2 * Copyright (c) 1983 Regents of the University of California.
5 * Redistribution and use in source and binary forms are permitted
6 * provided that: (1) source distributions retain this entire copyright
7 * notice and comment, and (2) distributions including binaries display
8 * the following acknowledgement: ``This product includes software
9 * developed by the University of California, Berkeley and its contributors''
10 * in the documentation or other materials provided with the distribution
11 * and in all advertising materials mentioning features or use of this
12 * software. Neither the name of the University nor the names of its
13 * contributors may be used to endorse or promote products derived
14 * from this software without specific prior written permission.
15 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
20 #define VERSION "5.6 (Cygnus)"
24 "@(#) Copyright (c) 1983 Regents of the University of California.\n\
25 All rights reserved.\n";
29 static char sccsid[] = "@(#)gprof.c 5.6 (Berkeley) 6/1/90";
39 * things which get -E excluded by default.
41 char *defaultEs[] = { "mcount" , "__mcleanup" , 0 };
43 int discard_underscores = 1; /* Should we discard initial underscores? */
44 int bsd_style_output = 0; /* As opposed to FSF style output */
58 while ( *argv != 0 && **argv == '-' ) {
73 debug |= atoi( *argv );
76 printf("[main] debug = %d\n", debug);
78 printf("%s: -d ignored\n", whoami);
83 addlist( Elist , *argv );
85 addlist( elist , *argv );
89 addlist( elist , *++argv );
94 addlist( Flist , *argv );
96 addlist( flist , *argv );
100 addlist( flist , *++argv );
104 addlist( kfromlist , *++argv );
105 addlist( ktolist , *++argv );
111 case 'T': /* "Traditional" output format */
112 bsd_style_output = 1;
115 printf ("gprof version %s\n", VERSION);
123 Usage: %s [-a] [-b] [-c] [-d[num]] [-E function-name] [-e function-name]\n\
124 [-F function-name] [-f function-name] [-k from to] [-s] [-T] [-z]\n\
125 [image-file] [profile-file...]\n", whoami);
134 a_outname = A_OUTNAME;
143 * turn off default functions
145 for ( sp = &defaultEs[0] ; *sp ; sp++ ) {
147 addlist( Elist , *sp );
149 addlist( elist , *sp );
152 * how many ticks per second?
153 * if we can't tell, report time in ticks.
158 fprintf(stderr, "time is in ticks, not seconds\n");
161 * get information about a.out file.
165 * get information about mon.out file(s).
168 getpfile( gmonname );
172 } while ( *argv++ != 0 );
174 * dump out a gmon.sum file if requested
180 * assign samples to procedures
184 * assemble the dynamic profile
186 timesortnlp = doarcs();
188 if (bsd_style_output) {
189 printgprof( timesortnlp ); /* print the dynamic profile */
190 printprof(); /* print the flat profile */
192 printprof(); /* print the flat profile */
193 printgprof( timesortnlp ); /* print the dynamic profile */
203 * Set up string and symbol tables from a.out.
204 * and optionally the text space.
205 * On return symbol table is sorted by value.
211 abfd = bfd_openr (a_outname, NULL);
218 if (!bfd_check_format (abfd, bfd_object)) {
219 fprintf (stderr, "%s: %s: bad format\n", whoami, a_outname);
223 /* getstrtab(nfile); */
225 gettextspace( abfd );
226 qsort(nl, nname, sizeof(nltype), valcmp);
229 if ( debug & AOUTDEBUG ) {
232 for (j = 0; j < nname; j++){
233 printf("[getnfile] 0X%08x\t%s\n", nl[j].value, nl[j].name);
240 * Read in symbol table
249 i = get_symtab_upper_bound (abfd); /* This will probably give us more
250 * than we need, but that's ok.
252 syms = (asymbol**)xmalloc (i);
253 nosyms = bfd_canonicalize_symtab (abfd, syms);
256 for (i = 0; i < nosyms; i++) {
257 if (!funcsymbol (syms[i]))
263 fprintf(stderr, "%s: %s: no symbols\n", whoami , a_outname );
267 nl = (nltype *) calloc( askfor , sizeof(nltype) );
269 fprintf(stderr, "%s: No room for %d bytes of symbol table\n",
270 whoami, askfor * sizeof(nltype) );
274 /* pass2 - read symbols */
277 for (i = 0; i < nosyms; i++) {
278 if (!funcsymbol (syms[i])) {
280 if ( debug & AOUTDEBUG ) {
281 printf( "[getsymtab] rejecting: 0x%x %s\n" ,
282 syms[i]->value, syms[i]->name);
287 /* Symbol offsets are always section-relative. */
288 npe->value = syms[i]->value + syms[i]->section->vma;
289 npe->name = syms[i]->name;
291 /* If we see "main" without an initial '_', we assume
292 names are *not* prefixed by '_'. */
293 if (npe->name[0] == 'm' && discard_underscores
294 && strcmp(npe->name, "main") == 0)
295 discard_underscores = 0;
298 if ( debug & AOUTDEBUG ) {
299 printf( "[getsymtab] %d %s 0x%08x\n" ,
300 nname , npe -> name , npe -> value );
310 * read in the text space of an a.out file
321 texsec = bfd_get_section_by_name (abfd, ".text");
322 if (texsec == NULL) {
326 textspace = (u_char *) malloc( texsec->_cooked_size );
328 if ( textspace == 0 ) {
329 fprintf( stderr , "%s: ran out room for %d bytes of text space: " ,
330 whoami , texsec->_cooked_size);
331 fprintf( stderr , "can't do -c\n" );
334 bfd_get_section_contents (abfd, texsec, textspace, texsec->filepos,
335 texsec->_cooked_size);
338 * information from a gmon.out file is in two parts:
339 * an array of sampling hits within pc ranges,
348 struct veryrawarc rawarc;
350 pfile = openpfile(filename);
353 * the rest of the file consists of
354 * a bunch of <from,self,count> tuples.
