Draft document-rendering tiling API to be called from viewer apps

[Michael Meeks]

Hi guys,

On Wed, 2012-06-06 at 16:33 +0100, Iain Billett wrote:
> I haven't experimented but I've heard that not using power-of-two
> lengths is seriously detrimental to performance.

	Must make a mental note to ask a GPU hacker; in fact - lets do that :-)
Luc - sorry to bother you with a quick question. For all these Android /
portable 3D chipsets that you deal with - is it really true that we need
to allocate power-of-two sized, preferably square textures ? [ we're
working on rendering documents to bitmaps to manipulate them for a
LibreOffice android viewer ].

>  Considering we'll be using lots of resources to begin I think we
> should make every reasonable optimisation

	Sure; but say we have a Tablet, mine has 1280x800 resolution, are you
seriously suggesting rounding 1280 up to 2048 ? and 800 up to 1024 ?
giving a 2Mpixel texture instead of a 1Mpixel texture ;-) is that really
going to be more fast/efficient. Presumably having a sane rowstride (eg.
1280) makes a ton of sense, but a power of two in each dimension ?
Anyhow - Luc can give an authoritative answer :-)

> Also, in this arrangement, It would be nice to have the dimensions of
> the "document" area of the texture. Then if the rest of the bitmap is
> fully transparent and the plane is too we can render a "page" alone
> and keep track of dimensions. 

	I guess.

> My main concern with all this is in fact dealing with user interaction
> i.e. mapping measurements in pixels to OpenGL coordinates. This is
> more of a problem for editing, though. 

	Yep.

> Also, are we intending to have continuous scroll documents ( with all
> the pages available via scroll - like most desktop applications ) or
> would we be happy to have users flick (or otherwise) through pages. I
> think the latter would make it easier to manage resources for large
> documents.

	I think we'll want to flick through pages when we have them :-) For
spreadsheets of course we don't really have that (hence liking doubles
for positioning the view-pane incidentally).

	Thanks !

		Michael.

-- 
michael.meeks at suse.com  <><, Pseudo Engineer, itinerant idiot