#asahi-gpu · 2023-04-08 17:03
alyssaand the coefficient register store is written by the hardware once for each triangle for each 32x32 tile#asahi-gpu · 2023-04-08 17:03
alyssaIt's just copying the coefficient registers from the coefficient register store#asahi-gpu · 2023-04-08 17:02
alyssaI bet ldcf isn't doing any work at all#asahi-gpu · 2023-04-08 17:02
alyssaooh#asahi-gpu · 2023-04-08 16:43
alyssahow can this work what#asahi-gpu · 2023-04-08 16:43
alyssathat would require the provoking vertex colour to be the colour at a tile edge (i.e. interpolated) though so err that doesn't work either#asahi-gpu · 2023-04-08 16:42
alyssaand since tiles are at most 32x32 that works out#asahi-gpu · 2023-04-08 16:42
alyssaor I guess ldcf returns something different in those tiles#asahi-gpu · 2023-04-08 16:41
alyssawhat if the triangle is bigger than 32x32 how can this possibly work#asahi-gpu · 2023-04-08 16:41
alyssastill doesn't explain the weird 32x32 tile thing#asahi-gpu · 2023-04-08 16:41
alyssaseemingly#asahi-gpu · 2023-04-08 16:41
alyssameaning, the third register is the colour at the provoking vertex and the first two registers are the delta vectors#asahi-gpu · 2023-04-08 16:40
alyssaso ldcf is loading the iid in reverse order from NIR#asahi-gpu · 2023-04-08 16:19
alyssajust some abstract 32x32 tile used for barycentric calculations, I guess#asahi-gpu · 2023-04-08 16:19
alyssanope#asahi-gpu · 2023-04-08 16:16
alyssahold on wat, is this relative to the tile? is that what's happening? oh boy.#asahi-gpu · 2023-04-08 16:10
alyssawouldn't be the first vendor to do ieee 754 crimes#asahi-gpu · 2023-04-08 16:10
alyssaunsure what bit 0 of A means, maybe they're passing some sideband info there#asahi-gpu · 2023-04-08 16:09
alyssa(so we can have just one of them denorm, apparently?)#asahi-gpu · 2023-04-08 16:09
alyssaok so seemingly it's considered valid if bit 0 of A is clear, or if A and B are not *both* denorm#asahi-gpu · 2023-04-08 16:08
alyssayeah#asahi-gpu · 2023-04-08 15:59
alyssapretty sure I blew a sign somewhere#asahi-gpu · 2023-04-08 15:59
alyssatook a lot of poking through the assembly to get there, but#asahi-gpu · 2023-04-08 15:59
alyssaOk, most of the complexity is to handle subnormal deltas#asahi-gpu · 2023-04-08 14:57
alyssameh, 24 is short enough I might as well expand it out#asahi-gpu · 2023-04-08 14:56
alyssa24 without perspective correction#asahi-gpu · 2023-04-08 14:50
alyssa~40 instructions#asahi-gpu · 2023-04-08 14:50
alyssaI think Mesa has this code somewhere#asahi-gpu · 2023-04-08 14:50
alyssaInterpolating a single float with a constant offset requires a huge pile of instructions.. this is a bit annoying for r/e but whatever#asahi-gpu · 2023-04-08 14:47
alyssa> Unknown 3:0: 0x1#asahi-gpu · 2023-04-08 14:47
alyssawe do have#asahi-gpu · 2023-04-08 14:47
alyssainterestingly, the CF bindings are the same (i.e. for Gouraud shading not flat shading)#asahi-gpu · 2023-04-08 14:44
alyssacf1 is the thing we want to interpolate, cf0 is w#asahi-gpu · 2023-04-08 14:44
alyssa 4c: a1010140 TODO.ldcf r0_r1_r2, cf1#asahi-gpu · 2023-04-08 14:44
alyssa 44: a18d004000400000 TODO.ldcf r3_r4_r5, cf0#asahi-gpu · 2023-04-08 14:44
alyssabarycentric coordinates#asahi-gpu · 2023-04-08 14:44
alyssaheart eyes#asahi-gpu · 2023-04-08 14:44
alyssaI hoped this was so!#asahi-gpu · 2023-04-08 14:44
alyssaOOOoooOOOOOoo#asahi-gpu · 2023-04-08 14:41
alyssa(disasm is wrong obviously)#asahi-gpu · 2023-04-08 14:40
alyssawhere the r1l is the 2 here#asahi-gpu · 2023-04-08 14:40
alyssa^ this is "interpolate at sample ID r1l"#asahi-gpu · 2023-04-08 14:40
alyssa 2e: 6185014002400300 TODO.iter r1, perspective, cf1, cf0, 2, no_forward, sample, centroid, 0b0#asahi-gpu · 2023-04-08 14:40
alyssaunclear if there's an extension for it#asahi-gpu · 2023-04-08 14:40
alyssathe sample ID starts at bit 32 in either case#asahi-gpu · 2023-04-08 14:40
alyssa11 -- register sample ID#asahi-gpu · 2023-04-08 14:40
alyssa10 -- pixel centroid#asahi-gpu · 2023-04-08 14:39
alyssa01 -- immediate sample ID#asahi-gpu · 2023-04-08 14:39
alyssa00 - pixel centre#asahi-gpu · 2023-04-08 14:39
alyssasample/centroid together are actually a 2-bit enum, I guess