Thanks, Phil, for reminding me of this. I've committed a fix (0ad5aa8)
for that. There was a tail-recursive function in your example that
returned a tuple. Poly/ML puts these on the stack where it can but in
this case it was causing the function to cease to be tail-recursive,
eating up the stack. The reason for the regression was a change in the
way tuples on the stack are handled in the 32-in-64 bit version. Your
example now works as it used to.
Michael, it would be worth trying your code with the latest version to
see if this has fixed it. If it hasn't then presumably it's a different
problem and I'll have to investigate further.
David
On 16/06/2020 09:58, Phil Clayton wrote:
I also found that 5.8 can consume a huge amount of memory compared to
5.7.1. In my case, it was a very small example doing simple arithmetic
operations (demonstrating Fermat's method to find pairs of factors) so
it may be related. I have raised an issue for it:
https://github.com/polyml/polyml/issues/121
Phil
On 16/06/20 02:30, Norrish, Michael (Data61, Acton) wrote:
I have some HOL code that runs to completion in ~24 seconds user time.
On 5.7.1 it runs in about that much wall clock time too.
(Unfortunately, this test is running on a different machine,
necessarily, but this is an old Linux desktop which is far less
powerful than the new Macbook Pro which is running 5.8.)
On 5.8 (updated for MacOS Catalina to the SHA above), wall clock time
is 8 minutes.
The relevant code is the selftest in examples/miller/miller which is a
big arithmetic normalisation.
On the Mac, memory consumption (as reported by the Activity Monitor)
gets as high as 20GB; on the Linux machine running 5.7.1, the
consumption remains very slight (top reports 3% in its %MEM column;
the machine has 16GB)
Michael
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
_______________________________________________
polyml mailing list
polyml@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml
