Preferred targets were failing because we were looking them up by
Microarchitecture object, not by string.
- [x] Add a call to `str()` to fix target lookup.
- [x] Add a test to exercise this part of concretization.
- [x] Add documentation for setting `target` in `packages.yaml`
* microarchitectures: zen starts from x86_64, not from excavator
* Unit tests: fixed a test that is wrong with the new modeling
* microarchitectures: fixed features and inheritance for 15h family
bulldozer doesn't inherit from barcelona (10h) + added xop, lwp and tbm
instruction sets to the 15h family (it distinguish the family from 17h)
Addresses #12804
This PR adds the creation of the remaining (16) templates to ensure we can create them with expected content. The goal is to facilitate catching during testing.
Spack doesn't need `requests`, and neither does `jsonschema`, but
`jsonschema` tries to import it, and it'll succeed if `requests` is on
your machine (which is likely, given how popular it is). This commit
removes the import to improve Spack's startup time a bit.
On a mac with SSD, the import of requests is ~28% of Spack's startup time
when run as `spack --print-shell-vars sh,modules` (.069 / .25 seconds),
which is what `setup-env.sh` runs.
On a Linux cluster where Python is mounted from NFS, this reduces
`setup-env.sh` source time from ~1s to .75s.
Note: This issue will be eliminated if we upgrade to a newer `jsonschema`
(we'd need to drop Python 2.6 for that). See
https://github.com/Julian/jsonschema/pull/388.
- This is needed to support Cray machines -- we need an architecture
mic_knl > x86_64
- We used Cray's naming scheme for this target to make it work seamlessly
with the module-based detection sccheme on Cray. mic_knl is pretty
much dead, so this will be the last succh target. We will need to work
wtih Cray and other vendors in the future.
Seamless translation from 'target=<generic>' to either
- target.family == <generic> (in methods)
- 'target=<generic>:' (in directives)
Also updated docs to show ranges in directives.
Spack can now:
- label ppc64, ppc64le, x86_64, etc. builds with specific
microarchitecture-specific names, like 'haswell', 'skylake' or
'icelake'.
- detect the host architecture of a machine from /proc/cpuinfo or similar
tools.
- Understand which microarchitectures are compatible with which (for
binary reuse)
- Understand which compiler flags are needed (for GCC, so far) to build
binaries for particular microarchitectures.
All of this is managed through a JSON file (microarchitectures.json) that
contains detailed auto-detection, compiler flag, and compatibility
information for specific microarchitecture targets. The `llnl.util.cpu`
module implements a library that allows detection and comparison of
microarchitectures based on the data in this file.
The `target` part of Spack specs is now essentially a Microarchitecture
object, and Specs' targets can be compared for compatibility as well.
This allows us to label optimized binary packages at a granularity that
enables them to be reused on compatible machines. Previously, we only
knew that a package was built for x86_64, NOT which x86_64 machines it
was usable on.
Currently this feature supports Intel, Power, and AMD chips. Support for
ARM is forthcoming.
Specifics:
- Add microarchitectures.json with descriptions of architectures
- Relaxed semantic of compiler's "target" attribute. Before this change
the semantic to check if a compiler could be viable for a given target
was exact match. This made sense as the finest granularity of targets
was architecture families. As now we can target micro-architectures,
this commit changes the semantic by interpreting as the architecture
family what is stored in the compiler's "target" attribute. A compiler
is then a viable choice if the target being concretized belongs to the
same family. Similarly when a new compiler is detected the architecture
family is stored in the "target" attribute.
- Make Spack's `cc` compiler wrapper inject target-specific flags on the
command line
- Architecture concretization updated to use the same algorithm as
compiler concretization
- Micro-architecture features, vendor, generation etc. are included in
the package hash. Generic architectures, such as x86_64 or ppc64, are
still dumped using the name only.
- If the compiler for a target is not supported exit with an intelligible
error message. If the compiler support is unknown don't try to use
optimization flags.
- Support and define feature aliases (e.g., sse3 -> ssse3) in
microarchitectures.json and on Microarchitecture objects. Feature
aliases are defined in targets.json and map a name (the "alias") to a
list of rules that must be met for the test to be successful. The rules
that are available can be extended later using a decorator.
