I need a "Do not use debian stable for GPU development" sign to tap on.
Dave Airlie
mastodon 4.7.3GPUs and Linux
at that stage of of shader optimization where everything I try, makes it slower.
once again I find myself debugging code written by a version of me who seems to have a superior intellect.
it was all bit 1 of NV_VIRTUAL_FUNCTION_PRIV_MMU_INVALIDATE_PDB fault. nouveau on spark runs some tests now.
0 days since I did sudo service gdm stop on the wrong machine.
sometimes you just have to throw a bigger workgroup at it.
Reached the point of development cycle with nouveau on the spark where I want to throw it through the window, so I should be pretty close to getting display lit up
new pahole record for me
/* XXX 24624 bytes hole, try to pack */
there comes a time around 100,000 SSA indexes where inline the shit out of it all starts to become questionable.
sometimes it's easier to just add new printks than to work out how to switch on the pre-existing ones that are buried in driver macros and dynamic debug.
HDMI FRL support on nouveau, please test if you have HDMI2.1 and high refresh monitor https://lore.kernel.org/nouveau/20260423004552.3289884-1-airlied@gmail.com/T/#t
deviceName = NVK GB20B
though CTS is a failfest due to some sync or coherency issue, and display isn't working yet.
once again gcm was not the answer I was looking for
took a wrong turn and now I'm in the mesa vulkan meta blit/resolve paths and wondering if I should just back away slowly.
Okay I have llama.cpp coopmat2 paths running on radv for the first time, they are 10 times slower that coopmat but gotta start somewhere
I have a lot of laptops, one of the turing ones, Lenovo p53 the display stopped turning on yesterday, I decided to open it up and battery was swelling bad, just dropped the battery off at a disposal place, but the laptop is pretty toast. Can't see a dislodged cable, but I probably need an ifixit kit to dig deeper.