[FFmpeg-devel] [PATCH][VAAPI][2/6] Add common data structures and helpers (take 5)

[Gwenole Beauchesne]

On Tue, 10 Mar 2009, Gwenole Beauchesne wrote:

> On Sat, 7 Mar 2009, Michael Niedermayer wrote:
>
>>>>> +    }
>>>>> +    memcpy(rds->slice_data + rds->slice_data_size, buffer, size);
>>>>> +    rds->slice_data_size += size;
>>>> 
>>>> also whats the point in allocating a buffer and memcpy into it that is
>>>> compared to using the original?
>>> 
>>> We can be memcpy()'ing into mapped memory from HW accelerator. For the
>> 
>> i dont mind memcpy() into memory from the accelerator, i do mind if this
>> could be memcpy into normal memory, it simply seems unneeded in that case
>> ...
>
> Dropped all memcpy() but there is no striking difference in performance...
>
> On a 1080p H.263 sample (Grey.ts):
> - before: A:40347.6 V:40348.9 A-V: -1.317 ct:  1.247 904/904 13%  1%  3.1% 8 
> 0
> - after : A:40347.6 V:40348.9 A-V: -1.317 ct:  1.247 904/904 13%  1%  3.0% 8 
> 0
>
> On a 1080p MPEG-2 sample:
> - before: A:24956.4 V:24956.5 A-V: -0.047 ct: -0.862 828/828 18%  2%  4.7% 5 
> 0
> - after : A:24956.4 V:24956.5 A-V: -0.046 ct: -0.862 828/828 19%  1%  4.7% 6 
> 0

Here is another one:
- Dropped calls to alloca()
- Simplify slice buffer IDs management

I don't think I can make it smaller.