Commit | Line | Data |
---|---|---|
86b66dcd BJ |
1 | /* |
2 | * prof | |
3 | */ | |
4 | #include <stdio.h> | |
5 | #include <sys/types.h> | |
6 | #include <sys/stat.h> | |
7 | #include <a.out.h> | |
8 | #include <pagsiz.h> | |
9 | ||
10 | typedef short UNIT; /* unit of profiling */ | |
11 | #define A_OUTNAME "a.out" | |
12 | #define MON_OUTNAME "mon.out" | |
13 | #define MON_SUMNAME "mon.sum" | |
14 | ||
15 | /* | |
16 | * The symbol table; | |
17 | * for each external in the specified file we gather | |
18 | * its address, the number of calls and compute its share of cpu time. | |
19 | */ | |
20 | struct nl { | |
21 | char *name; | |
22 | unsigned value; | |
23 | float time; | |
24 | long ncall; | |
25 | } *nl; | |
26 | int nname; | |
27 | struct nl *np; | |
28 | struct nl *npe; | |
29 | ||
30 | /* | |
31 | * The header on the mon.out file. | |
32 | * Mon.out consists of one of these headers, an array of ncount | |
33 | * cnt structures (as below) and then an array of samples | |
34 | * representing the discretized program counter values. | |
35 | */ | |
36 | struct hdr { | |
37 | UNIT *lowpc, *highpc; | |
38 | int ncount; | |
39 | } h; | |
40 | ||
41 | /* | |
42 | * Each counter has an address and a number of calls. | |
43 | */ | |
44 | struct cnt { | |
45 | unsigned cvalue; | |
46 | long cncall; | |
47 | } *cbuf; | |
48 | ||
49 | /* | |
50 | * Each discretized pc sample has | |
51 | * a count of the number of samples in its range | |
52 | */ | |
53 | unsigned UNIT *samples; | |
54 | ||
55 | FILE *pfile, *nfile; | |
56 | ||
57 | unsigned lowpc, highpc; /* range profiled */ | |
58 | double ransca, ranoff; /* scaling for blowing up plots */ | |
59 | unsigned sampbytes; /* number of bytes of samples */ | |
60 | int nsamples; /* number of samples */ | |
61 | double totime; /* total time for all routines */ | |
62 | double maxtime; /* maximum time of any routine (for plot) */ | |
63 | double scale; /* scale factor converting samples to pc | |
64 | values: each sample covers scale bytes */ | |
65 | char *strtab; /* string table in core */ | |
66 | off_t ssiz; /* size of the string table */ | |
67 | struct exec xbuf; /* exec header of a.out */ | |
68 | ||
69 | int aflg; | |
70 | int nflg; | |
71 | int vflg; | |
72 | int lflg; | |
73 | int zflg; | |
74 | int sflag; | |
75 | ||
76 | char *namfil; | |
77 | ||
78 | int timcmp(), valcmp(), cntcmp(); | |
79 | ||
80 | main(argc, argv) | |
81 | char **argv; | |
82 | { | |
83 | int lowpct, highpct; | |
84 | ||
85 | /* | |
86 | * Use highpct and lowpc as percentages, temporarily | |
87 | * for graphing options involving blow-up | |
88 | */ | |
89 | lowpct = -1; | |
90 | highpct = -1; | |
91 | argv++; | |
92 | while ( *argv != 0 && **argv == '-' ) { | |
93 | *argv += 1; | |
94 | if (**argv == 'l') | |
95 | lflg++; | |
96 | else if (**argv == 'a') | |
97 | aflg++; | |
98 | else if (**argv == 'n') | |
99 | nflg++; | |
100 | else if (**argv == 'z') | |
101 | zflg++; | |
102 | else if (**argv == 'v') | |
103 | vflg++; | |
104 | else if ( **argv == 's' ) | |
105 | sflag++; | |
106 | else if (**argv >= '0' && **argv <= '9') { | |
107 | int i = atoi(*argv); | |
108 | if (lowpct == -1) | |
109 | lowpct = i; | |
110 | else | |
111 | highpct = i; | |
112 | } | |
113 | argv++; | |
114 | } | |
115 | if ( *argv != 0 ) { | |
116 | namfil = *argv; | |
117 | argv++; | |
118 | } else { | |
119 | namfil = A_OUTNAME; | |
120 | } | |
121 | if (lowpct >= 100) | |
122 | lowpct = 0; | |
123 | if (highpct <= lowpct || highpct > 100) | |
