Gabriel
LightNews LightNews
banner
dssgabriel.bsky.social
Gabriel
@dssgabriel.bsky.social
18 followers 85 following 18 posts
PhD candidate, HPC Software Engineering @cea.fr / DAM
MSc HPC & Simulation from @univparissaclay.bsky.social
Architecture, microbenchmarking & SIMD sorcery. Research on distributed computing, data structures & memory layouts at exascale.

RTFM 👹
Posts Replies Media Videos
Reposted by Gabriel
matt.godbolt.org
Day 9: Why does the compiler keep an "expensive" multiply in the loop instead of using clever addition tricks? The slower operation enables something more valuable. What's the hidden benefit?

xania.org/202512/09-in...
youtu.be/vZk7Br6Vh1U
#AoCO2025
Induction variables and loops — Matt Godbolt’s blog
Compilers can rewrite loops to avoid expensive calculations
xania.org
December 9, 2025 at 12:56 PM
Reposted by Gabriel
nextsilicon.com
The Spectra supercomputer at Sandia National Laboratories is now live - powered by 128 #NextSilicon Maverick-2 accelerators.

A first-of-its-kind system built on our Intelligent Compute Architecture for faster, more efficient HPC.

🔗 Full report here:
tinyurl.com/yc7fy5x8
December 8, 2025 at 3:43 PM
Reposted by Gabriel
shafik.bsky.social
Cursed code of the day

#programming
godbolt screen shot Code: #include <stdio.h> volatile double d = -32.0; int main(void) { printf("%f\n%lx\n%lx\n", d, (unsigned long)d, (unsigned long)-32.0); } clang armv8 output: -32.000000 0 ffffe8ed40f8 gcc ARM64 output: -32.000000 0 0 Clang x86-64 output: -32.000000 ffffffffffffffe0 7ffde9ce61f8 gcc x86-64 output: -32.000000 ffffffffffffffe0 ffffffffffffffe0
December 8, 2025 at 4:51 PM
Reposted by Gabriel
hgpu.bsky.social
Microbenchmarking NVIDIA’s Blackwell Architecture: An in-depth Architectural Analysis

#PTX #CUDA #Benchmarking #Blackwell #HPC

hgpu.org?p=30437
Microbenchmarking NVIDIA’s Blackwell Architecture: An in-depth Architectural Analysis
As GPU architectures rapidly evolve to meet the overcoming demands of exascale computing and machine learning, the performance implications of architectural innovations remain poorly understood acr…
hgpu.org
December 7, 2025 at 9:30 PM
Reposted by Gabriel
matt.godbolt.org
Day 8 of Advent of Compiler Optimisations! 🔄

Index-based for vs pointer while vs range-for vs std::accumulate—which is fastest? Three produce identical assembly, but one doesn't!

xania.org/202512/08-go...
youtu.be/FB8Hgj3TpJM

#AoCO2025
Going loopy — Matt Godbolt’s blog
Exploring the ways optimisers deal with loop constructs
xania.org
December 8, 2025 at 1:10 PM
Reposted by Gabriel
matt.godbolt.org
Day 7: Dividing by 10 compiles to... no division at all!

How does the compiler pull this off? Magic constants, bit shifts, and a clever arithmetic trick.

xania.org/202512/07-di...
youtu.be/V9Pvv1tkocM

#AoCO2025
Multiplying our way out of division — Matt Godbolt’s blog
How compilers avoid expensive division with multiplication tricks
xania.org
December 7, 2025 at 11:07 AM
Reposted by Gabriel
instlatx64.bsky.social
#Intel released the "Intel #Xeon6 vs #AMD #EPYC Competitive Infographic" pdf:
cdrdv2-public.intel.com/859022/intel...
December 7, 2025 at 7:27 PM
Reposted by Gabriel
matt.godbolt.org
Day 6 of Advent of Compiler Optimisations! Divide by 512—just a shift, right? But the compiler adds extra instructions. Why? A subtle difference between what you asked and what you meant!

