* Add openfoam to aws-pcluster-neoverse_v1 stack
* Add more apps to aws-pcluster-x86_64_v4 stack
* Remove WRF while hdf5 cannot build in buildcache at the moment
* Update comment
* Add more apps for aws-pcluster-neoverse_v1 stack
* Remove apps that currently do not build
* Disable those packages that won't build
* Modify syntax such that correct cflags are used
* Changing syntax again to what works with other packages
* Fix overriding packages.yaml entry for gettext
* Use new `prefer` and `require:when` clauses to clarify intent
* Use newer spack version to install intel compiler
This removes the need for patches and makes sure the `prefer` directives in
`package.yaml` are understood.
* `prefer` not strong enough, need to set compilers
* Revert "Use newer spack version to install intel compiler"
This reverts commit ecb25a192c.
Cannot update the spack version to install intel compiler as this changes the
compiler hash but not the version. This leads to incompatible compiler paths. If
we update this spack version in the future make sure the compiler version also updates.
Tested-by: Stephen Sachs <stesachs@amazon.com>
* Remove `prefer` clause as it is not strong enough for our needs
This way we can safely go back to installing the intel compiler with an older
spack version.
* Prefer gcc or oneapi to build gettext
* Pin gettext version compatible with system glibc-headers
* relax gettext version requirement to enable later versions
* oneapi cannot build older gettext version
* ParaView: Update version 5.12.0
Add 5.12.0 release
Update default to 5.12.0
* Add patch for building ParaView 5.12 with kits
* Drop VTKm from neoverse
* Changes to re-enable aws-pcluster pipelines
- Use compilers from pre-installed spack store such that compiler path relocation works when downloading from buildcache.
- Install gcc from hash so there is no risk of building gcc from source in pipleine.
- `packages.yam` files are now part of the pipelines.
- No more eternal `postinstall.sh`. The necessary steps are in `setup=pcluster.sh` and will be version controlled within this repo.
- Re-enable pipelines.
* Add and
* Debugging output & mv skylake -> skylake_avx512
* Explicilty check for packages
* Handle case with no intel compiler
* compatibility when using setup-pcluster.sh on a pre-installed cluster.
* Disable palace as parser cannot read require clause at the moment
* ifort cannot build superlu in buildcache
`ifort` is unable to handle such long file names as used when cmake compiles
test programs inside build cache.
* Fix spack commit for intel compiler installation
* Need to fetch other commits before using them
* fix style
* Add TODO
* Update packages.yaml to not use 'compiler:', 'target:' or 'provider:'
Synchronize with changes in https://github.com/spack/spack-configs/blob/main/AWS/parallelcluster/
* Use Intel compiler from later version (orig commit no longer found)
* Use envsubst to deal with quoted newlines
This is cleaner than the `eval` command used.
* Need to fetch tags for checkout on version number
* Intel compiler needs to be from version that has compatible DB
* Install intel compiler with commit that has DB ver 7
* Decouple the intel compiler installation from current commit
- Use a completely different spack installation such that this current pipeline
commit remains untouched.
- Make the script suceed even if the compiler installation fails (e.g. because
the Database version has been updated)
- Make the install targets fall back to gcc in case the compiler did not install
correctly.
* Use generic target for x86_64_vX
There is no way to provision a skylake/icelake/zen runner. They are all in the
same pools under x86_64_v3 and x86_64_v4.
* Find the intel compiler in the current spack installation
* Remove SPACK_TARGET_ARCH
* Fix virtual package index & use package.yaml for intel compiler
* Use only one stack & pipeline per generic architecture
* Fix yaml format
* Cleanup typos
* Include fix for ifx.cfg to get the right gcc toolchain when linking
* [removeme] Adding timeout to debug hang in make (palace)
* Revert "[removeme] Adding timeout to debug hang in make (palace)"
This reverts commit fee8a01580489a4ea364368459e9353b46d0d7e2.
* palace x86_64_v4 gets stuck when compiling try newer oneapi
* Update comment
* Use the latest container image
* Update gcc_hashes to match new container
* Use only one tag providing tags per extends call
Also removed an unnecessary tag.
