1 #define PERL_NO_GET_CONTEXT
7 /*#define DBG_TIMER 1 */
10 # define DBG_SUB_NOTIFY(A,B) warn(A, B)
12 # define DBG_SUB_NOTIFY(A,B) /* nothing */
16 # define DBG_TIMER_NOTIFY(A) warn(A)
18 # define DBG_TIMER_NOTIFY(A) /* nothing */
21 /* HZ == clock ticks per second */
23 # define HZ ((I32)CLK_TCK)
25 # include <starlet.h> /* prototype for sys$gettim() */
26 # define Times(ptr) (dprof_times(aTHX_ ptr))
30 # define HZ ((I32)CLK_TCK)
35 # ifdef OS2 /* times() has significant overhead */
36 # define Times(ptr) (dprof_times(aTHX_ ptr))
37 # define INCL_DOSPROFILE
38 # define INCL_DOSERRORS
40 # define toLongLong(arg) (*(long long*)&(arg))
41 # define DPROF_HZ g_dprof_ticks
43 # define Times(ptr) (times(ptr))
48 XS(XS_Devel__DProf_END); /* used by prof_mark() */
50 /* Everything is built on times(2). See its manpage for a description
55 clock_t tms_utime; /* cpu time spent in user space */
56 clock_t tms_stime; /* cpu time spent in system */
57 clock_t realtime; /* elapsed real time, in ticks */
63 typedef union prof_any PROFANY;
67 char* out_file_name; /* output file (defaults to tmon.out) */
68 PerlIO* fp; /* pointer to tmon.out file */
69 long TIMES_LOCATION; /* Where in the file to store the time totals */
70 int SAVE_STACK; /* How much data to buffer until end of run */
71 int prof_pid; /* pid of profiled process */
72 struct tms prof_start;
74 clock_t rprof_start; /* elapsed real time ticks */
94 #ifdef PERL_IMPLICIT_CONTEXT
101 prof_state_t g_prof_state;
103 #define g_dprof_ticks g_prof_state.dprof_ticks
104 #define g_out_file_name g_prof_state.out_file_name
105 #define g_fp g_prof_state.fp
106 #define g_TIMES_LOCATION g_prof_state.TIMES_LOCATION
107 #define g_SAVE_STACK g_prof_state.SAVE_STACK
108 #define g_prof_pid g_prof_state.prof_pid
109 #define g_prof_start g_prof_state.prof_start
110 #define g_prof_end g_prof_state.prof_end
111 #define g_rprof_start g_prof_state.rprof_start
112 #define g_rprof_end g_prof_state.rprof_end
113 #define g_wprof_u g_prof_state.wprof_u
114 #define g_wprof_s g_prof_state.wprof_s
115 #define g_wprof_r g_prof_state.wprof_r
116 #define g_otms_utime g_prof_state.otms_utime
117 #define g_otms_stime g_prof_state.otms_stime
118 #define g_orealtime g_prof_state.orealtime
119 #define g_profstack g_prof_state.profstack
120 #define g_profstack_max g_prof_state.profstack_max
121 #define g_profstack_ix g_prof_state.profstack_ix
122 #define g_cv_hash g_prof_state.cv_hash
123 #define g_total g_prof_state.total
124 #define g_lastid g_prof_state.lastid
125 #define g_default_perldb g_prof_state.default_perldb
126 #define g_depth g_prof_state.depth
127 #ifdef PERL_IMPLICIT_CONTEXT
128 # define g_THX g_prof_state.aTHX
131 # define g_frequ g_prof_state.frequ
132 # define g_start_cnt g_prof_state.start_cnt
136 dprof_times(pTHX_ struct tms *t)
