tensor.zeros to use async memset #1806
Merged
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
This is akin to _full_usm_ndarray, but does not take fill_value, hence does not require castings. It dispatches straight to handler::memset.
|
Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞 |
|
Array API standard conformance tests for dpctl=0.18.0dev0=py310hdf72452_344 ran successfully. |
|
@oleksandr-pavlyk |
Bitwise zero values, and 1-byte wide types now use memset, instead of using fill. ``` In [1]: import dpctl.tensor as dpt, dpctl.tensor._tensor_impl as ti In [2]: res = dpt.empty(10**6, dtype="i8") In [3]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(0, dst=res, sycl_queue=res.sycl_queue)[0].wait() 243 µs ± 22.6 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [4]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(0, dst=res, sycl_queue=res.sycl_queue)[0].wait() 229 µs ± 14 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [5]: %timeit -n 2000 -r 11 ti._zeros_usm_ndarray(dst=res, sycl_queue=res.sycl_queue)[0].wait() 227 µs ± 23 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [6]: %timeit -n 2000 -r 11 ti._zeros_usm_ndarray(dst=res, sycl_queue=res.sycl_queue)[0].wait() 233 µs ± 25.9 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [7]: %timeit -n 2000 -r 11 ti._zeros_usm_ndarray(dst=res, sycl_queue=res.sycl_queue)[0].wait() 301 µs ± 54.1 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [8]: %timeit -n 2000 -r 11 ti._zeros_usm_ndarray(dst=res, sycl_queue=res.sycl_queue)[0].wait() 236 µs ± 17.2 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [9]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(0, dst=res, sycl_queue=res.sycl_queue)[0].wait() 240 µs ± 35.2 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [10]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(1, dst=res, sycl_queue=res.sycl_queue)[0].wait() 243 µs ± 17.6 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [11]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(1, dst=res, sycl_queue=res.sycl_queue)[0].wait() 263 µs ± 39.9 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [12]: %timeit -n 2000 -r 11 ti._full_usm_ndarray(0, dst=res, sycl_queue=res.sycl_queue)[0].wait() 239 µs ± 26.4 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) In [13]: %timeit -n 2000 -r 11 ti._zeros_usm_ndarray(dst=res, sycl_queue=res.sycl_queue)[0].wait() 224 µs ± 18.1 µs per loop (mean ± std. dev. of 11 runs, 2,000 loops each) ```
oleksandr-pavlyk
force-pushed
the
fixme-async-memset
branch
from
da954f9 to
f0d926a
Compare
|
Array API standard conformance tests for dpctl=0.18.0dev0=py310hdf72452_345 ran successfully. |
ndgrigorian
reviewed
Aug 22, 2024
ndgrigorian
reviewed
Aug 22, 2024
Co-authored-by: ndgrigorian <[email protected]>
ndgrigorian
approved these changes
Aug 22, 2024
|
Array API standard conformance tests for dpctl=0.18.0dev0=py310hdf72452_346 ran successfully. |
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.
Addressed outstanding FIXME note to use async call to populate array with zeros.
Added optimization for
_full_usm_ndarrayto usehandler::memsetinstead ofhandler::fillfor 1-byte wide types and for other types when fill value is bitwise zero.