Reducing the priority of this request to reflect the lack of an actual POC and the age of the request without further follow-up. Requests with "live" submitters that can be interacted with to refine the idea and evaluate satisfaction with solution proposals get much farther in active development. Another alternative is to fix the code oneself and submit a working patch from which to start.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The main issue here is that the functions are generated from descriptions
but the current implementation does not appear to have support for multiple
names providing a single high-level PDL function or operator.
Since this functionality is Core and critical to PDL functioning, it will be
difficult to fix without significant risk to PDL correctness. Given that computation
is much cheaper than memory access on modern processors, a "hack around"
might be to implement with a pre-cast to double and a post-conversion back
to float. Of course you then have to handle out-of-range errors and more.
--Chris
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
This actually works already: get_convertedpdl creates a data-flowing conversion to eg a "double" ndarray that will the data-flow changes back afterwards on destruction.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Reducing the priority of this request to reflect the lack of an actual POC and the age of the request without further follow-up. Requests with "live" submitters that can be interacted with to refine the idea and evaluate satisfaction with solution proposals get much farther in active development. Another alternative is to fix the code oneself and submit a working patch from which to start.
The main issue here is that the functions are generated from descriptions
but the current implementation does not appear to have support for multiple
names providing a single high-level PDL function or operator.
Since this functionality is Core and critical to PDL functioning, it will be
difficult to fix without significant risk to PDL correctness. Given that computation
is much cheaper than memory access on modern processors, a "hack around"
might be to implement with a pre-cast to double and a post-conversion back
to float. Of course you then have to handle out-of-range errors and more.
--Chris
This actually works already: get_convertedpdl creates a data-flowing conversion to eg a "double" ndarray that will the data-flow changes back afterwards on destruction.