Using numba kernel caching for numba-dpex spirv kernel #815
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
Contributor
chudur-budur
commented
Nov 4, 2022
chudur-budur
commented
- Have you provided a meaningful PR description?
- WIP for Implementing a more robust caching mechanism #814
- 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?
- 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`.
added 9 commits
November 5, 2022 12:46
- 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`.
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
from
151b913 to
29b5c46
Compare
chudur-budur
commented
Nov 9, 2022
numba_dpex/caching.py
Outdated
| """ | ||
| Returns the unserialized CompileResult | ||
| """ | ||
| return compiler.CompileResult._rebuild(target_context, *payload) |
There was a problem hiding this comment.
@diptorupd Is this going to work with numba_dpex?
numba_dpex/caching.py
Outdated
| if not self._impl.check_cachable(data): | ||
| return | ||
| self._impl.locator.ensure_cache_path() | ||
| # key = self._index_key(sig, data.codegen) |
There was a problem hiding this comment.
@diptorupd This data is a compiled object from spirc_kernel.py (here). But the compiled kernel doesn't have codegen.
numba_dpex/caching.py
Outdated
| self._impl.locator.ensure_cache_path() | ||
| # key = self._index_key(sig, data.codegen) | ||
| key = self._index_key(sig) | ||
| # data = self._impl.reduce(data) |
There was a problem hiding this comment.
@diptorupd This reduce doesn't work on the compiled kernel either. I am getting this:
python driver.py
kernel: 2
-----> Dispatcher.__init__()
/localdisk/work/akmkhale/numba-dpex/driver.py:28: DeprecationWarning: The [] (__getitem__) method to set global and local ranges for launching a kernel is deprecated. Use the execute function instead.
data_parallel_sum[(100,)](a, b, c)
/localdisk/work/akmkhale/numba-dpex/numba_dpex/core/kernel_interface/dispatcher.py:334: UserWarning: Use of __getitem__ to set the global_range attribute is deprecated. Use the keyword argument "global_range" of __call__ method to set the attribute.
warn(
/localdisk/work/akmkhale/numba-dpex/numba_dpex/core/kernel_interface/dispatcher.py:359: UserWarning: Kernel to be submitted without a local range letting the SYCL runtime select a local range. The behavior can lead to suboptimal performance in certain cases. Consider setting the local range value for the kernel execution.
The local_range keyword may be made a required argument in the future.
warn(
-----> dispatcher.kernel_name: data_parallel_sum
-----> spriv_kernel._compile()
-----> caching.load_overload()
-----> spirv_kernel._compile().cres == None
Traceback (most recent call last):
File "/localdisk/work/akmkhale/numba-dpex/driver.py", line 36, in <module>
main()
File "/localdisk/work/akmkhale/numba-dpex/driver.py", line 28, in main
data_parallel_sum[(100,)](a, b, c)
File "/localdisk/work/akmkhale/numba-dpex/numba_dpex/core/kernel_interface/dispatcher.py", line 427, in __call__
kernel.compile(
File "/localdisk/work/akmkhale/numba-dpex/numba_dpex/core/kernel_interface/spirv_kernel.py", line 167, in compile
cres = self._compile(
File "/nfs/site/home/akmkhale/.conda/envs/numba-dpex/lib/python3.9/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
return func(*args, **kwargs)
File "/localdisk/work/akmkhale/numba-dpex/numba_dpex/core/kernel_interface/spirv_kernel.py", line 120, in _compile
self._cache.save_overload(cres.signature, cres)
File "/localdisk/work/akmkhale/numba-dpex/numba_dpex/caching.py", line 95, in save_overload
data = self._impl.reduce(data)
File "/localdisk/work/akmkhale/numba-dpex/numba_dpex/caching.py", line 28, in reduce
return cres._reduce()
File "/nfs/site/home/akmkhale/.conda/envs/numba-dpex/lib/python3.9/site-packages/numba/core/compiler.py", line 183, in _reduce
libdata = self.library.serialize_using_object_code()
File "/nfs/site/home/akmkhale/.conda/envs/numba-dpex/lib/python3.9/site-packages/numba/core/codegen.py", line 922, in serialize_using_object_code
data = (self._get_compiled_object(),
File "/nfs/site/home/akmkhale/.conda/envs/numba-dpex/lib/python3.9/site-packages/numba/core/codegen.py", line 630, in _get_compiled_object
raise RuntimeError("no compiled object yet for %s" % (self,))
RuntimeError: no compiled object yet for <Library 'data_parallel_sum' at 0x7f1f1c4fbcd0>
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
from
bf4dda5 to
b43201c
Compare
…Python/numba-dpex into refactor/kernel_interfaces
… SpirvKernel attributes in reduce()/rebuild() Moving caching.py into numba_dpex/core
|
@diptorupd The caching mechanism is correct now, it's using all of the numba machinery. I just need to add backend and device type into the key. |
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
from
f99df29 to
b43201c
Compare
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
from
b43201c to
7ee0f83
Compare
5 tasks
diptorupd
force-pushed
the
refactor/kernel_interfaces
branch
4 times, most recently
from
17676d2 to
b68a0a9
Compare
This was referenced Dec 8, 2022
|
superceded by #843 |
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.