[FFmpeg-devel] FFMpeg on SH4 (STi7109)

[Måns Rullgård]

M?ns Rullg?rd wrote:
> "Jean-Michel Hautbois" <jhautbois at gmail.com> writes:
>
>> 2008/11/18 Vitor Sessak <vitor1001 at gmail.com>
>>
>>> Hi
>>>
>>> Jean-Michel Hautbois wrote:
>>>
>>>> It runs, on the CPU. Now, the big work will start !
>>>>
>>>
>>> Nice. Does "make test" pass?
>>
>> I don't know, I will try next week I think (and hope).
>>
>>> The "right" way is to make the optimizations in qpel.c compile and work.
>>> Did you try something in the line of the attached patch (completely
>>> untested, I do not have a cross-compiler installed)?
>>
>> I tried something like that, an it didn't solve everything.
>>
>> I will go really further, inclunding QPel optimizations, before
>> submitting anything.
>
> I have fixed the code so it builds out of the box now.  It also works
> to the extent qemu allows me to test it (qemu sh4 support is buggy).

*Now* it should be working.  I fixed a few more bugs, both in FFmpeg
and qemu.

> Could I ask you to run some simple benchmarks with and without the
> functions from qpel.c and dsputil_align.c?  I have the feeling those
> optimisations are not all that optimal when all comes around.  A
> comment in qpel.c talks about *p++ vs. p[n] to access arrays, and most
> functions in this file seem to be the plain C functions rewritten to
> use *p++ access.  There is, however, very little difference in the
> generated code.

I'd still like to see some benchmarks.

-- 
M?ns Rullg?rd
mans at mansr.com