Use different data models for DpnpNdArray Type for kernel and dpjit targets #1118
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 2 commits
August 20, 2023 00:25
- Previously, a dpnp.ndarray argument passed to a kernel function was
passed as an instance of numba_dpex.core.types.USMNdArray.
The USMNdArray type uses the numba_dpex.core.datamodel.models.USMArrayModel
data model where pointers have an explicit address space qualifier.
The changed code now registers the dpnp.ndarray-specific DpnpNdArray type
with a USMArrayModel for the DpexKernelTarget and with the
numba.core.datamodel.models.ArrayModel for the CPUTarget. By doing so, we now
correctly recognize a dpnp.ndarray passed to a kernel as such instead of a generic
USMNdArray object. The change unblocks the creation of a Numba-specific CPython
wrapper for kernel functions as we need to use the correct Numba type to
properly unbox and box a dpnp.ndarray.
- The dpnp.ndarray overload for getitem_arraynd_intp involves
updating the sycl_queue object when __getitem__ returns a
view. The existing implementation was trying to do so even
when __getitem__ returns a scalar.
Adds an extra check to avoid the problem.
diptorupd
force-pushed
the
fix/dpnp_data_model
branch
from
addf4dc to
12c5c23
Compare
github-actions bot
added a commit
that referenced
this pull request
Aug 20, 2023
Use different data models for DpnpNdArray Type for kernel and dpjit targets 2f006b9
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, a
dpnp.ndarrayargument passed to akernelfunction was passed as an instance ofnumba_dpex.core.types.USMNdArray. TheUSMNdArraytype uses thenumba_dpex.core.datamodel.models.USMArrayModeldata model where pointers have an explicit address space qualifier.The PR now registers the
dpnp.ndarray-specificDpnpNdArraytypes with aUSMArrayModelfor theDpexKernelTargetand with thenumba.core.datamodel.models.ArrayModelfor theCPUTarget. By doing so, we now correctly recognize adpnp.ndarraypassed to a kernel as such instead of a genericUSMNdArrayobject. The change unblocks the creation of a Numba-specific CPython wrapper for kernel functions as we need to use the correct Numba type to properly unbox and box adpnp.ndarray.Additionally, fixes a bug in
numba_dpex.dpnp_iface.arrayobj.getitem_arraynd_intpthat was found during testing.