#blas — Public Fediverse posts

Микроядро SME2 sgemm: 1024 умножения-сложения за проход

В этой части мы возьмём первый из пропусков BLIS — микроядро sgemm ; сначала зафиксируем форму аккумулятора 32×32 в четырёх тайлах ZA, затем разберём горячий цикл по K, потом эпилог с alpha , beta и ловушкой NaN, а в конце посмотрим, какие числа производительности даёт именно это ядро. Следующая часть оставит структуру почти той же, но заменит геометрию тайла и условия существования ядра.

https://habr.com/ru/articles/1071462/

#BLIS #GEMM #SME2 #Apple_Silicon #микроядро #кэшблокинг #упаковка_данных #высокопроизводительные_вычисления #линейная_алгебра #BLAS

BLIS: недостающую среднюю ступеньку построили тридцать лет назад

В этой части мы начнём с тупика, в который приводит голый цикл FMOPA ; затем покажем, какую часть GEMM BLIS уже построил за нас; после этого разберём пять циклов BLIS и место микроядра внутри них. Финал главы должен сделать дальнейший план конкретным: что именно нужно добавить для Apple SME, а что уже относится к готовой «мебели» фреймворка.

https://habr.com/ru/articles/1069058/

#BLIS #GEMM #SME2 #Apple_Silicon #микроядро #кэшблокинг #упаковка_данных #высокопроизводительные_вычисления #линейная_алгебра #BLAS

Re-reading the old BLAS level 3 and #FLAME papers about microkernels and performance for performant #GEMM implementations in #BLAS, #GotoBLAS and #BLIS. I had forgotten how well written they are!

Question for the #stats folks working with #mgcv and #gam / #bam:
I'm dealing with a dataset with a high temporal autocorrelation, so after a lengthy discussion with the author of `gratia` (my current PI), we concluded that I should move away from AR(1) models and rather work with #NCV.
No problems with the setup, but the run time is forever.
My first model has been running for 20h by now, with no indication of how long it will take.
I have already switched over to #OpenBLAS, but it seems weird that my CPU load is constantly at only 15-25%. The only indication that something is happening is that RAM usage is slowly but steadily increasing.
I'm dealing with ~40k data points across ~300 time series, and a NCV window of 7 data points, so nothing crazy. The model includes two- and three-way interactions, and random smooths, but all of those things are biologically relevant.

I'd happily take any suggestions on how to speed things up, how to get proper CPU usage, or at least get an estimate of how far along the calculations are.

I might eventually need to switch over to #twlss, from the current `tw()`, and that will make things worse as that is not supported by `bam()`.

I'm currently on Windows, but I might be able to run things on an #HPC if needed, but not sure if I can easily fiddle with #BLAS there.

Did you recently update your #rstats to 4.6? Is your matrix algebra slow? Not #BLAS ting? `sessionInfo()` shows which BLAS library R uses. Switching from reference BLAS to Apple Accelerate is easily a 50x speedup.

Let me know if this worked for you. 🚀

Post: uatz.org/blog/2026-06-04-r-blas-macos/ #rstats

#Copilot and I are about 30% away from creating a #Pascal version of #LAPACK using #BLAS. We are about two days away from achieving 80% of LAPACK. Then we will tweak it using some GPU acceleration to make its speed comparable to some python libraries like Numpy.

It is important to note that one must be very disciplined in keeping clean documentations, a thorough and tight testing cycle, a rigid workflow pattern, or an AI will tend to skip tests, become sloppy and lose focus.

#AI #LLM

#AI illiteracy is real. While still arguing with a bunch of AI haters, #Copilot and I just finished our #Pascal #BLAS level 1-3 Implementation plus eigenvalue, cholesky, and sparse #matrix, so we will never need #python, #C, C#, #Rust, ... for our Small Language Project. We will expand our Pascal Numeric Library (PNL) v1.0 to something like #Numpy and #Pytorch, but with static arrays, deterministic data structure, no referencing, no pointer arithmetic.

#LLM #programming #computer

While arguing with some AI haters, #Copilot and I created this Pure #Pascal #BLAS (Level 1,2,3 Core) Implementation in less than 1 day. We encountered many serious problems, including drifting of workflow pattern, getting stuck in a Delphi error loop, overhauling our original design... But as long as you understand AI, keep good documentations, maintain the core structure of the problem,.. you will be able to work with AI successfully. Don't hesitate to use more than one #AI at a time.