xania.org/202512/06-di...
youtu.be/7Rtk0qOX9zs

#AoCO2025
Division — Matt Godbolt’s blog
Division doesn't have to be slow with some clever tricks
xania.org
December 6, 2025 at 12:49 PM
Reposted by Gabriel
matt.godbolt.org
Day 5 of Advent of Compiler Optimisations! x86 has LEA, but ARM has the barrel shifter—instructions can shift operands cheaply. The compiler uses this to multiply without multiplying!

xania.org/202512/05-ba...
youtu.be/TZubUyr2UEY

#AoCO2025
ARM's barrel shifter tricks — Matt Godbolt’s blog
The ARM architecture has a cool feature, and compilers know how to use it
xania.org
December 5, 2025 at 12:05 PM
Reposted by Gabriel
matt.godbolt.org
Day 4 of Advent of Compiler Optimisations! Multiply by constants—which ones use actual multiply? The compiler has tricks to avoid it, then saves you from your own clever hacks.

xania.org/202512/04-mu...
youtu.be/1X88od0miHs

#AoCO2025
xania.org
December 4, 2025 at 12:07 PM
Reposted by Gabriel
phoronix-poster.bsky.social
NVIDIA Releases CUDA 13.1 With New "CUDA Tile" Programming Model - https://www.phoronix.com/news/NVIDIA-CUDA-13.1
NVIDIA Releases CUDA 13.1 With New "CUDA Tile" Programming Model
NVIDIA just released CUDA 13.1 for what they claim is "the largest and most comprehensive update to the CUDA platform since it was invented two decades ago." The most notable addition with the CUDA 13.1 release is CUDA Tile as a new tile-based programming model...
www.phoronix.com
December 4, 2025 at 10:56 PM
Reposted by Gabriel
steveklabnik.com
#jj-vcs 0.36.0 us out! Including finally moving the documentation to our own domain, docs.jj-vcs.dev

github.com/jj-vcs/jj/re...
Release v0.36.0 · jj-vcs/jj
About jj is a Git-compatible version control system that is both simple and powerful. See the installation instructions to get started. Release highlights The documentation has moved from https:/...
github.com
December 4, 2025 at 11:41 PM
Reposted by Gabriel
matt.godbolt.org
Day 3 of Advent of Compiler Optimisations! A while loop, recursion, direct addition—all compile to one instruction. The compiler sees the pattern beneath the code.

xania.org/202512/03-mo...
youtu.be/wHg9lYPMvvE

#AoCO2025
You can't fool the optimiser — Matt Godbolt’s blog
Pattern recognition can see through obfuscated code to find the right instruction
xania.org
December 3, 2025 at 12:00 PM
Reposted by Gabriel
matt.godbolt.org
Day 2: Adding two integers on x86? Not with `add`! The compiler uses a completely different instruction—one designed for memory addressing. Why? xania.org/202512/02-ad... youtu.be/BOvg0sGJnes #AoCO2025
Addressing the adding situation — Matt Godbolt’s blog
We learn why adding on x86 isn't as obvious as you might think
xania.org
December 2, 2025 at 11:28 AM
Reposted by Gabriel
matt.godbolt.org
Ever wonder why compilers use `xor eax, eax` to zero registers? It's smaller AND faster—CPUs optimise it out entirely!

Day 1 of Advent of Compiler Optimisations: xania.org/202512/01-xo...
Video: youtu.be/eLjZ48gqbyg

#AoCO2025
Why xor eax, eax? — Matt Godbolt’s blog
Why do compilers love xor-ing registers so much?
xania.org
December 1, 2025 at 12:12 PM
Reposted by Gabriel
glennklockwood.com
I wrote up my notes from #SC25. Have a look: blog.glennklockwood.com/2025/12/sc25...

I’ll keep picking away at the editing, but would love to hear more from others about what stood out to them. I wasn’t at the conference itself as much this years as in the past, so I know I missed a lot.

