CppCon 2016: Klaus Iglberger “The Blaze High Performance Math Library'

—
—
In this presentation we introduce the Blaze C++ math library, a hot contender for the linear algebra performance throne. Blaze is an open-source, high-performance library for dense and sparse arithmetic. It combines elegance and ease of use with HPC-grade performance, making it one of the most intuitive and at the same time fastest C++ math libraries available.

We demonstrate its basic linear algebra functionality by means of several BLAS level 1 to 3 operations and explain why Blaze outperforms even well established linear algebra libraries. Additionally, we present some advanced features that enable users to adapt Blaze to special circumstances: custom data structures, custom operations, and the customizable error reporting mechanism.

—
Klaus Iglberger
CD-adapco
Senior Software Engineer
Nürnberg Area, Germany
—

*-----*
*-----*

[franciscogerardohernandezr]

I used it for my PhD thesis and it is quite easy to translate from Matlab to C++ with this library. Surprisingly, it only performed marginally faster, about 10% faster than paralleled Matlab code (it is my fault for sure), but the most important lesson was that it allowed me to find bugs that Matlab tends to cover, such as faulty memory allocation and typecasting.

[ultradude]

Sounds like it's easier to use than most of the other libraries, and quite an improvement in "performance per effort"!
This is an exciting library for sure!

[laurentvioleau]

As demonstrated in the presentation, Blaze has some super fast kernels. Why not contributing those kernels to an other library, say Eigen, rather than making a new library? Is there something fundamental to Blaze that would make it impossible or impractical to add the code to other linear algebra C++ libraries?

[CC]

Maybe it's a skill issue, but I'm getting better performance using Eigen for simple linear regression of a few 100s of observations (Blaze is faster for 10s of observations).

[maxmustermann]

Back than he coulnt assume people have more than 4 cores at home and 7 years later cheap CPUs come with 12