05:32 <icecream95> alyssa: Remember how you thought you solved the bug around some 16-bit instructions causing timeouts?

05:32 <icecream95> I'm not entirely sure yet, but I may have some bad news for you...

09:50 <icecream95> Oh wait, that caused INSTR_INVALID_ENC, not just timeouts. Guess I'll just have to rely on bisect for finding what this regression is

09:57 <icecream95> 4636f9e11a1 ("panfrost: Initialise the blend equation in create_blend_state") is the first bad commit

09:58 <icecream95> Is this the first time I've bisected to one of my own commits?..

10:10 <icecream95> Forcing allow_forward_pixel_to_kill to false at least works around the timeouts (this is for STK)

10:10 <icecream95> The commit I bisected to caused allow_forward_pixel_to_kill to be set to true in some cases where it was previously false

10:57 <icecream95> alyssa: Have you found that pow seems to return the wrong values for some inputs through your unit testing yet?

12:33 <HdkR> How far along is panvk so far? :>

12:35 <HdkR> Really what I'm asking is, how far away from running DXVK/VKD3D on panvk is it? :D

12:42 <HdkR> https://twitter.com/FEX_Emu/status/1423974617857748998 Because...I neeeed it

13:22 <alyssa> icecream95: `Forcing allow_forward_pixel_to_kill to false at least works around the timeouts (this is for STK) `

13:22 <alyssa> this is... odd. FPK flags aren't supposed to have any constraints, in the sense that setting them wrong can break rendering or slow down the app, but not ever cause faults.

13:22 <alyssa> `Have you found that pow seems to return the wrong values for some inputs through your unit testing yet? `

13:23 <alyssa> ...yes and? floating point in GLES has super minimal precision requirements

13:23 <alyssa> the OpenCL code paths are supposed to be a lot more expensive

13:23 <alyssa> (and OpenCL has `native_pow` and `half_pow` for faster GLES precision computations)

13:23 <alyssa> Dunno how this is handled in clover+NIR

13:47 <macc24> alyssa: did supertuxkart break again?

13:56 <alyssa> idk

14:11 <macc24> i only need a mt8173 laptop and then i will have all the infinity stones to turn half of x86 machines into ash