124 | highpct = 100; | |
125 | ransca = 100./(highpct-lowpct); | |
126 | ranoff = 2040. + 40.8*lowpc*ransca; | |
127 | /* | |
128 | * get information about a.out file. | |
129 | */ | |
130 | getnfile(); | |
131 | /* | |
132 | * get information about mon.out file(s). | |
133 | */ | |
134 | if ( *argv == 0 ) { | |
135 | getpfile( MON_OUTNAME ); | |
136 | } else { | |
137 | do { | |
138 | getpfile( *argv ); | |
139 | argv++; | |
140 | } while ( *argv != 0 ); | |
141 | } | |
142 | asgnsamples(); /* assign samples to procedures */ | |
143 | #ifdef plot | |
144 | if (vflag) | |
145 | plotprof(); /* a plotted or ... */ | |
146 | else | |
147 | #else | |
148 | printprof(); /* a printed profile */ | |
149 | #endif | |
150 | if ( sflag != 0 ) { | |
151 | putprof(); | |
152 | } | |
153 | done(); | |
154 | } | |
155 | ||
156 | printprof() | |
157 | { | |
158 | double time, actime; | |
159 | ||
160 | actime = 0; | |
161 | printf(" %%time cumsecs #call ms/call name\n"); | |
162 | if (!lflg) | |
163 | qsort(nl, nname, sizeof(struct nl), timcmp); | |
164 | for (np = nl; np<npe-1; np++) { | |
165 | if (zflg == 0 && np->time == 0 && np->ncall == 0) | |
166 | continue; | |
167 | time = np->time/totime; | |
168 | actime += np->time; | |
169 | printf("%6.1f%9.2f", 100*time, actime/60); | |
170 | if (np->ncall != 0) | |
171 | printf("%7ld %8.2f", | |
172 | np->ncall, np->time/(np->ncall*.06)); | |
173 | else | |
174 | printf("%7.7s %8.8s", "", ""); | |
175 | printf(" %s\n", np->name); | |
176 | } | |
177 | } | |
178 | ||
179 | /* | |
180 | * Set up string and symbol tables from a.out. | |
181 | * On return symbol table is sorted by value. | |
182 | */ | |
183 | getnfile() | |
184 | { | |
185 | ||
186 | nfile = fopen(namfil,"r"); | |
187 | if (nfile == NULL) { | |
188 | perror(namfil); | |
189 | done(); | |
190 | } | |
191 | fread(&xbuf, 1, sizeof(xbuf), nfile); | |
192 | if (N_BADMAG(xbuf)) { | |
193 | fprintf(stderr, "%s: bad format\n", namfil); | |
194 | done(); | |
195 | } | |
196 | getstrtab(); | |
197 | getsymtab(); | |
198 | qsort(nl, nname, sizeof(struct nl), valcmp); | |
199 | } | |
200 | ||
201 | getstrtab() | |
202 | { | |
203 | ||
204 | fseek(nfile, N_SYMOFF(xbuf) + xbuf.a_syms, 0); | |
205 | if (fread(&ssiz, sizeof (ssiz), 1, nfile) == 0) { | |
206 | fprintf(stderr, "%s: no string table (old format?)\n", namfil); | |
207 | done(); | |
208 | } | |
209 | strtab = (char *)calloc(ssiz, 1); | |
210 | if (strtab == NULL) { | |
211 | fprintf(stderr, "%s: no room for %d bytes of string table", | |
212 | namfil, ssiz); | |
213 | done(); | |
214 | } | |
215 | if (fread(strtab+sizeof(ssiz), ssiz-sizeof(ssiz), 1, nfile) != 1) { | |
216 | fprintf(stderr, "%s: error reading string table\n", namfil); | |
217 | done(); | |
218 | } | |
219 | } | |
220 | ||
221 | /* | |
222 | * Read in symbol table | |
223 | */ | |
224 | getsymtab() | |
225 | { | |
226 | register int i; | |
227 | ||
228 | /* pass1 - count symbols */ | |
229 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
230 | nname = 0; | |
231 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
232 | struct nlist nbuf; | |
233 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
234 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
235 | continue; | |
236 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
237 | continue; | |
238 | nname++; | |
239 | } | |
240 | if (nname == 0) { | |
241 | fprintf(stderr, "%s: no symbols\n", namfil); | |
242 | done(); | |
243 | } | |
244 | nl = (struct nl *)calloc((nname+1), sizeof (struct nl)); | |
