IRC Logs of #dri-devel on for 2014-10-21

Previous dayChoose dateNext day Show menu

08:05 #dri-devel: < danvet> airlied, oh dear the ioremap guy is fun
08:06 #dri-devel: < danvet> the leftover mapping is page for config space
08:06 #dri-devel: < danvet> hilarious
08:06 #dri-devel: < danvet> *1 page
08:07 #dri-devel: < ryao> Is this an early April Fool's joke?
08:07 #dri-devel: < simpson> "This is the first time an ISA for a GPU supported by the Android ecosystem has been made public." Haha, what.
08:08 #dri-devel: < ryao> They claim that the full ISA for PowerVR Rogue is published. I wonder if there is a NDA.
08:08 #dri-devel: < glennk> well, i guess considering metal pretty much exposes it
08:09 #dri-devel: < ryao> Do initialization sequences count as part of the ISA?
08:09 #dri-devel: < imirkin_> what actual hardware does that ISA map to? gma500? or even weirder things?
08:09 #dri-devel: < glennk> only if you count dead chickens as part of the ISA
08:09 #dri-devel: < glennk> imirkin, its a successor to it, similar but different
08:10 #dri-devel: < ryao> imirkin_: I do not know. It is supposed to be their latest stuff.
08:10 #dri-devel: < imirkin_> does it exist inside actual silicon that is inside of purchasable devices, or is it theoretical at this point?
08:11 #dri-devel: <+anholt> I've heard it's basically theoretical at this point. unlike the other 2 vendors with shipping hw with android and fully documented 3d chips already.
08:11 #dri-devel: < ryao> I have seen these names before:
08:12 #dri-devel: < glennk> its in iphone 5 and later
08:12 #dri-devel: < simpson> There is not an NDA, but the docs are only available with an installer which asks that the user not use any part of the SDK for a handful of purposes including interoperability and RE.
08:12 #dri-devel: < ryao> I wonder if this is a response to Samsung developing their own GPU because Samsung internally hated PowerVR because only 1 ir 2 guys were allowed to read the documentation.
08:12 #dri-devel: < glennk> don't know which android devices (if any) though
08:14 #dri-devel: < ryao> Google found this for the G6430:
08:16 #dri-devel: < imirkin_> neat
08:18 #dri-devel: < robclark> simpson, asks, as in "pretty please, we'd prefer if you didn't"?
08:18 #dri-devel: < robclark> otherwise that sounds like an NDA..
09:08 #dri-devel: < daniels> ryao: i 100% do not buy the samsung-no-docs thing. img have always been quite free with docs, as long as you sign an nda.
09:09 #dri-devel: < daniels> the rgx has shipped on a couple of non-iphone platforms (one running wayland), but very very few
09:13 #dri-devel: < ryao> daniels: Guy Martin of Samsung told me that the way things work internally isthat only 1 or 2 guys are allowed to read the documentation and everyone else has to ask them questions.
09:14 #dri-devel: < daniels> ryao: that's samsung's failing then - img are by _far_ the most generous/loose of all the proprietary gpu vendors
09:46 #dri-devel: < zachr> bwidawsk, ajax, I was looking through the mailing list archives for information on VGEM and why it never went any further. I can't find anything past Feb 26th, 2012. Could you tell me why it ended so abruptly? I'm looking to resurrect it.
09:47 #dri-devel: <+ajax> zachr: no reason beyond i had other things to work on, really.
09:48 #dri-devel: <+ajax> though the modesetting driver's dri support ought to be most of the way to the same goal
09:49 #dri-devel: <+ajax> if you can convince llvmpipe to draw to the server's buffers instead of doing PutImage all the time.
09:51 #dri-devel: < zachr> ajex: To which server are you referring?
09:51 #dri-devel: <+ajax> there's only the one in my world...
09:51 #dri-devel: <+ajax> X is what i meant, i guess it'd be true of wayland too
09:52 #dri-devel: <+ajax> although in wayland the server doesn't allocate the buffer anyway, nm
