We're not paranoid; that resistribution restriction on our driver didn't come from us (I work for PowerVR). As a rule of thumb, it's the device or SoC vendor that calls those shots and controls redistribution.

I interviewed once for a graphics driver programming position for Imagination Tech. (who own PowerVR) and my impression was that they were indeed absolutely paranoid about open source. The first thing the guy had a go at me for was my pretty meager contributions to open source. Apparently this was a solid indicator that I would give out all the company secrets!

If the interviewer had a problem with your open source contributions then I'm sorry, since that really shouldn't be any problem for us whatsoever. There is absolutely no feeling at the company that open source equals the leak of secrets, and I find it hard to believe your story because of that.

Thanks for clarifying. I admit that I was very put off when I downloaded the Nexus drivers and had to scroll though all the legalese just to use them (I was actually messing with a Samsung S5P SoC on a different board and I was hoping that I might be able to reuse those drivers there - no dice as per the EULA.).

Is there any reason why PowerVR don't release userland binary libraries for download themselves? Is it just that the libraries are too closely wedded to specific kernel modules and the like? Or specific hardware revisions? (I noticed breaking changes between different version of the pvr kernel module sources.)

Or is it more just that there's no real incentive to bother?

Purely technically, it wouldn't be too difficult for us to release something. The driver tends to be coupled to the SoC, rather than the graphics core (and core revision), due to integration with the SoC's display hardware, so we'd release drivers for a 'platform' rather than our graphics IP. That's what Google did for Nexus S, so we'd follow that model.

The big thing stopping us is support and maintaining the code that does the graphics IP to SoC coupling. We could dump that burden on the SoC/platform vendor (like with Google), but eventually code support just swims upstream to us and we don't have the resources right now to handle that for a large number of SoCs.

Apologies that the EULA prevented you from using those drivers on another S5PC110 platform, I really wish that wasn't the case.

Not all manufacturers are so paranoid. Qualcomm has released the Adreno 2xx GPU drivers for ICS (which is very exciting for those of us with Snapdragons):

https://developer.qualcomm.com/develop/mobile-technologies/g...

I'm actually using these now (courtesy of the developer of my custom ROM) and they are a big improvement over leaked drivers I had been using.

> I don't know why GPU manufacturers are so paranoid.

I've read the reason for the paranoia around GPU drivers has something to do with modern GPUs needing so much controlling software in the driver that the competition could glean details of the hardware design from the driver source code.

I'm unable to find the source of this information. Can anyone privy to these sorts of details confirm or deny this?

Well, I didn't even mean source code (although that would be great.) It's hard enough just to find binary GL blobs that you're legally allowed to pass around.

With respect to the source code, I suspect you're right - and I'd also suspect FUD around the possibility of patent issues when other people can see more of how things are implemented. But IANAL or a GPU manufacturer so I'm just guessing.