06:53airlied: DavidHeidelberg: I had an open MR to increase that to pass spec
06:54airlied: not sure if it's still oen
08:01Ermine: are there docs on pre-atomic kms driver structure?
08:09dinoOSdone: It does not really matter if data access does not work, there is always a chanche to have divide by two path, it's not intended to be done on accelerator, dma burst does it very fast the encoding and decoding and data access is done on other device , with unaligned access shifting the bits back to divide by two. It's not a nuclear science, you are just terroring me, it's not like i can
08:09dinoOSdone: not knock an australian porn fart down, i do that very efficiently and quick, it's i prefer to keep my criminal record clean as it is. We spawn other authorities to deal with jack and other terrorists just like they were spawned to treat the terrorslut.
08:31dinoOSdone: It's the computer engineers have done everything in their power that things work well, of course the dma is and should be default path to offload the encoder and decoder too, while intrinsics work on the cpu/gpu/npu whatever, it's only that you have to get the correct format out from DMA bursts, it would work amazingly in theory, i am not sure if you disagree with the combination theory or
08:31dinoOSdone: you ever did, but that is the way it works, they can work without intrinsics too, however intrinsics keep the memory footprint even lower.
08:57karolherbst: DavidHeidelberg: https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#opencl-embedded-profile "Max number of image objects arguments of a kernel declared with the read_only qualifier. The minimum value is 8 if CL_DEVICE_IMAGE_SUPPORT is CL_TRUE, the value is 0 otherwise."
09:56jonasstahl: There is nothing i want from you cause i go for my own code anyhow, first you do whatever shift based decoder/encoder to test linux is good at compiling non-performantly any code , dma/ddr one can be realized in many ways and those pages were given, you can control the burst to selectively overflow in all ways possible, cause the fifo descriptor is what you yourself define, but also dma
09:56jonasstahl: channel muxing does the same , consult the docs, i need to deal with learning traffic now, i explained enough that you are wrong to tell me to hang myself , cause i have been enough correct on theory, i won the debates, i am not sure what happened in domededov bombing if authorities did not care what can i do, since i do not have time for investigating that , it's not on my paycheck , i
09:56jonasstahl: can not do miracles, however i think russians were at my side, and i do not like them to be destroyed, but again i am not sure on this, it all seems so that ukranians started killing of russians and putin reacted , they can and russians can't? Maybe you were unfair to sleeping giant, as a neighbor i do not want conflicts with them, and so should ukranians think.
20:35mivanchev: hey, could someone help me regarding my statically compiled mesa
20:35mivanchev: if GLX doesn't use driver loading anymore, how are drivers loaded now?
20:36mivanchev: I mean this change in particular https://gitlab.freedesktop.org/mesa/mesa/-/commit/a43eefa7182d41b7c5797dcbb31f77d337ebf65c#7f64c03d67073cec4ab15b95a45de7c0ad80436d