Colorspace adjustment

[Alex Deucher]

On Mon, Oct 20, 2014 at 3:01 PM, David Harks <dave at dwink.net> wrote:
>
> On Oct 20, 2014, at 1:09 PM, Alex Deucher <alexdeucher at gmail.com> wrote:
>
> On Fri, Oct 17, 2014 at 7:36 PM, David Harks <dave at dwink.net> wrote:
>
> All,
>
> In the proprietary driver, it’s possible to set the colorspace used for
> output, with 4 possible options on my hardware: YCbCr 4:4:4, YCbCr 2:2:2,
> RGB Limited, and RGB Full. For a system I’m working on, we have been using
> the YCbCr 4:4:4 color space for compatibility reasons with some of the
> devices we output to, and I’d like to get the same feature going on the
> radeon driver.
>
> From what I’ve read
> (http://lists.freedesktop.org/archives/xorg/2014-March/056478.html), this
> feature isn’t yet part of the radeon driver, but since it seems to be a
> register setting it may be possible to add. So I took a look at the code in
> dce5_crtc_load_lut() and modified it to use NI_OUTPUT_CSC_YCBCR_709,
> compiled, and tried it.
>
> It didn’t seem to change the output at all.
>
> So I grabbed a copy of radeontool, and tried to see if I could inspect the
> registers to see if my changes were taking effect, but I think I’m missing
> something because the values I get don’t seem to be what I’d expect:
>
> avivotool regmatch 0x68f0 # NI_OUTPUT_CSC_CONTROL +
> radeon_crtc->crtc_offset
> 0x68f0 0x01000020 (16777248)
>
> I would expect the output to be something with 0x33, since
> (NI_OUTPUT_CSC_GRPH_MODE(NI_OUTPUT_CSC_YCBCR_709) |
> NI_OUTPUT_CSC_OVL_MODE(NI_OUTPUT_CSC_YCBCR_709)) should be 0x33.
>
> Is there some other register I might need to touch in order to make these
> changes take effect? Any pointers to what else I should be looking for?
>
>
> What asic are you using?  Note that register bar moved on Sea Islands
> and newer asics and I'm not sure if the version of avivotool you are
> using handles that correctly yet.  Also note that these registers are
> instanced (hence the radeon_crtc->crtc_offset being added to the
> register offset), so make sure you are reading the appropriate
> instance for the display controller you are using.
>
>
> it’s an AMD E2-1800 APU, AKA a Radeon HD 7340 (Wrestler), which I think gets
> classified as a PALM. So I suppose it actually falls under the “Evergreen”
> moniker, which branches to the dce4_crtc_load_lut() code, which in turn
> looks to have a completely different machinery for this purpose, if any.
>
> Under reg_srcs/evergreen, I see a CB_COLOR_CONTROL entry, but that’s just a
> wild guess. Any pointers to where else I should be looking?

I think it was a little more complex on DCE4.  I'll have to try and
dig up the info and get back to you.  The CB_* registers are part of
the 3D pipeline rather than the display pipe so they are not relevant
here.

Alex