[
https://issues.apache.org/jira/browse/SINGA-303?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15884037#comment-15884037
]
ASF subversion and git services commented on SINGA-303:
-------------------------------------------------------
Commit 3417dd9d638331e2451e767c05b12fff20484a8c in incubator-singa's branch
refs/heads/master from [~flytosky]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=3417dd9 ]
SINGA-303 Create conda packages
Combine with travis:
1. singa at travis/apache does normal building and test (apt-get and homebrew)
2. singa at travis/nusdbsystem creates conda packages for linux and osx, and
upload them to anaconda (channel: nusdbsystem)
Create GPU conda packages using nvidi-docker. package name is like
singa-cudaxx-cudnnyy
Update cmakefile to disable test by default
TODO: update jenkin's setting to create conda GPU packages, and replace wheel
packages which may result in runtime linking errors, e.g., on osx
NOTE:
1. conda's protobuf (3.0.0) was compiled using gcc 4; if we compile
singa using gcc 5 (abi != 0), then there will be linking error;
Therefore, we are using gcc4.8 to compile singa. ubuntu14.04 comes with
gcc 4.8 and ubuntu16.04 comes with gcc 5.
2. use -stdlib=libc++ is necessary to for the cxxflag in osx
3. nvidia-docker puts cudnn into different system folders for diff
docker images. cmake should be able to find them, since they are in
system paths. hence, no need to export cmake paths.
> Create conda packages
> ---------------------
>
> Key: SINGA-303
> URL: https://issues.apache.org/jira/browse/SINGA-303
> Project: Singa
> Issue Type: New Feature
> Reporter: wangwei
> Assignee: wangwei
>
> conda can build packages for software written in c++, python, etc.
> It provides its own building environment and thus can avoid the path mess
> problem. It is also cross platform.
> This ticket will add scripts and configurations to build packages for SINGA
> using conda tools. The packages will be uploaded to anaconda. Then users can
> install singa via
> {code}
> conda install -c nusdbsystem singa
> {code}
> We will use travis SINGA-302 to create the packages for osx and linux (CPU
> version) and use jenkins to create linux packages for GPU. Windows packages
> will be considered later.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