* Move generic setup script out of individual stack
* Cleanup from last commit
* Enable checking signature for packages available on the container
* Remove commented packages / Add comment for palace
* Enable openmpi@5 which needs pmix>3
* don't look for intel compiler on aarch64
* e4s: new packages: glvis, laghos
* gl: require: osmesa
* be explicit: glvis ^llvm so that llvm-amdgpu not chosen
* glvis fails on oneapi stack due to issue 42839
* gitlab: remove requests for unreferenced packages
The packages removed in this commit are not built by any of
our current GitLab CI stacks.
* gitlab: update memory requests for "huge" packages
* gitlab: reduce memory requests for overprovisioned packages
* gitlab: more memory for py-torch (again)
* gitlab: update memory but keep CPU the same
It is useful to enable/disable stacks in order to handle turning
specific stacks on/off based on runner availability, stack stability,
testing requirements, etc.
The disabled stack list takes precedence over the enable stack list. The
assumption is that stacks that are disabled are so due to some
functionality missing or broken for that stack.
The enable stack list implicitly disables all stacks not listed in the
enable list.
The main script body is over-written for power. Putting thet timing
aggregation in the after script allows it to be called on all of the
current pipelines.
This "breaks" the deprecated schema by allowing unknown attributes
to the attributes section of the job types. The breaking change here is
that deprecated stacks will no longer ignore attributes that are unknown
but rather assume the new CI schema behavior of injecting them into the
generated CI configuration. This change is required to secure
authentication in Spack CI.
* gitlab: remove commented-out duplicate entries
* gitlab: reclassify some packages from "huge" to "large"
Our observed max memory usage for these packages is as follows:
hipblas: 7.7G
qt: 6.6G
visit: 9.7G
All of these should fit within a "large" request (currently 12G).
* gitlab: remove pango from list of huge packages
This package is not currently built by any of our CI stacks.
* gitlab: update requests for high memory packages
Refine resource requests for memory-intensive packages based on
max memory usage data.
Currently requirements allow to express "strong preferences" and "conflicts" from
configuration using a convoluted syntax:
```yaml
packages:
zlib-ng:
require:
# conflict on %clang
- one_of: ["%clang", "@:"]
# Strong preference for +shared
- any_of: ["+shared", "@:"]
```
This PR adds syntactic sugar so that the same can be written as:
```yaml
packages:
zlib-ng:
conflict:
- "%clang"
prefer:
- "+shared"
```
Preferences written in this way are "stronger" that the ones documented at:
- https://spack.readthedocs.io/en/latest/packages_yaml.html#package-preferences
* e4s ci: use latest intel/hpckit 2024 based image
* use latest container image: ecpe4s/ubuntu22.04-runner-amd64-oneapi-2024.0.0:2023.12.01
* comment out failing specs
* update to use patched container
* remove generalized package preference for intel-oneapi-mkl@2023
* change packages commented out
* Add rust-analyzer as variant to rust build
* Expose cargo module only when +cargo
* rust: add v1.74.0 and v1.75.0 and remove variants in favor of +dev
* [@spackbot] updating style on behalf of alecbcs
* Fix variant typo
---------
Co-authored-by: alecbcs <alecbcs@users.noreply.github.com>
The gcc-runtime package adds a separate node for gcc's dynamic runtime
libraries.
This should help with:
1. binary caches where rpaths for compiler support libs cannot be
relocated because the compiler is missing on the target system
2. creating "minimal" container images
The package is versioned like `gcc` (in principle it could be
unversioned, but Spack doesn't always guarantee not mixing compilers)
This PR changes the default behavior of `spack config get` and `spack config blame`
to print a flattened version of the entire spack configuration, including any active
environment, if the commands are invoked with no section arguments.
The new behavior is used in Gitlab CI to help debug CI configuration, but it can also
be useful when asking for more information in issues, or when simply debugging Spack.
* developer tools stack try 2
This version is actually in use locally and has largely stabilized, at
least on x86. Some packages are still a challenge on ppc64le, but maybe
worth keeping this working as a set.
* add packages, try to get container with newer gcc
* remove reuse: true
* try to get cmake to build on medium, 25 minutes is too long
* add lsd package and add to dev tools stack
* clean up fzf dependency and sorting
* Update share/spack/gitlab/cloud_pipelines/stacks/developer_tools/spack.yaml
* cuda: add 12.3.0 (#40827)
* Switch to dashes
* yet more underscores
---------
Co-authored-by: Paul R. C. Kent <kentpr@ornl.gov>
* oneapi 2024.0.0 release
* oneapi v2 directory support and some cleanups
* sycl abi change requires 2024 compilers for packages that use sycl
---------
Co-authored-by: Robert Cohn <robert.s.cohn@intel.com>
* e4s ci: add exago +cuda, +rocm builds
* exago: rename 5-18-2022-snapshot to snapshot.5-18-2022
* disable exago +rocm for non-external rocm ci install
* note that hiop +rocm fails to find hip libraries when they are spack-installed
* ci: don't register detectable compilers
Cause they go out of sync...