09:54 #dri-devel: < zachr> Ah. I'm working without X or wayland.
09:58 #dri-devel: < robclark> daniels, img is generous w/ docs because their customers have to fix the driver enough to ship a product :-P
09:59 #dri-devel: < ghallberg> oh right I was going to do that fifo-log thingie.
09:59 #dri-devel: < glennk> robclark, similar also for the hardware itself
10:00 #dri-devel: < robclark> heheh
10:00 #dri-devel: < glennk> the same being true for any ip vendor of course
10:01 #dri-devel: < daniels> robclark: so very much less so than, say, viv
10:02 #dri-devel: < robclark> heh, I suppose I should be glad I never had to deal w/ viv
10:04 #dri-devel: < mlankhorst> is it worse than pvr? :P
10:05 #dri-devel: < danvet> it is pvr
10:06 #dri-devel: < robclark> danvet, viv == vivante..
10:06 #dri-devel: < danvet> oh, I've thought mlankhorst asked about the img stuff
10:06 #dri-devel: * danvet mixed up things
10:06 #dri-devel: < glennk> i think nintendo ds is the only physical thing i've seen vivante in?
10:06 #dri-devel: < robclark> glennk, freescale imx6.. and I think some marvell stuff..
10:08 #dri-devel: < glennk> last i dealt with imx was imx51 which was ati z180 and z400
10:09 #dri-devel: < robclark> yeah, would be nice if someone got freedreno going on imx5.. although a2xx (aka z4xx) hasn't really gotten much love for a while..
10:09 #dri-devel: < robclark> but otoh, a3xx is where it started getting fast enough to be interesting ;-)
10:09 #dri-devel: < glennk> shouldn't be hard, its identical to adreno 200
10:10 #dri-devel: < robclark> right
10:10 #dri-devel: < glennk> adreno 200 is as slow as the gpu in the first android device though
10:10 #dri-devel: * robclark assumed a200 *was* the gpu in the first android device..
10:10 #dri-devel: < glennk> adreno 130, gles 1.1
10:10 #dri-devel: < robclark> ahh
10:10 #dri-devel: < glennk> in the g1
10:10 #dri-devel: < robclark> wow, even older school..
10:11 #dri-devel: < glennk> i have good memories of the g1 :-)
10:11 #dri-devel: < mlankhorst> ah
10:13 #dri-devel: < daniels> mlankhorst: that's actually hugely, hugely, hugely, hugely insulting to pvr
10:14 #dri-devel: < robclark> vivante should probably just give up on their own driver and work with the etnaviv folks ;-)
10:14 #dri-devel: < daniels> mlankhorst: pvr isn't going to win any beauty pageants, but does everything you'd hope it would, and does so reliably
10:14 #dri-devel: < daniels> viv, otoh, is an absolute nightmare
10:15 #dri-devel: < daniels> any kind of sync? nah. eglimage recounting? sure, but delay cleanup/destroy until eglTerminate. buffer sharing? here's a physical address! etc
10:16 #dri-devel: <+ajax> eglTerminate? you're shitting me.
10:16 #dri-devel: < glennk> hey, at least they don't consider reboot as the only method of cleanup
10:16 #dri-devel: < daniels> oh yeah, and even though they delay eglimage destruction - presumably as there's no sync with job completion anywhere (all driven by userspace), they copy on TargetTexture2D
10:17 #dri-devel: < daniels> so we had a compositor that was both spending 16ms per frame in TargetTexture2D, and leaking memory out the nose
10:17 #dri-devel: < daniels> ...
10:17 #dri-devel: <+ajax> at least it wasn't losing one frame per frame for very long then
10:17 #dri-devel: < daniels> oh yeah, and the eglBindWaylandDisplayWL impl was literally return EGL_TRUE
10:18 #dri-devel: < daniels> and if you wanted, y'know, wayland support, you could write your own fsl/viv-specific code
10:18 #dri-devel: * daniels twitches
10:25 #dri-devel: < simpson> I'm gonna go make my own GPU, with blackjack and hookers.
10:25 #dri-devel: < robclark> the marketing/bis-dev guys would love that :-P

Written by Christoph Brill © 2007-2014

Valid XHTML 1.0 Transitional

Available in Android Market