Imagination Releases Full ISA Documentation For PowerVR Rogue GPUs

[_SXX_]

So, some day we will be able to install a (usable) GNU/Linux distro on android phones?

Yes. If Intel's plan to conquest this market would be successful.

[robclark]

It is the first if you use a half dozen qualifiers to make it so.

It is the first generally available ARM based GPU to have opened up ISA documentation that is found in consumer handheld hardware running Android.

one more qualification: opened up ISA documentation *by the vendor* ;-)

there are at least 5 more ISA's for arm/SoC gpus that have been r/e'd (vivante + mali VS + FS + two generations of adreno)

[IsacDaavid]

I think you mean lima driver developer

I was referring to this statement I read in etnaviv's repo; correct me if I'm wrong.

When their OpenGL driver is free software I'll go buy their stuff eagerly to reinforce good behavior. Releasing your ISA just means you aren't a complete asshole. I mean, imagine a CPU without a public ISA spec. I can, and they all died in the 1980s like they deserved.

so much this

[IsacDaavid]

So, some day we will be able to install a (usable) GNU/Linux distro on android phones?

We are able to install proper and native GNU/Linux distros on most Android devices already. All you need is root rights and an unlock-able bootloader. That doesn't mean you'll be able to use all the sensors, radios and graphical accelerators though. When writing their proprietary blobs, manufacturers seldom care about interoperability with non-Android Linux systems; so not even going full subjugated with binary blobs is a solution. libhybris is an effort to solve this problem, as far as using nonfree software is regarded as a solution.

Now, on the GPU side, some FOSS reverse-engineered drivers have become surprisingly performant to the point of being mainlined into Linux and praised by Dolphin developers for its unbelievable superiority over the official drivers, as is the case of freedreno. It's a shame Qualcomm's boards are ridden with other freedom issues, not to mention Qualcomm-based phones have poor isolation between broadband's CPU and the normal CPU, making it virtually impossible to guarantee security and privacy of the host OS.

Yes. If Intel's plan to conquest this market would be successful.

I have higher hopes with AMD. To be fair, AMD has shown stronger commitment to GNU/Linux and free (libre) software and FLOSS in general (coreboot, their CPUs don't require proprietary ucode, proprietary memory initialisation software, no remote kill switch). Their interest in ARM is prominent, I think.

[mlau]

I think you mean lima driver developer. And you are welcome as yes, it is that convoluted.

But! This is just the isa, and they are doing this to finally get some traction for Rogue on OpenCL and AArch64. The convolutedness will still mean that this will not lead to a proper free driver, and yes, no-one should bother trying to create one: the actual hindrance is still there, just as before.

Imgtec really should contact some good open source folks and task them directly, but that does not seem to be their agenda here.

This "release" sounds more like a ploy to get applications optimized for PowerVR, i.e. to lock people in, while keeping all the secrets
secret (or like "our driver is crappy, so here's some specs, please do all the work and optimize your app for our gpu").

[Philip]

It is the first if you use a half dozen qualifiers to make it so.

It is the first generally available ARM based GPU to have opened up ISA documentation that is found in consumer handheld hardware running Android.

[...]

And the open Broadcom driver that the Raspi uses is on an ARM system, but that dev board is not a mobile device.

The driver that Broadcom released was actually from the BCM21553, used in some horrible low-end Android phones (e.g. Samsung Galaxy Ace Duos).

The Rasperry Pi has a significantly different SoC architecture, though it's got basically the same (but much faster) GPU block, so it was possible to port the driver across. Some other more respectable mid-range Android phones (e.g. Samsung Galaxy Grand) have the same GPU too (though yet another significantly different SoC architecture). And Broadcom released documentation of the GPU architecture (including the ISA for the shader units) which applies to all those devices.

So it was definitely "a GPU supported by the Android ecosystem", and I think you'd need to add a few more qualifiers if you want IMG's statement of firstness to be true.

Still, even if they're not first and even if they're only releasing a small amount of information, it's a step in the right direction

[curaga]

This "release" sounds more like a ploy to get applications optimized for PowerVR, i.e. to lock people in, while keeping all the secrets
secret (or like "our driver is crappy, so here's some specs, please do all the work and optimize your app for our gpu").

People will happily do so as long as there's PowerVR in iToys.

[wizard69]

Wow, two pages of massive negativity here! This is a step in the right direction folks, that should be applauded. Does it solve every PowerVR related issue, no - but it is a start.

[robclark]

Now, on the GPU side, some FOSS reverse-engineered drivers have become surprisingly performant to the point of being mainlined into Linux and praised by Dolphin developers for its unbelievable superiority over the official drivers, as is the case of freedreno. It's a shame Qualcomm's boards are ridden with other freedom issues, not to mention Qualcomm-based phones have poor isolation between broadband's CPU and the normal CPU, making it virtually impossible to guarantee security and privacy of the host OS.

fwiw, at least if you stick w/ the qcom chips which do not have an integrated modem, this is not really a concern. afaiu for external modem everyone uses some sort of high speed serial connection (chip to chip uart or usb) since it is cheap (ie. doesn't require many traces on the pcb, doesn't occupy many pins coming off the chip)

[robclark]

Wow, two pages of massive negativity here! This is a step in the right direction folks, that should be applauded. Does it solve every PowerVR related issue, no - but it is a start.

Since pvr does so much in the shader (including the micro-kernel / firmware, which is a big part of what makes it's architecture such a convoluted mess), it is actually a pretty useful step. *Assuming* it doesn't require a click through agreement that would prevent r/e or working on an open src driver (and I have heard this is not the case). To do much of anything on pvr you would need to disassemble and re-write the micro-kernel firmware.

(but.. negativity should not be too much of a surprise.. img has many (device) generations of bad feelings in the community to overcome)