What logic should be behind this?
So, what does a dumb user like myself, who compiles this stuff nightly, need to do?
Switch branches to radeon-testing? Stick with drm-next?
I guess drm-radeon-testing is what I want...
@Qaridarium: KDE 4.4 (released in about two months) does not really use take advantage of OpenGL 2 yet. KDE 4.5 however will shine with OpenGL2/3 and that stuff is already heavily under development.
No, it works for radeon as well.
Originally Posted by DanL
Since radeonhd doesn't work with KMS it's likely that radeon + KMS will be the driver to use to get the most GL features.
Firefox has a built-in spell checker, you may want to give it a try...
Originally Posted by Qaridarium
DanL was probably trying to say "it works for HD-xxxx hardware", ie r6xx and up.
Originally Posted by legume
I believe similar work is underway for 3xx-5xx, however -- see nha's r300 work back in early October.
Last edited by bridgman; 12-03-2009 at 12:08 PM.
Ahh, of course, sorry DanL I can't see the word radeonhd without thinking of the driver.
Originally Posted by bridgman
Richard is working full time on GLSL. Like most developers, he would much prefer to polish and finalize the code before committing it to a public repo, but if he pushes out his latest working code at least once a week then I don't bug him for weekly status reports. Writing code is always more fun than writing status reports, so once a week out comes the new code
Originally Posted by monraaf
There's a large element of "learn as you go" here, so status reports would be "here's what works" rather than "here's what has to be done before I'm finished" anyways. Most of the work here is in that nasty little area between "it's finished in theory" and "it's finished in practice", more so than with most development.
The roadmap is pretty simple though...
- get the simple tests and demos working properly
- get the rest of the tests and demos working properly
- if application bugs occur try to replicate the problem in a simple test then fix, hopefully leaving a test behind for next time
- if all else fails, debug application issues directly
I guess we're somewhere in the second step right now.
One thing I haven't had time to do is build estimation models specific to open source development work. So far I have been using models from commercial development projects, which seem to be pretty good at a "whole project" level but don't work so well between milestones. My impression is that initial progress (from start to "testable") tends to be slower for open source drivers because the pre-existing code base is often both older and less well documented, but there's a point where community developers are able to pile on and start investigating specific issues so progress zooms ahead much more quickly than you would see with a fixed-team-size commercial project.
The other estimating challenge right now is that there are a lot of interesting things going on with the driver stack, so community developers are spread across more projects than usual. I don't have a good model for "shininess" (ie project X being sufficiently appealing that community developers work on X rather than Y or Z), but that seems to be one key to making useful schedule estimates. We're also seeing some very capable new developers appear seemingly out of nowhere and start diving into very complex stuff -- and I don't have an model for the spontaneous arrival of new developers either
If one person does all the work these things take a *long* time, so we try to focus our developers on areas where working inside AMD makes a difference, ie getting new hardware functionality working for the first time and leaving working code and/or documentation so the community can pick up and run with it.
Last edited by bridgman; 12-03-2009 at 01:02 PM.