* remove intel compiler, it can be detected too
* Do not run spack compiler find since compilers are registered in concretize job already
* trilinos: work around +stokhos +cuda +superlu-dist bug due to EMPTY macro
* gromacs +cp2k: build in CI
* libxsmm: x86 only
* attempt to fix dbcsr + new mpich
* use c11 standard
* gromacs: does not depend on dbcsr
* cp2k: build with cmake in CI, s.t. dbcsr is a separate package
* cp2k: cmake patches for config files and C/C++ std
* cp2k: remove unnecessary constraints due to patch
Improve how mirrors are used in gitlab ci, where we have until now thought
of them as only a string.
By configuring ci mirrors ahead of time using the proposed mirror templates,
and by taking advantage of the expressiveness that spack now has for mirrors,
this PR will allow us to easily switch the protocol/url we use for fetching
binary dependencies.
This change also deprecates some gitlab functionality and marks it for
removal in Spack 0.23:
- arguments to "spack ci generate":
* --buildcache-destination
* --copy-to
- gitlab configuration options:
* enable-artifacts-buildcache
* temporary-storage-url-prefix
* Allow branching out of the "generic build" unification set
For cases like the one in https://github.com/spack/spack/pull/39661
we need to relax rules on unification sets.
The issue is that, right now, nodes in the "generic build" unification
set are unified together with their build dependencies. This was done
out of caution to avoid the risk of circular dependencies, which would
ultimately cause a very slow solve.
For build-tools like Cython, however, the build dependencies is masked
by a long chain of "build, run" dependencies that belong in the
"generic build" unification space.
To allow splitting on cases like this, we relax the rule disallowing
branching out of the "generic build" unification set.
* Fix issue with pure build virtual dependencies
Pure build virtual dependencies were not accounted properly in the
list of possible virtuals. This caused some facts connecting virtuals
to the corresponding providers to not be emitted, and in the end
lead to unsat problems.
* Fixed a few issues in packages
py-gevent: restore dependency on py-cython@3
jsoncpp: fix typo in build dependency
ecp-data-vis-sdk: update spack.yaml and cmake recipe
py-statsmodels: add v0.13.5
* Make dependency on "blt" of type "build"
* e4s amd64 gcc ci stack: sync with e4s-23.08
* e4s amd64 oneapi ci stack: sync with e4s-23.08
* e4s ppc64 gcc ci stack: sync with e4s-23.08
* add new ci stack: e4s amd64 gcc w/ external rocm
* add new ci stack: e4s arm gcc ci
* updates
* py-scipy: -fvisibility issue is resolved in 2023.1.0: #39464
* paraview oneapi fails
* comment out pkgs that fail to build on power
* fix arm stack name
* fix cabana +cuda specification
* comment out failing spces
* visit fails build on arm
* comment out slepc arm builds due to make issue
* comment out failing dealii arm builds
* e4s cray rhel stack: expand to full spec list
* comment out gasnet; require %gcc for unzip
* require openblas@0.3.20 to get around %cce error; follow up with issue report
* comment out failing specs;
* comment out axom and xyce due to errors
* improve clarity of failing specs
* Add OIDC tokens to gitlab-ci jobs
This should allow us to start issuing just-in-time generated
credentials for CI jobs that need to modify binary mirrors. The "aud"
claim of the token describes what the token is allowed to do. The
claim is verified against a set of rules on the IAM role using signed
information from GitLab. See spack-infrastructure for the claim
verification logic.
---------
Co-authored-by: Scott Wittenburg <scott.wittenburg@kitware.com>
Currently the configuration of the pipeline is such that
there are multiple "optimal" solutions. This is due to
the pipeline making ~lld the default for LLVM, but leaving
+libomptarget from package.py
Since LLVM recipe has a:
conflicts("~lld", "+libomptarget")
clingo is forced to change one of the two defaults, and
the penalty for doing so is the same for each. Hence, it
is random whether we'll get +libomptarget+lld
or ~libomptarget~lld.
