[MPlayer-dev-eng] [PATCH] remove mp3lib

xi ctrl.alt.sup at free.fr
Thu Sep 22 13:03:08 CEST 2011


Reimar Döffinger wrote:
> On Thu, Sep 22, 2011 at 02:34:21AM +0200, xi wrote:
>> Erik Auerswald a écrit :
>>> Hi Xavier,
>>>
>>> On Wed, Sep 21, 2011 at 11:35:31AM +0200, xi wrote:
>>>> Dan Oscarsson wrote:
>>>>> ons 2011-09-21 klockan 00:18 +0000 skrev Carl Eugen Hoyos:
>>>>>> Reimar Döffinger<Reimar.Doeffinger<at>    gmx.de>    writes:
>>>>>>
>>>>>>> I am still interested in seeing performance numbers and statistics from
>>>>>>> ffmp3 vs. mp3lib when ffmp3 is slower,
>>>>>> When I tested (after FFmpeg optimisations were committed), ffmp2float was faster
>>>>>> on systems where it does not matter (including 64bit C, SSE2 and Altivec), but
>>>>>> slower than internal libmp3 on systems where I believe it does make a
>>>>>> difference, namely 32bit MMX and SSE.
>>>>> Does it make a difference that is significant?
>>>>> I have used mplayer on a 32bit system for many years together with
>>>>> vdpau. As vdpau does take nearly no CPU time, there was always lots of
>>>>> CPU available for audio decoding. It may be different if you do not have
>>>>> hardware video decoding, but sometime we have to decide if the work to
>>>>> support old hardware is worth the extra mess in the code.
>>>>> There are a lot of ancient drivers that would be nice to get rid of.
>>>>>
>>>> Hey, please don't drop support for old hardware. MPlayer has always been
>>>> the best for playing the videos on my "old" hardware ; as a MPlayer
>>>> user, I wouldn't like this to change.
>>> Could you test the speed of mp3 decoding on your systems?
>>>
>>> Options to use:
>>>
>>> # for mp3lib
>>> -benchmark -vo null -ao null -afm mp3lib
>>>
>>> # for ffmp$WHATEVER
>>> -benchmark -vo null -vo null -afm ffmpeg
>>>
>>> You might need to additionally add -format s16le to force the same output
>>> format for a fair comparison.
>>>
>> Ok, here are the results :
>
> I would welcome performance measurements from you, too
> (perf record/perf report for example).
> Carl will need a while longer before he can provide any.

No problem, however I don't know at all how to use the "perf" tool (I 
have just discovered it).

I suppose I have to run "perf record mplayer -benchmark..." ?
And then, what about perf record which seems to open perf.data file ? 
(An example would be helpful).

You need it for each of the 6 tests from yesterday ?

Xavier


More information about the MPlayer-dev-eng mailing list