[FFmpeg-devel] [PATCH] remove gcc 3.3 workaround in swscale_template.c

[Benoit Fouet]

On 2009-09-22 16:10, Diego Biurrun wrote:
> On Tue, Sep 22, 2009 at 03:53:19PM +0200, Benoit Fouet wrote:
>> On 2009-09-21 00:16, Diego Biurrun wrote:
>>> On Sun, Sep 20, 2009 at 09:02:46PM +0200, Reimar D?ffinger wrote:
>>>> On Sun, Sep 20, 2009 at 08:00:40PM +0200, Diego Biurrun wrote:
>>>>> On Sun, Sep 20, 2009 at 07:12:44PM +0200, Reimar D?ffinger wrote:
>>>>>> On Sun, Sep 20, 2009 at 06:53:36PM +0200, Diego Biurrun wrote:
>>>>>>> There's an ugly preprocessor gcc 3.3 workaround in swscale_template.c:
>>>>>>>
>>>>>>>   /* GCC 3.3 makes MPlayer crash on IA-32 machines when using "g" operand here,
>>>>>>>      which is needed to support GCC 4.0. */
>>>>>>>   #if ARCH_X86_64 && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 4))
>>>>>>>                   :: "m" (src1), "m" (dst), "g" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask),
>>>>>>>   #else
>>>>>>>                   :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask),
>>>>>>>   #endif
>>>>>>>
>>>>>>> At the very least it should be updated to use the AV_GCC_VERSION_AT_LEAST
>>>>>>> macro from libavutil.  However, I would prefer to get rid of it
>>>>>>> completely.  If I understand the comment correctly, deleting the whole
>>>>>>> #else clause would be the way to achieve this.  Since I know little
>>>>>>> enough assembler to have no real idea what "g" and "m" operands are all
>>>>>>> about I'd like to hear an informed opinion.
>>>>>> Huh? No, the upper line should give faster code but causes crashes due
>>>>>> to miscompilation with gcc 3.3
>>>>>> The lower line does not compile with gcc 4.0 on x86_64 but produces correct code
>>>>>> with 3.3.
>>>>>> At least that's what the comment says.
>>>>> OK, then I propose getting rid of the workaround completely.  I don't
>>>>> think x86_64 users with gcc 3.3 exist, so it's not worth the ugliness on
>>>>> our side.
>>>> Hm? Which do you consider a workaround? I think you might still need two cases,
>>>> one for x86_32 and one for x86_64 for gcc 4.x, you could only get rid of the
>>>> gcc version check.
>>> OK, I propose the attached patch.
>> which results in code as ugly, but with no support for gcc < 3.3 on x86_64
> 
> Excuse me, but how can 
> 
>   #if ARCH_X86_64
> 
> be as ugly as
> 
>   #if ARCH_X86_64 && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 4))
> 
> ?
> 
> You must be deeply confused.
> 

why, no, but thank you for feeling concerned...
it's an ifdeffery anyway; whether it is 15 or 77 characters long, I
don't care...

>> I'd say, if you really want to change that code, change it to use
>> AV_GCC_VERSION_AT_LEAST, and that should do, no ?
> 
> No, that only works for gcc, which sucks.

I don't understand you; isn't the following code working ?
#ifdef ARCH_X86_64 && AV_GCC_VERSION_AT_LEAST(3, 4)

>  Also, we don't know if the
> problem exists for gcc 3.3 only or for gcc <3.3 as well..
> 
> Note that this is the last occurrence of __GNUC__ outside of appropriate
> places.
> 

which can be fixed by my previous remark, unless I'm missing something.

Ben