144 if (CheckOSError(DosTmrQueryFreq(&g_frequ)))
145 croak("DosTmrQueryFreq: %s", SvPV(perl_get_sv("!",TRUE),n_a));
147 g_frequ = g_frequ/DPROF_HZ; /* count per tick */
148 if (CheckOSError(DosTmrQueryTime(&cnt)))
149 croak("DosTmrQueryTime: %s",
150 SvPV(perl_get_sv("!",TRUE), n_a));
151 g_start_cnt = toLongLong(cnt);
154 if (CheckOSError(DosTmrQueryTime(&cnt)))
155 croak("DosTmrQueryTime: %s", SvPV(perl_get_sv("!",TRUE), n_a));
157 return (t->tms_utime = (toLongLong(cnt) - g_start_cnt)/g_frequ);
161 /* Get wall time and convert to 10 ms intervals to
162 * produce the return value dprof expects */
163 # if defined(__DECC) && defined (__ALPHA)
166 _ckvmssts(sys$gettim(&vmstime));
168 retval = vmstime & 0x7fffffff;
170 /* (Older hw or ccs don't have an atomic 64-bit type, so we
171 * juggle 32-bit ints (and a float) to produce a time_t result
172 * with minimal loss of information.) */
173 long int vmstime[2],remainder,divisor = 100000;
174 _ckvmssts(sys$gettim((unsigned long int *)vmstime));
175 vmstime[1] &= 0x7fff; /* prevent overflow in EDIV */
176 _ckvmssts(lib$ediv(&divisor,vmstime,(long int *)&retval,&remainder));
178 /* Fill in the struct tms using the CRTL routine . . .*/
179 times((tbuffer_t *)t);
180 return (clock_t) retval;
181 # else /* !VMS && !OS2 */
188 prof_dumpa(pTHX_ opcode ptype, U32 id)
190 if (ptype == OP_LEAVESUB) {
191 PerlIO_printf(g_fp,"- %"UVxf"\n", (UV)id);
193 else if(ptype == OP_ENTERSUB) {
194 PerlIO_printf(g_fp,"+ %"UVxf"\n", (UV)id);
196 else if(ptype == OP_GOTO) {
197 PerlIO_printf(g_fp,"* %"UVxf"\n", (UV)id);
199 else if(ptype == OP_DIE) {
200 PerlIO_printf(g_fp,"/ %"UVxf"\n", (UV)id);
203 PerlIO_printf(g_fp,"Profiler unknown prof code %d\n", ptype);
208 prof_dumps(pTHX_ U32 id, char *pname, char *gname)
210 PerlIO_printf(g_fp,"& %"UVxf" %s %s\n", (UV)id, pname, gname);
214 prof_dumpt(pTHX_ long tms_utime, long tms_stime, long realtime)
216 PerlIO_printf(g_fp,"@ %ld %ld %ld\n", tms_utime, tms_stime, realtime);
220 prof_dump_until(pTHX_ long ix)
224 clock_t realtime1, realtime2;
226 realtime1 = Times(&t1);
229 opcode ptype = g_profstack[base++].ptype;
230 if (ptype == OP_TIME) {
231 long tms_utime = g_profstack[base++].tms_utime;
232 long tms_stime = g_profstack[base++].tms_stime;
233 long realtime = g_profstack[base++].realtime;
235 prof_dumpt(aTHX_ tms_utime, tms_stime, realtime);
237 else if (ptype == OP_GV) {
238 U32 id = g_profstack[base++].id;
239 char *pname = g_profstack[base++].name;
240 char *gname = g_profstack[base++].name;
242 prof_dumps(aTHX_ id, pname, gname);
245 U32 id = g_profstack[base++].id;
246 prof_dumpa(aTHX_ ptype, id);
250 realtime2 = Times(&t2);
251 if (realtime2 != realtime1 || t1.tms_utime != t2.tms_utime
252 || t1.tms_stime != t2.tms_stime) {
253 g_wprof_r += realtime2 - realtime1;
254 g_wprof_u += t2.tms_utime - t1.tms_utime;
