Don't know why I'm realizing this only today, but #fosdem is an impressive OSS event with its 1102 presentations https://fosdem.org/2025/schedule/events/
=> More informations about this toot | View the thread
Rider 2025.1 is coming with mixed-mode debugging https://blog.jetbrains.com/dotnet/2025/01/08/rider-2025-1-roadmap/#mixed-mode-debugging π₯³
That plus hopefully, later this year (or next?), a rework of the debugging windows to use a table instead of unaligned variables and I could start to switch from VS+R#! Hope that the debugger will allow to also investigate raw/managed memory directly and debug steps at the assembler level π
[#]dotnet
=> More informations about this toot | View the thread
I've been often explaining around me the clear distinction to make between complex & complicated. Asking ChatGPT to formulate this with a few catchy phrases and it's pretty damn good:
=> More informations about this toot | View the thread
Great, now I can monthly contribute to https://opencollective.com/alpinelinux π₯³
=> More informations about this toot | View the thread
Making some progress of my ARM64 assembler, generating automatically all encoding of instructions (still some operands not handled) and I can directly write an assembler line in C# and gets everything resolved to a const! β¨
=> More informations about this toot | View the thread
Finally, 10468 tests passing for my ARM 64 disassembler - I'm reusing part of Capstone unit tests, which were based on LLVM-MC tests π₯³
The decoding of operands was a lot more manual than I thought, as the ARM XML specs don't formalize this but via a free form text, which makes it super laborious. There are also several instructions that require manual dispatching based on conditions. And it doesn't even cover SME/SVE/SVE2 (it doubles everything!)
Anyway, next step: the assembler! π
[#]dotnet
=> More informations about this toot | View the thread
I'm getting there with my ARM64 assembler/disassembler .NET lib, zero allocation for the decoding and quite fast! π
Not visible in the screenshot, but there is a significant amount of work involved to get that out! π
Still plenty of work remaining for the disasm to be fully complete, handling other operands (e.g. label, extend...), prettify them when in groups (e.g. labels) ...etc. π€―
And then, generate a strongly typed assembler API π»
Should be pretty cool once finished! π
[#]dotnet #csharp
=> More informations about this toot | View the thread
It's a kind of a weird marketing moment for the Switch 2 to just get out with a trailer about mainly its formfactor, and nothing else. So much wait from the market for this? π€
=> More informations about this toot | View the thread
@nietras I'm waiting for the Phi-4 ONNX version to be available (tried optimum to convert and failed installing it). I could go with LLamaSharp but dunno if it is worth it. With which runtime representation do you mostly work with these days?
=> More informations about this toot | View the thread
Me, last week: oh oh, detecting ARM64 instruction from a plain uint is easy, the rest should be similar. π
Me, this week: shit, automating the decoding of operands is way trickier than I expected π€―
Firstly, because operands are not formalized in the ARM64 XML specs apart through some pseudo-structured text that you have to parse.
Secondly, the encoding of some operands can be super convoluted.
Hope I'll have enough patience to finish this. I thought that I could do it quicker & move on π¨
=> More informations about this toot | View the thread
Fuck all the enablers of Trump and especially the billionaires π€¬
=> More informations about this toot | View the thread
I'm pretty excited by the NVIDIA DIGITS GB10 Gracewell https://nvidianews.nvidia.com/news/nvidia-puts-grace-blackwell-on-every-desk-and-at-every-ai-developers-fingertips for doing neural networks experiment at scale, right from home, but that form factor is so ugly and ridiculous. It looks like a Luxury/LVMH GPU box while I just really need a dev kit π
=> More informations about this toot | View the thread
First version of the ARM64 ISA decoding is working! π₯³
4241+ instructions (including SVE/SVE2/SME...etc.) decoded with an automatically generated decoding table that is less than 50KB. It might be the most compact decoding table for ARM64 ISA!
Decoding speed is 10M/s, dunno if it is fast enough or not, will check that later
[#]dotnet
=> More informations about this toot | View the thread
In the middle of my holidays π
I'm playing with the idea to build a C# library to assemble/disassemble ARM64 ISA with hopefully 99% of it generated from ARM ISA XML specifications. π€
But the weirdest part is that... I'm not entirely sure why I want to build that π
At least, I feel the challenge is interesting π€
We will see! βΊοΈ
[#]dotnet #csharp
=> More informations about this toot | View the thread
When travelling with my laptop, I'm always feeling inefficient and unproductive when coding on it, missing double big screens, a proper mouse and keyboard. It's probably a matter of getting use to it, but I can't π
=> More informations about this toot | View the thread
Not completely surprising in this post about "Performance of llama.cpp on Snapdragon X Elite/Plus" https://github.com/ggerganov/llama.cpp/discussions/8273#discussioncomment-11356628
This NPU seems to run significantly slower even than its CPU. Only 1.5% of the claimed 45 TOPS performance. They should be sued for false advertising - like probably many others. π
=> More informations about this toot | View the thread
Spent the weekend trying to make EventSource/EventPipe working with NativeAOT on macOS, but failed, starting a session with a provider doesn't enable the EventSource... I will have to create a simpler repro, try also on Linux and log an issue otherwise. Really not happy with the energy lost to make it work π
[#]dotnet
=> More informations about this toot | View the thread
@ncopa hey, I have a question regarding alpine: are frame pointers enforced by apkbuild? (Or not removed basically)
=> More informations about this toot | View the thread
In another news, I'm starting to add support for (native callstacks) profiling to ultra on macOS. I have to say that Windows with its ETW Kernel session is so much better equipped than macOS. It's absurd that I have to do my own Thread.Sleep/suspend/stackwalk/resume + dylib injection to get such basic stuffs. π
At least I'm going to be able to rely on .NET (Core) tracing infrastructure which is pretty damn good for collecting data from a remote process. π
[#]dotnet
=> More informations about this toot | View the thread
CTRL+β/β vs Command+β/β
or CTRL+X/C/V vs Command+X/C/V
are killing my productivity whenever I have to switch between Windows and macOS, I don't know if there are many folks constantly switching during the day between these OS, but it is one of these annoyances that I find super hard and I understand why one would stick to one OS for good for daily use. Maybe that's just because I'm getting old. βΊοΈ
In the end, I have to develop on Windows, and git sync my work to mac to test it! π
=> More informations about this toot | View the thread
=> This profile with reblog | Go to xoofx@mastodon.social account This content has been proxied by September (3851b).Proxy Information
text/gemini