Queru: Another year of AOSP [LWN.net]

Queru: Another year of AOSP

Posted Jan 3, 2012 17:44 UTC (Tue) by wookey (guest, #5501)
In reply to: Queru: Another year of AOSP by zlynx
Parent article: Queru: Another year of AOSP

Of course people care. Firmware is code too, and it's often quite broken. Being able to fix it would be useful.

The graphics situation is much worse than simple firmware blobs. There are great dollops of proprietary code running on the main CPU making kernel upgrades and porting difficult/impossible, and removing the opportunity to fix or improve or re-use stuff.

All the manufacturers have got themselves into a right mess, scared of patents, even though their infringement is exactly the same whatever licence the code uses - either they are infringing or they aren't (they are of course, obviously, because the whole are is rife with idiotic patents).

Some believe that their graphics drivers really are so good that no-one outside could help improve things. Which is an odd thing to believe as we know how hard it is to employ all the smart people in any given area.

They've all got this weird idea that publishing CPU instruction sets and registers so that people can write compilers and optimised code and understand how their device works is good, but publishing GPU instruction sets so that people can do the same things is wrong and bad and must be discouraged. his seems to me to be demonstrable nonsense, but I'm not quite sure how we fix it.

to post comments

Queru: Another year of AOSP

Posted Jan 3, 2012 20:38 UTC (Tue) by zlynx (guest, #2285) [Link]

Well, one reason the graphics companies hate publishing their details is that they hate to support customers. They'd rather not deal with them at all.

If the CPU and register sets were published, they would be under the gun to support that. If their CPU instructions had a flaw, like the Pentium division problem, they'd have to fix it.

But when everything has to funnel through their drivers they can just release a software update to silently rewrite flawed operations. Its a miracle! No need for expensive hardware recalls!

It also makes it very difficult to reverse engineer their driver code. Are they cheating at a benchmark? Are they silently speeding up a game by only running particular shaders every third frame? Who can tell?

Look at all the crap Intel got because their compiler used the CPUID check for GenuineIntel to enable proper SSE optimizations. No one would have ever known if you couldn't read the instruction set.