mirror of
https://github.com/fulpstation/fulpstation.git
synced 2025-12-10 01:57:01 +00:00
05b76b123e
RESULTS:
sorting 10 random lists of length 3 to 303 in increments of 3
(Meh, I forgot to refresh this one, there were only 338 trials rather than 1010, can't be bothered to recode the test)
Profile results (total time)
Proc Name Self CPU Total CPU Real Time Calls
------------------------------------------------- --------- --------- --------- ---------
/proc/sortList 0.672 16.141 16.243 171226 <--TG's current mergesort(recursive, hence the higher number of calls)
/proc/sortTim 0.008 3.278 3.274 338 <--TimSort
/proc/sortMerge 0.011 2.839 2.855 338 <--new mergesort
/proc/sortInsert 0.010 2.124 2.103 338 <--binary insertion
Sorting 10 presorted lists with 3 inversions (3 elements shuffled up), Lists of length 3 to 303 (increments of 3)
Profile results (total time)
Proc Name Self CPU Total CPU Real Time Calls
------------------------------------------------- --------- --------- --------- ---------
/proc/sortList 1.290 23.056 23.254 308050 <--rather cataclysmic
/proc/sortMerge 0.015 4.077 4.068 1010 <--
/proc/sortInsert 2.639 3.472 3.464 1010 <--
/proc/sortTim 0.014 1.567 1.576 1010 <--TimSort is faaar more effective in these cases,
Timsort can exploit runs effectively
sorting 10 presorted lists which have been reversed
Profile results (total time)
Proc Name Self CPU Total CPU Real Time Calls
------------------------------------------------- --------- --------- --------- ---------
/proc/sortList 1.234 23.193 23.295 308050
/proc/sortMerge 0.023 4.681 4.686 1010
/proc/sortInsert 2.875 3.750 3.765 1010
/proc/sortTim 0.020 3.294 3.284 1010 //This can be lower by using a different comparison method
*Corrected: /proc/sortTim 0.017 0.665 0.663 1010 //Using a non-strictly ascending comparison
sorting 10 presorted lists
Profile results (total time)
Proc Name Self CPU Total CPU Real Time Calls
------------------------------------------------- --------- --------- --------- ---------
/proc/sortList 1.199 21.391 21.517 308050
/proc/sortMerge 0.018 3.724 3.729 1010
/proc/sortInsert 2.497 3.302 3.309 1010
/proc/sortTim 0.024 0.586 0.584 1010
Summary, all the new procs are faster than the old ones. TimSort is ever so slightly slower than Insertion and Merging on random lists. But on lists with natural runs (partially sorted data) it is far faster than all others.
The old merge sort was removed and replaced with timSort. Other algorithms are provided as alternatives.
All algorithms use a central datum, so accept many of the same parameters. For instance, setting associative=1 will make them sort associative lists by their associated values, rather than keys.
They also accept a cmp argument. This allows sorting of lists of datums, text, numbers or whatever. The pre-existing helpers in lists.dm were rewritten as examples.
16 lines
475 B
Plaintext
16 lines
475 B
Plaintext
//merge-sort - gernerally faster than insert sort, for runs of 7 or larger
|
|
/proc/sortMerge(list/L, cmp=/proc/cmp_numeric_asc, associative, fromIndex=1, toIndex)
|
|
if(L && L.len >= 2)
|
|
fromIndex = fromIndex % L.len
|
|
toIndex = toIndex % (L.len+1)
|
|
if(fromIndex <= 0)
|
|
fromIndex += L.len
|
|
if(toIndex <= 0)
|
|
toIndex += L.len + 1
|
|
|
|
sortInstance.L = L
|
|
sortInstance.cmp = cmp
|
|
sortInstance.associative = associative
|
|
sortInstance.mergeSort(fromIndex, toIndex)
|
|
|
|
return L |