Re: [Linux-fbdev-devel] Re: [PATCH] fbdev: colormap fixes

[Jon Smirl <jonsmirl@gmail.com>]

On 7/29/05, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> On Fri, 29 Jul 2005, Antonino A. Daplas wrote:
> > Jon Smirl wrote:
> > > On 7/28/05, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > > > On Thu, 28 Jul 2005, Jon Smirl wrote:
> > > > > I've verified now that all ATI R300+ chips have 10bit cmaps. These are
> > > > > pretty common so I'd be in favor of making this into a binary
> > > > > attribute where I can get/set the whole table at once. Given that
> > > > > OpenGL is already supporting 12 and 16 bits these tables are only
> > > > > going to get much larger.
> > > > >
> > > > > 1024 entries * 5 fields * 2 bytes = 10KB -- too big for a text
> > > > > attribute.
> > > > >
> > > > > 65536 entries * 5 fields * 2 bytes = 655KB -- way too big for a text
> > > > > attribute.
> > > > >
> > > > > The bits_per_pixel sysfs attribute is an easy way to tell how many
> > > > > entries you need. You can just set it at 4, 8, 10, etc until you get
> > > > > an error. Now you know the max. 2^n and you know how many entries.
> > > > No, bits_per_pixel can be (much) larger than the color map size. E.g. a
> > > > simple
> > > > ARGB8888 directcolor mode has bits_per_pixel = 32 and color map size =
> > > > 256.
> > >
> > > So I have the bits_per_pixel attribute wrong in sysfs. It needs to be
> > > bits_per_color and then let the driver sort it out.  Otherwise there
> > > is no way to set ARGB8888 versus ARGB2101010. With bits per color you
> > > would set 8 or 10.
> >
> > No, you have to add another attribute for {transp|red|green|blue}.{len,offset}
> > and another attribute for the pixelformat. Then using those, one can
> > easily deduce the cmap size.
> 
> Indeed. One bits_per_color cannot handle e.g. RGB565 (or RGBA{10,10,10,2} :-).
> 
> > > If that isn't good enough I can switch the attribute to take strings
> > > like ARGB8888.
> > >
> >
> > Please no.
> 
> Ack.

> 
> > > What do you think, should I just switch to fbconfig names and a binary
> > > cmap attribute?
> >
> > Does a binary attribute not have the same buffer size limitation as
> > the text attribute?  I really don't know, just asking.
> 
> Yes it has, but since binary data is more compact, you can fit more data in
> PAGE_SIZE.

Binary attributes are not limited to page size.

-- 
Jon Smirl
jonsmirl@gmail.com