Mailinglist Archive: radeonhd (265 mails)

[Matthias Hopf <mhopf@xxxxxxx>]

On Jan 02, 09 01:05:17 -0800, Yang Zhao wrote:
> I've been looking into the LUT loading code over the last few days so
> that Argyll CMS doesn't make a kaleidoscope out of my monitor every
> time I try to calibrate it. Patches are available, but I'm still a
> little confused about the extensions that allow this.
>
> Here's what I've found:
>  - colormap data isn't range masked properly when being written to
> registers: this means funky colours when >10 bits per channel of
> precision is being used in the input
>  - radeonhd advertises 8-bit colormap precision instead of the supported 10

I don't think I known enough about this area - but I assume that Luc
didn't want to dig into details whether 10 bit colormaps are actually
supported correctly by Xorg... CC'ing Luc here.

But from the rest of the comments it looks to me like if you have more
knowledge about that issue than anyone else :-]

> The question is, how is this precision mismatch supposed to be
> handled? Relatedly, what other types of applications would try to call
> LoadPalette() or XF86VidModeSetGammaRamp(), specifically with only a
> partially defined map?
>
> Currently, my approach is to verify that no rows of the input map has
> a value larger than 10 bits, otherwise assuming 16-bit input and
> truncating the LS 6 bits. This seems to work fine for xgamma, as the
> generated colormap uses the correct advertised precision. As for
> Argyll, it seems to always assume 16-bit precision and emit complete
> tables, meaning a row with more than 10 bits of data is always
> present; XF86-VidModeExtension doesn't provide any facilities for
> exposing allowed colormap precision, and I couldn't find how this can
> be queried.

I don't like heuristics that much, that said, I don't see an easy
solution to this dilemma.

> Since the older xf86 extensions never seem to be well documented, I
> have no idea if this is a valid approach. This heuristic breaks if
> only a partial map is given as input, but this is not specifically
> disallowed in the spec that is available. It would also break down if
> none of the rows contain values larger than 0x3FF when 16bit precision
> is assumed, but I'm not sure if this is a practical situation to worry
> about.
>
> I've yet to look at what other drivers are doing.
>
> In any case, at least for 24 bit depth mode, colour calibration with
> Argyll is working.

Thanks

Matthias

-- 
Matthias Hopf <mhopf@xxxxxxx>      __        __   __
Maxfeldstr. 5 / 90409 Nuernberg   (_   | |  (_   |__          mat@xxxxxxxxx
Phone +49-911-74053-715           __)  |_|  __)  |__  R & D   www.mshopf.de
-- 
To unsubscribe, e-mail: radeonhd+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: radeonhd+help@xxxxxxxxxxxx