| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
* update spack package from axom's main repo
* flake8, plus loop over possible clangformat paths
|
|
|
|
|
|
* update rocm-bandwidth-test,rocm-dbgapi,rocprofiler for rocm 3.9.0
* bump up rocm-gdb for rocm-3.9.0
|
|
|
|
|
|
|
|
|
|
|
|
* Add new package: abi-compliance-checker
* refine dependencies
* change exuberant-ctags to universal-ctags
|
|
* libusb: disable udev
spack has no libudev/systemd package currently
* convert to AutotoolsPackage
* remove spack import
|
|
|
|
The previous patch left out the SGE.script file.
|
|
|
|
* bump up version for 3.9.0 release
* update version of rocminfo for rocm-3.9.0
* bump up rocm-cmake version for rocm-3.9.0
* bump up rocm-smi and rocmdevice-libs for 3.9.0
* bumpup comgr version for rocm_ 3.9.0
* bump rocm-clang-ocl for rocm-3.9.0
* bump hipify-clang for rocm-3.9.0
|
|
without this, the library does not have a dependency
on the fft lib it was built with.
|
|
* py-json-get: new package at 1.1.1
* py-json-get: new package at 1.1.1
Co-authored-by: las_djorton <las_djorton@build.las.iastate.edu>
|
|
|
|
* amrex: new version 20.11
* Fix formatting issues
|
|
|
|
fix to allow Gtk and others work from 'import gi' etc.
|
|
|
|
|
|
|
|
|
|
|
|
In this case, the `zen2` architecture is unknown; Spack must use the
LLVM-style `znver2` target instead.
|
|
* Trilinos: Add STRUMPACK dependency
* break long lines, flake8 cleanup
* Use spec['strumpack'].libs.directories[0]
instead of spec['strumpack'].prefix.lib
because libraries may be in lib or lib64.
Likewise use headers.directories[0] iso prefix.include.
Suggested by adamjstewart
|
|
|
|
* range-v3: add version 0.11.0
This release drops support for llvm-3.9 per the release notes.
https://github.com/ericniebler/range-v3/releases/tag/0.11.0
* range-v3: rename 'develop' version to 'master'
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
|
|
|
|
and superlu-dist future versions] (#19686)
|
|
|
|
|
|
|
|
* Add new package: py-boom
* change name to py-boom-boot-manager
|
|
|
|
Running v3.1.2 with Python 3.7 returns a SyntaxError close to `async=True`.
|
|
* Add new package: abi-dumper
* refine dependencies
|
|
* Updated cuDNN package to check to make sure that target directory
exists before linking it.
* Fixed flake8
* Fixed Flake8
|
|
* allows UCX since v1.7 to build with more recent version of gdrcopy (v2.X)
* Update var/spack/repos/builtin/packages/ucx/package.py
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
|
|
There was an error introduced in #19209 where `full_hash()` and
`build_hash()` are called on older specs that we've read in from the DB;
older specs may not be able to compute these hashes (e.g. if they have
removed patches used in computing the full_hash).
When serializing a Spec, we want to generate the full/build hash when
possible, but we need a mechanism to skip it for Specs that have
themselves been read from YAML (and may not support this).
To get around this ambiguity and to fix the issue, we:
- Add an attribute to the spec called `_hashes_final`, that is `True`
if we can't lazily compute `build_hash` and `full_hash`.
- Set `_hashes_final` to `False` for new specs (i.e., lazily
computing hashes is ok)
- Set `_hashes_final` to `True` for concrete specs read in via
`from_node_dict`, as it may be too late to recompute hashes.
- Compute and write out all hashes in `node_dict_with_hashes` *if
possible*.
Effectively what this means is that we can round-trip specs that are
missing `_build_hash` and `_full_hash` without recomputing them, but for
all new specs, we'll compute them and store them. So Spack should work
fine with old DBs now.
|
|
|
|
|
|
* root: update to built-in CMakePackage functions
* root: Disable options from missing variants
* Remove modification of CMAKE_PROGRAM_PATH
|
|
Co-authored-by: Scott McMillan <smcmillan@nvidia.com>
|
|
fixes #19625
|
|
Currently fails with:
Error: NoLibrariesError: Unable to recursively locate rocm-device-libs libraries
|
|
* hip: rocminfo is a runtime requirement
* hip: +setup_run_environment, +setup_dependent_run_environment
* hip: run environment: get lib dir using libs.directories[0], not prefix.lib
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
|
|
This fixes sbang relocation when using old binary packages, and updates
code in `relocate.py`.
There are really two places where we would want to handle an `sbang`
relocation:
1. Installing an old package that uses `sbang` with shebang lines like
`#!/bin/bash $spack_prefix/sbang`
2. Installing a *new* package that uses `sbang` with shebang lines like
`#!/bin/sh $install_tree/sbang`
The second case is actually handled automatically by our text relocation;
we don't need any special relocation logic for new shebangs, as our
relocation logic already changes references to the build-time
`install_tree` to point to the `install_tree` at intall-time.
Case 1 was not properly handled -- we would not take an old binary
package and point its shebangs at the new `sbang` location. This PR fixes
that and updates the code in `relocation.py` with some notes.
There is one more case we don't currently handle: if a binary package is
created from an installation in a short prefix that does *not* need
`sbang` and is installed to a long prefix that *does* need `sbang`, we
won't do anything. We should just patch the file as we would for a normal
install. In some upcoming PR we should probably change *all* `sbang`
relocation logic to be idempotent and to apply to any sort of shebang'd
file. Then we'd only have to worry about which files to `sbang`-ify at
install time and wouldn't need to care about these special cases.
|
