |
598 | } | |
599 | lastx += 500.; | |
600 | if(lastx > 2000.) | |
601 | lastx = 50.; | |
602 | } | |
603 | } | |
604 | #endif |