SDSU CS 662: Theory of Parallel Algorithms
Batcher's Sort

[To Course Home Page]
San Diego State University -- This page last updated March 27, 1995
----------

8*8 Odd-Even Merging Circuit

Translating Circuits to Code

Straighten out some wires
Makes the indexing easier




8        merge      6    regroup      6         merge     4         merge     1                       
   1                   1                   1                   1              2                       
      5                   4           4                   5                   3                       
         3                   3             3                   2              4                       
6                   8                 8                   6                   5                       
   2                   2                   2                   3              6                       
      4                   5           5                   8                   8                       
         9                   9             9                   9              9

Batcher's Merge-Exchange Sort


Input: N and array A[1:N] of items to sort

set T = ceiling(lg(N))
for P = 2**(T-1), 2**(T-2), 2**(T-3), ..., 1 do

	R = 0, D = P

	for Q = 2**(T-1), 2**(T-2), 2**(T-3), ..., P do

		for (K = 1 to N - D ) and  ((K-1) and P) = R do in parallel
			if A[K] > A[K + D] then 
				swap(A[K], A[K + D ])
			end if
		end for

		D = Q - P
		R = P
	end for
end for


(K + 1) and P means logical and of K and P

If number of processors is less than N than the swap becomes a merge