[FFmpeg-devel] Hacks on FATE
Mon Jun 22 07:57:39 CEST 2009
David Conrad wrote:
> On Jun 18, 2009, at 2:09 AM, Stefan Gehrer wrote:
>> Robert Swain wrote:
>>> On Wed, 17 Jun 2009 23:00:33 +0100, M?ns Rullg?rd <mans at mansr.com>
>>>> Some of the FATE configurations are using decidedly non-standard
>>>> configure options. Some even apply PATCHES to the code. This is not
>>>> the way we want it to be.
>>>> x86_32 MacOSX: --disable-decoder=cavs
>>>> Compiler bug?
>>> Lack of registers and poor register allocation as I recall.
>> Apart from not having the platform to reproduce this, I am
>> also lacking the skills to fix operand constraints in
>> the inline assembly, and I think this is where the problem
>> It seems to come down to the constraints in the QPEL_CAVSVNUM
>> macro in the libavcodec/x86/cavsdsp_mmx.c file.
>> I would be grateful for any comments about them.
> This issue should have been worked around in r16158, at least I can't
> reproduce it with gcc version 4.0.1 (Apple Inc. build 5490) anymore.
Indeed. I just checked it out and --disable-decoder=cavs is no longer
necessary. I have just updated the Mac OS X/x86_32 config so that it
doesn't do that.
Thanks to Mans for doing this audit on all the FATE configs.
More information about the ffmpeg-devel