ANN: The JuliaPro distribution by Julia Computing

announcement
juliacomputing

#41

@Harsha, thank you for your information.

I wish there was a portable mode which doesn’t do any of this and keeps all configuration within the Julia Installation folder.

That would be perfect.

Thank You.


#42

Fwiw the web page also says that commercial use isn’t permitted with the free version
https://juliacomputing.com/products/juliapro.html


#43

Fixed it by removing that line.

-viral


#44

Any progress with Portable Version?

That would be amazing to have.


And now... Julia 0.6.2!
#45

Hello,
I’m very excited by the 0.5.1.1 JuliaPro environment. On windows it feels again like the prehistoric Julia 0.2 that used to have the great “Batteries included” sticker on the box. IJulia and Juno work the first time without any collision with the various external python/atom installs.

I’d like to ask where I should write the error issues (PRs?) that are specific to JuliaPro (on windows). Into JuliaLang GitHub? Or is there a different repo for this purpose?

Currently I’ve encounter problem with the use of %HOMEDRIVE%%HOMEPATH% constructs in the Juno.bat and IJulia.bat batch files. This environmental variables do cause troubles on corporate, domain connected windows machines, where they use to point to some network mapped drives (location definitely not used by developpers for storing any important config data) that without the network connection to domain do not exist.
So my first step is allways replace %HOMEDRIVE%%HOMEPATH% with %USERPROFILE%. (Juno.bat and IJulia.bat) on any new instalation.

%USERPROFILE% points allways to the system drive and this is the right location to store user-specific configurations. Btw. this is the path returned by Base.homedir() function via libuv (and where juliarc is looked for).

Petr


And now... Julia 0.6.2!
#46

Hi. I’m also encounter problems installing the 5.11 distro. Does not appear to be a juliarc.jl issue. I’d be glad to post the install log or send it privately. Really excited by this!


#47
  1. Any the Atom editor update is blocking? - if so you should warn on the such issues! I spent a hour to struggle with…
  2. Vote for the portable installation mode from the bundle ones more!

#48

Hi,

2 questions:

  1. In the latest blog post on Julia Computing, Julia features in Intel’s Parallel Universe Magazine, you refer to Intel’s Parallel Universe Magazine. In Issue 29 on page 23 there is an article on Julia.
    In the article it says the following:

JuliaPro will soon ship with Intel® Math Kernel Library (Intel® MKL) for accelerated BLAS operations and optimizations for multicore and the latest Intel® processors.

Are they talking only on the Enterprise version or will it hold for the free version of Julia Pro?

  1. Any chance for a Julia Portable version? Maybe a collaboration with PortableApps?

Thank You.


Building Julia 0.6.0 with MKL on OSX 10.12.5
#49

@viral, Any comment on this - ANN: The JuliaPro distribution by Julia Computing?

Thank You.


#50

The MKL version was available in the 0.5 series, but is now taken down. We are hoping to bring back MKL in a point release of in the 0.6 series. It is going to be in the free download.


#51

@viral, This is amazing!
Best news Julia could have.
Can’t wait to have this.

What about Portable Mode?
It will be really appreciated.

Thank You.


#52

The install is mostly portable, except for paths that can written into .ji files. I guess if you just delete all the .ji files and recreate them, it should just work. Maybe to start with, we should just write this in the FAQ.

-viral


#53

Hi. I see the following files in the JuliaPro v0.6 download:

  • JuliaPro_v0.6.0.1_mkl_build-9.exe
  • JuliaPro_JuliaInXL_v0.6.0.1_mkl_build-9

and guess the former is the build for using MKL and perhaps the latter is to link MKL to v0.6.

My questions:

  • Is the guess above correctly?
  • If so, is there an easy way to verify that the MKL is properly linked? (any command to check the version or something? I have already installed the files.)

I couldn’t find any documentation about those files either on the internet or in the official online Julia documentation. Will appreciate any clarification. Thanks.


#54

Is the version info below sufficient to confirm that MKL is properly linked?