356 while ( fread( &rawarc , sizeof rawarc , 1 , pfile ) == 1 ) {
357 arc.raw_frompc = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_frompc);
358 arc.raw_selfpc = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_selfpc);
359 arc.raw_count = bfd_get_32 (abfd, (bfd_byte *) rawarc.raw_count);
361 if ( debug & SAMPLEDEBUG ) {
362 printf( "[getpfile] frompc 0x%x selfpc 0x%x count %d\n" ,
363 arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
382 if((pfile = fopen(filename, "r")) == NULL) {
386 if (sizeof(struct rawhdr) != fread(&raw, 1, sizeof(struct rawhdr), pfile))
388 fprintf(stderr, "%s: file too short to be a gmon file\n", filename);
391 tmp.lowpc = (UNIT *)bfd_get_32 (abfd, (bfd_byte *) &raw.lowpc[0]);
392 tmp.highpc = (UNIT *)bfd_get_32 (abfd, (bfd_byte *) &raw.highpc[0]);
393 tmp.ncnt = bfd_get_32 (abfd, (bfd_byte *) &raw.ncnt[0]);
395 if ( s_highpc != 0 && ( tmp.lowpc != h.lowpc ||
396 tmp.highpc != h.highpc || tmp.ncnt != h.ncnt ) ) {
397 fprintf(stderr, "%s: incompatible with first gmon file\n", filename);
401 s_lowpc = (unsigned long) h.lowpc;
402 s_highpc = (unsigned long) h.highpc;
403 lowpc = (unsigned long)h.lowpc / sizeof(UNIT);
404 highpc = (unsigned long)h.highpc / sizeof(UNIT);
405 sampbytes = h.ncnt - sizeof(struct rawhdr);
406 nsamples = sampbytes / sizeof (UNIT);
408 if ( debug & SAMPLEDEBUG ) {
409 printf( "[openpfile] hdr.lowpc 0x%x hdr.highpc 0x%x hdr.ncnt %d\n",
410 h.lowpc , h.highpc , h.ncnt );
411 printf( "[openpfile] s_lowpc 0x%x s_highpc 0x%x\n" ,
412 s_lowpc , s_highpc );
413 printf( "[openpfile] lowpc 0x%x highpc 0x%x\n" ,
415 printf( "[openpfile] sampbytes %d nsamples %d\n" ,
416 sampbytes , nsamples );
428 parentp = nllookup( rawp -> raw_frompc );
429 childp = nllookup( rawp -> raw_selfpc );
431 && onlist( kfromlist , parentp -> name )
432 && onlist( ktolist , childp -> name ) ) {
435 childp -> ncall += rawp -> raw_count;
437 if ( debug & TALLYDEBUG ) {
438 printf( "[tally] arc from %s to %s traversed %d times\n" ,
439 parentp -> name , childp -> name , rawp -> raw_count );
442 addarc( parentp , childp , rawp -> raw_count );
446 * dump out the gmon.sum file
451 register nltype *nlp;
452 register arctype *arcp;
456 if ( ( sfile = fopen ( sumfile , "w" ) ) == NULL ) {
461 * dump the header; use the last header read in
463 if ( fwrite( &h , sizeof h , 1 , sfile ) != 1 ) {
470 if (fwrite(samples, sizeof (UNIT), nsamples, sfile) != nsamples) {
475 * dump the normalized raw arc information
477 for ( nlp = nl ; nlp < npe ; nlp++ ) {
478 for ( arcp = nlp -> children ; arcp ; arcp = arcp -> arc_childlist ) {
479 arc.raw_frompc = arcp -> arc_parentp -> value;
480 arc.raw_selfpc = arcp -> arc_childp -> value;
481 arc.raw_count = arcp -> arc_count;
482 if ( fwrite ( &arc , sizeof arc , 1 , sfile ) != 1 ) {
487 if ( debug & SAMPLEDEBUG ) {
488 printf( "[dumpsum] frompc 0x%x selfpc 0x%x count %d\n" ,
489 arc.raw_frompc , arc.raw_selfpc , arc.raw_count );
500 if ( p1 -> value < p2 -> value ) {
503 if ( p1 -> value > p2 -> value ) {
516 samples = (int *) calloc (nsamples, sizeof(int));
518 fprintf( stderr , "%s: No room for %d sample pc's\n",
523 for (i = 0; i < nsamples; i++) {
527 fread(raw, sizeof (raw), 1, pfile);
528 value = bfd_get_16 (abfd, (bfd_byte *) raw);
535 "%s: unexpected EOF after reading %d/%d samples\n",
536 whoami , --i , nsamples );
542 * Assign samples to the procedures to which they belong.
