VBCC -prof profiling - requirements?

[Warty]

I'm trying to out -prof option in VBCC, and I get tons of errors like this:

tmp.0.OCp6aN.o: In "_Action_FindThisAction":
Error 21: tmp.0.OCp6aN.o (CODE+0x28): Reference to undefined symbol __startprof.
tmp.0.OCp6aN.o: In "l1":
Error 21: tmp.0.OCp6aN.o (CODE+0x74): Reference to undefined symbol __endprof.

vlink -bamigahunk -x -Bstatic -Cvbcc -nostdlib -Z -mrel "$VBCC/targets/m68k-kick13/lib/startup.o" "/var/tmp/tmp.0.w8T6Rt.o" "/var/tmp/tmp.1.J7m1ZV.o" "/var/tmp/tmp.2.xtwhPA.o" "/var/tmp/tmp.3.dnhtFw.o" "/var/tmp/tmp.4.obH6Aw.o" "/var/tmp/tmp.5.BfnXn6.o" "/var/tmp/tmp.6.MN1aV2.o" "/var/tmp/tmp.7.NcrnbQ.o" "/var/tmp/tmp.8.O1caEi.o" "/var/tmp/tmp.9.DDYsC6.o" "/var/tmp/tmp.10.tf1Q2B.o" "/var/tmp/tmp.11.YPW6mC.o" "/var/tmp/tmp.12.asIG0s.o" "/var/tmp/tmp.13.zc9bao.o" "/var/tmp/tmp.14.12oTsZ.o" "/var/tmp/tmp.15.bOFQKr.o" "/var/tmp/tmp.16.LkmuBB.o" "/var/tmp/tmp.17.kWshnf.o" "/var/tmp/tmp.18.zHdTEt.o" "/var/tmp/tmp.19.RZljYQ.o" "/var/tmp/tmp.20.b3O5z4.o" "/var/tmp/tmp.21.Jlz3gP.o" "/var/tmp/tmp.22.1XIEgL.o" "/var/tmp/tmp.23.isbbTC.o" "/var/tmp/tmp.24.V3vNou.o" "/var/tmp/tmp.25.Jk8XIs.o" "/var/tmp/tmp.26.BZDcSw.o" "/var/tmp/tmp.27.cPl1i0.o" "/var/tmp/tmp.28.VvuMRo.o" "/var/tmp/tmp.29.iXMYbt.o" "/Users/micahbly/dev/bbedit-workspace-amiga/LichKing-Amiga/build_vbcc/ptplayer.o"   -lauto -lamiga -lreqtoolsnb -lm13 -L"$VBCC/targets/m68k-kick13/lib" -L"$VBCC/ndk13/Include-Strip1.3/lib" -lvc -o /Users/micahbly/dev/bbedit-workspace-amiga/LichKing-Amiga/build_vbcc/LichKing failed

are __endprof and __startprof in vc.lib? I can see something in there with that name, but only the vc.lib/vcs.lib that is in the m68k-amigaos target, not in the kick13 target. Is the profiling maybe not supported yet in kick13 target?

[phx]

Originally Posted by Warty

but only the vc.lib/vcs.lib that is in the m68k-amigaos target, not in the kick13 target. Is the profiling maybe not supported yet in kick13 target?

Correct. And good that you remind me about it, because I had completely forgotten.
The code from m68k-amigaos cannot be used, because it depends on timer.device

UNIT_ECLOCK

and the

ReadEClock()

function, which are only present since V36. The profiler needs the time since program start in micro seconds.

Basically, the only thing I have to do is to write a Kickstart 1.x version for this:

#include <exec/types.h>
#include <exec/execbase.h>
#include <devices/timer.h>
#include <proto/exec.h>
#include <stdio.h>
#include <stdint.h>
#include <string.h>

static uint32_t _ReadEClock(__reg("a0") uint64_t *,
                            __reg("a6") void *) = "\tjsr\t-60(a6)";

static void *timerbase;
static struct timerequest timerequest;
static uint64_t start,eclock_msfactor;
static uint32_t eclock_freq;


int _prof_timerinit()
{
  if (SysBase->LibNode.lib_Version < 36)
    return 0;

  /* hack a timerequest, without CreateMsgPort() and CreateIORequest() */
  memset(&timerequest,0,sizeof(struct timerequest));
        timerequest.tr_node.io_Message.mn_Node.ln_Type = NT_REPLYMSG;
        timerequest.tr_node.io_Message.mn_Length = sizeof(struct timerequest);

  if (OpenDevice(TIMERNAME,UNIT_ECLOCK,(struct IORequest *)&timerequest,0)
      != 0) {
    fprintf(stderr,"_prof_inittimer: Can't get " TIMERNAME " for profiling!\n")
    return 0;
  }

  timerbase = timerequest.tr_node.io_Device;
  eclock_freq = _ReadEClock(&start,timerbase);
  eclock_msfactor = (1000000ULL << 16) / eclock_freq;
  return 1;
}


void _prof_timerexit()
{
  if (timerbase) {
    CloseDevice((struct IORequest *)&timerequest);
    timerbase = NULL;
  }
}


uint32_t _prof_time()
{
  uint64_t now;

  _ReadEClock(&now,timerbase);
  return (uint32_t)(now-start);
}


uint32_t _prof_micros(uint32_t ecval)

{
  /* convert EClock value to microseconds */
  return (eclock_msfactor * (uint64_t)ecval) >> 16;
}

[Warty]

ok, thanks for letting me know. Thanks for all you do for us, by the way.

[patrik]

If your own code is of primary concern for profiling, you could always compile and run with profiling on 2.0+.

[Warty]

yep, good point. I also have AProf working, but sometimes the results are a bit weird (impossible numbers), and virtually always there is a lot of variation in the time lengths recorded (which, I dunno, might be true anytime running functions with such short run times).

Not in any hurry...