245 | if (nl == 0) { | |
246 | fprintf(stderr, "prof: No room for %d bytes of symbol table\n", | |
247 | (nname+1) * sizeof (struct nlist)); | |
248 | done(); | |
249 | } | |
250 | ||
251 | /* pass2 - read symbols */ | |
252 | fseek(nfile, N_SYMOFF(xbuf), 0); | |
253 | npe = nl; | |
254 | nname = 0; | |
255 | for (i = xbuf.a_syms; i > 0; i -= sizeof(struct nlist)) { | |
256 | struct nlist nbuf; | |
257 | fread(&nbuf, sizeof(nbuf), 1, nfile); | |
258 | if (nbuf.n_type!=N_TEXT && nbuf.n_type!=N_TEXT+N_EXT) | |
259 | continue; | |
260 | if (aflg==0 && nbuf.n_type!=N_TEXT+N_EXT) | |
261 | continue; | |
262 | npe->value = nbuf.n_value/sizeof(UNIT); | |
263 | npe->name = strtab+nbuf.n_un.n_strx; | |
264 | npe++; | |
265 | nname++; | |
266 | } | |
267 | npe->value = -1; | |
268 | npe++; | |
269 | } | |
270 | ||
271 | /* | |
272 | * information from a mon.out file is in two parts: | |
273 | * the counters of how many times each procedure was called, | |
274 | * if it was called at all; | |
275 | * and an array of sampling hits within pc ranges. | |
276 | * the counters must be dealt with on a file-by-file basis, | |
277 | * since which procedures are represented may vary. | |
278 | * the samples ranges are fixed, but must be summed across | |
279 | * files, and then distributed amoung procedures, because | |
280 | * of the wierd way the plotting is done. | |
281 | */ | |
282 | getpfile( filename ) | |
283 | char *filename; | |
284 | { | |
285 | openpfile( filename ); | |
286 | readcntrs(); | |
287 | asgncntrs(); /* assign counts to procedures */ | |
288 | readsamples(); | |
289 | closepfile(); | |
290 | } | |
291 | ||
292 | openpfile( filename ) | |
293 | char *filename; | |
294 | { | |
295 | struct stat stb; | |
296 | ||
297 | if((pfile = fopen( filename ,"r")) == NULL) { | |
298 | perror( filename ); | |
299 | done(); | |
300 | } | |
301 | fstat(fileno(pfile), &stb); | |
302 | fread(&h, sizeof(struct hdr), 1, pfile); | |
303 | lowpc = h.lowpc - (UNIT *)0; | |
304 | highpc = h.highpc - (UNIT *)0; | |
305 | sampbytes = | |
306 | stb.st_size - sizeof(struct hdr) - h.ncount*sizeof(struct cnt); | |
307 | nsamples = sampbytes / sizeof (unsigned UNIT); | |
308 | } | |
309 | ||
310 | closepfile() | |
311 | { | |
312 | ||
313 | fclose( pfile ); | |
314 | free( cbuf ); | |
315 | } | |
316 | ||
317 | /* | |
318 | * read in procedure call counters | |
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 | 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 | /* | |
376 | * read pc samples | |
377 | */ | |
378 | readsamples() | |
379 | { | |
380 | register i; | |
381 | unsigned UNIT sample; | |
382 | ||
383 | if ( samples == 0 ) { | |
384 | samples = (unsigned UNIT *) | |
385 | calloc( sampbytes , sizeof (unsigned UNIT) ); | |
386 | if ( samples == 0 ) { | |
387 | printf( "prof: No room for %d sample pc's\n" , | |
388 | sampbytes / sizeof (unsigned UNIT) ); | |
389 | done(); | |
390 | } | |
391 | } | |
392 | for ( i = 0 ; ; i++ ) { | |
393 | fread( &sample , sizeof (unsigned UNIT) , 1 , pfile ); | |
394 | if ( feof( pfile ) ) { | |
395 | break; | |
396 | } | |
397 | samples[ i ] += sample; | |
398 | } | |
399 | if ( i != nsamples ) { | |
400 | fprintf( stderr , | |
401 | "prof: unexpected EOF after reading %d/%d samples\n" , | |
402 | --i , nsamples ); | |
403 | done(); | |
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 | } | |
476 | } | |
477 | h.ncount++; | |
478 | fwrite( &h , sizeof (struct hdr) , 1 , sfile ); | |
479 | /* | |
480 | * write out the counters | |