544 * There are three cases as to where pcl and pch can be
545 * with respect to the routine entry addresses svalue0 and svalue1
546 * as shown in the following diagram. overlap computes the
547 * distance between the arrows, the fraction of the sample
548 * that is to be credited to the routine which starts at svalue0.
554 * +-----------------------------------------------+
556 * | ->| |<- ->| |<- ->| |<- |
558 * +---------+ +---------+ +---------+
562 * pcl pch pcl pch pcl pch
564 * For the vax we assert that samples will never fall in the first
565 * two bytes of any routine, since that is the entry mask,
566 * thus we give call alignentries() to adjust the entry points if
567 * the entry mask falls in one bucket but the code for the routine
568 * doesn't start until the next bucket. In conjunction with the
569 * alignment of routine addresses, this should allow us to have
570 * only one sample for every four bytes of text space and never
571 * have any overlap (the two end cases, above).
578 unsigned long pcl, pch;
580 unsigned long overlap;
581 unsigned long svalue0, svalue1;
583 /* read samples and assign to namelist symbols */
584 scale = highpc - lowpc;
585 scale /= nsamples - 1;
587 for (i = 0, j = 1; i < nsamples; i++) {
591 pcl = lowpc + scale * i;
592 pch = lowpc + scale * (i + 1);
595 if ( debug & SAMPLEDEBUG ) {
596 printf( "[asgnsamples] pcl 0x%x pch 0x%x ccnt %d\n" ,
601 for (j = j - 1; j < nname; j++) {
602 svalue0 = nl[j].svalue;
603 svalue1 = nl[j+1].svalue;
605 * if high end of tick is below entry address,
611 * if low end of tick into next routine,
612 * go for next routine.
616 overlap = min(pch, svalue1) - max(pcl, svalue0);
619 if (debug & SAMPLEDEBUG) {
620 printf("[asgnsamples] (0x%x->0x%x-0x%x) %s gets %f ticks %d overlap\n",
621 nl[j].value/sizeof(UNIT), svalue0, svalue1,
623 overlap * time / scale, overlap);
626 nl[j].time += overlap * time / scale;
631 if (debug & SAMPLEDEBUG) {
632 printf("[asgnsamples] totime %f\n", totime);
657 * calculate scaled entry point addresses (to save time in asgnsamples),
658 * and possibly push the scaled entry points over the entry mask,
659 * if it turns out that the entry point is in one bucket and the code
660 * for a routine is in the next bucket.
664 register struct nl *nlp;
665 unsigned long bucket_of_entry;
666 unsigned long bucket_of_code;
668 for (nlp = nl; nlp < npe; nlp++) {
669 nlp -> svalue = nlp -> value / sizeof(UNIT);
670 bucket_of_entry = (nlp->svalue - lowpc) / scale;
671 bucket_of_code = (nlp->svalue + UNITS_TO_CODE - lowpc) / scale;
672 if (bucket_of_entry < bucket_of_code) {
674 if (debug & SAMPLEDEBUG) {
675 printf("[alignentries] pushing svalue 0x%x to 0x%x\n",
676 nlp->svalue, nlp->svalue + UNITS_TO_CODE);
679 nlp->svalue += UNITS_TO_CODE;
688 extern char *strtab; /* string table from a.out */
689 extern int aflag; /* if static functions aren't desired */
695 * must be a text symbol,
696 * and static text symbols don't qualify if aflag set.
703 if (aflag && (symp->flags&BSF_LOCAL)) {
705 fprintf (stderr, "%s(%d): %s: not a function\n", __FILE__, __LINE__, symp->name);
710 symprefix = bfd_get_symbol_leading_char (abfd);
711 i = bfd_decode_symclass (symp);
712 #if defined(DEBUG) && 0
713 if (i != 'T' && i != 't')
714 fprintf (stderr, "%s(%d): %s is of class %c\n", __FILE__, __LINE__, symp->name, i);
718 * Any external text symbol should be okay. (Only problem would be
719 * variables in the text section.)
726 * 't' is static text; -a says to ignore it. So if it's not
727 * a static text symbol, *or* it is and the user gave -a, we
731 if (i != 't' || aflag)
735 * can't have any `funny' characters in name,
736 * where `funny' includes `.', .o file names
737 * and `$', pascal labels.
742 for (name = symp->name; *name; name++) {
743 if ( *name == '.' || *name == '$' ) {
748 /* On systems where the C compiler adds an underscore to all names,
749 * static names without underscores seem usually to be labels in
750 * hand written assembler in the library. We don't want these
751 * names. This is certainly necessary on a Sparc running SunOS 4.1
752 * (try profiling a program that does a lot of division). I don't
753 * know whether it has harmful side effects on other systems.
754 * Perhaps it should be made configurable.
757 if (symprefix && symprefix != *symp->name)