Adds a package for mallocMC, a memory allocator for many core
architectures.
This project provides a framework for fast memory managers on
many core accelerators. Currently, it supports NVIDIA GPUs of
compute capability sm_20 or higher through the ScatterAlloc
algorithm. mallocMC is header-only.
Downstream Dependencies
PIConGPU and other projects of HZDR's
*Computational Radiation Physics* group
References
- Markus Steinberger, Michael Kenzel, Bernhard Kainz and Dieter Schmalstieg.
*ScatterAlloc: Massively parallel dynamic memory allocation for the GPU*,
Innovative Parallel Computing (InPar), 2012,
https://doi.org/10.1109/InPar.2012.6339604
- Carlchristian Eckert.
*Enhancements of the massively parallel memory allocator ScatterAlloc and its adaption to the general interface mallocMC*,
Junior Thesis (grosser Beleg), 2014,
http://dx.doi.org/10.5281/zenodo.34461
This adds a maintained version of the (since 2012)
stalled original project.
https://github.com/ComputationalRadiationPhysics/cuda_memtest
Nvidia's NVML (via the GPU deployment kit) could also be
added, providing serial number output of failing GPUs
for multi-GPU nodes.
* Fixes patch paths for earlier boost versions
The directory structure of boost changed at version 1.56.0, so the patch
being used for python support did not work on earlier versions. This
adds another patch that matches earlier versions.
* Removes commented out code
* Fixes needed to build htop
* Build libncursesw.so, not libncurses.so
* Remove htop hack now that it is picking up newer ncurses
* Tell packages to link to new ncursesw libraries
- clean up fetch urls
- inherit from AutotoolsPackage instead of Package
- explicit fPIC flags look suspicious, but leave them in there.
- in scalasca, use the cube version that is implicit from
the scorep dependency instead of specifying it twice.
* The beginnings of a package for shiny-server
Just stashing a WIP.
This doesn't work.
This goes for a while and/but blows up with some OpenSSL related issue.
* Make it work!
Yay!
* shiny-server needs R with X support
My environment gets this for "free" from my packages.yaml, but it
should be explicity.
* Address feedback
- python version
- gcc dependency
- Flake8 comment rules
* Richer caveats and warnings.
* Convert to CMakePackage and fix python version typo
* Fix typo: noqab -> noqa
* Ensure proper build location, clean up comments
* Update tinyxml
* url seems to have changed
* spack replaces "_2_6_2" with "_2.6.2" unless the url is explicitly
given
* copyfile is no longuer available by default
* URL version-parsing should work as of #2972
* Remove fake URLs from Spack
* Ignore long lines for URLs that start with ftp:
* Preliminary changes to version regexes
* New redesign of version regexes
* Allow letters in version-only
* Fix detection of versions that end in Final
* Rearrange a few regexes and add examples
* Add tests for common download repositories
* Add test cases for common tarball naming schemes
* Finalize version regexes
* spack url test -> spack url summary
* Clean up comments
* Rearrange suffix checks
* Use query strings for name detection
* Remove no longer necessary url_for_version functions
* Strip off extraneous information after package name
* Add one more test
* Dot in square brackets does not need to be escaped
* Move renaming outside of parse_name_offset
* Fix versions for a couple more packages
* Fix flake8 and doc tests
* Correctly parse Python, Lua, and Bio++ package names
* Use effective URLs for mfem
* Add checksummed version to mitos
* Remove url_for_version from STAR-CCM+ package
* Revert changes to version numbers with underscores and dashes
* Fix name detection for tbb
* Correctly parse Ruby gems
* Reverted mfem back to shortened URLs.
* Updated instructions for better security
* Remove preferred=True from newest version
* Add tests for new `spack url list` flags
* Add tests for strip_name_suffixes
* Add unit tests for version separators
* Fix bugs related to parseable name but in parseable version
* Remove dead code, update docstring
* Ignore 'binary' at end of version string
* Remove platform from version
* Flip libedit version numbers
* Re-support weird NCO alpha/beta versions
* Rebase and remove one new fake URL
* Add / to beginning of regex to avoid picking up similarly named packages
* Ignore weird tar versions
* Fix bug in url parse --spider when no versions found
* Less strict version matching for spack versions
* Don't rename Python packages
* Be a little more selective, version must begin with a digit
* Re-add fake URLs
* Fix up several other packages
* Ignore more file endings
* Add parsing support for Miniconda
* Update tab completion
* XFAILS are now PASSES for 2 web tests
- _spider in web.py was actually failing to spider deeper than a certain
point.
- Fixed multiprocessing pools to not use daemons and to allow recursive
spawning.
- Added detailed tests for spidering and for finding archive versions.
- left some xfail URL finding exercises for the reader.
