It turns out std::linalg has a couple small wording issues.
#308
The one that practically matters for implementations is LWG 4302.
https://cplusplus.github.io/LWG/lwg-active.html#4302
The reference implementation of vector_sum_of_squares must be fixed.
- The default algorithm should not attempt rescaling. (Don't try to rescale integers.)
- Constrain the rescaling algorithm to floating-point numbers.
If LWG 4302 is accepted, that would also make it permissible to use the most recent (post-2021) xLASSQ algorithm, as the scaling factor would no longer be required to be the max absolute value.
It turns out std::linalg has a couple small wording issues.
#308
The one that practically matters for implementations is LWG 4302.
https://cplusplus.github.io/LWG/lwg-active.html#4302
The reference implementation of
vector_sum_of_squaresmust be fixed.If LWG 4302 is accepted, that would also make it permissible to use the most recent (post-2021) xLASSQ algorithm, as the scaling factor would no longer be required to be the max absolute value.