[mythtv] ffmpeg sync
reidjr at btconnect.com
Sat Oct 20 20:00:51 UTC 2007
> Janne Grunau wrote:
>> On Sunday 14 October 2007 23:36:26 John wrote:
>>> Once I got over the apparent dependency on gcc version (gcc-2.95 and
>>> gcc-4.1 both failed to compile the ffmpeg patch for some reason,
>>> seems like a known issue) and compiled with gcc-3.4, I used both the
>>> main, and additional patch.
>> It builds here with gcc 4.1.2 (32 bit and 64 bit). It would be nice
>> if you can give details of your system and the build failure.
> the failure message was as I posted above, and seemed to be a known
> ffmpeg /compiler problem.
> can't find a register in class `GENERAL_REGS' while reloading `asm'
> I was originally set up with gcc-4.1 (GCC) 4.1.2 20061115 (prerelease)
> (Debian 4.1.1-20), and was trying to compile it on my backend which is
> a DURON1600 (athlonXP). The bug report seemed to suggest that it was
> non-determenistic which seems bizarre.
> I was just about to see if I could reuse your skiploopfilter patch
> from a few months back, but was allready at the "too difficult" stage.
> Any chance that you could get that into any commit. Always good to be
> able to ease of the cpu if possible.
> Thanks for all your efforts, much appreciated.
The gcc version may have been a red herring ... or maybe not, I havn't
tried all possible combinations yet.
After much messing around I realised I was using --enable-proc-opt to
try and wring the last clock cycle out of my machines. If I compiled on
my backend (Duron1600) it was compiling with -march=athlon , which would
fail, whereas the core2duo was -march=K8. -march=K8 would compile, and
-march=athlon would not. (and -march=K8 dies with illegal instruction on
my Duron back end -- surprise). The default -march=pentiumpro works on
all my machines, and I cant see any performance difference on playing
back h264. I had expected a measurable difference when optimized against
the newer processors.
More information about the mythtv-dev