Instead of requiring setting up a mock CI environment with the official
image locally. While such docs would be handy anyways, requiring that
entire setup to get a small piece of information from a build is
unnecessary.

Useful to avoid setting up the CI locally to get a single extra message.

tejas authored 2 years ago and Brad King committed 2 years ago

- With this change we can use e.g. ImageMagick::Magick++ directly
  in targt_link_libraries.
- This change also adds CFLAGS which was missing before.
- Also adds example on how to use the targets.

According to NVIDIA's container image support policy [1], they will
remove tags for older images over time.  In order to preserve our
ability to rebuild CI images based on them in the future, base them on
tags we control.

[1] https://gitlab.com/nvidia/container-images/cuda/blob/85fbd45b92/doc/support-policy.md

These patches now support the `-MF` output, so remove the `none` support
added just for the old patchset which did not use it.

Also update the flag name to `-fmodule-output=`.

Due to the new Clang module mapper flag, use a new experimental support
UUID as well.

Robert Maynard authored 2 years ago and Brad King committed 2 years ago

Some package managers provide a minimal set of static libraries,
which historically has caused problems for FindCUDAToolkit.
Resolve this issue by creating a CUDA image with only the minimal
set of static libraries.

All our CI hosts now have Linux kernel version 3.17 or higher,
so we no longer need a dedicated tag.

This backports commit c212c7dc (gitlab-ci: drop unnecessary linux
kernel version tag, 2022-11-30) to the 3.25 release branch.

The latter tags are now available on our CI runners to more explicitly
distinguish `linux-x86_64` from `linux-aarch64` and to distinguish
`{macos,windows}-x86_64` from `{macos,windows}-arm64`.

This backports commit 33fe37a6 (gitlab-ci: replace '{macos,windows}'
tags with '{macos,windows}-x86_64', 2022-11-30) and commit 94d82d91
(gitlab-ci: replace 'linux' tags with 'linux-x86_64', 2022-11-30) to
the 3.25 release branch.

Fixes: #24225

Kyle Edwards authored 2 years ago and Brad King committed 2 years ago

And use sccache and Ninja.

Kyle Edwards authored 2 years ago and Brad King committed 2 years ago

The pre-build step will allow actions to be executed before the
CMake build but after Ninja and sccache have been downloaded, so
they can be used in the pre-build step. The env step is meant for
setting up environment variables anyway, rather than performing
actions.

The post-build step will allow extra artifacts to be computed after
the build is complete.

The job added by commit 16a5c367 (gitlab-ci: add nightly job testing
NVHPC with Ninja on Linux, 2022-11-21) started failing after CI hosts
were updated from NVIDIA driver version 520 to version 525.  Somehow
this causes the NVHPC multi-cuda `nvcc` wrapper to select CUDA 10.2
instead of 11.7 as with older driver versions.

As a workaround, add the preferred version of `nvcc` to the PATH.

Issue: #24225

Visual Studio does not currently have a place to insert the logic
required to implement this test, so skip it if possible.

These were missed by commit 94d82d91 (gitlab-ci: replace 'linux'
tags with 'linux-x86_64', 2022-11-30).

Robert Maynard authored 2 years ago and Brad King committed 2 years ago

Fixes: #24188

All our CI hosts now have Linux kernel version 3.17 or higher,
so we no longer need a dedicated tag.

The latter tags are now available on our CI runners to more explicitly
distinguish `{macos,windows}-x86_64` from `{macos,windows}-arm64`.

The latter tag is now available on our CI runners to more explicitly
distinguish `linux-x86_64` from `linux-aarch64`.

This groups their definitions nicely, and makes their purpose more
clear from reference sites.

The '_builder' part of tag entry names is not useful with our current
naming conventions.

Instead, just set the variables for how scanning works since that is
part of the compiler mechanisms.

Fixes: #24198

We already have `t:debian10-{makefiles,ninja}-clang` jobs covering old
versions of Clang with Makefile and Ninja generators.  Complement them
with new `t:fedora37-{makefiles,ninja}-clang` jobs covering recent
versions of Clang.

Fedora 37 now packages LLVMFlang, so enable Fortran tests with that
compiler too.

Issue: #22387

Also explicitly install `clang`.

Issue: #22387

Since Apple does not distribute the OpenMP header and library
with its AppleClang compiler, download them as part of the jobs.

Visual Studio 17.4 now contains official support for what CMake needs.

Add Windows ARM64 native CI jobs to build and test with Ninja,
and to test with VS.

Issue: #23624