255 g_wprof_s += t2.tms_stime - t1.tms_stime;
257 PerlIO_printf(g_fp,"+ & Devel::DProf::write\n");
258 PerlIO_printf(g_fp,"@ %"IVdf" %"IVdf" %"IVdf"\n",
259 /* The (IV) casts are one possibility:
260 * the Painfully Correct Way would be to
262 (IV)(t2.tms_utime - t1.tms_utime),
263 (IV)(t2.tms_stime - t1.tms_stime),
264 (IV)(realtime2 - realtime1));
265 PerlIO_printf(g_fp,"- & Devel::DProf::write\n");
266 g_otms_utime = t2.tms_utime;
267 g_otms_stime = t2.tms_stime;
268 g_orealtime = realtime2;
274 prof_mark(pTHX_ opcode ptype)
277 clock_t realtime, rdelta, udelta, sdelta;
283 SV *Sub = GvSV(PL_DBsub); /* name of current sub */
286 if (g_profstack_ix + 5 > g_profstack_max) {
287 g_profstack_max = g_profstack_max * 3 / 2;
288 Renew(g_profstack, g_profstack_max, PROFANY);
292 realtime = Times(&t);
293 rdelta = realtime - g_orealtime;
294 udelta = t.tms_utime - g_otms_utime;
295 sdelta = t.tms_stime - g_otms_stime;
296 if (rdelta || udelta || sdelta) {
298 g_profstack[g_profstack_ix++].ptype = OP_TIME;
299 g_profstack[g_profstack_ix++].tms_utime = udelta;
300 g_profstack[g_profstack_ix++].tms_stime = sdelta;
301 g_profstack[g_profstack_ix++].realtime = rdelta;
303 else { /* Write it to disk now so's not to eat up core */
304 if (g_prof_pid == (int)getpid()) {
305 prof_dumpt(aTHX_ udelta, sdelta, rdelta);
309 g_orealtime = realtime;
310 g_otms_stime = t.tms_stime;
311 g_otms_utime = t.tms_utime;
319 cv = INT2PTR(CV*,SvIVX(Sub));
320 svp = hv_fetch(g_cv_hash, (char*)&cv, sizeof(CV*), TRUE);
324 sv_setiv(*svp, id = ++g_lastid);
325 pname = ((GvSTASH(gv) && HvNAME(GvSTASH(gv)))
326 ? HvNAME(GvSTASH(gv))
329 if (CvXSUB(cv) == XS_Devel__DProf_END)
331 if (g_SAVE_STACK) { /* Store it for later recording -JH */
332 g_profstack[g_profstack_ix++].ptype = OP_GV;
333 g_profstack[g_profstack_ix++].id = id;
334 g_profstack[g_profstack_ix++].name = pname;
335 g_profstack[g_profstack_ix++].name = gname;
337 else { /* Write it to disk now so's not to eat up core */
338 /* Only record the parent's info */
339 if (g_prof_pid == (int)getpid()) {
340 prof_dumps(aTHX_ id, pname, gname);
344 PL_perldb = 0; /* Do not debug the kid. */
353 if (g_SAVE_STACK) { /* Store it for later recording -JH */
354 g_profstack[g_profstack_ix++].ptype = ptype;
355 g_profstack[g_profstack_ix++].id = id;
357 /* Only record the parent's info */
358 if (g_SAVE_STACK < g_profstack_ix) {
359 if (g_prof_pid == (int)getpid())
360 prof_dump_until(aTHX_ g_profstack_ix);
362 PL_perldb = 0; /* Do not debug the kid. */
366 else { /* Write it to disk now so's not to eat up core */
368 /* Only record the parent's info */
369 if (g_prof_pid == (int)getpid()) {
370 prof_dumpa(aTHX_ ptype, id);
374 PL_perldb = 0; /* Do not debug the kid. */
379 # define defstash PL_defstash
382 /* Counts overhead of prof_mark and extra XS call. */
