Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

openblas compilation errors #1183

Closed
svaksha opened this issue May 11, 2017 · 9 comments
Closed

openblas compilation errors #1183

svaksha opened this issue May 11, 2017 · 9 comments

Comments

@svaksha
Copy link

svaksha commented May 11, 2017

Hi,
While installing Julia on Ubuntu 17.04, i ran into some compilation errors: JuliaLang/julia#13842 (comment)
Tried to overcome it by not using the system openblas but installing from source failed too: https://gist.github.com/svaksha/037cb9776737d103383b9fa742d2a25a

@martin-frbg
Copy link
Collaborator

Seems you had a spelling error (missing "n") in your first attempt:

make FC=gfortra
OpenBLAS: Detecting fortran compiler failed. Cannot compile LAPACK. Only compile BLAS.

and you did not do "make clean" before the second attempt (with make TARGET=NEHALEM) which may have led to a mixed build with some parts of the code expecting lapack functions to be present.
What is your CPU ? I would expect anything from two years ago (when that julia issue was new) to be detected correctly by a recent OpenBLAS so a TARGET option should not be necessary.

@brada4
Copy link
Contributor

brada4 commented May 11, 2017

You can plant very fresh OpenBLAS as system libblas.so.3 at disposal of julia, numpy and friends:
https://github.com/xianyi/OpenBLAS/wiki/Faq#debianlts
Or do nothing as long as both products are recent from distribution (they know their ways with compilers better)

@svaksha
Copy link
Author

svaksha commented May 17, 2017 via email

@martin-frbg
Copy link
Collaborator

That would seem to fit my speculation that the build system should be able to figure out the CPU type (and the fortran compiler) automatically. If OpenBLAS is built as part of the Julia installation, either someone had the foresight to issue a make clean at the start of the build process "just in case", or the checkout also fetched a fresh copy of OpenBLAS, oblitterating any broken bits that may have been left over from your previous attempt.

@svaksha
Copy link
Author

svaksha commented May 17, 2017 via email

@martin-frbg
Copy link
Collaborator

The i7-7600U (as you mentioned in the julia issue comment) should be recognized by the current git "develop" source tree, though not yet by the 0.2.19 release. (HASWELL target should be more appropriate for that cpu rather than the much older and less advanced NEHALEM, unless you intend to use the same library build on older machines as well). Not sure about the earlier issues but glad that they appear to have gone away :-)

@brada4
Copy link
Contributor

brada4 commented May 17, 2017

You can export OPENBLAS_CORETYPE=Haswell with ububtu-supplied OpenBLAS or supplant it with develop OpenBLAS. Either will be significantly quicker than old cpu architecture builds.

@martin-frbg
Copy link
Collaborator

@svaksha was this sufficient to solve your problem(s), or would you like further assistance ?

@svaksha
Copy link
Author

svaksha commented Jun 22, 2017

@brada4, thanks. " make OPENBLAS_TARGET_ARCH=HASWELL" fixed the build failures. (<-- for folks having similar issues).
@martin-frbg, solved now, thanks!

@svaksha svaksha closed this as completed Jun 22, 2017
@svaksha svaksha reopened this Jun 22, 2017
@svaksha svaksha closed this as completed Jun 22, 2017
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants