[RFC 0/8] rmk's Dove DRM/TDA19988 Cubox driver

[Russell King - ARM Linux]

On Fri, May 17, 2013 at 07:40:23PM +0200, Jean-Francois Moine wrote:
> Maybe I did not explain correctly: the colored cursor maybe RGB888 +
> transparency (64x64) or full ARGB (64x32 or 32x64). I coded the first
> case. And, yes, I better like a hardware cursor: it asks for less
> computation, and I get it immediately at graphic starting time!

Having looked at this now, using the RGB+transparency is less than ideal
because we're having to reduce an alpha channel down to a simple on/off
transparency.  X cursors really are alphablended components!

So, the options here are:
(a) use "software" rendered cursor
    + correct and expected cursor size
    + correct rendering
    + possible to use the GPU (I believe mine does)
    - maybe time consuming as it has to be removed/replaced on the screen
(b) use RGB+transparency for 64x64 hardware cursor
    + correct and expected cursor size
    + does not have to be removed/replaced when screen contents change
    - incorrect rendering due to reducing the alpha channel to a simple
      on/off transparency mask
    - has to be reloaded when the pointer is close to the edges of the
      screen which is CPU intensive
    - cursor image data passed into DRM is required to be ARGB (I've
      discussed this with David Airlie last night.)  This means we have
      to do translation to RGB+T in the kernel which is *not* nice.
(c) use ARGB 32x64 or 64x32 hardware cursor
    + does not have to be removed/replaced when screen contents change
    + correct rendering of cursor
    - unexpected cursor size; user clients do not expect to be restricted
      to 32 rows or 32 lines of cursor
    - can only select maximum cursor size on initialization of hardware
      cursor; can't dynamically switch between 32x64 and 64x32 sizes
    - has to be reloaded when the pointer is close to the edges of the
      screen which is CPU intensive

While I would like to have hardware cursor support, I don't think it's
worth the effort.  The one which tips it for me is the need to reload
the cursor near the edges - note the restriction in the documentation
that the cursor position + cursor size can't be outside the active
size.  Also, cursors generally have 7 or so transparent pixels to the
left of them (which of course changes depending on the cursor shape.)

It's also less CPU intensive to (probably) allow the GPU to render it
than it is for the CPU to deal with these hardware restrictions.

I would've liked to see hardware cursor support, but I think what I'm
going to be doing is stripping the cursor code out of the main driver
into a separate optional patch which will ultimately be dropped.