Summary: | dev-python/numpy-1.15.4 USE=mkl - fails at either build time or run time for multiple undisclosed reasons | ||
---|---|---|---|
Product: | Gentoo Linux | Reporter: | Joel Berendzen <joel> |
Component: | Current packages | Assignee: | Gentoo Science Related Packages <sci> |
Status: | UNCONFIRMED --- | ||
Severity: | normal | CC: | frp.bissey, python |
Priority: | Normal | Keywords: | EBUILD |
Version: | unspecified | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Attachments: | ebuild for numpy 1.15.4 with mkl |
Smells of my original numpy-1.15.2 ebuild in the sage-on-gentoo overlay. Minus the patch to remove blas hardcoding. The stuff about the doc is a give away. There is a more up to date ebuild for 1.15.4 in the sage-on-gentoo overlay and a patch for blas (same stuff as always). https://github.com/cschwan/sage-on-gentoo/blob/master/dev-python/numpy/numpy-1.15.4.ebuild https://github.com/cschwan/sage-on-gentoo/blob/master/dev-python/numpy/files/numpy-1.15.2-no-hardcode-blas.patch Not sure if it helps with mkl though. I don't test with that. |
Created attachment 557026 [details] ebuild for numpy 1.15.4 with mkl eselecting mkl for {blas,cblas,lapack} fails at either build time or run time for multiple reasons. Here's a recipe that works for me to get mkl goodness running for numpy: 1. Emerge sci-libs/mkl-18.0.2.199 from the science overlay. 2. Install the attached ebuild. 3. Enable the mkl USE flag numpy. 4. Run some benchmarks and enjoy the huge performance increase. Yes I know the ebuild has problems: 1. It breaks numpy builds for other BLAS implementations. 2. It has hard-coded paths for the MKL library.