[Ffmpeg-devel] [PATCH] SSE counterpart of ff_imdct_calc_3dn2

Guillaume POIRIER gpoirier
Tue Sep 19 17:22:47 CEST 2006


Michael Niedermayer a ?crit :
> On Tue, Sep 19, 2006 at 01:25:14PM +0200, Guillaume POIRIER wrote:
>> On 9/19/06, Zuxy Meng <zuxy.meng at gmail.com> wrote:
>>> 2006/9/19, Guillaume Poirier <gpoirier at mplayerhq.hu>:
>>>> Diego Biurrun wrote:
>>>>> On Thu, Sep 14, 2006 at 11:02:48AM +0200, Guillaume POIRIER wrote:
>>>>>> On 9/14/06, Zuxy Meng <zuxy.meng at gmail.com> wrote:
>>>>>> Well, it still needs -fomit-frame-pointer to get compiled at all, so I
>>>>>> don't think it's good to go, unless we add the requirement that FFmpeg
>>>>>> needs to be compiled with -fomit-frame-pointer.
>>>>>
>>>>> Wasn't this just on your specific gcc version?
>>>> No, I've checked with GCC 3.3, 4.0.3, 4.2.x (from svn) and they all
>>>> fail on the same code on IA32.
>>> IIRC another thread has pointed out that even ff_imdct_3dn2.c itself
>>> won't compile without -fomit-frame-pointer....
>> That's possible, but I can't test. I don't have a 32-bits machine with
>> 3dnow2 support in it...
> 
> you dont need one for _compiling_

Woops, yeah, Off course. :-(

I just test it, and this code doesn't need -fomit-frame pointer to get 
compiled, so I suppose compiling -fPIC shouldn't be a problem.


>> Michael, what's your opinion about it?
>>
>> Should lavc's makefile append '-fomit-frame-pointer' to compile
>> libavcodec/i386/fft_3dn2.c as well as ff_imdct_calc_sse.diff?
>>
>> It's a bit of a hack, but it would work.
> 
> hmm, dunno, what about PIC?
> 
> if noPIC works and PIC needs -fomit-frame-pointer then ive no objections
> but if noPIC needs -fomit-frame-pointer then PIC as it has 1 register less
> shouldnt work even with -fomit-frame-pointer and thats IMHO not so ideal


Ok, I'm not sure I fully understand, but here's the situation put in a 
nutshell:
I can't get the code to compile _at all_ with -fPIC, with or without 
-fomit-frame-pointer.
The code compiles _without_ -fPIC _only_ if I append -fomit-frame-pointer.

Michael, do you see a way to fix this (or work around it)?

Guillaume




More information about the ffmpeg-devel mailing list