03:24 <alyssa> Now passing all of dEQP-GLES3.functional.fragment_out.array.fixed.*

03:25 <alyssa> Of course most 'interesting' MRT is with floating-point/integer framebuffers

03:25 <alyssa> unfortunately that's raising a DATA_INVALID_FAULT, something about the blend shader ...

03:26 <alyssa> otoh blend shaders w/ MRT don't make much sense.

03:32 <alyssa> They seem to be doing a lot of redundant computation and it isn't at all clear why

03:36 <alyssa> It's possible something isn't well optimized but I'm not convinced there isn't more to the story

03:38 <alyssa> I sort of suspect the r1.z is used in the blend shader anyway

03:38 <alyssa> so if MRT is enabled and you don't set it, you get junk due to internal details and kerplutz

03:40 <HdkR> There is probably some spec statement somewhere stating if you don't fully write a render target then you get undefined data anyway

03:41 <HdkR> If blend mode aligns to the missing channels anyway

03:41 <alyssa> Weeeeeeeeee

04:21 <alyssa> robher: Is panfrost_mmu_map_fault_addr a known panic?

14:33 <robher> alyssa: maybe. There was a race fixed when closing the device fd.