[Ffmpeg-devel] [PATCH] wma optimizations

[Loren Merritt]

On Sun, 20 Aug 2006, Christian Iversen wrote:
> On Sunday 20 August 2006 19:57, Michael Niedermayer wrote:
>> On Sun, Aug 20, 2006 at 07:47:23PM +0200, Christian Iversen wrote:
>>>
>>> So it's 0.1 clock cycle?
>>
>> yes
>>
>>> I somehow doubt that :-)
>>
>> heretic ...
>
> Well... eh.
>
> I'm just wondering why you don't count in, say, "12000 cycles"?

If it is 12000, there's no reason to use dezi. But if the code being 
profiled is 10 cycles, then premature rounding adds an extra 5% error. 
Note that the timer routines measure an average, so its not guaranteed to 
be a whole number of cycles.

--Loren Merritt