picrin/etc/R7RS/inputs/mperm.input

36 lines
1.5 KiB
Plaintext
Raw Normal View History

; The traditional parameters for this benchmark are 10:9:2:1,
; but that's too small for modern computers.
;
; The new parameters for this benchmark are 20:10:2:1.
; M: N:K:L
;
; N=10 means the benchmark starts by generating a list of all
; 10! = 3628800 permutations of the first 10 integers, allocating
; 13492889 pairs (a little over 100 megabytes on 32-bit machines
; with two-word pairs), all of which goes into the generated list.
; (That is, the first phase of the benchmark generates absolutely
; no garbage.) This represents a savings of about 63% over the
; storage that would be required by an unshared list of permuations.
; The generated permutations are in order of a gray code that bears
; no obvious relationship to a lexicographic order.
;
; Then M*(K-L) = 20*(2-1) = 20 more such lists are allocated.
;
; The live storage peaks at K=2 times the storage occupied by a
; single list of all N! permutations.
;
; At the end of each of the M=20 iterations, the oldest L/K = 1/2
; of the peak storage becomes garbage. Object lifetimes (measured
; in bytes or pairs allocated) are distributed uniformly between
; L/K times the peak storage and the peak storage itself.
20 ; M (number of iterations)
10 ; N (length of each permutation)
2 ; K (size of queue)
1 ; L (number of old copies removed when queue is filled)
; Note: the result below is ignored, since it can be
; computed from N above.
16329600 ; result (/ (* N (+ N 1) (factorial N)) 2)