This fixes it by changing also the default for libomptarget.
Instead of pointing to the image on DockerHub, which rate limits us and
causes pipeline failures durying busy times, use the version at ghcr.
And we might as well use the ghcr version everywhere else too.
* Write timing information for installs from cache
* CI: aggregate and upload install_times.json to artifacts
* CI: Don't change root directory for artifact generation
* Flat event based timer variation
Event based timer allows for easily starting and stopping timers without
wiping sub-timer data. It also requires less branching logic when
tracking time.
The json output is non-hierarchical in this version and hierarchy is
less rigidly enforced between starting and stopping.
* Add and write timers for top level install
* Update completion
* remove unused subtimer api
* Fix unit tests
* Suppress timing summary option
* Save timers summaries to user_data artifacts
* Remove completion from fish
* Move spack python to script section
* Write timer correctly for non-cache installs
* Re-add hash to timer file
* Fish completion updates
* Fix null timer yield value
* fix type hints
* Remove timer-summary-file option
* Add "." in front of non-package timer name
---------
Co-authored-by: Harmen Stoppels <harmenstoppels@gmail.com>
Co-authored-by: Harmen Stoppels <me@harmenstoppels.nl>
Setting the undocumented variable SPACK_CONCRETIZER_REQUIRE_CHECKSUM
now causes the solver to avoid accounting for versions that are not checksummed.
This feature is used in CI to avoid spurious concretization against e.g. develop branches.
* VTK: Add patch for python 3.8 support
* CI: Re-enable VisIt in CI
* Configure spec matrix for stack with VisIt
* Add pugixml dep for 8.2.0
* Make VTK and ParaView consistent on proj dep
* OpenMPI 3: provides MP support by default
* Add details on proj dep in ParaView
* Add python 3.8 to test mock repo
* Patches to get VisIt VTK interface
* CI: Disable VisIt with GUI in DAV
* AMReX: 23.06+ Multi-Dim Support
This updated the Spack package to allow to install AMReX, modules of
AMReX in E4S deployments and dependent packages with support for
multiple dimensions. Due to an upstream change in AMReX, we do not
longer need to ship three, binary incompatible package variants.
* [E4S] oneAPI AMReX < 23.06 Variant
Work-around the auto-concretization to the multi-dim of `dimensions`,
which only in 23.06+ became a multi-variant.
* e4s cray rhel ci: temporarily disable amrex build until spurious ci failure can be resolved
---------
Co-authored-by: eugeneswalker <eugenesunsetwalker@gmail.com>
Refactor gitlab ci configs so that mac and cray jobs can reuse as much higher level
configuration as possible.
* CI: remove redundant sections
* CI: Include base linux CI configs in cray stacks
Relocation and runner mapping is consistent between cray and linux runners.
* Export user cache path in before script
* CI: add GPG root for mac runners
* Disable user configs
Metal runners share a ~ directory
* Disable user config and add configs in activate env
* ci: run spack list in power ci
Let's see if Spack itself is the bottleneck in CI...
* rebuild curl in CI
* more of the same please!
* drop the profiler
* undo rebuildme test in ci variant
* add comment for posterity
* enable profiling
* trigger CI
* See how it goes now that perf regressions are fixed on develop
* try shorter poll intervals
* Revert "try shorter poll intervals"
This reverts commit d60c34ad3eceead0c13a5277cf8e783fd42b7458.
* Remove spec.format call in Database._get_matching_spec_key
* once more in ci please
* undo irrelevant changes
* run spack list in before script
* test in ci
* -:
* Undo CI testing
* openssl: prefer 3.x
This PR is not intended to be merged immediately, but it would be good
to see what packages fail to build in CI so that we can get proper
version constraints on openssl (before all packages update and support
both openssl 1 and 3)
* Disable assembly for 3.x %oneapi
* cmake: depend on spack curl, to deal with curl - openssl compat
* also make zlib external
* remove overly strict & unsafe requirement on py-cryptographty patch version number
* update openssl compat bounds in py-cryptography
* smaller diff
* Make libssh2 an autotools/cmake package
* fix weird upperbound in libssh2 as there is not openssl v2
* libssh2: pc file lists plain -lssl -lcrypto w/o leading -L flag, confusing libgit2 parsing of pkg-config output
* Actually fix the issue in libssh2: its pc file looks broken