#LLM

What is #BLAS?

BLAS is a set of fast matrix routines originally written in #Fortran.
If you’re tired of dynamic types, hidden references, ownership rules, and endless “stream” abstractions, Free #Pascal + BLAS gives you old‑school, deterministic HPC #programming with none of the modern noise.

#Copilot and I will be using Free Pascal and BLAS for our Small Language Model project #SLM. No more #C, #python, #Rust, or C#

#AI #LLM #computer

Why do people use #python, a glue language, which is so slow? The only reason is the AI ecosystem.

#Copilot and I just tested Free Pascal and BLAS for its speed without using #numpy or #pytorch. The result is amazing. It took less than a second to do a 1024x1024 #matrix multiplication.

We will be using Free #Pascal and #BLAS to write our Small Language Model #SLM using #NNUE.

#AI #LLM

The plot thickens #BLAS #rstats #lapack
(When one is about to rip through 10s of millions of medical records, one must profile the tools if the project is to finish before one's retirement)
FlexiBLAS makes this benchmarks a breeze

The plot thickens #BLAS #rstats #lapack (When one is about to rip through 10s of millions of medical records, one must profile the tools if the project is to finish before one's retirement) FlexiBLAS makes this benchmarks a breeze

I wonder if the #lapack that comes with #AOCL is being picked up by flexiblas in #rstats. The things I have to do for the love of electronic health records analytics #bigdata #blas

I wonder if the #lapack that comes with #AOCL is being picked up by flexiblas in #rstats.
The things I have to do for the love of electronic health records analytics #bigdata #blas

Por san #Blas, la #cigüeña verás.

#cigüeñablanca #ciconiaciconia #fedipajareo #ColmenarViejo

Developing a BLAS Library for the AMD AI Engine [pdf]

https://uni.tlaan.nl/thesis/msc_thesis_tristan_laan_aieblas.pdf

#HackerNews #Developing #BLAS #AMD #AI #Engine #AI #Library #PDF #Research

Another post on #Quansight PBC blog: "BLAS/LAPACK #packaging"

https://labs.quansight.org/blog/blas-lapack-packaging

"""
#BLAS and #LAPACK are the standard libraries for linear algebra. The original implementation, often called Netlib LAPACK, developed since the 1980s, nowadays serves primarily as the origin of the standard interface, the reference implementation and a conformance test suite. The end users usually use optimized implementations of the same interfaces. The choice ranges from generically tuned libraries such as OpenBLAS and BLIS, through libraries focused on specific hardware such as Intel® oneMKL, Arm Performance Libraries or the Accelerate framework on macOS, to ATLAS that aims to automatically optimize for a specific system.

The diversity of available libraries, developed in parallel with the standard interfaces, along with vendor-specific extensions and further downstream changes, adds quite a bit of complexity around using these libraries in software, and distributing such software afterwards. This problem entangles implementation authors, consumer software authors, build system maintainers and distribution maintainers. Software authors generally wish to distribute their packages built against a generically optimized BLAS/LAPACK implementation. Advanced users often wish to be able to use a different implementation, more suited to their particular needs. Distributions wish to be able to consistently build software against their system libraries, and ideally provide users the ability to switch between different implementations. Then, build systems need to provide the scaffolding for all of that.

I have recently taken up the work to provide such a scaffolding for the Meson build system; to add support for BLAS and LAPACK dependencies to Meson. While working on it, I had to learn a lot about BLAS/LAPACK packaging: not only how the different implementations differ from one another, but also what is changed by their respective downstream packaging. In this blog post, I would like to organize and share what I have learned.
"""

#CondaForge #Debian #Fedora #Gentoo

Wspominałem już może, że pracuję nad przejściem #Gentoo z na wpół zepsutego eselect-ldso dla #BLAS / #LAPACK, na #FlexiBLAS. Oznacza to również, że czeka nas okres przejściowy, w czasie którego obydwa rozwiązania będą wspierane.