481 | */ | |
482 | for ( np = nl ; np < npe - 1 ; np++ ) { | |
483 | if ( np -> ncall > 0 ) { | |
484 | kp.cvalue = np -> value * sizeof (unsigned UNIT); | |
485 | kp.cncall = np -> ncall; | |
486 | fwrite( &kp , sizeof (struct cnt) , 1 , sfile ); | |
487 | } | |
488 | } | |
489 | kp.cvalue = 0; | |
490 | kp.cncall = 0; | |
491 | fwrite( &kp , sizeof (struct cnt) , 1 , sfile ); | |
492 | /* | |
493 | * write out the samples | |
494 | */ | |
495 | fwrite( samples , sizeof (unsigned UNIT) , nsamples , sfile ); | |
496 | fclose( sfile ); | |
497 | } | |
498 | ||
499 | min(a, b) | |
500 | { | |
501 | if (a<b) | |
502 | return(a); | |
503 | return(b); | |
504 | } | |
505 | ||
506 | max(a, b) | |
507 | { | |
508 | if (a>b) | |
509 | return(a); | |
510 | return(b); | |
511 | } | |
512 | ||
513 | valcmp(p1, p2) | |
514 | struct nl *p1, *p2; | |
515 | { | |
516 | ||
517 | return(p1->value - p2->value); | |
518 | } | |
519 | ||
520 | timcmp(p1, p2) | |
521 | struct nl *p1, *p2; | |
522 | { | |
523 | float d; | |
524 | ||
525 | if (nflg && p2->ncall != p1->ncall) | |
526 | return (p2->ncall - p1->ncall); | |
527 | d = p2->time - p1->time; | |
528 | if (d > 0.0) | |
529 | return(1); | |
530 | if (d < 0.0) | |
531 | return(-1); | |
532 | return(strcmp(p1->name,p2->name)); | |
533 | } | |
534 | ||
535 | cntcmp(p1, p2) | |
536 | struct cnt *p1, *p2; | |
537 | { | |
538 | ||
539 | return(p1->cvalue - p2->cvalue); | |
540 | } | |
541 | ||
542 | done() | |
543 | { | |
544 | ||
545 | #ifdef plot | |
546 | if(vflg) { | |
547 | point(0, -2040); | |
548 | closepl(); | |
549 | } | |
550 | #endif | |
551 | exit(0); | |
552 | } | |
553 | ||
554 | #ifdef plot | |
555 | plotprof() | |
556 | { | |
557 | double time, lastx, lasty, lastsx; | |
558 | register i; | |
559 | ||
560 | openpl(); | |
561 | erase(); | |
562 | space(-2048, -2048, 2048, 2048); | |
563 | line(-2040, -2040, -2040, 2040); | |
564 | line(0, 2040, 0, -2040); | |
565 | for(i=0; i<11; i++) | |
566 | line(-2040, 2040-i*408, 0, 2040-i*408); | |
567 | lastx = 0.; | |
568 | lasty = ranoff; | |
569 | scale = (4080.*ransca)/(sampbytes/sizeof(UNIT)); | |
570 | lastsx = 0.0; | |
571 | for(i = 0 ; i < nsamples ; i++ ) { | |
572 | unsigned UNIT ccnt; | |
573 | double tx, ty; | |
574 | ccnt = samples[i]; | |
575 | time = ccnt; | |
576 | tx = lastsx; | |
577 | ty = lasty; | |
578 | lastsx =- 2000.*time/totime; | |
579 | lasty =- scale; | |
580 | if(lasty >= -2040. && ty <= 2040.) { | |
581 | line((int)tx, (int)ty, (int)lastsx, (int)lasty); | |
582 | if (ccnt!=0 || lastx!=0.0) { | |
583 | tx = lastx; | |
584 | lastx = -time*2000./maxtime; | |
585 | ty =+ scale/2; | |
586 | line(0, (int)ty, (int)tx, (int)ty); | |
587 | } | |
588 | } | |
589 | } | |
590 | scale = (4080.*ransca)/(highpc-lowpc); | |
591 | lastx = 50.; | |
592 | for(np = nl; np<npe; np++) { | |
593 | if(np->value < lowpc) | |
594 | continue; | |
595 | if(np->value >= highpc) | |
596 | continue; | |
597 | time = np->time/totime; | |
598 | lasty = ranoff - (np->value - lowpc)*scale; | |
599 | if(lasty >= -2040. && lasty <= 2040.) { | |
600 | char bufl[BUFSIZ], *namp; | |
601 | register j; | |
602 | line(0, (int)lasty, 50, (int)lasty); | |
603 | line((int)(lastx-50),(int)lasty,(int)lastx,(int)lasty); | |
604 | point((int)(lastx+30), (int)(lasty+10)); | |
605 | sprintf(bufl, "%s\n", np->name + np->name[0] == '_'); | |
606 | label(bufl); | |
607 | } | |
608 | lastx =+ 500.; | |
609 | if(lastx > 2000.) | |
610 | lastx = 50.; | |
611 | } | |
612 | } | |
613 | #endif |