[FFmpeg-devel] release

David Conrad lessen42
Fri Jan 30 22:42:01 CET 2009


On Jan 30, 2009, at 4:32 PM, Diego Biurrun wrote:

> On Fri, Jan 30, 2009 at 04:30:27PM -0500, David Conrad wrote:
>> On Jan 30, 2009, at 4:23 PM, Diego Biurrun wrote:
>>
>>> On Fri, Jan 30, 2009 at 09:10:59PM +0000, Robert Swain wrote:
>>>> 2009/1/26 Diego Biurrun <diego at biurrun.de>:
>>>>> I say we make a list of things that need to be done before
>>>>> releasing.
>>>>> This has to be a short list.  We fix those issues, hold off on
>>>>> committing
>>>>> non-bugfix changes for a few days, test what we can and then we
>>>>> make a
>>>>> release.
>>>>
>>>> Regressions tests fail on OS X.
>>>>
>>>> I use i686-apple-darwin9-gcc-4.2.1 (GCC) 4.2.1 (Apple Inc. build
>>>> 5566)
>>>> and Mike uses gcc 4.0.1, build 5484 from Apple. I experience the  
>>>> same
>>>> errors.
>>>>
>>>> http://fate.multimedia.cx/index.php?test_result=7861639 (x86_64)
>>>> http://fate.multimedia.cx/index.php?test_result=7861135 (x86_32)
>>>>
>>>> How can I help?
>>>
>>> Find out which revision broke it.
>>
>> For reference, this is due to floating point rounding errors.
>> Adding -mfpmath=x87 to cflags should cause make test to pass on OS X/
>> x86.
>
> Rob, can you verify this?
>
> Does this flag have any bad side-effects?


hm, I meant -mfpmath=387 of course. OS X uses -mfpmath=sse by default  
(as should x86_64 in all cases according to gcc docs, so I dunno why  
Linux/x86_64 regtests are passing on fate...)

Anyways, it should be slower (though probably not much since most of  
the speed-critical float code is asm) so I'd recommend only adding it  
to fate and not to configure. Of course the real solution is a better  
way for regression tests to check code using floating point ops...




More information about the ffmpeg-devel mailing list