[MPlayer-users] MPlayer 0.90rc4, radeon 9000, MC, IDCT

[D Richard Felker III]

On Sun, Feb 23, 2003 at 11:47:06PM +0100, Zbigniew ?uszpi?ski wrote:
> [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> Dnia nie 23. luty 2003 02:32, D Richard Felker III napisa?:
> > [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> >
> > On Tue, Feb 18, 2003 at 10:55:11PM +0100, Zbigniew ?uszpi?ski wrote:
> > > [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
> > > Hello !
> > >
> > > How should MPlayer be configured to use radeon's 9000 Motion Compensation
> > > and IDCT ? Vidix uses only YUV conversion and hardware scalling.
> >
> > It can't, and won't be able to. They're not useful.
> 
> <g> It is useful for people who do not have Pentium 9 with 50 THz clock and 
> ExtraMMX7ProExtended, with 512 TB RAM and 3000 kW power supply. <g>
> 
> To be serious, I think that manfacturers of graphics chips and programmers of 

I'm glad you weren't being serious with that last claim. I would agree
with you if you needed an insanely overpowered box for decoding...
Maybe you *DO* need that on Windows, since the player programs are all
horribly inefficient, but MPlayer is a plenty fast to play movies on a
high-end K6-2. And if your box is much slower than that, IT PROBABLY
DOES NOT HAVE AGP, in which case you can't use one of these silly
cards with hardware MC+IDCT!

> well known DVD players for windows  would not waste time for implementing 
> hardware acceleration of idct and mc if  it was redundant. Hardware 
> acceleration of idct and mc can save procesing power of CPU which can br used 
> for e.g. DTS or AC3 decoding, picture post-processing, to lower power 
                                ^^^^^^^^^^^^^^^^^^^^^^^

No, this is blatently NOT TRUE. You cannot do postproc if the MC and
IDCT are on the video card! Well, you *could*, but copying the picture
back from the video card to system mem would be 10x slower than just
doing the MC and IDCT in system mem to begin with, since reading from
video mem is prohibitively slow. This is one of the main reasons
hardware MC+IDCT is stupid.

> consumption. IDCT and MC steals 60% of processing power of CPU that is why it 

Must be a horrible inefficient implementation. Total video time on my
K6-3+ is usually under 60% for all of decoding, *and* the write to
video memory, for high-resolution 4:3 movies. On a more modern system,
or with wide-screen aspect (e.g. 640x304), this would be more like
5-20%.

> is accelerated by graphics chip. I know from radeon driver programmers that 
> writing idct and mc acceleration driver is very, very difficult. So I think 
> that in the future someone will face this task.

Not in MPlayer. It's nonsense and does not integrate with the
architecture of a full-featured, general-purpose player at all...

Rich