Update fannkuch.jl for ~61x speedup

Now based on Javascript. In general, the fastest js programs tend to be very C-like, and Julia maps C-like programs extremely fast, so I'm finding it's a good rule of thumb. Copy js and we come out really fast.

examples/shootout/fannkuch.jl

@@ -3,68 +3,74 @@
 # fannkuch-redux benchmark
 # https://shootout.alioth.debian.org/u32/performance.php?test=fannkuchredux
 #
-# Based on the Scala program
+# Based on the Javascript program
 #
-
-# flips the first a[1] elements of a, 
-# assuming a is a permutation of the set {1, ..., length(a)}
-flip(a) = [reverse(a[1:a[1]]), a[a[1]+1:end]]
-
 function fannkuch(n)
- perm = [1:n]
- counts = map(int, zeros(n))
- r = n+1
- flips = 0
- maxflips = 0
- nperm = 0
- checksum = 0
- while r > 1 
- while (r != 2)
- counts[r-1] = r-1
- r -= 1
- end
-
- # count flips
- flips = 0
- p = copy(perm)
- while (p[1] != 1)
- p = flip(p)
- flips += 1
- end
-
- if (flips > maxflips)
- maxflips = flips
- end
- if nperm % 2 == 0
- checksum += flips
- else
- checksum -= flips
- end
-
- go = true
- while go
- if r == n+1
- println(checksum)
- return maxflips
- end
- p0 = perm[1]
- for i = 1:r-1
- perm[i] = perm[i+1]
- end
- perm[r] = p0
- counts[r] -= 1
- if counts[r] > 0 
- go = false
- else
- r += 1
- end
- end
-
- nperm += 1
- end
- maxflips
+ p = Array(Int32,n)
+ for i = 1:n
+ p[i] = i
+ end
+ q = copy(p)
+ s = copy(p)
+ sign = 1; maxflips = sum = 0
+ while true
+ q0 = p[1]
+ if q0 != 1
+ for i = 2:n
+ q[i] = p[i]
+ end
+ flips = 1
+ while true
+ qq = q[q0] #??
+ if qq == 1
+ sum += sign*flips
+ flips > maxflips && (maxflips = flips)
+ break
+ end
+ q[q0] = q0
+ if q0 >= 4
+ i = 2; j = q0-1
+ while true
+ t = q[i]
+ q[i] = q[j]
+ q[j] = t
+ i += 1
+ j -= 1
+ i >= j && break
+ end
+ end
+ q0 = qq
+ flips += 1
+ end
+ end
+ #permute
+ if sign == 1
+ t = p[2]
+ p[2] = p[1]
+ p[1] = t
+ sign = -1
+ else
+ t = p[2]
+ p[2] = p[3]
+ p[3] = t
+ sign = 1
+ for i = 3:n
+ sx = s[i]
+ if sx != 1
+ s[i] = sx-1
+ break
+ end
+ i == n && return maxflips
+ s[i] = i
+ t = p[1]
+ for j = 1:i
+ p[j] = p[j+1]
+ end
+ p[i+1] = t
+ end
+ end
+ end
 end
-
 if length(ARGS) >= 1
  N = int(ARGS[1])
 else