Vorbis dev - Jun 2001 - gcc 2.95.2/irix/Laguerre_With_Deflation/inifinte loop

I built ogg vorbis from the rc1 cvs source on Irix 6.5.12
with gcc 2.95.2.  Using oggenc I encoded about 8,000 aiff files
but found about a dozen where oggenc would go into an infinite
loop.  I tracked the problem with Laguerre_With_Deflation() as far back
as
logmask being Inf in floor0_forward.  I'm now building gcc 3.0 with the
expectation this is a compiler issue.  If not, I'll back trace further.
Has this been observed by anyone else?  I didn't see it in a search of
the
archive.

        brian redman

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

Followup on the infinite loop problem.  Upgrading to gcc 3.0 had
no effect.

The code behaves different in floor0_forward() where val is first set.
Is rint(nan/130*1023) meaningful?  In one case (the one that doesn't
go into an infinite loop) it returns a negative number, in the other,
positive.  

    ber

  val=rint(amp/info->ampdB*maxval);

  linux/x86% ./oggenc
~/x.aiff                                                       
  Opening with aiff module: AIFF/AIFC file reader
  mapping0_forward: logmask[0] = inf
  floor0_forward: inf 130
  floor0_forward: amp = nan
  floor0_forward: info->ampdB = 130
  floor0_forward: maxval = 1023
  floor0_forward: val = -2147483648

  irix/r14k% ./oggenc ~/x.aiff
  Opening with aiff module: AIFF/AIFC file reader
  mapping0_forward: logmask[0] = inf
  floor0_forward: inf 130
  floor0_forward: amp = nan0x7fffffff
  floor0_forward: info->ampdB = 130
  floor0_forward: maxval = 1023
  floor0_forward: val = 2147483647

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

Fixed.  I hacked in a test of val for MAXINT (and included values.h)
but presumably assigning the return from rint() to a temp and checking
it for nan to set val to 0 is the right way to go.

For your amusement, the 4sec aiff sound byte that tripped across this
is at http://rineke.net/vorbis-dev/.

  brian

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

On Sat, Jun 23, 2001 at 07:24:04PM -0400, ber wrote:
> I built ogg vorbis from the rc1 cvs source on Irix 6.5.12
> with gcc 2.95.2.  Using oggenc I encoded about 8,000 aiff files
> but found about a dozen where oggenc would go into an infinite
> loop.  I tracked the problem with Laguerre_With_Deflation() as far back
> as
> logmask being Inf in floor0_forward.
OK, that's important to know.  All of it?

  I'm now building gcc 3.0 with the
> expectation this is a compiler issue.  If not, I'll back trace further.
> Has this been observed by anyone else?  I didn't see it in a search of
> the
> archive.
No, not seen.  And I'm not tripping it here using the file you
provided (although that could just be an effect of Linux/PowerPC not
tripping an exception where IRIX/MIPS would).

I'll try this with the processor configured to spew more tomorrow.

Monty

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.