#Blog #RISCV #SIMD

RISC-V Vector Basics and Development Environment

https://blog.yuuta.moe/2023/08/08/rvv-dev-env/

Our #SIMD capable #DNS zone file parser is still getting faster! It can now parse the .com zone in 16.344s on Jeroen’s laptop. That being said, optimising for the remaining record types is now getting seriously complicated. https://github.com/NLnetLabs/simdzone/issues/68

3 GB/s for #IPv4 parsing... 😳 #SIMD
https://github.com/NLnetLabs/simdzone/pull/73

📢 New ORC 0.4.34 release of the Optimised inner loop Runtime Compiler!

New parser API with better error reporting, thread-safety improvements around the orc codemem allocator and many smaller improvements, bug fixes and build fixes.

Enjoy!

https://lists.freedesktop.org/archives/gstreamer-announce/2023-May/000537.html

#GStreamer #Release #opensource #FreeSoftware #Multimedia #SIMD #Assembly

Most of the NLnet Labs crew is going to be at #RIPE86! Here's who to find for our #OpenSource #DNS projects:
🧑‍🔬 Interested about our #OpenStandards and #research work? Talk to @willem or @benno.
🏎️ Want to know how our super fast #SIMD zonefile parser is coming along? Chat with Jeroen.
🧷 Curious about our focus on #MemorySafety and DNS implementations in #rustlang ? Watch Philip's presentation in dns-wg to learn more...

We’re very excited that Daniel Lemire is contributing to our blazingly fast, #SIMD enabled #DNS zone file parser. #OpenSource https://lemire.me/en/ https://github.com/NLnetLabs/simdzone

Ever wanted to use #Rust to manipulate images like iterators?

image
.map(to_rgb)
.filter(|r, g, b| b > 100)
.collect();

Well, I don't quite have that, but I do have a #monadic-ish library called #obraztam.

https://framagit.org/dcz/obraztam

It's #immutable-first, it has .map() and .zip() and it optimizes to #SIMD instructions. Can be easily parallelized even more.

I used it to detect #laser beams in the picture!

#graphics #monad #iterator

As of now the closest to what I want is the #KNC instruction vpmadd231d zmm, zmm, zmm which can deal with EPI32

Idea is lowering power requirements while also taking advantage of the fp32 pipes extremely high performance, which also happens to have lower latency than the integer #SIMD unit (both CPI 0.5, but FP unit is lat 4 vs integer being lat10)

I need more than EPI16 native, but EPI32/64 is a waste of power and precision.

I'd also rather not do horrible things in the FP unit...

Our superfast #DNS zone file parser now has fallback functionality to support architectures for which no #SIMD
implementation is available yet. #OpenSource #clang https://github.com/NLnetLabs/simdzone/commit/88028c6e96ffe0fa85fb2c54fced7b4517a3c303

With just ONE line, decorate a #Python function with ‘@njit’ to parallelize it automatically. Multi-core (CPu and GPU) and SIMD vectorization supported. *jaw drops* 😲

https://numba.pydata.org/

#simd #parallelism #numba #jit #decorators

Hi everyone!

Was chatting with a few students and enthusiasts about the how part of why CPUs are so fast. It covers SIMD, multiple ALUs and Out of Order Execution.

Converted it into a Git Gist and added some more practical examples that should be approachable to anyone who's written a `for` or `while` loop at some point.

Please find it here: https://gist.github.com/FCLC/8322c7e0fd8746d256178c329e9d0de4

#CPU #SIMD #HPC #x86 #arm #C #OoO #OOE

Want to hear how we are targeting 1GB/s to parse a #DNS zone file? Jeroen's #FOSDEM23 talk is now live! Lots of details in the Q&A at the end. #FOSDEM #OpenSource #SIMD #SoftwareDevelopment https://fosdem.org/2023/schedule/event/dns_parsing_zone_files_really_fast/

This Saturday in the #DNS Devroom at @fosdem Jeroen will be presenting his work on simdzone, a fast and standards compliant DNS zone file parser. 🚀

Running simdzone on an Intel Core i7-1065G7 against an older .com zone file of 12482791271 bytes under Linux (Fedora 37).

GCC 12.2.1, release mode:

$ time ./parser ../../zones/com.zone
parsed 341535548 records

real 0m17.755s
user 0m16.602s
sys 0m1.105s

#clang #FOSDEM #FOSDEM23 #SoftwareDevelopment # #SIMD

https://github.com/NLnetLabs/simdzone

This Saturday at #FOSDEM we'll be presenting in the #DNS Devroom twice!

First up will be Jeroen, talking about parsing zone files really fast. Later in the session Philip will be presenting on Connectbyname and the Proxy Control option. #OpenSource #SoftwareDevelopment #OpenStandards #SIMD https://fosdem.org/2023/schedule/track/dns/

Testing SIMD instructions on ARM with Rust on Android https://buff.ly/402KQND #android #programming #simd

📍 Are GPUs FASTER than CPUs? (not exactly...)
— Low Level Learning
#GPU #CPU #SIMD #Computation
https://youtu.be/xi-wTlVUZsQ

#SIMD Instructions on ARM with Rust on #Android tested

Blog post
https://gendignoux.com/blog/2023/01/05/rust-arm-simd-android.html

A while ago kept a mini 4 day blog whilst hacking about on sort with AVX-512 #simd https://github.com/illume/renesortfun

Learnt some of avx-512 for the first time! Read a lot, including some good books and papers on optimization. Hoping to take it further some time. Some good speedups to be had. I love doing low level stuff occasionally. Whilst it is simpler it also requires a creativity, and a different sort of zen than other types of programming.

What do people use to analyze #SIMD code these days? I've got some #AVX2 code targeting Zen2 that I want to profile or statically analyze. AMD's uProf only samples at 1ms intervals, and I don't think microbenching is going to be much help.

Damn! The legendary @lynne of FFMPEG fame is here!

For those unaware, Lynne handles a lot of the incredible #SIMD acceleration found in #FFMPEG, including in x86_64 and Arm/aarch!