I profiled this simple program
Eval [
| n k r |
r := Random new.
n := 100000.
k := 500.
(1 to: n) select: [:x |
(r between: 1 and: n) <= k
ifTrue: [ n := n - 1. k := k - 1] ifFalse: [n := n - 1];
yourself ]
]
and I noticed that it spends 75% of the time generating random numbers.
Bytecode-wise it's not bad (30 bytecodes per random number), but
floating point causes a large number of garbage collections. Anybody
wants to give a shot at reimplementing Random with a good, fast rng?
Remember that on 32-bit machines SmallIntegers are only from -2^30 to
2^30-1.
FWIW, here is a "more optimized" version of the above:
o := OrderedCollection new.
1 to: n do: [:x |
((r between: 0 and: (n := n - 1)) < k
ifTrue: [o add: x. k := k - 1]) notNil ]
Either program is a good benchmark for your rng.
Paolo
_______________________________________________
help-smalltalk mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/help-smalltalk
