Sync-up master and staging

See merge request !99

Accelerate CI

See merge request !96

generic tests in test_layout.c. It is now possible to test new layouts by providing a constructor.

See merge request !91

Compared to previous versions, tests on non dense layout cover more cases.
Layouts are filled with element coordinates. It becomes easy to check the result from `deref()`, `slice()` etc
by looking at the content of the layout.

Backend detection

See merge request !94

Add macros for assessing compiled backends. Add function to assess backends usability at runtime. Use this capability in area cuda test to skip tests when no cuda device is present.

[refactor/fix] remove extra cpitch in layout_dense

See merge request !92

Cummulative pitches were kept up to the total storage size of the
layout, even though our code never uses this information. Remove a bug
due to bad array size arithmetic, and remove the extra element in the
array too.

This patch also removes the pitches provided by the user from the
struct.

Add ECP-CI at Argonne (Theta)

See merge request !71

Add job step in `.ecp-ci.yml` plus the right variables to launch CI job
on theta. Not that Theta has mom nodes, so it's a bit different.

Make use of only: variable to limit job definition depending on mirror
being used within ECP.

ECP-CI mirror from NMC

See merge request !72

`.ecp-ci.yml` is the file where we will define job steps for the ECP-CI
runners.

This steps must only be defined for the protected branches.

This commit introduce 3 steps, once to log the environment being used,
and two checking CUDA support.

The previous CUDA activation code was trying to differentiate between
building CUDA support and actually running the tests, but this isn't
really necessary: any platform with CUDA support should be able to run
the tests somewhere.

The configure.ac is modified to also check for all necessary cuda
runtime and headers at once.

We also introduce support for CUDA_HOME, which is an environment
variable available on some target systems.

Linux area code and associated test was making assumptions about a NUMA
bitmask always being contiguous, which is not necessarily the case on
complex topologies.

[ci] create a readthedocs environment

See merge request !90

Better tracking of the issue.

Use gitlab environment to keep track of doc building on the gitlab side.
Allow for staging on the readthedocs trigger to allow devs to checkout
the docs outside of master.

Remove unused and unsafe dist rule.

[doc/fix] doxygen typo

See merge request !89

in/out is not valid doxygen.

Improve CI with respect to documentation

See merge request !88

Add a configure option for building docs, and create a CI job that will
always run to validate it. Note that make install-data doesn't need
source builds, so the environment can be quite small.

This patch also fixes the uncovered issues, and make doc build fail on
warnings.

Tweak the language in the documentation

See merge request !84

Padded Tilings

See merge request !87

The new padded tilings uncovered bugs in the operator table and the
implementation of padded layouts in row-major. We've seen these issues
before, we should be a bit more careful in the future.

Continue the work of merging the new tilings API, now with padded
tilings.

[ci] rely on git to list repoquality files

See merge request !85

Instead of tracking by hand the list of files that should pass
repoquality, just use git. This should ensure that we don't miss new
files, or forget to update the repoquality listing when new files are
added.

[ci/style] activate stylecheck on unit tests

See merge request !83

Quite a few changes needed.

Merge new version of tilings

See merge request !82

The new tilings are a massive change compared to the current
benchmarking code. It is time we disable them and reimplement most of
them using the new APIs. This will happen in another MR.

Also removes va_args from the public API.

Refactor the tilings to become generic to N dimensions, and interfacing
with the newly added layouts.

The main idea for this version of tilings is to provide an index into a
partitioning of a source layout into sub-layouts of smaller sizes.

Fix bugs triggered by inconsistent argument order in layouts, and the
wrong native function being called when in row-major.

[feature] add padded layouts

See merge request !80

Merge of padded layouts implementation from @videau, which provides a
way to extend a layout in any dimension by using a neutral element.

[feature] extend dma to support custom operators

See merge request !79

Some operators will need an extra argument (like transforms).