2021-04-19
00:16bloom(Disclaimer: all the code for that is published but it's split across 3 repos and it'd probably be hard to reproduce as-is. This will be rectified over time..)
00:16bloomhttps://gist.github.com/alyssarosenzweig/4dc380184a4959fe609bfa26bf96dd6c are the shaders, from which you'll learn, knowing how to write a shader compiler does *not* mean you know how to write shaders that don't suck :p
00:32odmir has quit [Ping timeout: 240 seconds]
01:57odmir has joined #asahi-gpu
02:00odmir has quit [Remote host closed the connection]
02:47phiologe has quit [Ping timeout: 250 seconds]
02:48phiologe has joined #asahi-gpu
02:57odmir has joined #asahi-gpu
03:07zkrx has quit [Ping timeout: 240 seconds]
03:30zkrx has joined #asahi-gpu
03:45odmir has quit [Remote host closed the connection]
06:16M3zh[m] has joined #asahi-gpu
06:50radex has quit [Ping timeout: 260 seconds]
07:12radex has joined #asahi-gpu
09:00M3zh[m] is now known as ezh
09:04snalty has joined #asahi-gpu
09:28herbas has joined #asahi-gpu
09:29radexbloom: just curious, how do we know that AGX has "some form of out of order execution"? I thought OoO isn't really a thing in GPUs, and instead, heavy multithreading is used to hide latency, but I'm probably uninformed
09:50herbas has quit [Quit: herbas]
11:02herbas has joined #asahi-gpu
11:05Raqbit has joined #asahi-gpu
11:18herbas has quit [Quit: herbas]
12:21bloomradex: Heuristic argument.. we know GPUs need some form of perthread SIMD or VLIW or multi-issue or straight up superscalar for high perf
12:34tomtastic has quit [Quit: ZNC - https://znc.in]
12:34tomtastic has joined #asahi-gpu
12:58odmir has joined #asahi-gpu
13:05tomtastic has quit [Remote host closed the connection]
13:11radexbloom: huh, I thought "superscalar" just means "can issue multiple instructions in one cycle" and is orthogonal to OoO
13:19Yuzuit seems like the structures needed to support OoO would be expensive when architectural state is as large as it tends to be in GPUs
13:27Yuzuand if anything, the stuff in the linked presentation sounds in-order (the optimizations in slide 57, for instance, would be of fairly limited use in a true OoO machine, since independent ops would happily execute while waiting for blocked ops)
14:04vlixa has quit [Remote host closed the connection]
14:07vlixa has joined #asahi-gpu
14:16tomtastic has joined #asahi-gpu
15:56odmir has quit [Remote host closed the connection]
15:57odmir has joined #asahi-gpu
16:01odmir has quit [Ping timeout: 240 seconds]
16:28odmir has joined #asahi-gpu
16:59odmir has quit [Ping timeout: 240 seconds]
17:26odmir has joined #asahi-gpu
17:59odmir has quit [Ping timeout: 240 seconds]
18:05bloomyeah, dougall pointed out similar issues.. working theory is that it's dual-issue ALU, with memory loads as messagepassing (quick)
18:52odmir has joined #asahi-gpu
19:25odmir has quit [Ping timeout: 240 seconds]
19:54odmir has joined #asahi-gpu
20:27odmir has quit [Ping timeout: 268 seconds]
20:38taowa has quit [Ping timeout: 248 seconds]
20:38brinly has quit [Ping timeout: 248 seconds]
20:38NekomimiScience has quit [Ping timeout: 258 seconds]
20:39taowa has joined #asahi-gpu
20:40NekomimiScience has joined #asahi-gpu
20:40brinly has joined #asahi-gpu
20:49odmir has joined #asahi-gpu
21:39vlixa has quit [Remote host closed the connection]
21:43vlixa has joined #asahi-gpu
22:45odmir has quit [Remote host closed the connection]
22:46odmir has joined #asahi-gpu
22:51odmir has quit [Ping timeout: 240 seconds]
23:10odmir has joined #asahi-gpu
23:43pthariensflame has joined #asahi-gpu
23:45pthariensflame has quit [Client Quit]
23:47snalty has quit [Quit: ZNC 1.8.2 - https://znc.in]
23:47snalty has joined #asahi-gpu