384 test_time(pTHX_ clock_t *r, clock_t *u, clock_t *s)
386 CV *cv = perl_get_cv("Devel::DProf::NONESUCH_noxs", FALSE);
388 HV *oldstash = PL_curstash;
390 clock_t realtime1, realtime2;
391 U32 ototal = g_total;
392 U32 ostack = g_SAVE_STACK;
393 U32 operldb = PL_perldb;
395 g_SAVE_STACK = 1000000;
396 realtime1 = Times(&t1);
400 /* Disable debugging of perl_call_sv on second pass: */
401 PL_curstash = (k == 0 ? PL_defstash : PL_debstash);
402 PL_perldb = g_default_perldb;
405 g_profstack_ix = 0; /* Do not let the stack grow */
407 /* prof_mark(aTHX_ OP_ENTERSUB); */
409 PUSHMARK(PL_stack_sp);
410 perl_call_sv((SV*)cv, G_SCALAR);
412 /* prof_mark(aTHX_ OP_LEAVESUB); */
415 PL_curstash = oldstash;
416 if (k == 0) { /* Put time with debugging */
417 realtime2 = Times(&t2);
418 *r = realtime2 - realtime1;
419 *u = t2.tms_utime - t1.tms_utime;
420 *s = t2.tms_stime - t1.tms_stime;
422 else { /* Subtract time without debug */
423 realtime1 = Times(&t1);
424 *r -= realtime1 - realtime2;
425 *u -= t1.tms_utime - t2.tms_utime;
426 *s -= t1.tms_stime - t2.tms_stime;
431 g_SAVE_STACK = ostack;
436 prof_recordheader(pTHX)
440 /* g_fp is opened in the BOOT section */
441 PerlIO_printf(g_fp, "#fOrTyTwO\n");
442 PerlIO_printf(g_fp, "$hz=%"IVdf";\n", (IV)DPROF_HZ);
443 PerlIO_printf(g_fp, "$XS_VERSION='DProf %s';\n", XS_VERSION);
444 PerlIO_printf(g_fp, "# All values are given in HZ\n");
445 test_time(aTHX_ &r, &u, &s);
447 "$over_utime=%"IVdf"; $over_stime=%"IVdf"; $over_rtime=%"IVdf";\n",
448 /* The (IV) casts are one possibility:
449 * the Painfully Correct Way would be to
451 (IV)u, (IV)s, (IV)r);
452 PerlIO_printf(g_fp, "$over_tests=10000;\n");
454 g_TIMES_LOCATION = PerlIO_tell(g_fp);
456 /* Pad with whitespace. */
457 /* This should be enough even for very large numbers. */
458 PerlIO_printf(g_fp, "%*s\n", 240 , "");
460 PerlIO_printf(g_fp, "\n");
461 PerlIO_printf(g_fp, "PART2\n");
469 /* g_fp is opened in the BOOT section */
471 /* Now that we know the runtimes, fill them in at the recorded
477 prof_dump_until(aTHX_ g_profstack_ix);
479 PerlIO_seek(g_fp, g_TIMES_LOCATION, SEEK_SET);
480 /* Write into reserved 240 bytes: */
482 "$rrun_utime=%"IVdf"; $rrun_stime=%"IVdf"; $rrun_rtime=%"IVdf";",
483 /* The (IV) casts are one possibility:
484 * the Painfully Correct Way would be to
486 (IV)(g_prof_end.tms_utime-g_prof_start.tms_utime-g_wprof_u),
487 (IV)(g_prof_end.tms_stime-g_prof_start.tms_stime-g_wprof_s),
488 (IV)(g_rprof_end-g_rprof_start-g_wprof_r));
489 PerlIO_printf(g_fp, "\n$total_marks=%"IVdf, (IV)g_total);
497 check_depth(pTHX_ void *foo)
499 U32 need_depth = PTR2UV(foo);
500 if (need_depth != g_depth) {
501 if (need_depth > g_depth) {
502 warn("garbled call depth when profiling");
505 I32 marks = g_depth - need_depth;
507 /* warn("Check_depth: got %d, expected %d\n", g_depth, need_depth); */
