QuickSortExploiting the Principle of Exchanging Keys
' QuickSort. QuickSort, CombSort and ShellSort all exploit the principle of
' exchanging keys that are far apart in the list rather than adjacent.
' QuickSort does this most elegantly and rapidly. The approach is to choose a
' "pivot" value (ideally, the median key) and then to work from each end of the
' list toward the middle. A key at each end is compared to the pivot and
' nothing is done if the left key is less than the pivot or the right key is
' greater. When a left key greater than the pivot and a right key less than
' the pivot have been found, those keys (or their pointers) are swapped,
' and the process continues until the left and right pointers cross. We then
' recursively call QuickSort on the left and right sublists until the lists are
' small (and delegate final sorting to low overhead InsertionSort).
'
' QuickSort does not need any auxiliary arrays, but uses a modest amount of
' stack space for recursion. It is not stable (although its descendent Ternary
' QuickSort is). On average, it is the fastest of the O(N log N) sorts,
' but it suffers from rare "worst case" behavior where certain input orders of
' keys cause speed to deteriorate to O(N^2). Naive implementations of
' QuickSort that choose the middle key for pivot exhibit O(N^2) behavior on
' sorted lists. The version of QuickSort presented here makes worst case
' behavior very unlikely by choosing the median of the first,
' last and middle keys as pivot. Two versions are provided. pQuickSortS is
' set up for strings and can be adapted to doubles by changing the declaration
' of array A(). QuickSortL is set up for longs, or A() can be redeclared for
' integers.
'
' Reference: Robert Sedgewick, "Implementing Quicksort Programs",
' Comm. of the ACM 21(10):847-857 (1978).
'
' Speed: pQuickSortS sorts 500,000 random strings in 30.3 sec; sorts 100186
' library call numbers in 11.3 sec; sorts 25479 dictionary words in 2.0 sec
' (random order), 1.3 sec (presorted) or 1.8 sec (reverse sorted). QuickSortL
' sorts 500,000 random longs in 56 seconds. Timed in Excel 2001 on an 800 mhz
' PowerBook.
'
' Bottom line: contends with RadixSort for fastest; better adapted than Radix
' for non-string data, but not stable.
' Usage:
Dim S1(L To R) As String
Dim P1(L To R) As Long
Dim L1(L To R) As Long
For I = L To R
S1(I) = GetRandomString()
P1(I) = I
L1(I) = GetRandomLong()
Next I
pQuickSortS L, R, S1, P1
QuickSortL L, R, L1
' CODE:
Sub pQuickSortS(L As Long, R As Long, A() As String, P() As Long)
'We put "sentinel" values flanking the real keys to avoid an extra test in
' the inner loop.
A(L - 1) = MinStr
A(R + 1) = MaxStr
'We mostly sort the list with QuickSort.
pQuickS L, R, A(), P
'Then we finish up with low overhead InsertionSort
pInsertS L, R, A(), P
End Sub
Sub pQuickS(L As Long, R As Long, A() As String, P() As Long)
Dim MED As Long
Dim LP As Long
Dim RP As Long
Dim Pivot As String
Dim TMP As Long
'Sublists <= 12 keys will be finished by running the whole list once thru
' InsertionSort.
If R - L > 12 Then
'Get the median pointer...
MED = (L + R) \ 2
'and swap it to the leftmost position.
TMP = P(MED)
P(MED) = P(L)
P(L) = TMP
'Now compare the leftmost, next leftmost & rightmost to choose a median of
' 3...
If A(P(L + 1)) > A(P(R)) Then
TMP = P(L + 1)
P(L + 1) = P(R)
P(R) = TMP
End If
If A(P(L)) > A(P(R)) Then
TMP = P(L)
P(L) = P(R)
P(R) = TMP
End If
If A(P(L + 1)) > A(P(L)) Then
TMP = P(L + 1)
P(L + 1) = P(L)
P(L) = TMP
End If
'and use its key as our pivot.
Pivot = A(P(L))
'Now work inward from each end.
LP = L
RP = R + 1
Do
'Scan right for a pointer whose key >= Pivot. In case Pivot is the
' largest key, we have
'a sentinel value of MaxStr in A(R + 1) that will end a runaway loop.
' Using the sentinel
'avoids having a second test in the inner loop,
' so it can be as fast as possible.
Do
LP = LP + 1
Loop While A(P(LP)) < Pivot
'Scan left for a pointer whose key <= Pivot. Again,
' we have a sentinel value of MinStr
'in A(L - 1) to stop the loop if Pivot is the smallest value in the
' list.
Do
RP = RP - 1
Loop While A(P(RP)) > Pivot
'If the pointers have crossed we're done.
If RP <= LP Then Exit Do
'Otherwise, swap the pair we've identified.
TMP = P(LP)
P(LP) = P(RP)
P(RP) = TMP
Loop
'Swap the pointer of the Pivot value back into place.
TMP = P(L)
P(L) = P(RP)
P(RP) = TMP
'Sort the shorter sublist first so the recursion stack is limited to
' logarithmic depth.
If (RP - 1) - L <= R - LP Then
pQuickS L, RP - 1, A, P
pQuickS LP, R, A, P
Else
pQuickS LP, R, A, P
pQuickS L, RP - 1, A, P
End If
End If
End Sub
Sub pInsertS(L As Long, R As Long, A() As String, P() As Long)
Dim LP As Long
Dim RP As Long
Dim TMP As Long
Dim T As String
For RP = L + 1 To R
TMP = P(RP)
T = A(TMP)
For LP = RP To L + 1 Step -1
If T < A(P(LP - 1)) Then P(LP) = P(LP - 1) Else Exit For
Next LP
P(LP) = TMP
Next RP
End Sub
Sub QuickSortL(L As Long, R As Long, A() As Long)
A(L - 1) = MinStr
A(R + 1) = MaxStr
QuickL L, R, A
InsertL L, R, A
End Sub
Sub QuickL(L As Long, R As Long, A() As Long)
Dim MED As Long
Dim LP As Long
Dim RP As Long
Dim Pivot As String
Dim TMP As Long
If R - L > 12 Then
MED = (L + R) \ 2
TMP = A(MED)
A(MED) = A(L)
A(L) = TMP
If A(L + 1) > A(R) Then
TMP = A(L + 1)
A(L + 1) = A(R)
A(R) = TMP
End If
If A(L) > A(R) Then
TMP = A(L)
A(L) = A(R)
A(R) = TMP
End If
If A(L + 1) > A(L) Then
TMP = A(L + 1)
A(L + 1) = A(L)
A(L) = TMP
End If
Pivot = A(L)
LP = L
RP = R + 1
Do
Do
LP = LP + 1
Loop While A(LP) < Pivot
Do
RP = RP - 1
Loop While A(RP) > Pivot
If RP <= LP Then Exit Do
TMP = A(LP)
A(LP) = A(RP)
A(RP) = TMP
Loop
TMP = A(L)
A(L) = A(RP)
A(RP) = TMP
If (RP - 1) - L < R - LP Then
QuickL L, RP - 1, A
QuickL LP, R, A
Else
QuickL LP, R, A
QuickL L, RP - 1, A
End If
End If
End Sub
Sub InsertL(L As Long, R As Long, A() As Long)
Dim LP As Long
Dim RP As Long
Dim TMP As Long
For RP = L + 1 To R
TMP = A(RP)
For LP = RP To L + 1 Step -1
If TMP < A(LP - 1) Then A(LP) = A(LP - 1) Else Exit For
Next LP
A(LP) = TMP
Next RP
End Sub