//These are macros used to reduce on proc calls #define fetchElement(L, i) (associative) ? L[L[i]] : L[i] //Minimum sized sequence that will be merged. Anything smaller than this will use binary-insertion sort. //Should be a power of 2 #define MIN_MERGE 32 //When we get into galloping mode, we stay there until both runs win less often than MIN_GALLOP consecutive times. #define MIN_GALLOP 7 //This is a global instance to allow much of this code to be reused. The interfaces are kept separately var/datum/sortInstance/sortInstance = new() /datum/sortInstance //The array being sorted. var/list/L //The comparator proc-reference var/cmp = /proc/cmp_numeric_asc //whether we are sorting list keys (0: L[i]) or associated values (1: L[L[i]]) var/associative = 0 //This controls when we get *into* galloping mode. It is initialized to MIN_GALLOP. //The mergeLo and mergeHi methods nudge it higher for random data, and lower for highly structured data. var/minGallop = MIN_GALLOP //Stores information regarding runs yet to be merged. //Run i starts at runBase[i] and extends for runLen[i] elements. //runBase[i] + runLen[i] == runBase[i+1] //var/stackSize var/list/runBases = list() var/list/runLens = list() /datum/sortInstance/proc/timSort(start, end) runBases.Cut() runLens.Cut() var/remaining = end - start //If array is small, do a 'mini-TimSort' with no merges if(remaining < MIN_MERGE) var/initRunLen = countRunAndMakeAscending(start, end) binarySort(start, end, start+initRunLen) return //March over the array finding natural runs //Extend any short natural runs to runs of length minRun var/minRun = minRunLength(remaining) do //identify next run var/runLen = countRunAndMakeAscending(start, end) //if run is short, extend to min(minRun, remaining) if(runLen < minRun) var/force = (remaining <= minRun) ? remaining : minRun binarySort(start, start+force, start+runLen) runLen = force //add data about run to queue runBases.Add(start) runLens.Add(runLen) //maybe merge mergeCollapse() //Advance to find next run start += runLen remaining -= runLen while(remaining > 0) //Merge all remaining runs to complete sort //ASSERT(start == end) mergeForceCollapse(); //ASSERT(runBases.len == 1) //reset minGallop, for successive calls minGallop = MIN_GALLOP return L /* Sorts the specified portion of the specified array using a binary insertion sort. This is the best method for sorting small numbers of elements. It requires O(n log n) compares, but O(n^2) data movement (worst case). If the initial part of the specified range is already sorted, this method can take advantage of it: the method assumes that the elements in range [lo,start) are already sorted lo the index of the first element in the range to be sorted hi the index after the last element in the range to be sorted start the index of the first element in the range that is not already known to be sorted */ /datum/sortInstance/proc/binarySort(lo, hi, start) //ASSERT(lo <= start && start <= hi) if(start <= lo) start = lo + 1 for(,start < hi, ++start) var/pivot = fetchElement(L,start) //set left and right to the index where pivot belongs var/left = lo var/right = start //ASSERT(left <= right) //[lo, left) elements <= pivot < [right, start) elements //in other words, find where the pivot element should go using bisection search while(left < right) var/mid = (left + right) >> 1 //round((left+right)/2) if(call(cmp)(fetchElement(L,mid), pivot) > 0) right = mid else left = mid+1 //ASSERT(left == right) moveElement(L, start, left) //move pivot element to correct location in the sorted range /* Returns the length of the run beginning at the specified position and reverses the run if it is back-to-front A run is the longest ascending sequence with: a[lo] <= a[lo + 1] <= a[lo + 2] <= ... or the longest descending sequence with: a[lo] > a[lo + 1] > a[lo + 2] > ... For its intended use in a stable mergesort, the strictness of the definition of "descending" is needed so that the call can safely reverse a descending sequence without violating stability. */ /datum/sortInstance/proc/countRunAndMakeAscending(lo, hi) //ASSERT(lo < hi) var/runHi = lo + 1 if(runHi >= hi) return 1 var/last = fetchElement(L,lo) var/current = fetchElement(L,runHi++) if(call(cmp)(current, last) < 0) while(runHi < hi) last = current current = fetchElement(L,runHi) if(call(cmp)(current, last) >= 0) break ++runHi reverseRange(L, lo, runHi) else while(runHi < hi) last = current current = fetchElement(L,runHi) if(call(cmp)(current, last) < 0) break ++runHi return runHi - lo //Returns the minimum acceptable run length for an array of the specified length. //Natural runs shorter than this will be extended with binarySort /datum/sortInstance/proc/minRunLength(n) //ASSERT(n >= 0) var/r = 0 //becomes 1 if any bits are shifted off while(n >= MIN_MERGE) r |= (n & 1) n >>= 1 return n + r //Examines the stack of runs waiting to be merged and merges adjacent runs until the stack invariants are reestablished: // runLen[i-3] > runLen[i-2] + runLen[i-1] // runLen[i-2] > runLen[i-1] //This method is called each time a new run is pushed onto the stack. //So the invariants are guaranteed to hold for i= 2) var/n = runBases.len - 1 if(n > 1 && runLens[n-1] <= runLens[n] + runLens[n+1]) if(runLens[n-1] < runLens[n+1]) --n mergeAt(n) else if(runLens[n] <= runLens[n+1]) mergeAt(n) else break //Invariant is established //Merges all runs on the stack until only one remains. //Called only once, to finalise the sort /datum/sortInstance/proc/mergeForceCollapse() while(runBases.len >= 2) var/n = runBases.len - 1 if(n > 1 && runLens[n-1] < runLens[n+1]) --n mergeAt(n) //Merges the two consecutive runs at stack indices i and i+1 //Run i must be the penultimate or antepenultimate run on the stack //In other words, i must be equal to stackSize-2 or stackSize-3 /datum/sortInstance/proc/mergeAt(i) //ASSERT(runBases.len >= 2) //ASSERT(i >= 1) //ASSERT(i == runBases.len - 1 || i == runBases.len - 2) var/base1 = runBases[i] var/base2 = runBases[i+1] var/len1 = runLens[i] var/len2 = runLens[i+1] //ASSERT(len1 > 0 && len2 > 0) //ASSERT(base1 + len1 == base2) //Record the legth of the combined runs. If i is the 3rd last run now, also slide over the last run //(which isn't involved in this merge). The current run (i+1) goes away in any case. runLens[i] += runLens[i+1] runLens.Cut(i+1, i+2) runBases.Cut(i+1, i+2) //Find where the first element of run2 goes in run1. //Prior elements in run1 can be ignored (because they're already in place) var/k = gallopRight(fetchElement(L,base2), base1, len1, 0) //ASSERT(k >= 0) base1 += k len1 -= k if(len1 == 0) return //Find where the last element of run1 goes in run2. //Subsequent elements in run2 can be ignored (because they're already in place) len2 = gallopLeft(fetchElement(L,base1 + len1 - 1), base2, len2, len2-1) //ASSERT(len2 >= 0) if(len2 == 0) return //Merge remaining runs, using tmp array with min(len1, len2) elements if(len1 <= len2) mergeLo(base1, len1, base2, len2) else mergeHi(base1, len1, base2, len2) /* Locates the position to insert key within the specified sorted range If the range contains elements equal to key, this will return the index of the LEFTMOST of those elements key the element to be inserted into the sorted range base the index of the first element of the sorted range len the length of the sorted range, must be greater than 0 hint the offset from base at which to begin the search, such that 0 <= hint < len; i.e. base <= hint < base+hint Returns the index at which to insert element 'key' */ /datum/sortInstance/proc/gallopLeft(key, base, len, hint) //ASSERT(len > 0 && hint >= 0 && hint < len) var/lastOffset = 0 var/offset = 1 if(call(cmp)(key, fetchElement(L,base+hint)) > 0) var/maxOffset = len - hint while(offset < maxOffset && call(cmp)(key, fetchElement(L,base+hint+offset)) > 0) lastOffset = offset offset = (offset << 1) + 1 if(offset > maxOffset) offset = maxOffset lastOffset += hint offset += hint else var/maxOffset = hint + 1 while(offset < maxOffset && call(cmp)(key, fetchElement(L,base+hint-offset)) <= 0) lastOffset = offset offset = (offset << 1) + 1 if(offset > maxOffset) offset = maxOffset var/temp = lastOffset lastOffset = hint - offset offset = hint - temp //ASSERT(-1 <= lastOffset && lastOffset < offset && offset <= len) //Now L[base+lastOffset] < key <= L[base+offset], so key belongs somewhere to the right of lastOffset but no farther than //offset. Do a binary search with invariant L[base+lastOffset-1] < key <= L[base+offset] ++lastOffset while(lastOffset < offset) var/m = lastOffset + ((offset - lastOffset) >> 1) if(call(cmp)(key, fetchElement(L,base+m)) > 0) lastOffset = m + 1 else offset = m //ASSERT(lastOffset == offset) return offset /** * Like gallopLeft, except that if the range contains an element equal to * key, gallopRight returns the index after the rightmost equal element. * * @param key the key whose insertion point to search for * @param a the array in which to search * @param base the index of the first element in the range * @param len the length of the range; must be > 0 * @param hint the index at which to begin the search, 0 <= hint < n. * The closer hint is to the result, the faster this method will run. * @param c the comparator used to order the range, and to search * @return the int k, 0 <= k <= n such that a[b + k - 1] <= key < a[b + k] */ /datum/sortInstance/proc/gallopRight(key, base, len, hint) //ASSERT(len > 0 && hint >= 0 && hint < len) var/offset = 1 var/lastOffset = 0 if(call(cmp)(key, fetchElement(L,base+hint)) < 0) //key <= L[base+hint] var/maxOffset = hint + 1 //therefore we want to insert somewhere in the range [base,base+hint] = [base+,base+(hint+1)) while(offset < maxOffset && call(cmp)(key, fetchElement(L,base+hint-offset)) < 0) //we are iterating backwards lastOffset = offset offset = (offset << 1) + 1 //1 3 7 15 //if(offset <= 0) //int overflow, not an issue here since we are using floats // offset = maxOffset if(offset > maxOffset) offset = maxOffset var/temp = lastOffset lastOffset = hint - offset offset = hint - temp else //key > L[base+hint] var/maxOffset = len - hint //therefore we want to insert somewhere in the range (base+hint,base+len) = [base+hint+1, base+hint+(len-hint)) while(offset < maxOffset && call(cmp)(key, fetchElement(L,base+hint+offset)) >= 0) lastOffset = offset offset = (offset << 1) + 1 //if(offset <= 0) //int overflow, not an issue here since we are using floats // offset = maxOffset if(offset > maxOffset) offset = maxOffset lastOffset += hint offset += hint //ASSERT(-1 <= lastOffset && lastOffset < offset && offset <= len) ++lastOffset while(lastOffset < offset) var/m = lastOffset + ((offset - lastOffset) >> 1) if(call(cmp)(key, fetchElement(L,base+m)) < 0) //key <= L[base+m] offset = m else //key > L[base+m] lastOffset = m + 1 //ASSERT(lastOffset == offset) return offset //Merges two adjacent runs in-place in a stable fashion. //For performance this method should only be called when len1 <= len2! /datum/sortInstance/proc/mergeLo(base1, len1, base2, len2) //ASSERT(len1 > 0 && len2 > 0 && base1 + len1 == base2) var/cursor1 = base1 var/cursor2 = base2 //degenerate cases if(len2 == 1) moveElement(L, cursor2, cursor1) return if(len1 == 1) moveElement(L, cursor1, cursor2+len2) return //Move first element of second run moveElement(L, cursor2++, cursor1++) --len2 outer: while(1) var/count1 = 0 //# of times in a row that first run won var/count2 = 0 // " " " " " " second run won //do the straightfoward thin until one run starts winning consistently do //ASSERT(len1 > 1 && len2 > 0) if(call(cmp)(fetchElement(L,cursor2), fetchElement(L,cursor1)) < 0) moveElement(L, cursor2++, cursor1++) --len2 ++count2 count1 = 0 if(len2 == 0) break outer else ++cursor1 ++count1 count2 = 0 if(--len1 == 1) break outer while((count1 | count2) < minGallop) //one run is winning consistently so galloping may provide huge benifits //so try galloping, until such time as the run is no longer consistently winning do //ASSERT(len1 > 1 && len2 > 0) count1 = gallopRight(fetchElement(L,cursor2), cursor1, len1, 0) if(count1) cursor1 += count1 len1 -= count1 if(len1 <= 1) break outer moveElement(L, cursor2, cursor1) ++cursor2 ++cursor1 if(--len2 == 0) break outer count2 = gallopLeft(fetchElement(L,cursor1), cursor2, len2, 0) if(count2) moveRange(L, cursor2, cursor1, count2) cursor2 += count2 cursor1 += count2 len2 -= count2 if(len2 == 0) break outer ++cursor1 if(--len1 == 1) break outer --minGallop while((count1|count2) > MIN_GALLOP) if(minGallop < 0) minGallop = 0 minGallop += 2; // Penalize for leaving gallop mode if(len1 == 1) //ASSERT(len2 > 0) moveElement(L, cursor1, cursor2+len2) //else //ASSERT(len2 == 0) //ASSERT(len1 > 1) /datum/sortInstance/proc/mergeHi(base1, len1, base2, len2) //ASSERT(len1 > 0 && len2 > 0 && base1 + len1 == base2) var/cursor1 = base1 + len1 - 1 //start at end of sublists var/cursor2 = base2 + len2 - 1 //degenerate cases if(len2 == 1) moveElement(L, base2, base1) return if(len1 == 1) moveElement(L, base1, cursor2+1) return moveElement(L, cursor1--, cursor2-- + 1) --len1 outer: while(1) var/count1 = 0 //# of times in a row that first run won var/count2 = 0 // " " " " " " second run won //do the straightfoward thing until one run starts winning consistently do //ASSERT(len1 > 0 && len2 > 1) if(call(cmp)(fetchElement(L,cursor2), fetchElement(L,cursor1)) < 0) moveElement(L, cursor1--, cursor2-- + 1) --len1 ++count1 count2 = 0 if(len1 == 0) break outer else --cursor2 --len2 ++count2 count1 = 0 if(len2 == 1) break outer while((count1 | count2) < minGallop) //one run is winning consistently so galloping may provide huge benifits //so try galloping, until such time as the run is no longer consistently winning do //ASSERT(len1 > 0 && len2 > 1) count1 = len1 - gallopRight(fetchElement(L,cursor2), base1, len1, len1-1) //should cursor1 be base1? if(count1) cursor1 -= count1 moveRange(L, cursor1+1, cursor2+1, count1) //cursor1+1 == cursor2 by definition cursor2 -= count1 len1 -= count1 if(len1 == 0) break outer --cursor2 if(--len2 == 1) break outer count2 = len2 - gallopLeft(fetchElement(L,cursor1), cursor1+1, len2, len2-1) if(count2) cursor2 -= count2 len2 -= count2 if(len2 <= 1) break outer moveElement(L, cursor1--, cursor2-- + 1) --len1 if(len1 == 0) break outer --minGallop while((count1|count2) > MIN_GALLOP) if(minGallop < 0) minGallop = 0 minGallop += 2 // Penalize for leaving gallop mode if(len2 == 1) //ASSERT(len1 > 0) cursor1 -= len1 moveRange(L, cursor1+1, cursor2+1, len1) //else //ASSERT(len1 == 0) //ASSERT(len2 > 0) /datum/sortInstance/proc/mergeSort(start, end) var/remaining = end - start //If array is small, do an insertion sort if(remaining < MIN_MERGE) //var/initRunLen = countRunAndMakeAscending(start, end) binarySort(start, end, start/*+initRunLen*/) return var/minRun = minRunLength(remaining) do var/runLen = (remaining <= minRun) ? remaining : minRun binarySort(start, start+runLen, start) //add data about run to queue runBases.Add(start) runLens.Add(runLen) //Advance to find next run start += runLen remaining -= runLen while(remaining > 0) while(runBases.len >= 2) var/n = runBases.len - 1 if(n > 1 && runLens[n-1] <= runLens[n] + runLens[n+1]) if(runLens[n-1] < runLens[n+1]) --n mergeAt2(n) else if(runLens[n] <= runLens[n+1]) mergeAt2(n) else break //Invariant is established while(runBases.len >= 2) var/n = runBases.len - 1 if(n > 1 && runLens[n-1] < runLens[n+1]) --n mergeAt2(n) return L /datum/sortInstance/proc/mergeAt2(i) var/cursor1 = runBases[i] var/cursor2 = runBases[i+1] var/end1 = cursor1+runLens[i] var/end2 = cursor2+runLens[i+1] var/val1 = fetchElement(L,cursor1) var/val2 = fetchElement(L,cursor2) while(1) if(call(cmp)(val1,val2) < 0) if(++cursor1 >= end1) break val1 = fetchElement(L,cursor1) else moveElement(L,cursor2,cursor1) ++cursor2 if(++cursor2 >= end2) break ++end1 ++cursor1 //if(++cursor1 >= end1) // break val2 = fetchElement(L,cursor2) //Record the legth of the combined runs. If i is the 3rd last run now, also slide over the last run //(which isn't involved in this merge). The current run (i+1) goes away in any case. runLens[i] += runLens[i+1] runLens.Cut(i+1, i+2) runBases.Cut(i+1, i+2) #undef MIN_GALLOP #undef MIN_MERGE #undef fetchElement