Re: PPC r300 texture

[Jerome G. <j.g...@gm...>]

On Mon, 14 Feb 2005 10:02:07 -0500 (EST), Vladimir Dergachev
<vo...@mi...> wrote:
> 
> 
> On Mon, 14 Feb 2005, Jerome Glisse wrote:
> 
> >> On Mon, 14 Feb 2005 17:17:05 +1100, Paul Mackerras <pa...@sa...> wrote:
> >>> Vladimir Dergachev writes:
> >>>
> >>>>   I am a bit puzzled with your patch - do you really need to hard code
> >>>> format field in r300_state.c ?
> >
> > Ok i think i understand, Mesa tell us that the format is
> > MESA_FORMAT_RGBA8888_REV on big endiand and
> > MESA_FORMAT_RGBA8888 on little endian. Thus if i change
> > {0xffffff01, 0},
> > to
> > {R300_EASY_TX_FORMAT(Z, Y, X, W, W8Z8Y8X8), 0},
> >
> > it works. Vlad does this lead to some strange things on x86 ?
> 
> The reason these entries are as they are is because I (and probably no one
> else yet) was not able to trigger them with any test program.
> 
> Of course, it must be possible to debug them by writing a custom program,
> but I did not have time for that yet..

Yes but my understanding is that mesa will never give you
MESA_FORMAT_RGBA8888_REV on x86 because this tell you
that data are in reverse order (see mesa texformat). So i think
that all odd entry in this table are for REV order (which we will
found on big endian) and all even entry are for "normal" order
x86. I think that you can ask mesa to give you texture in REV
order on x86 but don't know how.

I am really want to commit this simple patch, maybe someone
with a better understand of mesa & who know radeon driver
could tell us if i am wrong. More over if you look at all table
(r200 table or table0 in r300) you will see that allmost all
odd entry have the _REV suffix.

So every even entry for you x86 and odd for us ppc ;)

By the way with which program do you test :
{0xffffff11, 0},
->{R300_EASY_TX_FORMAT(Y, Y, Y, X, Y8X8), 0},
{0xffffff13, 0},

the 0xffffff13 would be
{0xffffff11, 0},
{R300_EASY_TX_FORMAT(Y, Y, Y, X, Y8X8), 0},
{R300_EASY_TX_FORMAT(Y, Y, X, Y, Y8X8), 0},

But testing if the result is good is better :)

Jerome Glisse