Julia Version 0.6.0
Commit 903644385b* (2017-06-19 13:05 UTC)
Platform Info:
OS: Windows (x86_64-w64-mingw32)
CPU: Intel® Core™ i5-3317U CPU @ 1.70GHz
WORD_SIZE: 64
BLAS: mkl_rt
LAPACK: mkl_rt
LIBM: libopenlibm
LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)


#55

That looks good to me. You can also look at the output from Base.BLAS.vendor().


#56

Is anybody also reproducing that ARPACK and MKL are not built/linked in a compatible way?

eigs(randn(5,5)*im)

results in

signal (11): Segmentation fault: 11
while loading no file, in expression starting on line 0
mkl_blas_zdotc at /usr/local/intel//compilers_and_libraries_2016.2.146/mac/mkl/lib/libmkl_intel_thread.dylib (unknown line)
ZDOTC at /usr/local/intel//compilers_and_libraries_2016.2.146/mac/mkl/lib/libmkl_intel_ilp64.dylib (unknown line)
zneupd_ at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/julia/libarpack.dylib (unknown line)
neupd at ./linalg/arpack.jl:276
unknown function (ip: 0x11490b85a)
eupd_wrapper at ./linalg/arpack.jl:132
unknown function (ip: 0x1149099b5)
#_eigs#107 at ./linalg/arnoldi.jl:289
_eigs at ./linalg/arnoldi.jl:175
#eigs#100 at ./linalg/arnoldi.jl:91 [inlined]
eigs at ./linalg/arnoldi.jl:91
#eigs#99 at ./linalg/arnoldi.jl:90 [inlined]
eigs at ./linalg/arnoldi.jl:90
unknown function (ip: 0x1148f4bc2)
do_call at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
eval at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
jl_interpret_toplevel_expr at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
jl_toplevel_eval_flex at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
jl_toplevel_eval_in at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
eval at ./boot.jl:235
jlcall_eval_17499 at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
eval_user_input at ./REPL.jl:66
jlcall_eval_user_input_19792 at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/julia/sys.dylib (unknown line)
macro expansion at ./REPL.jl:97 [inlined]
#1 at ./event.jl:73
unknown function (ip: 0x1148f0bcf)
start_task at /Applications/JuliaPro-0.6.0.1.app/Contents/Resources/julia/Contents/Resources/julia/lib/libjulia.dylib (unknown line)
Allocations: 4575383 (Pool: 4573924; Big: 1459); GC: 8
Segmentation fault: 11

This is specific to eigs/ARPACK, since Base.BLAS.dotc is working without issues.


#57

Hello,

I’m now trying to install JuliaPro 0.6.2 on my Mac (El Capitan) and wondering
which of the OpenBLAS and MKL versions to install. My Mac already has
Julia v0.6 installed, which is placed in /Applications/Julia-0.6.app/.

  • First of all, is it possible to install JuliaPro on top of the existing “plain” Julia already
    installed in the above directory (which was downloaded from the official webpage)?
    Is there no interference problem because different installation directories will be used?

  • If so, is there any suggestion/recommendation about which of the OpenBLAS and
    MKL versions to use on Mac? If there significant performance or functionality differences?
    If possible, I would like to install both of them (to compare speed); is it also no problem
    if the installation directories are different (by default, say, /Application/JuliaPro-0.6.2/
    and /Application/JuliaPro-0.6.2-mkl/)?

  • Just out of curiosity, in the MKL version, is it possible (if necessary)
    to call library routines in MKL directly from Julia (e.g. via ccall)?

Thanks very much for any comments/suggestions!


#58

Go with MKL.
With all the respect to OpenBLAS, Intel MKL is faster.

I wish it wasn’t so.
As if it did and OpenBLAS was as efficient people could use AMD CPU for Scientific Programming without hurting performance.


#59

With all respect to MKL, do you have a source for this statement? I’ve heard that OpenBLAS is poorly tuned for small matrices, but don’t have any numbers myself. It might help push their developers to fix any problem areas if there’s an analysis of areas to target, where they are not at least matching MKL.


#60

Wasn’t there just yesterday or the day before yesterday a post about the comparison of MKL with OpenBLAS?