AMD R600/700 DRM Interrupts Support Pushed

[netkas]

Heh, it was dumb error, since last time .config had radeon & drm builtin. recompiled it as moduels and all is fine now.

Compiz runs with about 0.3 fps with vblank option enabled.

[agd5f]

Only kms support interrupts for r6xx+.

[Pfanne]

The last time I checked, voltages weren't being touched and the dynamic clocking wasn't as advanced as ATI's PowerPlay. Hopefully, we will see better power managment code when the 2.6.33 kernel's merge window opens/closes soon.

like i mentioned earlier in this thread, some of the powermanagement work is dependend on the new interrupt patches.

[DanL]

Are these drivers in a state yet where an HD4870 would run as cool as with fglrx?

The last time I checked, voltages weren't being touched and the dynamic clocking wasn't as advanced as ATI's PowerPlay. Hopefully, we will see better power managment code when the 2.6.33 kernel's merge window opens/closes soon.

[Zajec]

did u read the post ?

I applied it and compiled.

Whoops, I'm sorry! Here you have proper part of dmesg

[   52.431773] [drm] Loading RV620 Microcode
[   52.431780] platform radeon_cp.0: firmware: requesting radeon/RV620_pfp.bin
[   52.479415] platform radeon_cp.0: firmware: requesting radeon/RV620_me.bin
[   52.490488] platform radeon_cp.0: firmware: requesting radeon/R600_rlc.bin

[netkas]

Any reason you believe it to be in mainline?

It's not even in any git drm-2.6 branch. Apply it manually, wait for commiting or wait for .33-rc1.

did u read the post ?

I applied it and compiled.

[Zajec]

no. There are some PM features, but no dynamic reclocking based on workload. There are some experimental patches for that, but nothing went into mainline yet.

Real dynamic PM depends on KMS/DRI to ensure that all access to the hardware goes through the kernel layers (otherwise you wouldn't have dependable metrics about GPU load), so it wasn't feasible to implement while KMS was new.

You may see in patch comments I had some problems syncing with VBLANK interrupt. Now I hopefully know how to do this, so should provide new patch, almost ready to be applied.

[rohcQaH]

Are these drivers in a state yet where an HD4870 would run as cool as with fglrx?

no. There are some PM features, but no dynamic reclocking based on workload. There are some experimental patches for that, but nothing went into mainline yet.

Real dynamic PM depends on KMS/DRI to ensure that all access to the hardware goes through the kernel layers (otherwise you wouldn't have dependable metrics about GPU load), so it wasn't feasible to implement while KMS was new.

[Zajec]

just tried the patch with .32-rc8-git2

opengl apps still complain about IRQ's not enabled,

and there is no message in dmesg about loading new fw (yes, i installed it too)

Any reason you believe it to be in mainline?

It's not even in any git drm-2.6 branch. Apply it manually, wait for commiting or wait for .33-rc1.

[netkas]

just tried the patch with .32-rc8-git2

opengl apps still complain about IRQ's not enabled,

and there is no message in dmesg about loading new fw (yes, i installed it too)