Plus jest taki, że stan "po" jest kompatybilny pod względem ABI ze stanem "przed" (a przynajmniej powinien być — pracujemy z autorami, by poprawić ostatnie niedociągnięcia). Zastępujemy libblas.so, liblapack.so i inne biblitoteki dowiązaniami symbolicznymi, więc programy skompilowane przed zmianą po prostu zaczną używać FlexiBLAS.

Minus jest taki, że w drugą stronę nie jest tak łatwo. Po zastąpieniu biblitotek dowiązaniami, nowoskompilowane programy będą odczytywać SONAME z biblioteki docelowej, a więc zaczną się wiązać bezpośrednio z FlexiBLAS. Co za tym idzie, powrót do stanu poprzedniego będzie wymagał ich ponownej kompilacji.

Aby tego uniknąć, musielibyśmy zamiast dowiązań symbolicznych zastosować jakieś biblioteki pośredniczące, które miałyby "stare" SONAME, a korzystąły z funkcji FlexiBLAS. Niestety, nic prostego tu nie zadziała — musiałbym jakoś "wyeksportować" symbole z FlexiBLAS, i najlepiej podzielić je na odpowiednie biblioteki, żeby `-Wl,--as-needed` nic nie wycięło. Tylko jak to zrobić?

Cóż, eselect-ldso tworzy jakieś biblioteki, więc może uda się coś wykorzystać. No i szukam w źródłach, i nic nie mogę znaleźć. W końcu do mnie dociera, że cała logika dodana jest przez łatki Gentoo. A te łatki są po prostu paskudne. W OpenBLAS tworzymy dodatkowe biblioteki libblas.so, itp., które zawierają kopie obiektów z OpenBLAS i wiążą się z libopenblas, żeby pobrać brakujące zależności. Nawet nie wiążą się jedna z drugą, więc każda duplikuje sporo kodu niezależnie. Łatki dla BLIS są jeszcze gorsze — tu libblas.so i libcblas.so to praktycznie kopie libblis.so, z poszczególnymi "niepotrzebnymi" symbolami ukrytymi przy pomocy "visibility".

No cóż, można się było tego spodziewać po projekcie z #GSoC.

1. Zdobądź trochę wiedzy o paczkach #BLAS / #LAPACK w ramach bejmopracy.
2. Odkryj, że paczki #MKL w #Gentoo są mocno nieaktualne i ciut zepsute. Przejmij je, zaktualizuj, ulepsz.
3. Zainteresuj się #FlexiBLAS. Zacznij eksperymentować. Wrzuć paczkę do Gentoo.
4. Odkryj, że mechanizm dynamicznego przełączania BLAS / LAPACK niezbyt dobrze działa. Zaproponuj migrację do FlexiBLAS i przygotuj próbne zmiany.
5. Zauważ niespójności we wsparciu ILP64. Zaproponuj poprawki.
6. Odkryj, że wszystkie paczki BLAS / LAPACK w Gentoo są praktycznie bez opiekuna.

No więc wygląda na to, że jestem nowym opiekunem całego kompletu. Pracuję nad poprawkamj dla ILP64, a następnie będę musiał zaktualizować łatki dla migracji do FlexiBLAS.

1. Learn a bit about #BLAS / #LAPACK packaging for dayjob.
2. Learn that #MKL in #Gentoo is quite outdated. Take it over, bump it and improve the packaging.
3. Get curious about #FlexiBLAS. Start playing with it. Package it for #Gentoo.
4. Learn that runtime BLAS / LAPACK switching is quite broken. Come up with a FlexiBLAS transition plan and a proof-of-concept.
5. Notice inconsistency in ILP64 support flags. Propose unifying the behavior.
6. Learn that BLAS / LAPACK packages in Gentoo are pretty much unmaintained.

Well, looks like I'm the new maintainer of the whole stack, I'm working on consistent ILP64 support now, and then I'll have to rebase the FlexiBLAS transition bits.

Widzisz, że osoba z adresem e-mail #Debian .org opiekuje się paczkami #BLAS w #Gentoo, i myślisz sobie: "jak fajnie, że dystrybucje współpracują…"

A potem uświadamiasz sobie, że ta osoba wzięła tylko kasę z #GSoC w 2019, i zniknęła od razu po fakcie…

#WolneOprogramowanie