Am 13.03.19 um 18:33 schrieb Michel Dänzer: > [SNIP] >>>>>>> Copy how? Using a GPU engine? >>>>>> CPU maybe? Though I suppose that won't work if the buffer isn't CPU >>>>>> accesible :/ >>>>> Well we do have a debug path for accessing invisible memory with the >>>>> CPU. >>>>> >>>>> E.g. three registers: DATA and auto increment OFFSET_LO/HI. So you can >>>>> just read/write DATA over and over again if you want to access some >>>>> memory. >>>> Right. I assume that'll be very slow, but I guess it could do when the >>>> memory isn't directly CPU accessible. >>> Just made a quick test and reading 33423360 bytes (4096x2040x4) using >>> that interfaces takes about 13 seconds. >>> >>> IIRC we don't use the auto increment optimization yet, so that can >>> probably be improved by a factor of 3 or more. > I'd assume only writes are needed, no reads. I've played around with that for a moment and with a bit of optimization I can actually get about 20 MB/s write performance out of the debugging interface. This way overwriting a 4K framebuffer would take less than 2 seconds using this. It's not ideal, but I think for a panic screen perfectly reasonable. Christian. _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
