Adding caching and specialization to dpex.func #869
Closed
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
- The compiler module only contains the compiler pipeline
to compiler SpirvKernel objects.
- Creates a separate module for the unpack and pack
functions for kernel arguments.
- The new API is intended for use from the Dispatcher class.
- The concept of a kernel was decoupled from the notion of
dispatching of a kernel. The present implementation in
compiler.py intermixes both things, making hard the
separation of compute-follows-data based kernel launch
and legacy `dpctl.device_context` based behavior.
- Deprecates support for numpy arrays as kernel args.
- Deprecates support for the square bracket notation using
`__getitem__` to provide global and local ranges for a
kernel launch.
- Changes the behavior of specializing a kernel using only
a signature. The new way to specialize will require a
device type and a backend.
- Improvements to exception messages using custom exceptions.
- The new API is now inside `numba_dpex.core.kernel_interface`.
Fix docs rst
- Kernel specialization for specific signatures was fixed. - Specialization is only allowed for usm_ndarray. - Dispacther class was renamed as JitKernel. - decorators updated.
- print a better type name with keywords and all attributes.
- store a fully qualified filter string in the type.
- set the default address space to GLOBAL instead of None.
- fix the unify function to test on usm_type, device, address
space.
- Added unit tests for the kernel specialization feature.
- Fix kernel decorator to support list of signatures.
- Disallow JIT compilation of a specialized kernel and raise
an exception.
- Move sycl_usm_array_interface helper functions into core/utils.
- Move array type conversion function for unknown type with
sycl_usm_array_interface to UsmNdArray into core/typeconv.
- Fix how the UsmNdArray type is constructed for unknow
sycl_usm_array_interface arrays.
…ion using strings.
- Move compile_with_dpex into the core.compiler module.
- The target and typing contexts are now not set from dpex_target
inside the SpirvKernel.compile. They are passed in as args.
- Fix tests impacted by these changes.
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
5 times, most recently
from
637a04d
to
c74ea24
Compare
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.
[WIP] The numba_dpex
funcdecorator is not cached nor does it properly support specialization. Addressing #861add caching to
dpex.funcadd specialization to
dpex.funcHave you provided a meaningful PR description?
Have you added a test, reproducer or referred to an issue with a reproducer?
Have you tested your changes locally for CPU and GPU devices?
Have you made sure that new changes do not introduce compiler warnings?
If this PR is a work in progress, are you filing the PR as a draft?