- Implement subset semantics for comparing microarchitectures (treat
microarchitectures as a partial order, i.e. (a < b), (a == b) and (b <
a) can all be false.
- Implement logic to automatically demote the default target if the
compiler being used is too old to optimize for it. Updated docs to make
this behavior explicit. This avoids surprising the user if the default
compiler is older than the host architecture.
This commit adds unit tests to verify the semantics of target ranges and
target lists in constraints. The implementation to allow target ranges
and lists is minimal and doesn't add any new type. A more careful
refactor that takes into account the type system might be due later.
Co-authored-by: Gregory Becker <becker33.llnl.gov>
Add llnl.util.cpu_name, with initial support for detecting different
microarchitectures on Linux. This also adds preliminary changes for
compiler support and variants to control the optimizatoin levels by
target.
This does not yet include translations of targets to particular
compilers; that is left to another PR.
Co-authored-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Move verbose messages to debug level
get_patchelf should return None for test platform as well because create_buildinfo invokes patchelf to get rpaths.
Update command-line (CLI) parsing to understand references to yaml
files that store Spack specs. Where a file reference is encountered,
the full Spec in the file will be read in. A file reference may
appear anywhere that a spec could appear before. For example, if you
write "spack spec -y openmpi > openmpi.yaml" you may then install the
spec using the yaml file by running "spack install ./openmpi.yaml";
you can also refer to dependencies in this way (e.g.
"spack install foo^./openmpi.yaml").
There are two requirements for file references:
* A file path entered on the CLI must include a "/" even if the file
exists in your current working directory. For example, if you
create an openmpi.yaml file as above and run
"spack install openmpi.yaml" from the same directory, it will
report an error.
* A file path entered on the CLI must end with ".yaml"
This commit adds error messages to clearly inform the user of both
violations.
* implicit_rpaths are now removed from compilers.yaml config and are always instantiated dynamically, this occurs one time in the build_environment module
* per-compiler list required libraries (e.g. libstdc++, libgfortran) and whitelist directories from rpaths including those libraries. Remove non-whitelisted implicit rpaths. Some libraries default for all compilers.
* reintroduce 'implicit_rpaths' as a config variable that can be used to disable Spack insertion of compiler RPATHs generated at build time.
Fixes#12732Fixes#12767c22a145 added automatic detection and RPATHing of compiler libraries
to Spack builds. However, in cases where the parsing/detection logic
fails this was terminating the build. This makes the compiler library
detection "best-effort" and reports an issue when the detection fails
rather than terminating the build.
This is similar to #10191. The Ubuntu package for clang 8.0.0 displays
a very unusual version string, and we need this new regex to detect it
as just 8.0.0
Unit test have been complemented by the output that was failing
detection.
- Fix trailing whitespace missed by the bug described in #12755.
- Fix other style issues that have crept in over time (this can happen
when flake8 adds new checks with new versions)
E501 (line too long) exemptions are probably our most common ones -- we
add them for directives, URLs, hashes, etc. in packages. But we
currently add them even when a line *doesn't* need them, which can mask
trailing whitespace errors.
This changes `spack flake8` so that it will only add E501 exemptions if
the line is *actually* too long.
Co-Authored-By: Adam J. Stewart <ajstewart426@gmail.com>
mock_archive can now take multiple extension / tar option pairs (default matches old behavior).
url_fetch.test_fetch tests more archive types.
compression.EXTS split into EXTS and NOTAR_EXTS to avoid unwanted, non-meaningful combinatoric extensions such as .tar.tbz2.
- previously spec parsing didn't allow you to look up missing (but still
known) specs by hash
- This allows you to reference and potentially reinstall
force-uninstalled dependencies
- add testing for force uninstall and for reference by spec
- cmd/install tests now use mutable_database
* When cleaning the stage root, only remove directories that appear
to be used for staging Spack packages. Previously Spack was clearing
all directories in the stage root, which could remove content not
related to Spack if the user chose a staging root which contains
files/directories not managed by Spack.
