Better rolling buffer strategies (#77)

Author: konsumlamm

Control/Parallel/Strategies.hs

@@ -628,68 +628,33 @@ parFmap strat f = (`using` parTraversable strat) . fmap f
 -- --------------------------------------------------------------------------
 -- Strategies for lazy lists
 
--- List-based non-compositional rolling buffer strategy, evaluating list
--- elements to weak head normal form.
--- Not to be exported; used in evalBuffer and for optimisation.
-evalBufferWHNF :: Int -> Strategy [a]
-evalBufferWHNF n0 xs0 = return (ret xs0 (start n0 xs0))
-  where -- ret :: [a] -> [a] -> [a]
-           ret (x:xs) (y:ys) = y `pseq` (x : ret xs ys)
-           ret xs     _      = xs
-
-        -- start :: Int -> [a] -> [a]
-           start 0   ys     = ys
-           start !_n []     = []
-           start !n  (y:ys) = y `pseq` start (n-1) ys
-
--- | 'evalBuffer' is a rolling buffer strategy combinator for (lazy) lists.
--- Evaluation of the ith element induces evaluation of the (i+n)th element,
--- with the first n elements being evaluated immediately.
---
--- 'evalBuffer' is not as compositional as the type suggests. In fact,
--- it evaluates list elements at least to weak head normal form,
--- disregarding a strategy argument 'r0'.
---
--- > evalBuffer n r0 == evalBuffer n rseq
---
+-- | 'evalBuffer' is a rolling buffer strategy combinator for lazy lists.
+-- Pattern matching on the result of @evalBuffer n strat xs@ will evaluate the
+-- first @n+1@ elements of @xs@ using @strat@. Pattern matching on each
+-- additional list cons will evaluate an additional element using @strat@.
 evalBuffer :: Int -> Strategy a -> Strategy [a]
-evalBuffer n strat =  evalBufferWHNF n . map (withStrategy strat)
-
--- Like evalBufferWHNF, but sparks the list elements when pushing them
--- into the buffer.
--- Not to be exported; used in parBuffer and for optimisation.
-parBufferWHNF :: Int -> Strategy [a]
-parBufferWHNF n0 xs0 = return (ret xs0 (start n0 xs0))
-  where -- ret :: [a] -> [a] -> [a]
-           ret (x:xs) (y:ys) = y `par` (x : ret xs ys)
-           ret xs     _      = xs
-
-        -- start :: Int -> [a] -> [a]
-           start 0   ys     = ys
-           start !_n []     = []
-           start !n  (y:ys) = y `par` start (n-1) ys
-
-
--- | Like 'evalBuffer', but evaluates the list elements in parallel when
--- pushing them into the buffer.
--- Evaluation of the ith element induces parallel evaluation of the (i+n)th element,
--- with the first n elements being evaluated in parallel immediately.
---
--- > parBuffer n strat = evalBuffer n (rparWith strat)
+evalBuffer n0 strat xs0 = return (ret tied (drop n0 tied))
+  where
+    -- This is the heart of the strategy. The idea is to tie the evaluation
+    -- of each cons (to WHNF) to the evaluation of its contents (according
+    -- to strat). Walking the spine of the result will thus perform
+    -- the requested Eval actions.
+    tied = foldr go [] xs0
+      where
+        go x r = runEval ((: r) <$> strat x)
+
+    ret (x : xs) (_y : ys) = x : ret xs ys
+    ret xs       _         = xs
+
+-- | 'parBuffer' is a rolling buffer strategy combinator for lazy lists.
+-- Pattern matching on the result of @parBuffer n s xs@ sparks
+-- computations to evaluate the first @n+1@ elements of @xs@ using the
+-- strategy @s@. Pattern matching on each additional list cons will
+-- spark an additional computation.
+--
+-- @parBuffer n strat = 'evalBuffer' n ('rparWith' strat)@
 parBuffer :: Int -> Strategy a -> Strategy [a]
-parBuffer n strat = parBufferWHNF n . map (withStrategy strat)
--- Alternative definition via evalBuffer (may compromise firing of RULES):
--- parBuffer n strat = evalBuffer n (rparWith strat)
-
--- Deforest the intermediate list in parBuffer/evalBuffer when it is
--- unnecessary:
-
-{-# NOINLINE [1] evalBuffer #-}
-{-# NOINLINE [1] parBuffer #-}
-{-# RULES
-"evalBuffer/rseq"  forall n . evalBuffer  n rseq = evalBufferWHNF n
-"parBuffer/rseq"   forall n . parBuffer   n rseq = parBufferWHNF  n
- #-}
+parBuffer n strat = evalBuffer n (rparWith strat)
 
 -- --------------------------------------------------------------------------
 -- Strategies for tuples