21:45 <tfiga> liquidAcid: hi :)

21:46 <tfiga> all the Polish people working on Exynos are the same group :)

21:49 <liquidAcid> ?

21:49 <liquidAcid> ah, ok -- i see what you mean

21:49 <tfiga> unfortunately the issue with HDMI power domains is not clear and we didn't get any explanation, so the only thing sure to work correctly is that hack

21:49 <liquidAcid> i still don't understand though what rahul's and inki's patch are then exactly fixing

21:50 <tfiga> most likely the power down sequencing

21:51 <tfiga> but I'm not really sure those patches really fix it or rather reduce the probability of happenning...

21:53 <liquidAcid> i need to check on the coming weekend, haven't got any spare time this week (mini workshop)

21:55 <liquidAcid> tfiga, and someone pursuade jglisse to leave the userptr support in peace ;)

21:55 <tfiga> all I can say is that what Marek does is best we can do to get HDMI running right now

21:55 <tfiga> no, don't do it

21:55 <tfiga> or at least do it, but in a way making it hard to use in new userspace code

21:55 <tfiga> userptr is just a big Pandora's box

21:56 <tfiga> especially on systems without coherent memory systems

21:56 <liquidAcid> i would be happy to use it one day though, when iommu support arrives

21:57 <tfiga> what would be the use case?

21:58 <liquidAcid> i'm using the g2d to implement a video driver for retroarch (an emulator frontend)

21:58 <liquidAcid> the emulator core allocates memory for its emulator framebuffer itself and then passes it to the frontend

21:59 <liquidAcid> it would be nice to just use this malloced buffer then as input for the g2d

21:59 <tfiga> actually the G2D on Exynos4 can be coherent with CPU caches

21:59 <liquidAcid> also, almost all emulator cores just alloc one buffer at init and then never realloc it again -- which fits nicely with what userpte does

21:59 <tfiga> but I believe nobody tried to use this feature

22:00 <liquidAcid> is there any docu available for the g2d? so except for the libdrm test cases

22:00 <tfiga> I don't think there is anything public

22:00 <liquidAcid> hmm

22:01 <liquidAcid> i was trying to get some alpha blending stuff working

22:01 <liquidAcid> input is XRGB, but there seems to be a reg where you can define a alpha that is "added" to your source