[FFmpeg-devel] [FFmpeg-devel-irc] IRC log for 2010-09-08

Ronald S. Bultje rsbultje
Thu Sep 9 23:07:00 CEST 2010


Hi,

On Thu, Sep 9, 2010 at 5:03 PM, M?ns Rullg?rd <mans at mansr.com> wrote:
> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
>
>> On Thu, Sep 09, 2010 at 10:03:39PM +0200, Michael Niedermayer wrote:
>>> On Thu, Sep 09, 2010 at 01:00:59AM +0100, irc at mansr.com wrote:
>>> [...]
>>> > [11:05:50] <cartman> libmpcodecs/vf_yadif.c:262:9: error: ambiguous instructions require an explicit suffix (could be 'cmpb', 'cmpw', 'cmpl',
>>> > [11:05:54] <cartman> ? ? ? or 'cmpq')
>>> > [11:05:59] <cartman> somebody needs to tell michael
>>> > [11:07:03] * funman thinks somebody should be cartman
>>> > [11:07:46] <cartman> funman: he would flame me to hell
>>> > [11:07:47] <cartman> :P
>>>
>>> do you mean this line:
>>> ? ? 19836 ? ? lorenm ? ? ? ? ? ? "cmp ? ? ? $2, %[mode] \n\t"\
>>> ?
>>
>> Hm, I don't think this can work on a 64 bit system.
>> I guess it would be possible to change the constraint to "m" and
>> use cmpl, or use different code for 32 vs. 64 bit, but it all
>> seems ugly to me.
>
> The obvious solution is to move the code to yasm.

I'm starting to think I need to step up and object loudly and vocally
against any more inline asm being committed if the function isn't
actually inlined. Is the code being called here inlined and is the
inlining absolutely required for the performance gain it provides over
plain C?

If not, please port it to yasm before committing.

Ronald



More information about the ffmpeg-devel mailing list