Skip to content

Reuse reduction/dot buffer as sqrt output in linalg.norm #3062

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

Draft
antonwolfy wants to merge 2 commits into
base: master
Choose a base branch
from
+6 −4
Draft

Reuse reduction/dot buffer as sqrt output in linalg.norm #3062

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3d2ab4d
Reuse reduction/dot buffer as sqrt output in linalg.norm
antonwolfy Sep 8, 2026
ce5d3e4
Add changelog entry for linalg.norm sqrt out= reuse
antonwolfy Sep 8, 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 @@ -47,6 +47,7 @@ This release is compatible with NumPy 2.5.
* Linked the `dpnp_backend_c` library against only the MKL SYCL domains it uses (`BLAS`, `RNG`, `VM`) [#3012](https://github.com/IntelPython/dpnp/pull/3012)
* `dpnp` uses pybind11 3.1.0 [#3015](https://github.com/IntelPython/dpnp/pull/3015)
* Reworked the ASV benchmarks and added end-to-end workload benchmarks derived from dpBench [#2996](https://github.com/IntelPython/dpnp/pull/2996)
* Reduced allocations in `dpnp.linalg.norm` by reusing the reduction result as the `sqrt` output buffer in the 2-norm and Frobenius-norm branches [#3062](https://github.com/IntelPython/dpnp/pull/3062)

### Deprecated

Expand Down
9 changes: 5 additions & 4 deletions dpnp/linalg/dpnp_utils_linalg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1162,8 +1162,8 @@ def _norm_int_axis(x, ord, axis, keepdims):
return dpnp.abs(x).sum(axis=axis, keepdims=keepdims)
if ord is None or ord == 2:
# special case for speedup
s = (dpnp.conj(x) * x).real
return dpnp.sqrt(dpnp.sum(s, axis=axis, keepdims=keepdims))
s = dpnp.sum((dpnp.conj(x) * x).real, axis=axis, keepdims=keepdims)
return dpnp.sqrt(s, out=s)
if isinstance(ord, (int, float)):
absx = dpnp.abs(x)
absx **= ord
Expand Down Expand Up @@ -1215,7 +1215,8 @@ def _norm_tuple_axis(x, ord, row_axis, col_axis, keepdims):
row_axis -= 1
ret = dpnp.abs(x).sum(axis=col_axis).min(axis=row_axis)
elif ord in [None, "fro", "f"]:
ret = dpnp.sqrt(dpnp.sum((dpnp.conj(x) * x).real, axis=axis))
ret = dpnp.sum((dpnp.conj(x) * x).real, axis=axis)
ret = dpnp.sqrt(ret, out=ret)
elif ord == "nuc":
ret = _multi_svd_norm(x, row_axis, col_axis, dpnp.sum)
else:
Expand Down Expand Up @@ -2360,7 +2361,7 @@ def dpnp_norm(x, ord=None, axis=None, keepdims=False):
sqnorm = dpnp.dot(x_real, x_real) + dpnp.dot(x_imag, x_imag)
else:
sqnorm = dpnp.dot(x, x)
ret = dpnp.sqrt(sqnorm)
ret = dpnp.sqrt(sqnorm, out=sqnorm)
if keepdims:
ret = ret.reshape((1,) * ndim)
return ret
Expand Down
Loading