Skip to content

perf: skip numpy array creation in _compute_fwd_scale scalar path #384

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
antonwolfy merged 3 commits into from
Sep 30, 2026
Merged

perf: skip numpy array creation in _compute_fwd_scale scalar path #384

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
27f7037
perf: skip numpy array creation in _compute_fwd_scale scalar path
Sep 28, 2026
badf839
chore: improve
jharlow-intel Sep 28, 2026
e6ef92c
task: review
jharlow-intel Sep 29, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
* Multi-iterator constructors now return a status corresponding to allocation success or failure, and raise `MemoryError` instead of `ValueError` [gh-373](https://github.com/IntelPython/mkl_fft/pull/373)
* `_direct_fftnd` now also checks the status returned by the backend instead of discarding it [gh-373](https://github.com/IntelPython/mkl_fft/pull/373)
* Pinned Cython in the Coverity Scan workflow so generated code stays stable between scans, and added `coverity/README.md` documenting the known false-positive families and the scan review checklist [gh-374](https://github.com/IntelPython/mkl_fft/pull/374)
* Reduced Python overhead in `norm="forward"`/`"ortho"` scaling by computing the scale factor without `numpy.prod` [gh-384](https://github.com/IntelPython/mkl_fft/pull/384)

### Fixed
* Fixed `norm="forward"`/`"ortho"` scaling in `fftn`, `ifftn`, `rfftn`, `irfftn` and the `fft2` family when only a subset of axes is transformed: the scale used the full array shape instead of the transformed axes [gh-336](https://github.com/IntelPython/mkl_fft/issues/336), [gh-370](https://github.com/IntelPython/mkl_fft/pull/370)
Expand Down
12 changes: 11 additions & 1 deletion mkl_fft/_fft_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,8 @@
# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

import math

import numpy as np

# pylint: disable=no-name-in-module
Expand Down Expand Up @@ -70,7 +72,15 @@ def _compute_fwd_scale(norm, n, shape):
return 1.0

ss = n if n is not None else shape
nn = np.prod(ss)
# Avoid np.prod's array-creation overhead on the hot scalar (1-D) and
# sequence (N-D) paths; np.prod stays as the fallback for array-like
# `n`/`s` (e.g. np.array(8)) that math.prod can't handle.
if isinstance(ss, (int, np.integer)):
Comment thread
jharlow-intel marked this conversation as resolved.
nn = ss
elif isinstance(ss, (list, tuple)):
nn = math.prod(ss)
else:
nn = np.prod(ss)
fsc = 1 / nn if nn != 0 else 1
if norm == "forward":
return fsc
Expand Down
Loading