3 -----------------------------------------------------------------------------
6 -- Copyright : (c) 2001-2002 Manuel M T Chakravarty & Gabriele Keller
7 -- License : see libraries/base/LICENSE
9 -- Maintainer : Manuel M. T. Chakravarty <chak@cse.unsw.edu.au>
10 -- Stability : internal
11 -- Portability : non-portable (GHC Extensions)
13 -- Basic implementation of Parallel Arrays.
15 -- This module has two functions: (1) It defines the interface to the
16 -- parallel array extension of the Prelude and (2) it provides a vanilla
17 -- implementation of parallel arrays that does not require to flatten the
18 -- array code. The implementation is not very optimised.
20 --- DOCU ----------------------------------------------------------------------
22 -- Language: Haskell 98 plus unboxed values and parallel arrays
24 -- The semantic difference between standard Haskell arrays (aka "lazy
25 -- arrays") and parallel arrays (aka "strict arrays") is that the evaluation
26 -- of two different elements of a lazy array is independent, whereas in a
27 -- strict array either non or all elements are evaluated. In other words,
28 -- when a parallel array is evaluated to WHNF, all its elements will be
29 -- evaluated to WHNF. The name parallel array indicates that all array
30 -- elements may, in general, be evaluated to WHNF in parallel without any
31 -- need to resort to speculative evaluation. This parallel evaluation
32 -- semantics is also beneficial in the sequential case, as it facilitates
33 -- loop-based array processing as known from classic array-based languages,
36 -- The interface of this module is essentially a variant of the list
37 -- component of the Prelude, but also includes some functions (such as
38 -- permutations) that are not provided for lists. The following list
39 -- operations are not supported on parallel arrays, as they would require the
40 -- availability of infinite parallel arrays: `iterate', `repeat', and `cycle'.
42 -- The current implementation is quite simple and entirely based on boxed
43 -- arrays. One disadvantage of boxed arrays is that they require to
44 -- immediately initialise all newly allocated arrays with an error thunk to
45 -- keep the garbage collector happy, even if it is guaranteed that the array
46 -- is fully initialised with different values before passing over the
47 -- user-visible interface boundary. Currently, no effort is made to use
48 -- raw memory copy operations to speed things up.
50 --- TODO ----------------------------------------------------------------------
52 -- * We probably want a standard library `PArray' in addition to the prelude
53 -- extension in the same way as the standard library `List' complements the
54 -- list functions from the prelude.
56 -- * Currently, functions that emphasis the constructor-based definition of
57 -- lists (such as, head, last, tail, and init) are not supported.
59 -- Is it worthwhile to support the string processing functions lines,
60 -- words, unlines, and unwords? (Currently, they are not implemented.)
62 -- It can, however, be argued that it would be worthwhile to include them
63 -- for completeness' sake; maybe only in the standard library `PArray'.
65 -- * Prescans are often more useful for array programming than scans. Shall
66 -- we include them into the Prelude or the library?
68 -- * Due to the use of the iterator `loop', we could define some fusion rules
71 -- * We might want to add bounds checks that can be deactivated.
77 mapP, -- :: (a -> b) -> [:a:] -> [:b:]
78 (+:+), -- :: [:a:] -> [:a:] -> [:a:]
79 filterP, -- :: (a -> Bool) -> [:a:] -> [:a:]
80 concatP, -- :: [:[:a:]:] -> [:a:]
81 concatMapP, -- :: (a -> [:b:]) -> [:a:] -> [:b:]
82 -- head, last, tail, init, -- it's not wise to use them on arrays
83 nullP, -- :: [:a:] -> Bool
84 lengthP, -- :: [:a:] -> Int
85 (!:), -- :: [:a:] -> Int -> a
86 foldlP, -- :: (a -> b -> a) -> a -> [:b:] -> a
87 foldl1P, -- :: (a -> a -> a) -> [:a:] -> a
