#asahi-gpu · 2021-05-25 03:15
bloom`writeout` is probably a misnomer though#asahi-gpu · 2021-05-25 03:15
bloomi'm almost certain you get them offscreen too#asahi-gpu · 2021-05-25 03:14
dougall(i think i was doing offscreen rasterisation tests, and you don't get a writeout instruction there? it just stores to tile memory?)#asahi-gpu · 2021-05-25 03:13
bloompff#asahi-gpu · 2021-05-25 03:12
dougallyeah - having given it some thought, i definitely have no idea :P#asahi-gpu · 2021-05-25 03:05
bloomyeah, spec says prior to Metal 2.3 it was discard(), Metal 2.3+ it's demote()#asahi-gpu · 2021-05-25 03:03
bloomIIRC#asahi-gpu · 2021-05-25 03:03
bloomerrrr except Metal's discard is actually demote()#asahi-gpu · 2021-05-25 03:03
bloomif discard is used, that interacts (keep in mind `discard` is just a glorified `coverage mask = 0` write!)#asahi-gpu · 2021-05-25 03:03
bloomnah, end of frag#asahi-gpu · 2021-05-25 03:02
dougalloh, i'm getting mixed up i guess - i thought it was at the end of the vertex shaders?#asahi-gpu · 2021-05-25 03:02
bloomif fragment depth/stencil/coverage is written, that interacts#asahi-gpu · 2021-05-25 03:02
bloombut then it gets trickier#asahi-gpu · 2021-05-25 03:02
bloomin a frag shader with multiple outputs (render targets), two before the frist blend and one before each subsequent blend#asahi-gpu · 2021-05-25 03:02
bloomer, two before blend#asahi-gpu · 2021-05-25 03:02
bloomin a simple frag shader, right before `blend`#asahi-gpu · 2021-05-25 03:01
bloomit definitely related to writeout, in that I know /where/ it executes#asahi-gpu · 2021-05-25 03:01
bloomsince it executes /after/ the ld_vary instruction in usual circumstnaces#asahi-gpu · 2021-05-25 03:01
bloomuhm, inteprolation is unlikely#asahi-gpu · 2021-05-25 03:01
bloominstruction hot potato#asahi-gpu · 2021-05-25 03:01
bloom~~i thought you named it~~#asahi-gpu · 2021-05-25 02:58
dougall(not in any depth - i think roughly setting up the cf state for interpolation - but i think you named it and probably understand better)#asahi-gpu · 2021-05-25 02:55
bloom(Do we know what TODO.writeout actually does?)#asahi-gpu · 2021-05-25 02:55
bloomalso it has some weird interactions with TODO.writeout#asahi-gpu · 2021-05-25 02:55
bloomI haven't implemented discard yet since they don't have a discard_if which meant having if-stmts working was a prereq#asahi-gpu · 2021-05-25 02:54
bloom`discard` is separate#asahi-gpu · 2021-05-25 02:54
bloomwell#asahi-gpu · 2021-05-25 02:54
dougall:D#asahi-gpu · 2021-05-25 02:54
bloomdougall: it's everything :)#asahi-gpu · 2021-05-25 02:54
dougallhahaha that's awesome! does that imply most/all control flow is working, or do the loop tests avoid mixing other structures?#asahi-gpu · 2021-05-25 02:52
bloomdougall: ^ dEQP-GLES2.functional.shaders.loops.*#asahi-gpu · 2021-05-25 02:51
bloom Passed: 625/625 (100.0%)#asahi-gpu · 2021-05-25 02:29
bloomdougall: You did the hard part I just typed it in :p#asahi-gpu · 2021-05-25 02:04
dougall:o well done!#asahi-gpu · 2021-05-25 01:19
bloommore work remains but hey#asahi-gpu · 2021-05-25 01:18
PthariensFlame[m]Yay!#asahi-gpu · 2021-05-25 01:17
bloomthanks for the delicious notes#asahi-gpu · 2021-05-25 01:17
bloomdougall: Passed my first test with loops#asahi-gpu · 2021-05-24 22:16
choozyAh#asahi-gpu · 2021-05-24 22:09
bloomwrong window#asahi-gpu · 2021-05-24 21:06
choozy!ss ?#asahi-gpu · 2021-05-24 18:29
bloom!ss#asahi-gpu · 2021-05-24 14:19
yrlfon something with structured control-flow just jumping into the middle of something won't work I presume#asahi-gpu · 2021-05-24 14:19
yrlfon non-structured architectures you can avoid the duplicated condition by just jumping forward into the condition part of the loop (at the cost of an additional branch)#asahi-gpu · 2021-05-24 14:18
yrlfoh, that is... very interesting#asahi-gpu · 2021-05-24 12:49
bloomwith do...while as the loop priitives#asahi-gpu · 2021-05-24 12:49
bloomAGX is really, really weird in that its GPU asm is based on structured CF#asahi-gpu · 2021-05-24 12:49
bloomyrlf: nod#asahi-gpu · 2021-05-24 12:35
jixnow whether you can reliably know and use that within a compiler? no idea#asahi-gpu · 2021-05-24 12:34
jixyrlf: I think if the loop condition is complex, the loop usually has a low iteration count and the whole loop is executed frequently, in that case not wasting icache on the duplicated condition might make things faster