* disable SVM on Nvidia CL when running pytest

* undo previous changes

* disable in pytest actx

* PytatoPyOpenCLArrayContext: don't trust the arg limit reported by the
GPU

Apparently, CUDA doesn't like it when argument sizes get close
to the reported limit.
Avoids PTX/JIT errors of the type
CUDA_ERROR_INVALID_IMAGE: device kernel image is invalid
CUDA_ERROR_FILE_NOT_FOUND: file not found

* restructure and clarify

* add warning, clarify comment

* add comment about device naming

* LazilyPyOpenCLCompilingFunctionCaller: limit arg size for GPUs

* move limit

* also check for SVM presence

* get_target()

* memoize get_target

* UNDO BEFORE MERGE: use dev branches

* Hackety hack: SVM detection in actx constructor

* check whether passed allocator supports SVM

* undo loopy branch

* implement it for the base class

* subclass LoopyPyOpenCLTarget

* set actual limit

* undo pytato branch

* remove unused argument

* add type annotations

* add logging

* Refactor arg size passing to put less logic in the target

* flake8

* add a test

Co-authored-by: Andreas Kloeckner <inform@tiker.net>

Andreas Klöckner authored 2 years ago and Andreas Klöckner committed 2 years ago

Co-authored-by: Michael Campbell <mtcampbe@illinois.edu>

* support empty containers in array_equal

* support empty containers in array_equal (jax)

Alexandru Fikl authored 2 years ago and Andreas Klöckner committed 2 years ago

* adds an is_available to check is the array context works
* do not create a factory per CL device, since that does not
quite work for non-CL array contexts