On startup, julia 0.6 used to tell me which blas and lapack libraries it used. Alas, I think this is no longer so for julia 1.0. how do I find out? thanks for pointers.
PS: (on linux, do I need to install libatlas or anything like it to improve speed? on R, this used to be essential, or everything was dead slow…which means that 90% of all R users had a dead slow version. or is everything for julia in the box already?)
PS: I also peeked at an ldd julia, but could not see an *las library there. interestingly, it says it is linked to a libjulia.so.1 package, which it cannot find on the command line. so I presume that the binary julia than grabs it.
The precompiled Julia binaries include an up-to-date copy of OpenBLAS, which is the fastest open-source BLAS library and usually fairly competitive with Intel MKL and Apple Accelerate (the two most important proprietary BLAS libraries these days). So you don’t have to install external libraries to get a fast BLAS.
The BLAS version is reported by LinearAlgebra.versioninfo().
The standard system image loads LinearAlgebra, which links to the BLAS library during its initialization (after the minimal linking visible to ldd).
I wonder, have any of the embedded folks built system images which do useful things without linking BLAS?