Hi,

On Sat, Jan 26, 2013 at 10:01 AM, Daniel Kang <[email protected]> wrote:
> On Sat, Jan 26, 2013 at 3:23 AM, Diego Biurrun <[email protected]> wrote:
>> On Sat, Jan 26, 2013 at 12:32:16AM -0500, Daniel Kang wrote:
>>> --- a/libavcodec/x86/dsputil.asm
>>> +++ b/libavcodec/x86/dsputil.asm
>>> @@ -879,3 +884,984 @@ cglobal avg_pixels16, 4,5,4
>>> +
>>> +; HPEL mmxext
>>> +%macro PAVGB_OP 2
>>> +%if cpuflag(3dnow)
>>> +    pavgusb %1, %2
>>> +%else
>>> +    pavgb   %1, %2
>>> +%endif
>>> +%endmacro
>>
>> We have a macro for this in x86util.asm and it works the other way around.
>> I'm very suspicious of this doing the right thing on CPUs with mmxext and
>> 3dnow ...
>
> You're probably right. Fixed.
>
>>> +; mpeg4 qpel
>>> +
>>> +%macro MPEG4_QPEL16_H_LOWPASS 1
>>> +cglobal %1_mpeg4_qpel16_h_lowpass, 5, 5, 0, 8
>>
>> So it seems like dsputil.asm is becoming the new dumping ground for
>> functions of all kind.  It doubles in size after your patch and at
>> around 2k lines it starts to work against our current efforts of
>> splitting dsputil into sensibly-sized pieces.  If you continue your
>> porting efforts, it will probably end up around 5k lines or so.
>>
>> Whenever there is an opportunity to make dsputil less monolithic comes
>> up, we should exploit it.  That seems to be the case here.
>
> I was trying to avoid drama and bikeshedding re: file names and save
> that for another patch. I guess I could split it in this patch if you
> want.

While at it, please split hpel functions to a new file called
hpeldsp.asm. This will make my life slightly easier later on.

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to