On Wed, Jan 26, 2011 at 6:28 PM, Tim Bird <[email protected]> wrote: > On 01/25/2011 07:16 PM, Luke Kenneth Casson Leighton wrote: >> the proposal is therefore to not hang about: begin to reverse-engineer >> the "top" GPU, PowerVR, which is present in at least 75% of the >> top-of-the-range ARM embedded CPUs. > > Luke, > > I appreciate the proposal, and have a couple of items of feedback. > > 1) The desirability of an open source driver for PowerVR is > recognized within the industry. This was discussed at > both Embedded Linux summits that were held last fall > (in Tokyo and Cambridge). The companies represented decided > that one action we would take was an educational and > licensing pressure one, from multiple SOC consumers.
well... yes, you could play prisoner's dilemma, and ride on the back of the popularity of android etc. to help pile on the pressure (strategy mentioned in chris' talk). perhaps an additional way to pile on the pressure is to actually begin - and massively announce - the beginnings of a reverse-engineering project. in fact, with maciek's work that's actually begun _already_ but it just hasn't been announced. ... wait... _last fall_?? let me check the dates.... ok, right. september and october 2010. so... 5 months ago. can i ask... what concrete results have been achieved in that time? > 2) The undertaking you describe is indeed massive, and > is fraught with difficult issues, as you point out. > It might be outside the scope of what we can fund. crucially, in recognition of this, it's why i mentioned "you have to start somewhere" in the proposal. if you never begin, the status quo is maintained, and you are entirely at the mercy of the game of "prisoner's dilemma". you can't make concrete business decisions based on the "whims" of companies who really have no incentive to act, and the cynic in me (well, actually, repeating what many cynics have told _me_ on this topic :) says that their incentives are purely money-based. there's no money in "giving away the source code quotes for free quotes". the fact that a compelling business case can be made for leveraging state-of-the-art free software advances and techniques, especially around the massive advances made by utilising LLVM (and Gallium3D) where even apple have shown that LLVM can give you a significant performance increase when you have a multi-core / multi-threaded CPU behind that GPU, seems to have entirely escaped these proprietary 3D company's notice. and here's the rub: they either don't believe - or have not been informed compellingly - that the Free Software Community, despite evidence to the contrary (for the mainstream GPUs, NVidia, ATI, Intel) would "deliver the goods" and get better performance than they can, with their own proprietary libraries. perhaps... even.. they don't *want* to be "shown up" by a "bunch of aspies" :) soo... yeh. all quite amusing. what's particularly ironic / galling is that PowerVR is perfectly well capable of running the full OpenGL 3 spec, yet is restricted in the embedded space by the provision of a proprietary ES 2.0 library. that alone should be sufficient to justify the beginning of a reverse-engineering effort. remember, tim - i've done 2 major bits of strategic reverse-engineering. the one i'm well-known for is the Windows NT Domains protocols. that _was_ complex, and a lot of fun. after 3 years and 100,000 lines of hand-coded MSRPC packets and client-server architecture, there was enough for other companies, world-wide (not just the samba team but big companies like network appliances, EMC and so on) to "carry on". some of these big companies even privately licensed MS's IDL files _before_ the EU anti-trust case forced MS to make them available. the second bit of reverse-engineering was in 2005 (with a prior attempt back in 2000) - exchange 5.5. each time, the openchange team "copied" what i did (stripping out all mention of where they got the information from). after the 2nd attempt, which lasted about 4 months on my part at 14 hour days pretty much non-stop and unpaid, the openchange team's "copying" was sufficient for a group of people to continue working. within the openchange framework, money was made available (none of which went to me, because my name had been completely removed from the project) by interested parties who keenly wanted to make money from free software based around exchange 5.5 interoperability. so the point of mentioning this is: so it might not be possible to get everything done "in one hit" - but really, don't worry about it. you *know* that people are very very keen to see the 3D proprietary cartel broken wide open - you've met with them (and i haven't). so you can conclude, logically, that if at least *something* is available, they'll find *some* way to get behind it. but... yeah. my prior experience - and present financial situation which is such that i am prioritising paying for food for my family (not rent, bills or anything else) - says that this time there's no way in hell i can let people sponge off of my expertise again. quite simple, really. > However, having said the above I'll try to whip up a > wiki page for the proposal, and we'll at least give > it some discussion. appreciated. > -- Tim > > P.S. The list is about to go down for maintenance. Sorry about > the timing. It has been planned for a while, and it hits at a > bad time when this discussion is going on. ah well. you'll at least get a copy of this, i'll check the archives see if it got/gets through, re-post if necessary. l. _______________________________________________ Celinux-dev mailing list [email protected] http://tree.celinuxforum.org/mailman/listinfo/celinux-dev
