"Ronald S. Bultje" <[email protected]> writes:

> Hi,
>
> 2011/9/10 Måns Rullgård <[email protected]>:
>> "Ronald S. Bultje" <[email protected]> writes:
>>> 2011/9/10 Måns Rullgård <[email protected]>:
>>>> Luca Barbato <[email protected]> writes:
>>>>
>>>>> On 8/26/11 3:33 AM, Måns Rullgård wrote:
>>>>>> "Ronald S. Bultje"<[email protected]>  writes:
>>>>>>
>>>>>>> From: "Ronald S. Bultje"<[email protected]>
>>>>>>>
>>>>>>>  From 52.503s (~40fps) to 27.973sec (~80fps) decoding of 480p sintel
>>>>>>> trailer, i.e. a ~2x speedup overall, on a Nexus S.
>>>>>>
>>>>>> [...]
>>>>>>
>>>>>>> --- a/libavcodec/arm/asm.S
>>>>>>> +++ b/libavcodec/arm/asm.S
>>>>>>> @@ -97,6 +97,12 @@ T       add             \rn, \rn, \rm
>>>>>>>   T       ldr             \rt, [\rn]
>>>>>>>   .endm
>>>>>>>
>>>>>>> +.macro  ldr_dpren       rt,  rn,  rm:vararg
>>>>>>> +A       ldr             \rt, [\rn, -\rm]
>>>>>>> +T       sub             \rt, \rn, \rm
>>>>>>> +T       ldr             \rt, [\rt]
>>>>>>> +.endm
>>>>>>> +
>>>>>>>   .macro  ldr_post        rt,  rn,  rm:vararg
>>>>>>>   A       ldr             \rt, [\rn], \rm
>>>>>>>   T       ldr             \rt, [\rn]
>>>>>>> @@ -133,6 +139,12 @@ T       ldrh            \rt, [\rn]
>>>>>>>   T       add             \rn, \rn, \rm
>>>>>>>   .endm
>>>>>>>
>>>>>>> +.macro  ldrb_post       rt,  rn,  rm
>>>>>>> +A       ldrb            \rt, [\rn], \rm
>>>>>>> +T       ldrb            \rt, [\rn]
>>>>>>> +T       add             \rn, \rn, \rm
>>>>>>> +.endm
>>>>>>
>>>>>> Those macro names are very badly chosen.
>>>>>
>>>>> ldrb_post seems in line with ldr_post. ldr_dpren named as ldr_dpre
>>>>> seems equivalent to the other macros using sub. Which names you have
>>>>> in mind?
>>>>
>>>> I'm still trying to figure what "dpren" is supposed to denote.
>>>
>>> d is used for negative, pre is increment index before access. The dpre
>>> macro, however, still changes the argument, as if there were an
>>> exclamation mark (ldr r0, [r1, r2]!), whereas I want to use them
>>> without changing the pointer, so a "no change to pointer" is what the
>>> n stands for.
>>>
>>> d pre n = dpren. Other naming suggestions are welcome.
>>
>> Names should indicate what things do, not what they do not do.
>
> That's not a suggestion. If dpre is used for negative pre with change
> to ptr, isn't dpren useful for negative pre without change to ptr?

No.

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

Reply via email to