#asahi-gpu · 2021-07-24 16:08
xerpi[m]So if the HW doesn't support cubemap texture sampling you could always do it by SW by doing vector-to-plane intersections?#asahi-gpu · 2021-07-24 07:16
xerpi[m]Awesome :)#asahi-gpu · 2021-07-10 17:41
xerpi[m]Hmm but maybe you don't need to have CPU and GPU caches coherent for some buffers, so this "managed" mode can help reduce traffic#asahi-gpu · 2021-07-10 17:38
xerpi[m]Probably one of those magic numbers is related to this#asahi-gpu · 2021-07-10 17:38
xerpi[m]Btw this is interesting: https://developer.apple.com/documentation/metal/mtlstoragemode/managed#asahi-gpu · 2021-07-10 17:38
xerpi[m]:D#asahi-gpu · 2021-07-10 17:31
xerpi[m]Maybe the debug/developer macOS kernel printks a nice error message #asahi-gpu · 2021-07-10 17:29
xerpi[m]Would be nice if the GPU reported some kind of error code. Maybe macOS kernel doesn't expose it to userspace, but would mmio tracing help?#asahi-gpu · 2021-07-10 17:27
xerpi[m]Let's hope this was not the issue and that the GPU also snoops the L1/L2, otherwise it will get hairy, you would also need to invalidate BOs when reading data the GPU has written..#asahi-gpu · 2021-07-10 17:17
xerpi[m]Regarding this random hang/HW issue: are the CPU and GPU cache coherent? If so, only LLC or also L1/L2? Maybe you need to flush some shared stuff/descriptors up to LLC#asahi-gpu · 2021-06-05 21:08
xerpi[m]That's interesting, this means the fragment shader has both interpolated and flat varyings available to pick from, isn't this a waste of memory/regs? (I was expecting only one of them would be generated, depending on a flag)#asahi-gpu · 2021-06-05 20:35
xerpi[m]Flat shading + provoking vertex sounds fun!#asahi-gpu · 2021-06-05 20:21
xerpi[m]It seems like the fragment shader is reading the color varying from the wrong offset? Maybe some alignment is needed?#asahi-gpu · 2021-06-05 20:03
xerpi[m]num_descs is never touched#asahi-gpu · 2021-06-05 19:17
xerpi[m]Oh, I see that they are in screen space and not NDC#asahi-gpu · 2021-06-05 19:14
xerpi[m]gl_FragCoord are the coordinates of the current invoked fragment that came out of rasterization, right? So will they probably be preloaded in some register/special memory address?#asahi-gpu · 2021-06-05 19:10
xerpi[m]That's still amazingly fast progress, considering you are reverse engineering it :)#asahi-gpu · 2021-06-05 06:25
xerpi[m]<bloom "FWIW - intending to stream somet"> Nice! Looking forward to it :)#asahi-gpu · 2021-05-31 23:30
xerpi[m]It's getting late here, I will finish watching it tomorrow. Thanks again for streaming!#asahi-gpu · 2021-05-31 23:19
xerpi[m]Oh ok I see, thanks#asahi-gpu · 2021-05-31 23:18
xerpi[m]Shouldn't it be last_level - 1? #asahi-gpu · 2021-05-31 22:51
xerpi[m](never used mipmaps before, so it's quite new to me 😅)#asahi-gpu · 2021-05-31 22:51
xerpi[m]So with infinite mipmap levels the texture size gets doubled#asahi-gpu · 2021-05-31 22:46
xerpi[m]Makes sense that it's a "mipmapping enable" bit since the Levels field=0 means 1 level#asahi-gpu · 2021-05-31 22:30
xerpi[m]Depth stencil seems easier, right? Pass different formats to Metal and see how the bits change#asahi-gpu · 2021-05-31 22:16
xerpi[m]Icecream95 left a comment about this, misplaced (int) casting, missing parenthesis#asahi-gpu · 2021-05-31 20:36
xerpi[m]Thanks for streaming, it's being super interesting! 👍️#asahi-gpu · 2021-05-31 20:33
xerpi[m]Yay nice!#asahi-gpu · 2021-05-31 19:27
xerpi[m]I think .hw is missing when using it#asahi-gpu · 2021-05-31 19:13
xerpi[m]Oh sorry, it was the B that seemed like an 8, my bad#asahi-gpu · 2021-05-31 19:12
xerpi[m]And EAC RG11#asahi-gpu · 2021-05-31 19:12
xerpi[m]ON the pixel formats, ETC2#asahi-gpu · 2021-05-31 19:10
xerpi[m]Aren't there two 0x58?#asahi-gpu · 2021-05-31 18:19
xerpi[m]Are planar and YUV formats exposed in Metal?#asahi-gpu · 2021-01-23 13:07
xerpi[m]Do we know if the display engine supports scanout of tiled/compressed framebuffers (which I would guess it does), or is there a de-tiling/de-compressing job on the swapchain before presenting?