Closed vidraj closed 4 months ago
Thanks for spotting this. I'll take a closer look at the generated code and see if the compiler is still behaving. Also I think there are new instructions in POWER10 that might apply.
I verified that the comment/documentation changes are appropriate and should be committed. Basically repairing a late night pasto.
I verified that GCC is generating reasonable code for P8/P9/P10. The biggest issue to loading constants and P9/P10 targets will generated xxspltib as the best case.
Reviewed the new ops for P10 and found nothing that applies to isdigit/isalpha/isalnum. Most new ops are EOS handling, which should be a new and separate issue.
This correct and should be commited to master
Great!
The documentation comments for the two functions were swapped. It seems that the latency measurements were correct, though, as the more complex isalnum documents a longer latency, so these were kept as is.