[PATCH xserver v2] xf86cmap: Use old palette system for pseudocolour.

[Maarten Maathuis]

On Fri, Feb 11, 2011 at 7:53 PM, Keith Packard <keithp at keithp.com> wrote:
> On Fri, 11 Feb 2011 13:16:38 +0100, Maarten Maathuis <madman2003 at gmail.com> wrote:
>> 2011/2/11 Keith Packard <keithp at keithp.com>:
>
>> I'll admit i wasn't thinking about "odd" visuals when i did this and
>> at the time the review process of patches certainly wasn't what it is
>> today.
>
> I think the main issue is that the 'RandR' code lives in parallel with
> the 'classic' code and so we end up supporting essentially two driver
> stacks at this point. At some point, it would be lovely to figure out a
> plan to clean all of this up.
>
> The basic problem is that the driver now exposes two parallel colormap
> setting APIs -- the per-CRTC 'gamma_set' function and the per-screen
> 'LoadPalette' function. Neither of these is the right API -- gamma_set
> loads the whole colormap instead of just a piece (which would make
> StoreColors really slow). The DRM kernel interface also deals only with
> the whole colormap though.
>
> Go check out I830LoadPalette though -- as far as I can tell, it only
> works when called to write the whole colormap, so it should only work
> for StaticGray, StaticColor and TrueColor. Any other visual type will
> cause uninitialized values to get loaded into the LUT.

There are dynamic colormaps?

>
> To recap, it looks like fixing this requires:
>
>  1) Fixing the gamma_set driver API to deal with partial updates. I'd
>    suggest that we add a new driver API (set_lut) that provides for
>    partial LUT writes and emulating that with gamma_set for existing
>    drivers. We can consider renaming 'get_gamma' at some point, but
>    I don't see any reason to change its signature.
>
>  2) Emulate the new 'set_lut' function on drivers that don't support it
>    by calling the existing LoadPalette function when asked to write
>    values to the 'compat_output' crtc.
>
>  3) Rewrite the existing colormap code to do per-crtc LUT computations
>    and use the 'set_lut' function to write those to the hardware.
>
> Drivers can be migrated to the 'set_lut' API once we fix the kernel API
> to support partial LUT updates.
>
> I'd love to hear a simpler plan...
>
> --
> keith.packard at intel.com
>



-- 
Far away from the primal instinct, the song seems to fade away, the
river get wider between your thoughts and the things we do and say.