Giuseppe Cuccu

icon indicating a website link exascale.info/giuse icon indicating an email [email protected]

icon company Exascale Infolab, University of Fribourg icon location Switzerland icon location Bringing state-of-the-art machine learning into real world solutions

Results 21 comments of Giuseppe Cuccu

Optimize marshaling in parallel computation by allowing constructing Numo arrays from pre-allocated data pointers, such as coming from views.

Dask and similar methods aimed at distributed computing (e.g. [`ruby-spark`](https://github.com/ondra-m/ruby-spark)) should already work well with the current marshaling. For local parallelization though, accessing the underlying C pointer would constitute a...

Optimize marshaling in parallel computation by allowing constructing Numo arrays from pre-allocated data pointers, such as coming from views.

- Relationship between parallelization and C pointers: sometimes you want to compute complex objects in child processes to parallelize the execution. Then the result need to go back to the...

Optimize marshaling in parallel computation by allowing constructing Numo arrays from pre-allocated data pointers, such as coming from views.

- But yet the binary representation is returned in a String object. ```ruby dump = Numo::Int8.new(5).seq.marshal_dump # => [1, [5], 0, "\x00\x01\x02\x03\x04"] dump.last.class # => String ``` It is this...

Optimize marshaling in parallel computation by allowing constructing Numo arrays from pre-allocated data pointers, such as coming from views.

You are right, I don't know how I could forget the `\xnn` representation. Thank you for correcting me! This is important because I am working with one-byte-per-pixel images, so by...

Optimize marshaling in parallel computation by allowing constructing Numo arrays from pre-allocated data pointers, such as coming from views.

I found this article to be relevant: https://blog.rebased.pl/2017/12/27/writing-c-and-sharing-memory.html

RuntimeError: /usr/lib/libopenblas.so: undefined symbol: LAPACKE_dgesvd

Can confirm it works on Ubuntu 16.04, though the second file has an extra '.3' in the name: ```ruby Fiddle.dlopen("/usr/lib/libopenblas.so") Numo::Linalg::Lapack.dlopen("/usr/lib/liblapacke.so.3") ``` Any chance to make this the default when...

RuntimeError: /usr/lib/libopenblas.so: undefined symbol: LAPACKE_dgesvd

I confirm the short name causes an error on Ubuntu. Full trace: ``` giuse@free:~$ rvm gemset create test_numo_linalg ruby-2.4.2 - #gemset created /home/giuse/.rvm/gems/ruby-2.4.2@test_numo_linalg ruby-2.4.2 - #generating test_numo_linalg wrappers....... giuse@free:~$ rvm...

RuntimeError: /usr/lib/libopenblas.so: undefined symbol: LAPACKE_dgesvd

I am currently working to get Travis CI to pass. I noticed this line when it installs `libopenblas-base` which may be pertinent to this issue: ``` update-alternatives: using /usr/lib/openblas-base/libblas.so.3 to...

RuntimeError: /usr/lib/libopenblas.so: undefined symbol: LAPACKE_dgesvd

I confirm `numo-linalg 0.1.2` fixes this for me. (`[...]` stands for matrix content truncated. Notice no explicit call to `Fiddle::dlopen` nor `Numo::Linalg::Lapack::dlopen`) ```ruby 2.4.2 :001 > require 'numo/narray' => true...

RuntimeError: /usr/lib/libopenblas.so: undefined symbol: LAPACKE_dgesvd

**UPDATE:** it seems Travis CI is having the same issue as @mrkn -san. You can see the full log [on my Travis](https://travis-ci.org/giuse/machine_learning_workbench), errors 5 through 11. To see the complete...