* The documentation is updated with warnings about choosing a stage
directory that is only managed by Spack (although generally the
check added in this PR for "spack clean" should avoid removing
content that was not created by Spack)
* The default stage directory (in config.yaml) is now
$tempdir/$user/spack-stage and the logic is updated to omit the
$user portion of this path if $tempdir already contains a $user
directory.
* When creating stage root assign user read/write permissions to all
directories in the path under $user. Previously Spack was assigning
the permissions of the first existing parent directory
`spec.prefix` reads from Spack's database, and if you do this with
multiple consecutive read transactions, it can take a long time. Or, at
least, you can see the paths get written out one by one.
This uses an outer read transaction to ensure that actual disk locks are
acquired only once for the whole `spack find` operation, and that each
transaction inside `spec.prefix` is an in-memory operation. This speeds
up `spack find -p` a lot.
Refactor `spack.cmd.display_specs()` and `spack find` so that any options
can be used together with -d. This cleans up the display logic
considerably, as there are no longer multiple "modes".
This is another machine-readable version of `spack find`. Supplying the
`--json` argument causes specs to be written out as json records,
easily filered with tools like jq.
e.g.:
$ spack find --json python | jq -C ".[] | { name, version } "
[
{
"name": "python",
"version": "2.7.16"
},
{
"name": "bzip2",
"version": "1.0.8"
}
]
- spack find --format allows you to supply a format string and have specs
output in a more machine-readable way, without dedcoration
e.g.:
spack find --format "{name}-{version}-{hash}"
autoconf-2.69-icynozk7ti6h4ezzgonqe6jgw5f3ulx4
automake-1.16.1-o5v3tc77kesgonxjbmeqlwfmb5qzj7zy
bzip2-1.0.6-syohzw57v2jfag5du2x4bowziw3m5p67
...
or:
spack find --format "{hash}"
icynozk7ti6h4ezzgonqe6jgw5f3ulx4
o5v3tc77kesgonxjbmeqlwfmb5qzj7zy
syohzw57v2jfag5du2x4bowziw3m5p67
...
This is intended to make it much easier to script with `spack find`
When Spack installs a package it writes the package.py file and
patches to a separate repository (which reflects the state of the
package at the time it was installed). Previously, Spack only wrote
patches that were used at installation time. This updates the
archiving step to include all patch files that are relevant to the
package (in case that repository is used in another context).
This commit removes redundant calls to `libtoolize` and `aclocal`.
Some configurations, such as a Spack user using macOS with a
Homebrew-installed `libtool` added to their `packages.yaml`, have
`autoreconf` and GNU libtoolize installed as `glibtoolize`, but not
`libtoolize`. While Spack installations of `libtool` built from source
would install `glibtoolize` and symlink `libtoolize` to `glibtoolize`,
an external installation of GNU libtoolize as `glibtoolize` will not
have such a symlink, and thus the call `m.libtoolize()` will throw an
error because `libtoolize` does not exist at the path referenced by
`m.libtoolize()` (i.e.,
`self.spec['libtool'].prefix.bin.join('libtoolize')).
However, on these same systems, `autoreconf` runs correctly, and calls
`glibtoolize` instead of `libtoolize`, when appropriate. Thus,
removing the call to `libtoolize` should resolve the error mentioned
above.
The redundant call to `aclocal` is also removed in this commit because
the maintainers of GNU Automake state that "`aclocal` is expected to
disappear" and suggest that downstream users never call `aclocal`
directly -- rather, they suggest calling `autoreconf` instead.
Uses code from CMake to detect implicit link paths from compilers
System paths are filtered out of implicit link paths
Implicit link paths added to compiler config and object under `implicit_rpaths`
Implicit link paths added as rpaths to compile line through env/cc wrapper
Authored by: "Ben Boeckel <ben.boeckel@kitware.com>"
Co-authored by: "Peter Scheibel <scheibel1@llnl.gov>"
Co-authored by: "Gregory Becker <becker33@llnl.gov>"
c9e214f updated template creation by passing **kwargs to package
template classes but the template classes were not updated to accept
them; this adds **kwargs to package template initializers where they
are needed.
