Re: [PATCH v4 03/17] drm/uAPI: Add "active bpc" as feedback channel for "max bpc" drm property

From: Werner Sembach
Date: Wed Jul 14 2021 - 14:19:05 EST

Am 01.07.21 um 13:30 schrieb Werner Sembach:
Am 01.07.21 um 09:42 schrieb Pekka Paalanen:
On Wed, 30 Jun 2021 11:42:10 +0200
Werner Sembach <wse@xxxxxxxxxxxxxxxxxxx> wrote:

Am 30.06.21 um 10:21 schrieb Pekka Paalanen:
On Tue, 29 Jun 2021 13:02:05 +0200
Werner Sembach <wse@xxxxxxxxxxxxxxxxxxx> wrote:
Am 28.06.21 um 19:03 schrieb Werner Sembach:
Am 18.06.21 um 11:11 schrieb Werner Sembach:
Add a new general drm property "active bpc" which can be used by graphic
drivers to report the applied bit depth per pixel back to userspace.

While "max bpc" can be used to change the color depth, there was no way to
check which one actually got used. While in theory the driver chooses the
best/highest color depth within the max bpc setting a user might not be
fully aware what his hardware is or isn't capable off. This is meant as a
quick way to double check the setup.

In the future, automatic color calibration for screens might also depend on
this information being available.

Signed-off-by: Werner Sembach <wse@xxxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/drm_connector.c | 51 +++++++++++++++++++++++++++++++++
include/drm/drm_connector.h | 8 ++++++
2 files changed, 59 insertions(+)

diff --git a/drivers/gpu/drm/drm_connector.c b/drivers/gpu/drm/drm_connector.c
index da39e7ff6965..943f6b61053b 100644
--- a/drivers/gpu/drm/drm_connector.c
+++ b/drivers/gpu/drm/drm_connector.c
@@ -1197,6 +1197,14 @@ static const struct drm_prop_enum_list dp_colorspaces[] = {
* drm_connector_attach_max_bpc_property() to create and attach the
* property to the connector during initialization.
*
+ * active bpc:
+ * This read-only range property tells userspace the pixel color bit depth
+ * actually used by the hardware display engine on "the cable" on a
+ * connector. The chosen value depends on hardware capabilities, both
+ * display engine and connected monitor, and the "max bpc" property.
+ * Drivers shall use drm_connector_attach_active_bpc_property() to install
+ * this property.
+ *
Regarding "on the cable" and dithering: As far as I can tell, what the dithering option does, is setting a hardware
register here:

- https://elixir.bootlin.com/linux/v5.13/source/drivers/gpu/drm/i915/display/intel_display.c#L4534

- https://elixir.bootlin.com/linux/v5.13/source/drivers/gpu/drm/i915/display/intel_display.c#L4571

So dithering seems to be calculated by fixed purpose hardware/firmware outside of the driver?

The Intel driver does not seem to set a target bpc/bpp for this hardware so I guess it defaults to 6 or 8 bpc?
Never mind it does. This switch-case does affect the dithering output:
https://elixir.bootlin.com/linux/v5.13/source/drivers/gpu/drm/i915/display/intel_display.c#L4537
Hi,

I obviously do not know the intel driver or hardware at all, but
to me that just looks like translating from bits per pixel to bits per
channel in RGB mapping?
No, if i understand the documentation correctly: Writing bit depth here
with dithering enabled sets the dithering target bpc.
As found in this documentation p.548:
https://01.org/sites/default/files/documentation/intel-gfx-prm-osrc-lkf-vol02c-commandreference-registers-part2.pdf

So max bpc and active bpc are affecting/affected by the bpc after dithering.
By definition, if the cable carries N bpc, then dithering does not
change that. The cable still carries N bpc, but due to spatial or
temporal dithering, the *observed* color resolution may or may not be
higher than the cable bpc.
Yes, and max bpc and active bpc tell the cable bpc ist not the
*observed* bpc.
Of course, if the cable bpc is 8, and dithering targets 6 bpc, then 2
LSB on the cable are always zero, right?
I would assume that in this case only 6 bpc are actually send? Isn't the
whole thing of dithering that you can't send, for example, 8 bpc?
Maybe one would want to do that if the monitor has a 6 bit panel and it
simply ignored the 2 LSB, and the cable cannot go down to 6 bpc.
Is there dithering actually doing this? aka is my assumption above wrong?

AMD code that confused me before, is hinting that you might be right:
https://elixir.bootlin.com/linux/v5.13/source/drivers/gpu/drm/amd/display/dc/dce/dce_transform.c#L826

there is a set_clamp depth and a separate DCP_SPATIAL_DITHER_DEPTH_30BPP

So, what does "max bpc" mean right now?

It seems like dither on/off is insufficient information, one would also
need to control the dithering target bpc. I suppose the driver has a
policy on how it chooses the target bpc, but what is that policy? Is
the dither target bpc the cable bpc or the sink bpc?

Needless to say, I'm quite confused.
... We need someone who knows what dithering on intel and amd gpu
actually means.

But I don't want this to become a blocker for this patchset, because if
there is no dithering, which seems to be the norm, the active bpc
property is already really usefull as it is. So add a note to the docs
that the value might be invalid when dithering is active for now?
Hi,

not necessarily invalid. It all depends on how "max bpc" and "active
bpc" are defined.

If they are defined and implemented as "on the cable", then they are
both well-defined and always valid, regardless of what dithering or bit
clamping does, so this is the semantics I'd would prefer. It's clear,
but of course does not tell full story.

When better properties for dithering are added, those can then define
how it works on top of cable bpc, with no impact on "max bpc" or
"active bpc" properties.

So if we cannot tell what "max bpc" is, then "active bpc" should just
be defined as the same thing as "max bpc" affects, and leave the
precise definition of both for later.
But as long as I don't know exactly how dithering is affected by max bpc I can't tell for sure if active bpc is acting
the same. That's why I wrote it is "undefined" to not run into a trap where the actual behavior have to change after the
fact.
If the definition was observed bpc, then we would have problems and
would need to know everything right now. But you can't really make
promises of observed bpc anyway, because you don't know what the
monitor does to the video signal I suppose. Unless you define it "as if
observed through an ideal theoretical monitor" which then gets awkward
to explain.
Yes, that's why I think describing the "raw" signal is the best and leave it to the user to know what his or her Monitor
is making out of it.

Thanks,
pq
New idea: Instead of the "active"-properties with various if cases in the kernel code, there could just be blob properties exposing the hdmi infoframes, hdmi general control packages, dp misc0 and misc1 and dp vsc sdp.

Combined they have all the color information and it is made sure that it's what is actually send to the monitor (I would consider sending something differed then what is told in the infoframes a bug).

They also have built in version numbers, if in the future they contain more information.

Only disadvantage: We leave parsing for human readable output to the userspace.

Alternatively keep the "active"-properties but fill them from the infoframes.

I'm not entirely sure where to do that on amd, because there the infoframes are directly created in the dc code shortly before writing them to the hardware registers and immediately forgotten afterwards. But you still have access to the connector struct from that code so the property could be updated directly there.