Open Source LLM|Intel

OpenBLAS 0.3.11 Brings Many Changes From Zen 3 To POWER10 Support

OpenBLAS 0.3.11 is out as the newest major feature release for this BLAS linear algebra library. While the version number may not make it seem like a big update, it is especially when it comes to new CPU support.

When it comes to new processor support there is now optimized support for POWER10, Intel Cooper Lake Xeon support, auto-detection for Intel Comet Lake, auto-detection for AMD Renoir / Matisse / Zen 3 processors, preliminary support for Apple Vortex ARM processors, support

Open Source LLM

Systemd 247 Merges Systemd-OOMD For Improving Low-Memory/Out-Of-Memory Handling

Merged just minutes ago into systemd Git is the new systemd-oomd component pushed along by Facebook.

Systemd-oomd has been developed to improve the Linux out-of-memory / memory pressure behavior and based on Facebook's out-of-memory daemon code that's been extended to not only work for Linux servers but also desktop systems.

The systemd-oomd daemon polls for OOMD-enabled cgroups to monitor and will kill based on memory pressure or swap usage. The systemd-oomd behavior is configurable via

Open Source LLM

Apache Software Foundation Celebrates Two Decades Of OpenOffice

While the LibreOffice fork is much more popular than OpenOffice these days, the Apache Software Foundation does continue maintaining the OpenOffice codebase born out of Sun Microsystems' StarOffice.

It's been twenty years already since Sun initially open-sourced the office suite as OpenOffice. While the OpenOffice 1.0 release didn't come until two years later (2002) and wasn't until 2011 that Oracle transferred OpenOffice to the Apache Software Foundation, Apache is celebrating twenty years of OpenOffice.

Continue reading...

Open Source LLM

LLVM 11.0 Finally Available With Flang Fortran Compiler, Continued C++20 Work

After being one and a half months late, LLVM 11.0 is now tagged and ready to ship.

Release candidates not shipping on schedule and the need for additional RCs in recent weeks due to open bugs led to the LLVM 11.0 release schedule falling well off track. But in any case LLVM 11.0 is now available for download from LLVM's GitHub repository .

There are many improvements in LLVM 11.0 and Clang 11.0 including continued work bringing up C+

Open Source LLM

Intel Releases OpenVINO 2021.1 With Tigerlake Support, Expanded Capabilities

Ready to move past the tumultuous year that is 2020, Intel's open-source developers responsible for the OpenVINO toolkit today issued version 2021.1 with some big ticket additions.

While it's only October 2020, OpenVINO 2021.1 is out today with support for 11th Gen Intel Core (Tigerlake) processors and Xe Graphics, TensorFlow 2.2.x support, the OpenVINO Model Server for distributing models across cloud/edge environments,

Open Source LLM

Apache's TVM Deep Learning Compiler Picks Up WebAssembly, Better Rust Support

The first release candidate of TVM 0.7, the Apache incubator project providing a deep learning compiler stack, is now available.

Apache TVM is a compiler stack for deep learning systems in providing end-to-end compilation support for a variety of back-ends from all of the models from key deep learning frameworks. TVM supports a variety of targets including the leveraging of LLVM for supporting code generation on the major CPU architectures as well as the likes of NVIDIA CUDA. With the Apache 0.7 release candidate there

Open Source LLM

OpenCL 3.0 Specification Released With New Khronos Open-Source OpenCL SDK

Back in April was the provisional release of OpenCL 3.0 with making CL 2.x features optional while adding async DMA extensions and more . Today the finalized version of OpenCL 3.0 has been released plus also introducing an official Khronos OpenCL SDK.

The OpenCL 3.0 specification is now official, OpenCL 3.0 Conformance Tests are also now available, and the resources are on GitHub for interested developers. OpenCL 3.0 remains largely about making OpenCL 2 features optional so OpenCL 3.0 is suited for

Open Source LLM

Intel Has Been Working On OpenCL C 3.0 Support For Clang

Intel's compiler experts in Moscow have been working on OpenCL C 3.0 support for the LLVM Clang compiler front-end.

OpenCL 3.0 allows much greater flexibility in making OpenCL 2.x era features optional. With this on the compiler-side it's not very invasive when already supporting OpenCL 2.x functionality but a matter in part of just making very features optional.

Intel has already been making use of this OpenCL C 3.0 support within their own Clang downstream tree while Anton Zabaznov

Open Source LLM

Mesa 20.3 Can Now Consume SPIR-V Binaries Generated By LLVM's libclc

Microsoft's work on Mesa continues with a Microsoft engineer reviving earlier work carried out by David Airlie for allowing SPIR-V binaries produced by LLVM's libclc library to work with Mesa OpenCL.

Libclc is the LLVM library around OpenCL C programming language support and goes along with Clang's OpenCL front-end. Jesse Natalie of Microsoft has seen his two month old merge request land on Friday for being able to make use of libclc SPIR-V binaries that can be used by Mesa OpenCL code. Ultimately this code

Open Source LLM

GraalPHP Is A PHP JIT Implementation Built On GraalVM

Oracle's GraalVM has seen language support well outside of Java/OpenJDK from Ruby to WebAssembly to R and Python. The newest language seeing an experimental implementation built atop GraalVM is PHP.

Andrin Bertschi for his university thesis has been implementing PHP atop the GraalVM. A sizable subset of the PHP 7.4 programming language is implemented by this GraalPHP compiler and runtime.

Benchmark results of GraalPHP have been very promising against PHP 7 and other alternatives, including the upcoming PHP 8 with its own official JIT compiler:

Continue reading...