- Fix noqa annotations for some @when decorators
* add packages for EasyBuild and dependencies
* fix remarks w.r.t. docstrings & download urls, make dependency specs stricter w.r.t. required versions
* fix easybuild-* dep specs, should be open-ended
* correct dep specs for easybuild and easybuild-easyconfigs using @3.1.2[:]
* Run python setup.py test if --run-tests
* Attempt to import the Python module after installation
* Add testing support to numpy and scipy
* Remove duplicated comments
* Update to new run-tests callback methodology
* Remove unrelated changes for another PR
* perl: make extendable and add Module::Build package
* perl: allow 'spack create' to identify perl packages from their contents
* perl-module-build: fix indenting of package docstring
* perl: split install() method for extensions into phases
* perl: auto-detect build method (Makefile.PL vs Build.PL) and define a 'check' method
* PerlPackage: use import statements similar to those in AutotoolsPackage
* PerlModule: fix detection of Build.PL
* PerlPackageTemplate: remove extraneous lines to avoid flake8 warnings
* PerlPackageTemplate: split into separate templates for Makefile.PL and Build.PL
* PerlPackage: add cross-references to docstrings
* AutotoolsPackage: fix ambiguous cross-references to avoid errors in doc tests
* PerlbuildPackageTemplate: depend on perl-module-build if Build.PL exists
* ENH: add package for building OpenFOAM (1612) from www.openfoam.com
- provide 'openfoam' as virtual package.
- package as openfoam-com to reflect the distribution point.
This initial spack packaging for OpenFOAM supports a number of possible
variants and should handle 64-bit labels properly now that the scotch
package has been updated accordingly.
* ENH: update package for foam-extend (extend-project.de)
- provide 'openfoam' as virtual package.
- much of the build is now aligned with how the openfoam-com package
looks, with the aim of future refactoring.
- avoid installing intermediate targets.
- contains its own environment sourcing script for the build, for more
flexibility and robustness (doesn't touch the python build environ)
* ENH: added package for building from openfoam.org
- provide 'openfoam' as a virtual package.
- this is largely a direct copy of the openfoam-com package.
It has been supplied as a courtesy for users and to ensure maximum
consistency in quality and naming between the foam-extend,
openfoam-com and openfoam-org packages.
* CONFIG: add openfoam into bash completion providers list
* ENH: have openfoam-com use spack as USERMPI
- also simplify the generation of mplib/compiler rules
* ENH: have openfoam-org use spack as SYSTEMMPI
- this setup requires more environment settings than USERMPI
(openfoam-com), but is currently the only means of integration
for openfoam-org
- simplify generation of mplib/compiler rules
* ENH: simplify generation of mplib/compiler rules (foam-extend)
- rename mpi rules from SPACK,SPACKMPI to USER,USERMPI for consistency
with openfoam-com and to generalize for any build system.
* STYLE: record spack tree as a log file (openfoam)
- can be useful for future diagnostics and general record keeping
* libint: updated version, ported to AutotoolsPackage
* elpa: updated version, ported to AutotoolsPackage
* wannier90: updated version, ported to MakefilePackage
* pexsi: updated version, ported to MakefilePackage
* cp2k: updated version, removed wannier 90 from dependencies if cp2k@4.1
* plumed: updated version
Modifications:
* modules has been turned into a single variant, as different versions
have different modules available
* added dictionaries for patches
* netlib-scalapack: ported to CMakePackage
* Add Blast+ package.
Adds support for NCBI's blast+@2.6.0. I'll be adding a few historical versions in the near future.
It's a fairly direct transliteration of the [Homebrew Science recipe][hbs].
I skipped the mysql support.
There is a problem with HDF5 support. Blast's configure script test program defines a macro 'HOST' to a string value and one of the HDF5 bits has an variable named 'HOST'. Fun things happen. I've disabled support for now.
I've run very very basic tests. I'll get this out to my users and ensure that it works and fine tune the options and etc....
[hbs]: https://github.com/Homebrew/homebrew-science/blob/master/blast.rb
* Flake8 fixes
* Update w.t.r. mamelara's PR and cleanup
Add paths to things in the --with-* config args
Add add'l config flags from #3394.
Update patch to no-op and add comment.
Add add'l dependencies after studying `configure --help`.
* Flake8 cleanup
* Add support for blast+@2.2.30
Need to work around a configure bug. See https://github.com/Homebrew/homebrew-science/issues/2337#issuecomment-170011511.
* Re-wrap comment, line length not a problem.
I'd broken a comment across to line in a fit of flake8
induced fear. Adam pointed out that it's ok.
* Remove explicit prefix config
Remove explicit prefix config, the autotools package does it for us.
