16:00 DavidHeidelberg:resolved last crash in OpenCL piglit CI tests... now only some fails remaining! :)
17:43 karolherbst: nice
19:03 DemiMarie: banhammer?
19:03 karolherbst: DemiMarie: somebody is saying something?
19:04 karolherbst: ohh.. prolly just a matrix spam wave, enjoy :)
19:04 DemiMarie: karolherbst: someone posted a message on Matrix that seemed like a cryptocurrency scam
19:04 karolherbst: yeah.. I kicked one of them from #nouveau already
22:25 karolherbst: from a gallium pov, how does writing to a mipmap level of an image work? Is this even supported in GL or is that a VK only feature?
22:29 airlied: karolherbst: karolherbst: when you say image, what do you mean in gallium terms?
22:30 karolherbst: SHADER_IMAGE
22:30 karolherbst: pipe_image_view
22:31 airlied: you can't only render to one mipmap of an image at at time AFAIK and pipe_image_view has the level in it
22:31 karolherbst: currently looking into the cl_khr_mipmap_image_writes extension which enables shaders to specify the lod when doing write_image
22:31 karolherbst: airlied: okay, so atm this is a gallium limitation?
22:31 airlied: yeah I don't think GL supports that at all
22:31 karolherbst: or rather... no driver expects a lod at shader runtime
22:32 karolherbst: I see
22:32 karolherbst: well... at least the shaders seem to compile just fine
22:33 karolherbst: and I mostly just wanted to verify the compiler side is in order
22:36 karolherbst: the CL CTS doesn't test get_image_num_mip_levels ... uhh
22:37 airlied: karolherbst: https://registry.khronos.org/OpenGL/extensions/AMD/AMD_shader_image_load_store_lod.txt is that it?
22:37 karolherbst: sounds like it
22:38 karolherbst: if something uses it, it might even make sense to add it
22:38 airlied: only radv supports it on vulkan
22:38 karolherbst: mhhh...
22:38 airlied: maybe could be added to radeonsi/zink
22:38 karolherbst: not gonna force it if nothing uses it though
22:39 karolherbst: I just wanted to see if the nir/vtn bits are working alright here
22:39 karolherbst: but..
22:39 karolherbst: I'm fine with only implementing cl_khr_mipmap_image
22:39 karolherbst: alyssa just wanted to write internal shaders using msaa images in OpenCL C :D
22:40 karolherbst: and somehow all those cl extensions depend on each other
22:41 karolherbst: in cl we convert image loads to txl/txf anyway and that seems to be working just fine
22:41 airlied: well msaa doesn't have mipmaps :-)
22:41 karolherbst: yeah... but like...
22:41 karolherbst: the spirv llvm translator sets the ImageMipmap cap when you compile cl shaders using msaa images 🙃
22:42 karolherbst: I don't even know why
22:43 karolherbst: the cap is only needed if you set the lod to a non zero value, and I don't think it sets it to non 0 in that case, so prolly just a translator bug, but implementing mipmaps was like 3 hours of work, so whatever