On Mon, Sep 08, 2014 at 06:01:25PM +0200, Luc Verhaegen wrote: > On Mon, Sep 08, 2014 at 04:33:48PM +0200, Maxime Ripard wrote: > > On Mon, Sep 08, 2014 at 03:55:19PM +0200, Luc Verhaegen wrote: > > > > > > For those who do not know what Maxime is on about, here is the short > > > explanation: > > > http://linux-sunxi.org/Allwinner_SoC_Family#Naming_confusion > > > > > > I do not buy the modern allwinner sunXi naming. > > > > > > Retro-actively renaming half their line-up sun4i and the other half > > > sun8i, I have seen that before (cfr unichrome.sf.net). A few months in, > > > and that Vendor did another little pirouette. And again, and again. > > > > > > Where Allwinner drew the naming line now is quite arbitrary: the type of > > > the ARM core in there. The one component which is mostly generic and > > > reasonably universal. Sun4i suddenly became all ARM Cortex A8, sun8i > > > became all ARM Cortex A7. Now that they stuck a few A15 in A80, they > > > named that sun9i. > > > > Actually, I do find it consistent. > > For now.
Forgive me, I don't have the chance to own a crystal ball. > > > If the sunXiwYpZ scheme actually had something to do with the majority > > > of SoC specific component lineage, i would've totally bought into it: > > > that would put A20 close to A10, and would put A31 somewhere else > > > altogether, with A23 and A33 close together... > > > > Yeah, right, as if the sun4i/sun5i was in any way better.... > > It was a consistent scheme, which allwinner followed for many many > years. No, it was not. If it was to be a purely time-driven naming scheme, A10s should have been sun6i. > > > So no, this is purely a marketing driven decision, which will have > > > changed again in 6 months time. Where will Allwinners freshly announced > > > A83 land? Will they stick to their own naming scheme? What will happen > > > when Allwinner produces an ARMv8 chip, will they count beyond 9, or will > > > they rename everything sun7i/sun8i? > > > > > > Sun[4567]i was chronological, and initially sun[89]i were too. So I say > > > we just keep on counting for ourselves. > > > > A10s was released after A13, A31 and A20 at the same time. Not very > > chronological if you ask me... > > A10s is A13. [citation needed] > The A20 design was a stopgap, and the design started after A31 was > started. So it was chronological, just not chronological with the > release dates. Yeah, right. So it's chronological, but we have no way to tell, except by fully trusting you. The argument of (which?) authority is not really something that usually convince me. > > But I guess what it all boils down is this: do we want to keep > > allwinner's naming and be consistent with it or not. We've been so > > far, it's something I'd like to be kept. Especially since it's *very* > > easy to support (basically, keep the sun6i and sun7i, introduce sun8i > > for A23, A33, and whatever comes next, which should be the best if > > following your "connection" argument), and have sun9i for the A80. > > But this is not consistent with Allwinner naming. Being consistent with > allwinner naming would mean renaming most of the upstream code. And then > doing it again a few months from now. It is. You can find some reference to A31 being sun6i, and A20 being sun7i. Can you find any reference to A33 being sun10i? > Sun8i for A23 and A33 will only work until we see allwinner give out the > codename for A83, and whether the early marketing noise for it matches > reality and it really is a octal A7 with powervr. Something tells me > that this is going to be an A80, with less dram channels, less rogue > shaders, and with A7s instead of A15. I guess it could fit into both sun8i and sun9i. But A83 isn't really the topic here, is it? And even if Allwinner introduces a sun10i for A83, it will only proves that we were wrong by trying to stick it to A33. > > Introducing this sun10i out of nowhere, without any other > > code/documentation referring to it, and with the only argument that > > "some guy thought it was a better idea" is not that great. Especially > > when the time where they will have a new design and come up with > > sun10i chips. > > It will be a very short pain. Look closer into your crystal ball. > > > Another option is to completely switch to AW<chip id>, which would mimic > > > what i did for unichrome. > > > > > > A10 = AW1623 (sun4i) > > > A13/A10s = AW1625 (sun5i) > > > A31 = AW1633 (sun6i) > > > A20 = AW1651 (sun7i) > > > A23 = AW1650 (sun8i) > > > A80 = AW1635 (sun9i) > > > A33 = AW1667 (...) > > > > > > Now that's going to be real confusing. > > > > That could be an option, but like you said, it's pretty confusing to > > existing user of our code base. > > This is going to be a big pain. > > But ok, so what do you want to call A33, and do we rename A23 away from > sun8i, by adding suffixes there as well? > > The sun8i<suffix> "solution" is only a band-aid though. Perhaps A83 will > fit in this scheme, perhaps not. Something tells me that armv8 is going > to seriously kill this, and will require a more permanent solution to > this problem. A more permanent solution to a problem that isn't even a problem. Yeah, definitely looks way more important than a proper display driver. Maxime -- Maxime Ripard, Free Electrons Embedded Linux, Kernel and Android engineering http://free-electrons.com
signature.asc
Description: Digital signature
