MKL 2025.0 build for JavaCPP 1.5.12#1637
Conversation
|
There is no point in just changing the version of the presets. If you want to update the version of MKL though, please do it. |
|
I'll see whether I can update it, but I think even the version of the presets have some value, since mixing javacpp versions gives a warning (that could be safely ignored for now, but confuses people nevertheless). |
|
I see now that this was indeed built with a slightly newer version of MKL 2025.0.1, which is current on Arch Linux. When the Arch Linux repositories upgrade to 2025.1, I'll build it with then current JavaCPP, and create a PR for, I suppose, future 1.5.13-SNAPSHOT. After that, I hope oneMath will soon become a viable alternative (it seems to be Intel's rebranding of MKL, and a wrapper that supports many backends, similarly to how oneDNN works)... |
|
They've made a second release of MKL this year, so it doesn't look totally dead just yet. |
|
Thanks! |
I've just updated the MKL build for 2025.0 (the same as it was in JavaCPP 1.5.11) to fit into the rest of javacpp-presets which are nearing the 1.5.12 release.
I did a local build on my Arch Linux machine, and it went without any problems. (The Github Actions failed quickly due to, it was reported, a "outdated Ubuntu version"). So I hope it's a matter of updating the setup to a more recent stack (I don't know where to change this).
Windows build in GitHub Actions succeeded without a hitch.
I hope this could be included in JavaCPP 1.5.12, since MKL is not just a very featureful BLAS/LAPACK implementation, but also comes with DSP, sparse matrices, statistics, and load of other operations useful in number crunching, which are missing from the alternatives such as OpenBLAS.