* Added packages for Structural Simulation Toolkit and Dumpi trace
* sst-dumpi deps
Remove extraneous dependency from sst-dumpi
* Removed redundant function and fake-url from sst-dumpi package
* Fixed URL, Added git branches, requested changes
sst-dumpi version listed as 6.1, made an upstream change to 6.1.0
sst-macro and sst-dumpi can be installed as v6.1.0 and master branch
* Added bootstrap. Tweaked URLs and description
- drop old TCL support from paraview build.
- add +plugins variant to have include directories installed. This is
enabled by default since the additional diskspace for includes is
really minimal and since this also allows re-use of the VTK libraries
from ParaView without necessarily requiring a separate VTK
installation.
- +opengl2 is now the default. As per all newer VTK and paraview versions.
BUG: broken install for paraview-5.0.1 with includes and without python
- incorrect conditional for ui_pqExportStateWizard.h when python is
disabled and includes are to be installed.
gcc compiler detection patch.
These have both been fixed in paraview 5.3.0
ENH: refactor as a CMakePackage.
- Note that "spack install paraview" works as expected, but
"spack build paraview" fails in weird unrelated ways.
* Fix nettle dependency
GnuTLS requires a nettle from the 2.7 series, won't work
with 3.x (so sayeth configure when it crashes...).
* Fix configure symlinking GNUmakefile to itself
I'm confused about how this ever worked, but....
As configure runs, it does something like this:
```
config.status: linking /home/hartzelg/tmp/spack/var/spack/stage/gnutls-3.3.9-uq2u4ecucncv2l3akran765aystyvkys/gnutls-3.3.9/GNUmakefile to GNUmakefile
```
and since that is the cwd the build crashes with:
```
==> 'make' '-j28'
make: GNUmakefile: Too many levels of symbolic links
make: stat: GNUmakefile: Too many levels of symbolic links
make: *** No rule to make target `GNUmakefile'. Stop.
```
Since the comment in the configure script points out that the
GNUmakefile is "only for the maintainer" I've just commented out the
bit that leads to the symlinking.
Builds for me on CentOS 7. (I do have a recent-ish set of auto-*
things in my patch via linuxbrew)
* Wordsmith
* Fix symbolic link problem properly
Autotools magic. See #3565 for the details.
* Flake8 cleanup
* Add some headroom on the version number constraint
We really want < 3.0 but there's not a good way to express it.
We'd like to avoid having to update the file if they release 2.7.2.
This seems like a good compromise.
* Add url_for_version and depends_on('libiconv')
Fetching is broken. See https://github.com/LLNL/spack/issues/3554.
libgd, when used in gnuplot at least, needs a dependency
on libiconv.
* Undo change from debugging pokery
* pnfft: new package
* Convert some packages with overridden install from AutotoolsPackage to Package
* pnfft: fix URL
* Switch FFTW-derived packages back to AutotoolsPackage
* Disable unneeded build phases in FFTW and derived packages
* Separate build phases for FFTW and derived packages
* Fix broken merge
* fftw: pfft_patches for 3.3.6
* fftw: address @adamjstewart’s review comments
* Clean up the gcc package
* Add preliminary testing support
* Older versions of GCC do not depend on MPC
* DejaGnu 1.4.4 cannot be built in parallel
* Fix GCC test dir
* Add two more deps required to run the testsuite
there were various updated packages, both for the 2.x, 2.0.x,
and 1.10.x release branches of Open MPI.
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
(cherry picked from commit 3b671935783fc3af76dc642a54461c2f02ab8ebf)
* Replacing failing git clone by valid url
* Moving to previous version package (0.163)
* Removing automake and autoconf dependencies and adding url_list to pick up newer versions
* Corrected typo
* Reorder versions and removed libtool
* Add missing dependencies to gawk
* Fix make check
* Add mawk packages and awk virtual dependency
* Use mawk instead of gawk to build ncurses
* Rearrange order of patches and dependencies
* hpx5: Convert to AutotoolsPackage, several updates
- convert to AutotoolsPackage
- add several variants
- add several dependencies
- add new version 4.0.0
- don’t set compiler flags explicitly
* hpx5: Don’t import os
* hpx5: Clean up configure arguments
* hpx5: Correct libffi configure declaration
Also add new variant “+instrumentation”.
* hpx5: Correct CUDA, MPI, OpenCL configuration
* hpx5: Use built-in libffi
Spack’s libffi installs its headers in a strange place, and hpx5 can’t pick them up.
* libpciaccess: Convert to AutotoolsPackage, remove Darwin work-around
Other packages that depend on libpciaccess can do so via a conditional dependency (via a variant); we don’t need a dummy package for Darwin any more.
