18 Jul 2023 |
cwfitzgerald | Transmute is still big spook | 06:22:27 |
cwfitzgerald | I'm a pointer cast kinda guy | 06:23:07 |
m4b | are you saying you love the raw simplicity of pointer casting? 😎 | 06:23:51 |
m4b | the other terrible part of this is the crate is written entirely in a macro to emulate const addition/multiplication/bitshifting; i thought by now const generics would allow these things :( | 06:26:28 |
m4b | ah yea cwfitzgerald Box<T> has to implement copy to be put in [Option<Box<T>>; 10] :( | 07:00:53 |
m4b | probably i could do a MaybeUninit there, but at that point meh, raw pointer is fine; also doing stuff like this becomes awkward:
let ptr = array.get_mut_unchecked(offset).unwrap_unchecked(); // this is now an owned box, which is not what we want, as it will be dropped
| 07:02:43 |
cwfitzgerald | In reply to @m4b:matrix.org ah yea cwfitzgerald Box<T> has to implement copy to be put in [Option<Box<T>>; 10] :( No it doesn't? Modern array functions are your friend: std::array::from_fn doesn't require copy | 13:22:54 |
David Huculak | I still have yet to find a use case where I needed unsafe, but the stuff I'm writing isn't that low level. Do you find you need to use it often? Are the performance gains really that significant? | 13:56:32 |
cwfitzgerald | In reply to @davidhuculak:matrix.org I still have yet to find a use case where I needed unsafe, but the stuff I'm writing isn't that low level. Do you find you need to use it often? Are the performance gains really that significant? My day job has basically zero unsafe, outside of raw window handle - wgpu uses unsafe for performance reasons in very hot loops | 13:57:55 |
David Huculak | Nice | 13:58:33 |
David Huculak | I'm happy about not needing it because it means I get the safety cushion and still great speed. But I'm also sad about it because it creates a disincentive to learn those narsty details which seem important to know. Maybe I'll just read the nomicon | 14:00:27 |
cwfitzgerald | In reply to @davidhuculak:matrix.org I'm happy about not needing it because it means I get the safety cushion and still great speed. But I'm also sad about it because it creates a disincentive to learn those narsty details which seem important to know. Maybe I'll just read the nomicon It's not really that important tbh - most of this stuff comes up in justification for why the rules are the way they are. I mainly end up using unsafe for ffi, which is largely boring | 14:13:06 |
m4b | In reply to @cwfitzgerald:matrix.org No it doesn't? Modern array functions are your friend: std::array::from_fn doesn't require copy I am still stuck in around rust 2018 edition it seems :) | 14:50:24 |
| @neoncitylights:matrix.org left the room. | 17:05:35 |
19 Jul 2023 |
kvark | Looking at the Radeon GPU profiler output, it's quite annoying that the ISA code doesn't have association to source lines. If I'm trying to optimize a large shader program, it's hard to understand what to fix. | 04:57:43 |
cwfitzgerald | Even with spirv debug symbols (i.e. Not a Naga basted stack) ? | 05:06:52 |
kvark | pretty sure I enable SPIRV symbols. But even AMD's own documentation for this tool doesn't show any screenshots (or text) with symbols, so I'm out of hope | 05:53:53 |
cwfitzgerald | Hope you like reverse engineering unstructured contol flow 🥲 | 05:57:16 |
kvark | Ideally, it would show me the areas in the code that contribute to the VGPR/SGPR bottlenecks. In MY code, not just ISA. | 06:49:07 |
20 Jul 2023 |
jasperrlz | There's not really a great 1:1 mapping in a lot of cases | 01:49:19 |
| TAKAHASHI Tomoyuki joined the room. | 04:39:42 |
22 Jul 2023 |
| LaoLittle joined the room. | 01:32:34 |
24 Jul 2023 |
kvark | damn, that threading support had significantly worsen my Matrix experience
now there is a bunch of Unread that I'm not even able to quickly (or simetimes - ever) get rid of | 05:01:13 |
cwfitzgerald | Yup! | 05:01:32 |
cwfitzgerald | It's delightful lol | 05:01:36 |
jasperrlz | sometimes the Mark as Read button works, but only sometimes | 05:05:27 |
jasperrlz | I haven't really been impressed by anything Matrix has made software-wise, and even less impressed by their leadership. | 05:06:31 |
jasperrlz | https://github.com/vector-im/element-web/issues/24392 for reference, the issue has been open since February, and it doesn't seem like much of the software has been tested or verified against spec, considering how many issues still remain | 05:19:00 |
25 Jul 2023 |
| rajveermalviya changed their display name from rajveermalviya to rm. | 17:09:04 |
| rajveermalviya changed their display name from rm to rajveermalviya. | 17:11:10 |