#HPC
SC'25 recap
The annual SC conference was held last week, drawing over 16,000 registrants and 560 exhibitors to in St. Louis, Missouri to talk ab...
blog.glennklockwood.com
December 1, 2025 at 7:27 PM
Reposted by Gabriel
steveklabnik.com
From Zero to GitHub: Starting A New #jj-vcs Repo

www.visualmode.dev/from-zero-to...
From Zero to GitHub: Starting A New jj (Jujutsu) Repo
I document the steps and missteps that I take setting up a new jujutsu repo and pushing it to GitHub
www.visualmode.dev
December 3, 2025 at 8:32 PM
Reposted by Gabriel
eurohpc-ju.bsky.social
 📣 The procurement contract for #AliceRecoque, the new European #exascale supercomputer 🖥️ ⚡ located in #France, has been signed by @eurohpc-ju.bsky.social  and the selected vendor Eviden! It will be one of the backbones for Europe's network of #AIFactories

🔗 www.eurohpc-ju.europa.eu/contract-sig...
November 18, 2025 at 11:01 AM
Reposted by Gabriel
hpsf.bsky.social
We’re delighted to welcome Modules to the High Performance Software Foundation as an established project 🎊

Read the announcement ➡️ hpsf.io/blog/2025/hi...
November 10, 2025 at 2:25 PM
Reposted by Gabriel
instlatx64.bsky.social
#AMD #Zen6 znver6 ISA:
- #AVX512_BMM (CPUID.80000021.EAX[23], VBMACOR16x16x16, VBMACXOR16x16x16, VBITREV)
- #AVX512_FP16
- #AVX_NE_CONVERT
- #AVX_IFMA
- #AVX_VNNI_INT8
Source:
sourceware.org/pipermail/bi...
instlatx64.bsky.social InstLatX64 @instlatx64.bsky.social · Oct 29
#AMD CPUID refresh:
-- CPUID B80F00 is #Zen6 [1]
-- #Zen5c #Sorano CPUID is B10F00 (#EPYC 8005, SP6 socket) [2],[3]
-- #Zen5 #Grado CPUID is B40F40 (EPYC 4005, AM5 socket) [4]
GitHub:
github.com/InstLatx64/I...
1/2
November 8, 2025 at 1:11 AM
Reposted by Gabriel
phoronix-poster.bsky.social
NASM 3.00 Assembler Is Ready With Intel APX & AVX10 Support - https://www.phoronix.com/news/NASM-3.00-APX-AVX10
NASM 3.00 Assembler Is Ready With Intel APX & AVX10 Support
Slipping under my radar in October was the release of NASM 3.00 and the follow-up NASM 3.01 release shortly there after. This widely-used open-source assembler is now ready with support for Intel's Advanced Performance Extensions (APX) and AVX10...
www.phoronix.com
November 2, 2025 at 1:53 PM
Reposted by Gabriel
thephd.dev
Iiiiiit's published.

www.open-std.org/JTC1/SC22/WG...
N3694: Functions with Data - Closures in C (A Comprehensive Proposal Overviewing Blocks, Nested Functions, and Lambdas)
www.open-std.org
November 2, 2025 at 6:58 PM
Reposted by Gabriel
cea.fr
#anniversaire 🎂 | Le CEA fête ses 80 🕯️ .
Le 18/10/1945, le Général de Gaulle signe l’ordonnance fondatrice du CEA.

⚛️L'organisme a pour ambition d’offrir à la 🇫🇷 la maîtrise de l’atome dans différents domaines de la science, de l’industrie & de la défense nationale

👇Extrait
Extrait de l'ordonnance de création du CEA datant du 18 octobre 1945 © Archives Nationales \ Stéphane Méziache
October 18, 2025 at 6:58 AM
Reposted by Gabriel
lauriewired.bsky.social
Colleges do a terrible job of teaching C++.



It’s not “C with Classes”. Injected into curriculums as a demonstration of early CS concepts, it leaves many with a sour taste.



Students later immediately fall in love with the first language that *doesn’t* feel that way.
October 13, 2025 at 9:21 PM
Reposted by Gabriel
instlatx64.bsky.social
#AMD & #Intel unified future instructions:
#FRED #AVX10 #ChkTag #ACE (=ACE (Advanced Matrix Extensions for Matrix Multiplication): www.amd.com/en/blogs/202...
October 14, 2025 at 7:17 AM