* Disable libpciaccess dependency on non-Darwin platforms
* hwloc: Remove direct dependency on libpciaccess
* Added support for building Theano with GPU support. Added a package
for the libgpuarry libary needed for Theano.
* Added dependency on py-pexpect package which ipython requires.
* hwloc: Add missing dependencies
- add several missing dependencies
- add variants for various features
- add new version 1.11.6
* hwloc: Add descriptions to variants
* New package pocl
* pocl: Update dependencies, add self-test
* pocl: Don't require LLVM shared libraries
LLVM build fails with shared libraries.
* Add patch
* Update
* Update
* Make build and install work; install test still fails
* Split pocl into pocl proper and pocl-test
* Add debug output
* pocl: Update to 0.14-rc
* pocl: Correct flake8 error
* pocl: Heed code review recommendations
* pocl: Add newline at end of file
* pocl: Correct flake8 error
I don’t want to use an even longer line by putting the whole variant declaration onto a single line, nor do I think that having an overlong line and adding `# noqa` at the end is more readable than splitting a string over three lines.
* pocl: Correct dependency type for libtool
Previously, this would fail with a NoSuchMethodError:
class Package(object):
# this is the default implementation
def some_method(self):
...
class Foo(Package):
@when('platform=cray')
def some_method(self):
...
@when('platform=linux')
def some_method(self):
...
This fixes the implementation of `@when` so that the superclass method
will be invoked when no subclass method matches.
Adds tests to ensure this works, as well.
* Add paths to PKG_CONFIG_PATH to find cray .pc file
Adds two paths: /usr/lib64/pkgconfig and /usr/local/lib64/pkgconfig.
cray pkg-config looks at two different paths which are not included
in PKG_CONFIG_PATH environment variable. This PR addresses #3167.
* Flake8 fixes
* flake8 fix again
* cosmomc: New package CosmoMC
* cosmomc: Improve version numbering scheme
* planck-likelihood: New package
* cosmomc: Pacify flake8
* cosmomc: Code cleanup as requested in the review
* cosmomc: Various updates suggested during review
* cosmomc: Limit compiler versions (as documented in the build instructions)
* cosmomc: Correct flake8 error
* cosmomc: Use `libs` instead of `lapack_libs`
* planck-likelihood: Make makefile output more verbose and readable
* planck-likelilhood: Correct order of libraries when calling linker
* cosmomc: Clean up check for suitable compiler
* planck-likelihood: Use modern mechanism to set environment variables
* Cleanup list of offered versions (only the latest update release of each minor version is kept: 1.4.4, 1.5.2, 1.6.6, 1.7.1); introduced master revision (via Git). Slightly more selective installation of the documentation; copy README.md and LICENSE files to documentation folder. Removed outdated build options (make_args); generally stay with LIBXSMM's defaults (general purpose build). Introduced one build variant ("debug").
* Incorporated comments of PR #3371 (https://github.com/LLNL/spack/pull/3371).
* Fixed flake8 complaints.
* Made the full list of versions available starting with version 1.4 (there are earlier releases, however those have been released before the Spack specification became available; original spec. also started with 1.4).
* Made DBG and TRACE actually two separate arguments.
The tests fail on some systems (e.g. Comet at SDSC) that impose limits on the number of processes or threads one can run simultaneously on the head node. Thus the tests should not be run by default.
* ADIOS: change the invocation of python script that generates test files
The python script ADIOS*/utils/gpp/gpp.py is invoked directly in the
ADIOS*/tests/genarray/Makefile and the name of the python interpreter
can be quite long if it is one built under spack, i.e. longer than the
80 characters allowed by the #! line in bash. The name of the shbang
line is truncated at 80 characters and bash cannot find the python
interpreter specified.
This fix changes how the script is executed, by invoking it under
python. This way, the shbang line is ignored.
* adios: fixed flake8 errors from previous commit.
* adio: Switch to AutotoolsPackage class from Package class
Switched in order to add support for the ppc64le platform.
* Adding highwayhash package from Google
* Added the highwayhash package for Tensorflow
* Fixed the numerous flake8 issues
* fixed requested changes
* added conditionals for some installs
* forgot an import
* create install directories before you install anything in them
* Add new package for PaGMO/PyGMO
* Rename py-pygmo to pagmo
* Fix concretization bug
* Correct Python detection, use correct flags
* Install PyGMO to the pagmo prefix, not the Python prefix
This commits address build issues reported in issue #3268.
In particular, makes cleaner the dependency on parallel-netcdf
when +ncmpi variant is requested and adds CFLAGS=-D H5_USE_16_API
to handle the fact that Ior uses old 1.6 style APIs when +hdf5 is
requested.
These changes make Ior build fine and create modules for HDF5 versions
1.8.x and 1.10.0.
Even though build is successfull the modules haven't been tested in a
run.
