No subject

[Henry Cejtin]

fast: loop 10^8 times doing one Int.+ and one Int.compare against 0.

slow: same as above but IntInf.+ and IntInf.compare.

fast2: loop 10^8 times doing one Int.+ and one Int.= against 0.

slow2: same as above but IntInf.+ and IntInf.=.

slower: same as above, but instead of [0, 10^8), use [2^32, 2^32 + 10^8).

So here is a comparison of smlc-1998-12-17 with the latest on your machine:

        program smlc-1998-12-17 newest
        fast     5.03            5.04
        slow    11.54           23.19
        fast2    2.00            2.03
        slow2    8.54           19.15
        slower  19.41           17.15

I  blame  the  slowdown (more than a factor of 2) entirely on the right-shift
thing.  Could you try turning it back into a >> (for words) just  to  see  if
this is really the case?

Note,  the  real  cases  are  fast2  vs.  slow2.  Here we see that in the old
system, IntInf's cost a factor of 4.27 and in the new one they are  a  factor
of 9.43.