Hi all,

Some context on PR #29699: https://github.com/numpy/numpy/pull/29699

It integrates numpy-simd-routines
(https://github.com/numpy/numpy-simd-routines), a cross-platform SIMD
math library built on top of Google Highway — similar to Intel SVML
but supporting a wide range of architectures and independent from libm
to guarantee symmetric numeric behavior. It is still a WIP with more
kernels on the way.

Floating-point behavior is configurable at compile time per function.
Current defaults target ~1 ULP for double and ~4 ULP for single.
Build-time configuration will be exposed soon, and we may add limited
runtime control — likely just ULP tolerance — though that would
require extending NEP-54 or a new one.

Thanks!

--
Cheers,
Sayed

Sayed Awad
GitHub: seiko2plus
_______________________________________________
NumPy-Discussion mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3//lists/numpy-discussion.python.org
Member address: [email protected]

Reply via email to