More Radeon Driver Changes Queued For Linux 3.19

[curaga]

It might be true for integrated or dedicated GPUs in an laptop or motherboard.
It isn't always the performance that bothers me but the fact that some games doesn't run at all (runs great with catalyst), even games that run great on "real" graphics cards with radeon.
I do like the progress of AMD open source graphics driver and i use it as my main driver but for many games i still need catalyst.
I do think that the small integrated and dedicated GPU still needs much love.

Be the love you want to see in the driver!

[gutigen]

Radeon needs OpenGL compliance, power/fan management and maybe crossfile support a lot more than it needs more performance right now.

And on-disk shader cache to at least lower loading times which are quite huge in comparison to Catalyst and Nvidia.

[geearf]

And on-disk shader cache to at least lower loading times which are quite huge in comparison to Catalyst and Nvidia.

Intel was working on one: http://www.phoronix.com/scan.php?pag...tem&px=MTcwNzk
But what happened to it?

[gutigen]

Intel was working on one: http://www.phoronix.com/scan.php?pag...tem&px=MTcwNzk
But what happened to it?

I keep asking this question here and at reddit and no one knows :/

[Kraut]

And on-disk shader cache to at least lower loading times which are quite huge in comparison to Catalyst and Nvidia.

Should be really easy to implement if you have ARB_get_program_binary.
I'm just not sure what "GL_ARB_get_program_binary DONE (0 binary formats)" in http://cgit.freedesktop.org/mesa/mesa/tree/docs/GL3.txt is supposed to mean.

[geearf]

I keep asking this question here and at reddit and no one knows :/

Sad :/
It seemed like a good thing and patches were submitted, of course that does not mean they were good enough..

[smitty3268]

I keep asking this question here and at reddit and no one knows :/

That guy submitted some initial patches, and basically was told to completely rewrite how everything worked by the Mesa devs.

No one heard anything more for about 9 months, and then he came back with another big patchset.

Although it was a lot closer, the Mesa devs again gave him a lot of comments about changes they wanted.

And now no one has heard anything more for about 6 months, or whenever that came up.

So, presumably the guy is sort of working on it in his free time whenever he gets a week free, but that doesn't happen very often. Or maybe he's just given up. No one knows.

[smitty3268]

Should be really easy to implement if you have ARB_get_program_binary.
I'm just not sure what "GL_ARB_get_program_binary DONE (0 binary formats)" in http://cgit.freedesktop.org/mesa/mesa/tree/docs/GL3.txt is supposed to mean.

The spec says the method is allowed to say that the cached file is outdated, and therefore return NULL and the implementation will run the shader as normal.

That's been abused a bit by the Mesa devs, who have added the necessary method calls but simply always return NULL as if the cache was outdated.

That allows them to pass any checks for whether the extension is there, and work correctly, but it also means that nothing will ever get cached which is the entire point of the extension. So it's just a technicality letting them pass but not giving them any of the benefits.

I think the reasoning behind that was it's a required extension for GL ES 3 support, which they wanted, but no one wanted to actually spend the time implementing the hard parts if they didn't have to.

[geearf]

That guy submitted some initial patches, and basically was told to completely rewrite how everything worked by the Mesa devs.

No one heard anything more for about 9 months, and then he came back with another big patchset.

Although it was a lot closer, the Mesa devs again gave him a lot of comments about changes they wanted.

And now no one has heard anything more for about 6 months, or whenever that came up.

So, presumably the guy is sort of working on it in his free time whenever he gets a week free, but that doesn't happen very often. Or maybe he's just given up. No one knows.

Thanks for the information!
Though, I suppose it doesn't really give us a timeline in the end

[V10lator]

And on-disk shader cache to at least lower loading times which are quite huge in comparison to Catalyst and Nvidia.

Don't put all your hopes on that. A cache helps a bit but it doesn't seem to be the main bottleneck. I know that cause I'm one of the guys who tested the last round of patches and these where my results (loading time of Left 4 Dead 2) :

Empty cache: 54 secs
Cache on HDD: 52 secs
Cache on SSD: 48 secs
Cache on ramdisk: 48 secs

The tested system:
GPU: Sapphire Radeon HD 6950 (Dirt II Edition) with unlocked shaders
CPU: AMD Athlon(tm) II X3 455 overclocked to 3,5 GHz
HDD: WDC WD2002FAEX-007BA0
SSD: Corsair CSSD-R120GB2

See also: http://lists.freedesktop.org/archive...ne/061299.html