: Re: : Re: [MPlayer-users] BUG: mencoder of pre5 crashes everytime

[Martin Simmons]

>>>>> On Thu, 26 Aug 2004 09:38:28 +0200, Stefan Sassenberg <stefan.sassenberg at gmx.de> said:

  Stefan> Martin Simmons wrote:
  >>>>>>> On Tue, 17 Aug 2004 13:03:36 +0200, Stefan Sassenberg <stefan.sassenberg at gmx.de> said:
  >> 
  >> 
  Stefan> Martin Simmons wrote:
  >> 
  >> >> Hi there,
  >> >> 
  >> >> Very sorry for leaving it so long before replying.
  >> >> 
  >> >> The maps output just seems to add to the mystery, because the memory looks OK
  >> >> there and so I don't understand how the SIGSEGV can happen.  Unless gdb is
  >> >> broken or I've misanalysed it, discrepancies like this can only be caused by a
  >> >> bug in the OS or in the hardware.
  >> >> 
  >> >> The up command confused the last register dump, so please post one more crash
  >> >> with the gdb commands (assuming sad16_y2_mmx2 is the topmost function again):
  >> >> 
  >> >> bt
  >> >> disass $pc-32 $pc+32
  >> >> info all-registers
  >> >> disass sad16_y2_mmx2
  >> >> shell cat /proc/NNNNN/maps
  >> >> 
  >> >> __Martin
  >> >> 
  >> >> 
  >> 
  Stefan> Thanks, here you go.
  >> 
  >> I'm still not sure what is happening here.  The latest crash displays the
  >> usually "good" register contents that fail to explain the SIGSEGV.  I will
  >> keep looking.
  >> 
  >> Possibly your only hope for now is to reconfigure mplayer with --disable-mmx
  >> (or --disable-mmx2).  I've not tried that, but hopefully it will prevent
  >> sad16_y2_mmx2 from being called.
  >> 
  >> __Martin
  >> 
  >> 
  Stefan> I'll try that, thank you.

  Stefan> Yesterday I recorded 2hrs from tv with a precompiled 1.0pre3 without 
  Stefan> crash. Maybe you can tell from the differences between the versions 
  Stefan> where the problem is located, though the precompiled maybe doesn't use 
  Stefan> mmx2 code.

  Stefan> I can also download and test the 1.0pre4 or any other version if you 
  Stefan> think it might help.

The low level mmx2 code in 1.0pre3 is the same as 1.0pre5, but the code around
it is slightly different (it looks functionally the same though).  1.0pre4 is
almost identical to 1.0pre5.

Given the slightly random nature of the crash, it is possible that 1.0pre3
doesn't trigger it.  Good to hear you had some success though!

__Martin