Radeon Driver Gets Golden Registers In Linux 3.10

[phoronix]

Phoronix: Radeon Driver Gets Golden Registers In Linux 3.10

A second Linux 3.10 Radeon DRM driver pull request was submitted by AMD's Alex Deucher. The pull request sent to Red Hat's David Airlie for the DRM sub-system mentions the "golden registers" addition as being the highlight of this batch of new open-source AMD Linux graphics code...

http://www.phoronix.com/vr.php?view=MTM1Nzc

[agd5f]

So to get these golden registers.. will it be necessary to wait for kernel 3.10 or will an update to Xorg-edgers be sufficient? Might these golden registers fix choppy desktop effects on RS880/integrated 4250?

They only apply to r7xx asics and above so your asic is not affected. I doubt they will make a significant difference in performance. It sounds to me like you are hitting a slow path in the 3D driver. You'd probably have better luck updating to a newer version of the 3D driver. Note that some desktop environments default to picking MSAA visuals which use a huge amount of memory and bandwidth. I suspect you are seeing some variant of:
https://bugs.freedesktop.org/show_bug.cgi?id=61182

[Globule]

So to get these golden registers.. will it be necessary to wait for kernel 3.10 or will an update to Xorg-edgers be sufficient? Might these golden registers fix choppy desktop effects on RS880/integrated 4250?

[GreatEmerald]

@GreatEmerald: You can set other ones or in the same ones in a different order and things will still more or less work. But these are the registers that the hardware team themselves have said "Set these registers in this order and everything will work exactly how it should."

I say 'more or less work' because of the comment above about 'stranger behavior' or 'bugs' on certain cards. Its very possible that some or all of the card specific bugs the FOSS teams have hit are simply because registers are being plugged in the wrong order and hitting unknown and untested register configurations. If thats the case, then using these -correct- registers should automatically fix those bugs.

Clearer now?

Yeap, thanks. Sounds like a nice change.

[bridgman]

That's good to know. Though I'm still curious, what exactly makes those registers "golden"?

It's the values that are "golden", not the registers.

We call them golden because the HW team calls them golden

[Ericg]

golden = optimal

@GreatEmerald: You can set other ones or in the same ones in a different order and things will still more or less work. But these are the registers that the hardware team themselves have said "Set these registers in this order and everything will work exactly how it should."

I say 'more or less work' because of the comment above about 'stranger behavior' or 'bugs' on certain cards. Its very possible that some or all of the card specific bugs the FOSS teams have hit are simply because registers are being plugged in the wrong order and hitting unknown and untested register configurations. If thats the case, then using these -correct- registers should automatically fix those bugs.

Clearer now?

[agd5f]

That's good to know. Though I'm still curious, what exactly makes those registers "golden"?

golden = optimal

[GreatEmerald]

That's good to know. Though I'm still curious, what exactly makes those registers "golden"?

[jrch2k8]

i guess this could be a steping stone for APU reclock fixes later

[bridgman]

What is "d3 cold"?

It's an ACPI device state, as low as you can go IIRC. The device is completely powered down, loses all settings and needs to be completely re-initialized when it comes out of that state.