Introducing Pinch, a borrow-checked language that demonstrates the power of LLVM's Multi-Level Intermediate Representation (MLIR).
MLIR is one of the coolest new compiler projects. It allows multiple languages to share high level analysis and optimizations by transforming IR dialects.
As part of my advanced network security class I did a usability evaluation of some common tasks in the #Qubes operating system.
Qubes runs all applications in separate VMs, and does an excellent job designing their interface to maintain isolation. Give it a try!
If anyone needs help falling asleep you can find the paper here:
https://badland.io/static/A_Usability_Evaluation_of_QUBES_OS.pdf
To anyone using #vulkan with #freebsd on intel machines, you might need these two things:
Patch for Mesa:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=244877
The latest drm-kmod:
https://github.com/FreeBSDDesktop/kms-drm/pull/207/commits
Unfortunately, for now you'll have to run vulkan apps as root
weird observation: on my acer's trackpad if I click with two fingers libinput only detects one finger, but if I tap with two fingers it properly detects both.
In dtrace it looks like evdev is generating the correct events, and they just get ignored by libinput unless tap-to-click is enabled? Anyone know why this is?
It's always nice when rewriting something drops the CPU usage from 100% per thread to 6%.
It's funny that the best way to learn device standards like wifi 6 and architectures like Turing is to first read all of the previous specs and then see what's been added in the latest generation.
Bro I'm 22 and I need to read 22 years worth of documents before I'm up to speed on how to write drivers.
To get a rough idea about the impact of WPA2 crypto hardware offloading, tobhe@ has generated two flamegraphs with the new dt(4) dynamic tracer written by mpi@.
These graphs plot a "profile:hz:11" bt(5) probe before and after the above patch.
Before: https://tobhe.de/files/wifi.svg
After: https://tobhe.de/files/wifi-new.svg
It looks like iwm_tx() disappears in the second graph. Its overhead is reduced such that it never showed up whenever the profiler took its periodic kernel stack snapshot
That unreal 5 demo is incredible. I'm dying to know how they made streaming meshes so performant.
If streaming film quality assets requires a hyper-fast m.2 like in the PS5, will games even be able to expect players to have that kind of power? Next-gen consoles might but not all PC players will.
At the moment it's recommended to have state-of-the-art storage, in the future it might be mandatory for games that do this kind of streaming.
Introducing Pinch, a borrow-checked language that demonstrates the power of LLVM's Multi-Level Intermediate Representation (MLIR).
MLIR is one of the coolest new compiler projects. It allows multiple languages to share high level analysis and optimizations by transforming IR dialects.
Currently the utility can install some software actually... #WorkInProgress
As part of my advanced network security class I did a usability evaluation of some common tasks in the #Qubes operating system.
Qubes runs all applications in separate VMs, and does an excellent job designing their interface to maintain isolation. Give it a try!
If anyone needs help falling asleep you can find the paper here:
https://badland.io/static/A_Usability_Evaluation_of_QUBES_OS.pdf
Much like how Native Americans used every part of the buffalo, #llvm developers use every part of the C++ spec. This is making me reevaluate my prior assumption that I actually know this language.
@stsp I'm reading through the #FreeBSD rtwn driver that seems to be a descendent of your #OpenBSD driver and I'm curious what hardware documentation (if any) you referred to while writing it? I can't seem to find any datasheets or anything online, but maybe I'm not looking in the right place.
I have a netgear A6100 that is excruciatingly slow and I'm trying to figure out why.
To any other rustaceans, don't use mpsc::Receiver::recv_timeout:
https://github.com/rust-lang/rust/issues/39364
Use try_recv and add thread::sleep instead.
Here's to hoping the bug, which is originally from 2017, will be fixed soon.
If you're bored in quarantine and think compilers/GPUs are cool give this paper a read.
HPVM - Heterogeneous Parallel Virtual Machine. They create a virtual instruction set for parallel processors whose IR can be translated into GPU (Nvidia, intel) or multithreaded CPU code.
Does anyone know how mastodon efficiently aggregates the federated timeline? I don't know ruby so reading the source isn't going well.
I can find loads of docs on the API and how to use it, but none on how it works.
You would think that display planes in #vulkan would correspond to a drm plane in mesa, but nope. Instead drm connectors are used.
I think switching mesa to use drm planes would allow for hardware compositing, but would probably involve a significant amount of work. It would be extremely useful to #wayland compositors though, and would offer some real efficiency improvements.
#GhostBSD is a desktop distro of #FreeBSD, but it seems to consistently be ahead in benchmarks.
https://www.phoronix.com/scan.php?page=article&item=dfly58-freebsd121-ubuntu2004&num=9
They must have played with some lesser known compilation options. If anyone knows what they did I'd be very curious to hear.
Interested in climbing, Rust, FreeBSD, and all things computer graphics related.