509 prof_mark(aTHX_ OP_DIE);
511 g_depth = need_depth;
523 SV *Sub = GvSV(PL_DBsub); /* name of current sub */
525 #ifdef PERL_IMPLICIT_CONTEXT
526 /* profile only the interpreter that loaded us */
529 perl_call_sv(INT2PTR(SV*,SvIV(Sub)), GIMME | G_NODEBUG);
534 HV *oldstash = PL_curstash;
536 DBG_SUB_NOTIFY("XS DBsub(%s)\n", SvPV_nolen(Sub));
538 SAVEDESTRUCTOR_X(check_depth, (void*)g_depth);
541 prof_mark(aTHX_ OP_ENTERSUB);
543 perl_call_sv(INT2PTR(SV*,SvIV(Sub)), GIMME | G_NODEBUG);
544 prof_mark(aTHX_ OP_LEAVESUB);
552 #ifdef PERL_IMPLICIT_CONTEXT
556 prof_mark(aTHX_ OP_GOTO);
561 #endif /* for_real */
565 MODULE = Devel::DProf PACKAGE = DB
572 HV *oldstash = PL_curstash;
573 SV *Sub = GvSV(PL_DBsub); /* name of current sub */
574 /* SP -= items; added by xsubpp */
575 DBG_SUB_NOTIFY("XS DBsub(%s)\n", SvPV_nolen(Sub));
577 sv_setiv(PL_DBsingle, 0); /* disable DB single-stepping */
579 prof_mark(aTHX_ OP_ENTERSUB);
582 PL_curstash = PL_debstash; /* To disable debugging of perl_call_sv */
583 perl_call_sv(Sub, GIMME);
584 PL_curstash = oldstash;
586 prof_mark(aTHX_ OP_LEAVESUB);
588 /* PUTBACK; added by xsubpp */
593 MODULE = Devel::DProf PACKAGE = Devel::DProf
600 /* maybe the process forked--we want only
601 * the parent's profile.
604 #ifdef PERL_IMPLICIT_CONTEXT
607 g_prof_pid == (int)getpid())
609 g_rprof_end = Times(&g_prof_end);
610 DBG_TIMER_NOTIFY("Profiler timer is off.\n");
621 g_TIMES_LOCATION = 42;
622 g_SAVE_STACK = 1<<14;
623 g_profstack_max = 128;
624 #ifdef PERL_IMPLICIT_CONTEXT
628 /* Before we go anywhere make sure we were invoked
629 * properly, else we'll dump core.
632 croak("DProf: run perl with -d to use DProf.\n");
634 /* When we hook up the XS DB::sub we'll be redefining
635 * the DB::sub from the PM file. Turn off warnings
639 I32 warn_tmp = PL_dowarn;
641 newXS("DB::sub", XS_DB_sub, file);
642 newXS("DB::goto", XS_DB_goto, file);
643 PL_dowarn = warn_tmp;
646 sv_setiv(PL_DBsingle, 0); /* disable DB single-stepping */
649 char *buffer = getenv("PERL_DPROF_BUFFER");
652 g_SAVE_STACK = atoi(buffer);
655 buffer = getenv("PERL_DPROF_TICKS");
658 g_dprof_ticks = atoi(buffer); /* Used under OS/2 only */
664 buffer = getenv("PERL_DPROF_OUT_FILE_NAME");
665 g_out_file_name = savepv(buffer ? buffer : "tmon.out");
668 if ((g_fp = PerlIO_open(g_out_file_name, "w")) == NULL)
669 croak("DProf: unable to write '%s', errno = %d\n",
670 g_out_file_name, errno);
672 g_default_perldb = PERLDBf_NONAME | PERLDBf_SUB | PERLDBf_GOTO;
674 g_prof_pid = (int)getpid();
676 New(0, g_profstack, g_profstack_max, PROFANY);
677 prof_recordheader(aTHX);
678 DBG_TIMER_NOTIFY("Profiler timer is on.\n");
679 g_orealtime = g_rprof_start = Times(&g_prof_start);
680 g_otms_utime = g_prof_start.tms_utime;
681 g_otms_stime = g_prof_start.tms_stime;
682 PL_perldb = g_default_perldb;