On Friday 15 October 2010 14:59:18, Warren Harris wrote:
I trying to learn a bit about data parallel haskell, and started from the wiki page here: http://www.haskell.org/haskellwiki/GHC/Data_Parallel_Haskell. Two questions:
The examples express the dot product as:
dotp_double xs ys = sumP [:x * <http://haskell.org/ghc/docs/latest/html/libraries/base/Prelude.html#v:.
y | x <- xs | y <- ys:]
Unless I'm missing something, shouldn't this actually be:
dotp_double xs ys = sumP [:x * <http://haskell.org/ghc/docs/latest/html/libraries/base/Prelude.html#v:.
y | x <- xs, y <- ys:]
No, it's supposed to be a parallel list comprehension, the dot product is sum $ zipWith (*) xs ys and the { blah x y | x <- xs | y <- ys } syntax (where {, } stand in for [, ] in parallel list comprehensions and for [:, :] in parallel array comprehensions) means { blah x y | (x,y) <- zip xs ys }
Second, when I run Main with the prescribed 10000 element array, everything seems to work quite nicely. The task takes about 2 seconds on my 4 processor x86_64, and threadscope shows all processors nicely utilized. However, when bumping this to 100000 elements, rather than taking 10x longer as I expected, the process never terminates. During one run I even lost control of my machine and needed to do a hard reset. Are there known limits to the array sizes that can be handled with dph, or can someone suggest what might be going wrong here? Thanks,
Warren