[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: question on armhf color depth



On Monday 10 April 2017 07:21:45 Mark Morgan Lloyd wrote:

> On 10/04/17 10:30, Gene Heskett wrote:
> > On Monday 10 April 2017 04:55:39 Mark Morgan Lloyd wrote:
> >> On 10/04/17 02:30, Alan Corey wrote:
> >>> I think you can add entries to /etc/fb.modes but it's like making
> >>> old-style modelines, it takes lots of information.  And my old
> >>> buddy xvidtune doesn't work on a Pi.  But you're a tv guy.
> >>
> >> Video modes have been causing me a great deal of pain over the last
> >> few weeks. I could say much more but will keep it short with just
> >> two comments to start with:
> >>
> >> * Look at what's in your X log file and at
> >> /sys/class/graphics/fb0/modes (should apply to most systems that
> >> use standard video).
> >
> > pi@raspberrypi:~/linuxcnc $ cat /sys/class/graphics/fb0/modes
> > U:1366x768p-0
> > pi@raspberrypi:~/linuxcnc $
> >
> > From the log:
> >
> > [    12.657] (WW) Falling back to old probe method for modesetting
> > [    12.657] (EE) open /dev/dri/card0: No such file or directory
> > [    12.657] (WW) Falling back to old probe method for fbdev
> > --
> > [    12.780] (II) AIGLX: Screen 0 is not DRI2 capable
> > [    12.780] (EE) AIGLX: reverting to software rendering
> > [    14.243] (II) AIGLX: enabled GLX_MESA_copy_sub_buffer
> >
> > How can these be fixed? The video's update rate sux.
>
> You'll need to check with somebody who knows the specific RPi/Raspbian
> situation. I suspect it will require non-free drivers... and that's
> non-free as in beer as well as speech.
>
> >> * Use tvservice to get EDID info from your screen and feed it to
> >> edid-decode (systems other than RPi will have some different means
> >> of getting the binary EDID block, e.g. an Odroid uses a modified
> >> U-Boot).
> >>
> >> It's obviously important though to distinguish between
> >> resolution/sync configuration and the colour depth, the latter
> >> being largely determined by available memory.
> >
> > And 128 megs is not enough for 32 bit apparently. It (24 bit) also
> > slows rendering speed noticably.
>
> Stop flailing around and get the EDID info to see what refresh rate
> the screen will offer for any given resolution. Then start off at 16
> bits, try to get that working, and then try 24.
>
> The best I can get at 3840x2160 (?) is 24 fps from an RPi3. 16 bit
> video is OK, 24-bits is marginal (I suspect a problem with mouse
> pointer etc. code), 32 bits has a major impact on system stability
> (i.e. length of time it will run before it's using swap). HDMI is
> designed around a 24-bit data stream and I don't know to what extent
> trying to use 32 confers any advantage, and there's some unpleasant
> maximum pixel rates etc. to contend with. In principle it should be
> possible to tune the RPi modeline to emit sync rates etc. that don't
> exceed the limitations of the clock rate, in practice we've had no
> success with that yet but I'm still to start playing with full
> modelines... note that modeline format is not the classic XFree86 one.
>
> In practical terms the video capabilities of an Odroid C2 are
> substantially better, even if their Ubuntu-derived UI sucks and
> they've got package dependency problems.

I'll nominate that for the basic truth of the month.

I have a c2, and its video IS better, but they've never built an 
rt-preempt kernel, which is the minimum it takes to run LinuxCNC. And 
AFAIK, there has never been a usable spi driver written for it.

When I go on their forum and ask about those 2 items, I'm either ignored 
or told to sod off.

The SPI driver for the pi that one of the fellows on the LinuxCNC mailing 
list wrote, runs at 32 megabits/second, talks in 4 byte packets, and 
talks to a fpga based i/o card with 72 control lines, the first 30 or so 
can be made into fixed functions by loading the correct firmware into 
with the mesaflash utility. Those i/o pins not used become programable 
gpio pins.

That interface card can be programmed for either an EPP parport comm 
channel, or the SPI, which is even faster.  And this i/o card is 
relatively cheap at $62.  But its all 3.3 volt internally, with a 5 volt 
tolerance, and easily destroyed, so I have ordered a box to put the pi 
and the 7i90 in so I can separate the i/o from the motor drivers noises, 
which can exceed 20 volts of induced noise at frequencies of 100MHz or 
more. So line by line I am cobbling up filters and voltage clippers. 
When finished, I'll have the drivers and all their noise in one closed 
box, and the r-pi-3b and interface in a separate box just as well 
sealed.  And I have a wrist strap I put on before touching it as its 
backed up close to the garage door, which has been retro-fitted with an 
additional 2" of blue styro, If my butt slides along that stuff for 2 or 
3 inches, I'm up to several thousand volts, and can blow such stuff up 
instantly.  Wrist strap mandatory IOW.

 > I don't know to what extent 
> that indicates an inherent superiority of the Odroid's Mali graphics
> subsystem over the RPi's Broadcom one, or whether it's fair to say
> that the developer community understands Mali better that Broadcom, or
> if it's just me drawing some wrong conclusions.
>
> The driving force at this end is a colleague with sight problems who
> can only handle a spreadsheet etc. with a large fount and screen.
> Since very often I'm in the position of trying to set up his computer
> remotely when he's plugged it into an HDMI TV I've never seen I'm not
> exactly enjoying the experience.

I don't think I'd exactly enjoy that either. And a spreadsheet w/o a 
mono-spaced font isn't exactly enjoyable either.  We have a truely 
excellent mono-spaced font, called "hack". Just came out circa a year 
ago. Truetype I think, it scales up great, I've made yard sale signs 
with characters 3 inches high with it, and I'm using it right now in the 
TDE version of kmail-1.9.10.  Your colleague might find its very usable 
for his spreadsheet, or anything else for that matter. I think he'll 
like it, a lot.

re your suggestion to run the edid from tvservice thru edid-ecode, 
edid-decode doesn't exist on the pi. I'll see if synaptic runs any 
better on a 24 bit screen and see if I can find something similar.

If it will go thru the mail server, its also attached. 32 bytes.

Thank you Mark.

Cheers, Gene Heskett
-- 
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
Genes Web page <http://geneslinuxbox.net:6309/gene>

Attachment: aocedid
Description: Binary data


Reply to: