I re-ran the tests with your released 4.07 from GitHub, your 4.07 from this thread that checks for Escape every 10 evals, and a modified 4.07 where I changed the LDA so it checks every 255 evals. Tests run on a Master Turbo with MOS 3.50. I fixed up any negative values.I'll be interested to see the results on real hardware. The problem would affect all programs, and I'm sure a second 6502 was faster back in 1983!Results are not the same as a real system due to unemulated stuff, but the relative times should be about right. I'll give this a go on real hardware shortly.
No second processor:
4.07 - 30182 1267
4.07 0a - 21767 1265 (0.72x relative to 4.07)
4.07 ff - 20736 1265 (0.69x relative to 4.07)
Second processor (4 MHz 6502):
4.07 - 36343 592
4.07 0a - 12650 589 (0.35x relative to 4.07)
4.07 ff - 10082 590 (0.28x relative to 4.07)
A significant saving! - and with checking every 255 evals you're basically getting exactly the 2x speedup you'd expect when running in the 2nd processor.
Regarding checking less often: maybe there'd be problem cases where particular EVAL calls take a long time? But, running this specific test program a few times, I couldn't see any obvious delay when pressing ESCAPE even when checking every 255 evals.
--Tom
P.S. looking at my notes, I got "29588 1273" as output from Lisp 5.00 with MOS 3.50. So presumably 5.00 has some improvements somewhere (that might become even more effective combined with this fix?)
Statistics: Posted by tom_seddon — Tue Nov 19, 2024 11:21 pm