04:42 llyyr: could someone take a look at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39218 ?
07:15 mareko: cwabbott: let's ignore driver_location because that's a driver-specific thing that the frontend doesn't deal with
07:17 mareko: driver locations are also redundant because it should be possible to compute driver locations from a prefix bitmask of inputs_read/outputs_written in most cases
07:19 mareko: pipe_stream_output_info is only set to keep all gallium drivers working, and it's mainly used with nir_unlower_io_to_vars
07:20 mareko: it's OK to keep using variables because the GLSL compiler will do nir_lower_io unconditionally -> nir_opt_varyings -> nir_unlower_io_to_vars (if needed)
07:31 mareko: lowering lowered IO to driver-specific IO intrinsics or buffer loads/stores or shared loads/stores is what AMD drivers have been doing; let the standard lowered IO be the canonical form of IO for analysis/reasoning and optimizations, and let driver-specific intrinsics be the driver-lowered form of IO that gets translated to ISA
07:35 mareko: high-level NIR (close to source semantics) -> canonical NIR (best for analysis/optimizations) -> driver NIR (closer to ISA and still optimized by generic passes)
13:07 alyssa: I really wish gitlab wouldn't change the UI for no reason every version
13:11 alyssa: mareko: +1 on the strategy in that last message
13:57 karolherbst: I'm using driver_location inside rusticl to place kernel inputs within the byte addressed cb0 space. Though I could use location instead and make binding behave like "it's the nth argument"
13:58 karolherbst: but for most drivers it doesn't matter because it's just a load_ubo which doesn't need any of the variable infos anymore...
14:06 eric_engestrom: mesa devs: in light of various things (including collabora ci's outage and now the gitlab partial outage, but also several MRs that need more time), I'm delaying the branchpoint by a week, so instead of happening tomorrow it will be on next wednesday, january 21
14:07 eric_engestrom: I'll post a calendar update in a minute
14:07 alyssa: eric_engestrom: saved by the bell!
14:35 robclark: eric_engestrom: thx
14:38 eric_engestrom: (the calendar update is, ironically, not happening because of the gitlab issues 😅)
14:38 eric_engestrom: but given mupuf's progress report, I expect the migration will be done in probably 15-20min
14:39 eric_engestrom: hopefully once that's over, gitlab will be stable again