
LLVM, RISC-V and OpenTitan at lowRISC. Dabbles with hardware. @luismarques@mastodon.gamedev.place
Lisbon, Portugal
Joined September 2007
- Tweets356
- Following362
- Followers190
- Likes251
Cool. I just managed to optimize @nothings' stb_voxel_render voxel meshing approach from 20 bytes per face to 8 bytes per face.
1
279
Luís Marques retweeted
We made history!🎉The #OpenTitan® coalition, hosted by lowRISC, announced the 1st #opensource #silicon project to reach commercial availability with validated chips! Unprecedented support & investment by our project partners since 2018 made this possible. lowrisc.org/news/2024/02/ope…
Luís Marques retweeted
Three years ago, I got some footage that I couldn’t get out of my mind.
It showed the first police killing captured on body-worn cameras in NY.
The NYPD didn’t want anyone to see it.
It eventually launched me on the biggest story I’ve ever done, w/ @NYTmag
[THREAD]
Luís Marques retweeted
🤔 Want to know why we love OpenTitan?
It’s the first open source project building a transparent, high-quality reference design and integration guidelines for silicon RoT chips.
Check it out → goo.gle/3RYWF5M
Luís Marques retweeted
Antmicro is proud to present first #UVM testbenches running fully open source in #Verilator! Made possible by our collaboration with @westerndigital. Learn about our quest for this major milestone & the implementation technicalities: antmicro.com/blog/2023/10/ru… @CHIPSAlliance @Google
Luís Marques retweeted
Exciting news! The Sunburst project is paving the way for #CHERI technologies in #RISCV embedded platforms. Thanks to @UKRI_News' Digital Security by Design program via @innovateuk for the support. Discover our work on capability enhanced processors. sunburst-project.org/
Luís Marques retweeted
Mike's @MikeRogenmoser paper on "Trikarenos: A Fault-Tolerant RISC-V-based Microcontroller for CubeSats in 28nm" is online. Learn about this PULPissimo-based SoC with On-Demand Redundancy Grouping, a configurable lockstep approach with 3 Ibex cores here: arxiv.org/pdf/2310.02045.pdf
The Cranelift docs claim that the LLVM MC layer is used for branch relaxation. What's that about? 🤔github.com/bytecodealliance/…
71
Luís Marques retweeted
If you’re into @rustembedded you won’t want to miss this talk at #rustnationuk
We’re excited to have folks from @lowRISC both sponsoring and speaking. Check out the website schedule for talk details
#rustlang 🦀
I wonder at what age did baby @seanmcarroll start to believe the many spaces interpretation of reality. It’s a big ask to believe that mommy continues to exist after you’ve stopped seeing her…
Luís Marques retweeted
Leaving Google and starting working on mold was a bet, and it's going well so far. Now the idea to try to replicate the success of mold in C++ is growing on me. It feels like we might be able to write a 10x faster C++ compiler if we really focus on speed. Just thinking.
What began as a conflict over the transfer of consciousness from flesh to machine has escalated into a war that has decimated a million worlds. The RISC-V and the Arm have all but exhausted the resources of the entire galaxy in their struggle for domination.
Luís Marques retweeted
LLVM Weekly - #418, January 3rd 2022. 8th anniversary of LLVM Weekly (and no missed issues yet!), more 2021 LLVM Developers' Meeting videos posted, new Mips backend pass to avoid a hardware bug in the ~25 year old VR4300, and more. llvmweekly.org/issue/418
Luís Marques retweeted
=>
"A #RISCV in-network accelerator for flexible high-performance low-power packet processing", ISCA 2021 spcl.inf.ethz.ch/Publication…
@LucaBeniniZhFe @thoefler
First open-source sPIN implementation, based
on a multi-cluster @pulp_platform 22 nm
400 Gbit/s
sPIN nitter.gq/ogawa_tter/status/1311…
Replying to @ogawa_tter
=>
"General in-network processing - time is ripe!", @thoefler Keynote, High-Performance Interconnects Forum, HPC China, Sep 30, 2020
45:08 youtube.com/watch?v=t6jdjnnI…
PDF (10 MB) htor.inf.ethz.ch/publication…
sPIN on @pulp_platform
SC19 nitter.gq/ogawa_tter/status/1224…
nitter.gq/ogawa_tter/status/1309…
SC20
Luís Marques retweeted
Annoyed by a slow link step of a build? If so, check out my new linker github.com/rui314/mold. It's experimental but extremely fast on multicore machines.
Luís Marques retweeted
Algorithms become biased when we train them on biased humans. But we can train them to be better than humans. Here’s an example, algorithms that listen to patients rather than doctors. h/t @hilzoy
Medical puzzle:
Take Black vs. White patients with similar knee x-rays
Why do Black patients have more pain?
Our paper in @NatureMedicine provides one answer: go.nature.com/3i6zAuc
Algorithms see causes of knee pain in Black patients, that human radiologists miss.
🧵






























