On 16/01/2022 12:51, David Matthews wrote:
The long-promised update to the ARM code-generator is now in Git master. This
builds on the old version by adding the register allocation strategy borrowed
from the X86 version as well as low-level peephole optimisation. On my Apple
M1 processor it now seems to be faster than X86 code with Rosetta.
There are still some improvements to be made, particularly with floating
point, but no major changes are anticipated. It includes one optimisation
that isn't present on the X86 at the moment: small tuples are returned in
registers rather than on the stack.
This is really great.
I have now started to make some tests with Isabelle + AFP, and will come back
eventually with reports (and potential problems).
Makarius
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
