| 17 May 2025 |
 dramforever | In reply to @lotte:chir.rs
presumably both have working hardware aes, but the sha256 implementation seems to be 3x slower than on the cm4 NOPE | 17:42:31 |
| dramforever | no hardware AES on vf2 | 17:42:36 |
 @lotte:chir.rs | how the hell is the cm4 slower then | 17:42:55 |
| dramforever | i alsp doubt the crypto engine on vf2 is even supposed to be fast | 17:43:00 |
| dramforever | magic | 17:43:17 |
| dramforever | In reply to @dramforever:matrix.org
no hardware AES on vf2 okay tbc i mean aes instructions | 17:44:21 |
| dramforever | maybe they are using the crypto engine on vf2 and i was mistaken about its speed? | 17:46:00 |
| @lotte:chir.rs | i’m still sad about the B extension in riscv because it had some Very Interesting instructions | 17:46:33 |
| dramforever | zba_zbb is really nice | 17:47:06 |
| dramforever | i expect those to close up the difference with arm on basic operations | 17:48:01 |
| @lotte:chir.rs | zba is useful for indexing, zbb are instructions that really should have just been part of the base abi | 17:48:03 |
| @lotte:chir.rs | * zba is useful for indexing, zbb are instructions that really should have just been part of the base isa | 17:48:06 |
| dramforever | still no match for neon | 17:48:09 |
| dramforever | one day nixos rva23 will happen | 17:49:20 |
| dramforever | should happen ~the same time doing native builds become viable | 17:49:45 |
 Alex | It's already viable if you're willing to wait over a week :D | 17:50:30 |
| dramforever | In reply to @dramforever:matrix.org
should happen ~the same time doing native builds become viable (yes, qemu-user exists. good luck figuring out all the problems it causes.) | 17:51:30 |
| @lotte:chir.rs | it’s mostly just tests | 18:01:18 |
| Alex | The tests that have timeouts are especially annoying.
You spend 30min building a single package only for checkPhase to fail because the author was too strict with their timeouts. | 18:03:03 |
| @lotte:chir.rs | and there is mypy | 18:03:32 |
| @lotte:chir.rs | which takes like 10+ hours to build | 18:03:50 |
| @lotte:chir.rs | despite being a python application written in 100% python | 18:04:00 |
 Tristan Ross | In reply to @lotte:chir.rs
and there is mypy gfortran took like 14 hours | 18:13:25 |
| Tristan Ross | LLVM took like 8 I think lol | 18:13:36 |
| Alex | GHC takes 20h. Entirely justified. | 18:13:46 |
| Tristan Ross | Currently building the kernel | 18:13:47 |
| Tristan Ross | Lol, GHC on my Ampere Altra Max is done fast | 18:14:05 |
| Alex | I'd believe it, <2h on my desktop. | 18:14:20 |
| Tristan Ross | Takes like 15 minutes for me | 18:14:52 |
| Tristan Ross | Most packages on Ampere don't go over an hour. | 18:15:14 |
