Details On The PS4's Radeon GPU With Linux Driver Modifications

[chithanh]

I'm guessing that the DP is more capable than HDMI port (e.g. HDMI can't do the 3D that Sony needed} and that's why it's used. Or maybe they just wanted to do it weirdly.

After thinking some more, I now believe that the external HDMI encoder's purpose is to give Sony more flexibility without having to make changes to the APU.
For example, they could change it for an HDMI 2.0 capable version (which I think they did in the PS4 slim, not 100% sure).

I wonder though why they did not design the APU with only DisplayPort outputs then to begin with. It must have been a decision at some later stage.

He was asked if they wanted to upstream the code. He commented, that he knew that at least one AMD dev didn't want anything to do with the code. But then again AMD couldn't stop them from upstreaming this the kernel...

While AMD cannot prevent PS4 support code from going upstream, they can block it from entering the radeon/amdgpu driver (precedent here is GMA500/Poulsbo support rejected from i915).

So the code would have to be duplicated in the kernel, creating additional work and making the whole upstreaming proposition less attractive for the hackers.

[Hi-Angel]

precedent here is GMA500/Poulsbo support rejected from i915.

What's sad is that the article didn't have a link to discussion about why the driver was upstreamed separately of i915. I imagine there was really strong technical reasons to not share the code, because it's an open source, only technical reasons matters here.

[Tomin]

What's sad is that the article didn't have a link to discussion about why the driver was upstreamed separately of i915. I imagine there was really strong technical reasons to not share the code, because it's an open source, only technical reasons matters here.

Poulsbo is very different from Intel's other graphics chips since it's actually PowerVR based.
For more information: https://en.wikipedia.org/wiki/Intel_...erVR_GPU_based

[Tomin]

After thinking some more, I now believe that the external HDMI encoder's purpose is to give Sony more flexibility without having to make changes to the APU.
For example, they could change it for an HDMI 2.0 capable version (which I think they did in the PS4 slim, not 100% sure).

That would make sense.

[chithanh]

What's sad is that the article didn't have a link to discussion about why the driver was upstreamed separately of i915. I imagine there was really strong technical reasons to not share the code, because it's an open source, only technical reasons matters here.

Here is the link to the actual statement from Alan Cox:

https://lkml.org/lkml/2011/2/22/279

Poulsbo is very different from Intel's other graphics chips since it's actually PowerVR based.
For more information: https://en.wikipedia.org/wiki/Intel_...erVR_GPU_based

The 3D part is different. But the modesetting part is actually very similar to other Intel chips, which is why code sharing would be technically preferable, but was rejected presumably due to company politics.

[Hi-Angel]

Here is the link to the actual statement from Alan Cox:

https://lkml.org/lkml/2011/2/22/279

The 3D part is different. But the modesetting part is actually very similar to other Intel chips, which is why code sharing would be technically preferable, but was rejected presumably due to company politics.

As Alan said in the linked mail, devs didn't agree to share the code in fear of breaking i915 — from the statement, looks like i915 has to be partially rewritten to abstract away some details. I think, the question was whether it worth the efforts, given that the work on GMA500 is just a hobby of a single guy, who can easily stop the development.

It's hard to see from the mail whether there was anything more, but what I see doesn't quite sound as a political decision, it's about maintainment. Perhaps if Alan would start himself sending little patches for abstracting away details, it'd change something.

[Inopia]

After thinking some more, I now believe that the external HDMI encoder's purpose is to give Sony more flexibility without having to make changes to the APU.
For example, they could change it for an HDMI 2.0 capable version (which I think they did in the PS4 slim, not 100% sure).

It might also handle HDMI-CEC as that's one feature generally missing from AMD GPUs (or any consumer GPU for that matter).

[Tomin]

The 3D part is different. But the modesetting part is actually very similar to other Intel chips, which is why code sharing would be technically preferable, but was rejected presumably due to company politics.

Ah, of course. There is, as usual, hardly anything 3D related in the kernel.

This makes me wonder how much different the user space OpenGL driver would be from other PowerVR products. Probably not much. Then again, I don't think there is any open source driver for any PowerVR graphics so it doesn't matter.

[chithanh]

As Alan said in the linked mail, devs didn't agree to share the code in fear of breaking i915 — from the statement, looks like i915 has to be partially rewritten to abstract away some details. I think, the question was whether it worth the efforts, given that the work on GMA500 is just a hobby of a single guy, who can easily stop the development.

It's hard to see from the mail whether there was anything more, but what I see doesn't quite sound as a political decision, it's about maintainment. Perhaps if Alan would start himself sending little patches for abstracting away details, it'd change something.

I think not, Alan Cox was offered explanations along your lines but did not jump on any of these. Also it was not a matter of let's have this separate for now and look into integrating it later, it was a definite "instructed not to do so."

It might also handle HDMI-CEC as that's one feature generally missing from AMD GPUs (or any consumer GPU for that matter).

Yes, HDMI CEC would be one possible explanation. But AFAIK it is missing because of the extra license cost, and not because it is impossible to implement for AMD. So Sony as the APU customer could have specified HDMI CEC in their requirements - which presumably would have been cheaper than the external encoder chip (whose price still has to include royalties for the HDMI CEC license).

[suberimakuri]

Really interesting to watch. Mad clever. I should look up some presentation videos of the previous console work done. Last one I did was original Xbox. Any recommendations most appreciated!