16:47 Frogging101: I woke up my system from S3 and the sdma0 ring didn't start. RX 6900 XT, kernel version 6.0.12-arch1-1. http://www.fastquake.com/files/text/wakeup-amdgpu-sdma0-off.txt
16:56 Frogging101: Reported here. https://gitlab.freedesktop.org/drm/amd/-/issues/2335
17:48 MrCooper: agd5f: FYI, writing a Git commit hash (even abbreviated, so long as it's unambiguous) in a GitLab comment results in a clickable link (commits from another project in the same GitLab instance can be referenced as '<namespace>/<project>@<commit hash>')
19:01 agd5f: MrCooper, ah, good to know
19:02 haasn: Would it be possible to support more usage flags for planar images (e.g. G8_B8_R8_3PLANE_UNORM)?
19:02 haasn: I want to avoid memcpy if possible
19:02 haasn: But currently mesa not having any useful usage flags on these planar formats pretty much requires memcpying data in and out of them from more useful VkImages
19:03 haasn: for example, I would expect to be able to create a VkImageView of only a single plane of such a planar image, and then add this to a color attachment
19:03 haasn: But this requires specifying IMAGE_USAGE_COLOR_ATTACHMENT_BIT on the original (planar) image
19:03 haasn: Which mesa forbids by not listing it in the requirements
19:10 bnieuwenhuizen: haasn: so you want it in the ImageFormatProperties right (not FormatProperties, since we definitely can't render to the full format)
19:11 haasn: bnieuwenhuizen: I think so, yeah
19:11 haasn: let me double check
19:13 haasn: VkGetPhysicalDeviceImageFormatProperties returns VK_ERROR_FORMAT_NOT_SUPPORTED. yes
19:13 haasn: if I triy setting e.g. COLOR_ATTACHMENT_BIT
19:14 bnieuwenhuizen: did you set VK_IMAGE_CREATE_EXTENDED_USAGE_BIT?
19:15 haasn: ohh, nope
19:15 bnieuwenhuizen:wanted to apply a quick fix, only to notice somebody already did
19:16 haasn: yeah can confirm that fixes it :)
20:07 haasn: bnieuwenhuizen: what about BLIT_SRC/DST?
20:08 haasn: I can't blit into an vkimageview of a single plane
20:08 haasn: because VkFormatFeatureFlags does not have BLIT_DST_FLAG
20:09 haasn: and vkCmdBlitImage does not care about image views anyway
20:14 bnieuwenhuizen: haasn: doesn't VkFormatFeatureFlags2 or whatever have it?
20:15 bnieuwenhuizen: combined with CmdBlitImage2
20:20 haasn: bnieuwenhuizen: vulkaninfo https://bpa.st/raw/ETLDO
20:21 haasn: [0.081][e] vk Validation Error: [ VUID-vkCmdBlitImage-dstImage-02000 ] Object 0: handle = 0xa000000000a, name = ../src/tests/vulkan.c:165, type = VK_OBJECT_TYPE_IMAGE; | MessageID = 0x86bc2a78 | In vkCmdBlitImage, VkFormatFeatureFlags (0xc6c001) does not support required feature VK_FORMAT_FEATURE_2_BLIT_DST_BIT for format 1000156002 used by VkImage 0xa000000000a[../src/tests/vulkan.c:165] with tiling VK_IMAGE_TILING_OPTIMAL. The
20:21 haasn: Vulkan
20:21 haasn: spec states: The format features of dstImage must contain VK_FORMAT_FEATURE_BLIT_DST_BIT (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkCmdBlitImage-dstImage-02000)
20:21 haasn: and corresponding validation failure ^
20:22 haasn: I don't think using 1 or 2 would make a dif here
20:24 bnieuwenhuizen: haasn: is this about the combined format or the single-view format?
20:25 haasn: bnieuwenhuizen: combined format, since that's the only thing vkCmdBlitImage cares about (it does not take a VkImageView)
20:26 haasn: but actually
20:26 haasn: I don't think this is super critical anyway
20:26 haasn: because I just realized that you cannot blit from e.g. the color aspect of a non-planar image to a single plane of a planar image
20:26 haasn: the aspectMask must match for both
20:27 bnieuwenhuizen: haasn: that is what CmdBlitImage2 is for (supports ImageViews IIRC)
20:27 bnieuwenhuizen: oh nvm it doesnt?
20:27 bnieuwenhuizen: bleh
20:33 DrNick: is Steam's gldriverquery reliably crashing in llvm::PassInfo::createPass() for eglMakeCurrent(blah, 0, 0, 0) known?
20:38 haasn: bnieuwenhuizen: https://github.com/KhronosGroup/Vulkan-Docs/issues/2029
22:47 DrNick: ok, so gldriverquery is crashing because SDL is running eglMakeCurrent(blah, 0, 0, 0) in an atexit() handler and global destructors have already run
23:05 DrNick: lol SDL suggests registering SDL_Quit with atexit