refactor: storage, layout, and devices (mostly cartesian) #1880
Labels
gt4py.cartesian
Issues concerning the current version with support only for cartesian grids.
gt4py.next
Issues concerning the new version with support for non-cartesian grids.
module: storage
Related to storage subpackage
priority: low
Low priority task
Description
PRs #1843 and #1867 raised a couple of questions regarding storage, layouts, and device support/handling on the cartesian side. This is a collector issue in which I'd like to group a couple small PRs and maybe have discussions around these refactors.
So far, we've identified a potential for improvement in the following areas:
DeviceType
to what is actually used/supported. PR refactor: stick to supported device types #1879CUPY_DEVICE
(currently defined separately innext
andcartesian
) in_core
storage
andcartesian
, see refactor[cartesian]: useDeviceType
from core_defn inLayoutInfo
#1867 (comment)GT4PY_USE_HIP
environment variable since auto-detection of NVIDIA/AMD GPUs seems to work fine, see refactor[cartesian]: useDeviceType
from core_defn inLayoutInfo
#1867 (comment)Layout
intoBackend
, see refactor[cartesian]: useDeviceType
from core_defn inLayoutInfo
#1867 (comment)Consider this as a low-priority wish list for refactors on the side.
The text was updated successfully, but these errors were encountered: