Hello Claus, Sunday, March 11, 2007, 10:03:59 PM, you wrote:
both the array and strict list versions avoid some intermediate structures; for the arbitrarily invented, relatively small inputs i've tried, strict lists are the clear winner, thanks to lower memory traffic, but i'd like some feedback from the experts:
-are there any obvious inefficiencies in the array code?
obviously, arrays version should create no temporary cells. the problems was mainly due to 2 factors: 1) readArray m (i,j) 2) 'op' in 'l' which was passed as real closure and was not inlined due to weakness of ghc optimizer also, we should help strictness analyzer by marking all the variables used in tight loops as strict. after that is done, we got 1000 times less temporary data allocated and 5x faster execution. now it's a bit faster than strict lists -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com