In particular, CUDA 12 removes support for Kepler (3.x) entirely.
Unusually, however, it does not introduce any new architectures,
or even compute capabilities.
It turns out we don't generally need to proxy every compiler ever
through the top-level package. The number of times we directly poke at
one is negligible and direct imports are pretty clean.
Thanks to `ModuleInfo`, all modules are just named `foo.py` instead of
`unstable_foo.py`, which simplifies the import method a bit. This also
allows for accurate FeatureNew/FeatureDeprecated use, as we know when
the module was added and if/when it was stabilized.
We're going to do more with this in the next commit, but this just adds
the information for now. This allows the next commit have 100% mv
changes for some of the modules, which makes review easier
Instead of using FeatureNew/FeatureDeprecated in the module.
The goal here is to be able to handle information about modules in a
single place, instead of having to handle it separately. Each module
simply defines some metadata, and then the interpreter handles the rest.
All changes were created by running
"pyupgrade --py3-only"
and committing the results. Although this has been performed in the
past, newer versions of pyupgrade can automatically catch more
opportunities, notably list comprehensions can use generators instead,
in the following cases:
- unpacking into function arguments as function(*generator)
- unpacking into assignments of the form x, y = generator
- as the argument to some builtin functions such as min/max/sorted
Also catch a few creeping cases of new code added using older styles.
In commit 3340284805 the new ModuleObject
API got further updated to hide self.interpreter; at the time, the CUDA
module got transferred over to the wrapper which does provide it.
But it works fine without self.interpreter, so let's just use the best
base class.
Custom objects returned by modules must be subclass of ModuleObject and
have the state argument in its methods.
Add MutableModuleObject base class for objects that needs to be deep
copied on assignation.
Also, remove the possibility of passing in a compiler instance to
min_driver_version. This is because the NVCC compiler instance is,
as of CUDA Toolkit 11.0, no longer guaranteed to be versioned
identically to the toolkit itself.
D lang compilers have an option -release (or similar) which turns off
asserts, contracts, and other runtime type checking. This patch wires
that up to the b_ndebug flag.
Fixes#7082
David Seifert
Eli Schwartz
Olexa Bilaniuk
Alf Henrik Sauge
Dylan Baker
Daniel Mensinger
Xavier Claessens