88 scanlP, -- :: (a -> b -> a) -> a -> [:b:] -> [:a:]
89 scanl1P, -- :: (a -> a -> a) -> [:a:] -> [:a:]
90 foldrP, -- :: (a -> b -> b) -> b -> [:a:] -> b
91 foldr1P, -- :: (a -> a -> a) -> [:a:] -> a
92 scanrP, -- :: (a -> b -> b) -> b -> [:a:] -> [:b:]
93 scanr1P, -- :: (a -> a -> a) -> [:a:] -> [:a:]
94 -- iterate, repeat, -- parallel arrays must be finite
95 replicateP, -- :: Int -> a -> [:a:]
96 -- cycle, -- parallel arrays must be finite
97 takeP, -- :: Int -> [:a:] -> [:a:]
98 dropP, -- :: Int -> [:a:] -> [:a:]
99 splitAtP, -- :: Int -> [:a:] -> ([:a:],[:a:])
100 takeWhileP, -- :: (a -> Bool) -> [:a:] -> [:a:]
101 dropWhileP, -- :: (a -> Bool) -> [:a:] -> [:a:]
102 spanP, -- :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
103 breakP, -- :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
104 -- lines, words, unlines, unwords, -- is string processing really needed
105 reverseP, -- :: [:a:] -> [:a:]
106 andP, -- :: [:Bool:] -> Bool
107 orP, -- :: [:Bool:] -> Bool
108 anyP, -- :: (a -> Bool) -> [:a:] -> Bool
109 allP, -- :: (a -> Bool) -> [:a:] -> Bool
110 elemP, -- :: (Eq a) => a -> [:a:] -> Bool
111 notElemP, -- :: (Eq a) => a -> [:a:] -> Bool
112 lookupP, -- :: (Eq a) => a -> [:(a, b):] -> Maybe b
113 sumP, -- :: (Num a) => [:a:] -> a
114 productP, -- :: (Num a) => [:a:] -> a
115 maximumP, -- :: (Ord a) => [:a:] -> a
116 minimumP, -- :: (Ord a) => [:a:] -> a
117 zipP, -- :: [:a:] -> [:b:] -> [:(a, b) :]
118 zip3P, -- :: [:a:] -> [:b:] -> [:c:] -> [:(a, b, c):]
119 zipWithP, -- :: (a -> b -> c) -> [:a:] -> [:b:] -> [:c:]
120 zipWith3P, -- :: (a -> b -> c -> d) -> [:a:]->[:b:]->[:c:]->[:d:]
121 unzipP, -- :: [:(a, b) :] -> ([:a:], [:b:])
122 unzip3P, -- :: [:(a, b, c):] -> ([:a:], [:b:], [:c:])
124 -- overloaded functions
126 enumFromToP, -- :: Enum a => a -> a -> [:a:]
127 enumFromThenToP, -- :: Enum a => a -> a -> a -> [:a:]
129 -- the following functions are not available on lists
131 toP, -- :: [a] -> [:a:]
132 fromP, -- :: [:a:] -> [a]
133 sliceP, -- :: Int -> Int -> [:e:] -> [:e:]
134 foldP, -- :: (e -> e -> e) -> e -> [:e:] -> e
135 fold1P, -- :: (e -> e -> e) -> [:e:] -> e
136 permuteP, -- :: [:Int:] -> [:e:] -> [:e:]
137 bpermuteP, -- :: [:Int:] -> [:e:] -> [:e:]
138 bpermuteDftP, -- :: [:Int:] -> [:e:] -> [:e:] -> [:e:]
139 crossP, -- :: [:a:] -> [:b:] -> [:(a, b):]
140 indexOfP -- :: (a -> Bool) -> [:a:] -> [:Int:]
145 import GHC.ST ( ST(..), STRep, runST )
146 import GHC.Exts ( Int#, Array#, Int(I#), MutableArray#, newArray#,
147 unsafeFreezeArray#, indexArray#, writeArray# )
151 infix 4 `elemP`, `notElemP`
154 -- representation of parallel arrays
155 -- ---------------------------------
157 -- this rather straight forward implementation maps parallel arrays to the
158 -- internal representation used for standard Haskell arrays in GHC's Prelude
159 -- (EXPORTED ABSTRACTLY)
161 -- * This definition *must* be kept in sync with `TysWiredIn.parrTyCon'!
163 data [::] e = PArr Int# (Array# e)
166 -- exported operations on parallel arrays
167 -- --------------------------------------
169 -- operations corresponding to list operations
172 mapP :: (a -> b) -> [:a:] -> [:b:]
173 mapP f = fst . loop (mapEFL f) noAL
175 (+:+) :: [:a:] -> [:a:] -> [:a:]
176 a1 +:+ a2 = fst $ loop (mapEFL sel) noAL (enumFromToP 0 (len1 + len2 - 1))
177 -- we can't use the [:x..y:] form here for tedious
178 -- reasons to do with the typechecker and the fact that
179 -- `enumFromToP' is defined in the same module
184 sel i | i < len1 = a1!:i
185 | otherwise = a2!:(i - len1)
187 filterP :: (a -> Bool) -> [:a:] -> [:a:]
188 filterP p = fst . loop (filterEFL p) noAL
190 concatP :: [:[:a:]:] -> [:a:]
191 concatP xss = foldlP (+:+) [::] xss
193 concatMapP :: (a -> [:b:]) -> [:a:] -> [:b:]
194 concatMapP f = concatP . mapP f
196 -- head, last, tail, init, -- it's not wise to use them on arrays
198 nullP :: [:a:] -> Bool
202 lengthP :: [:a:] -> Int
203 lengthP (PArr n# _) = I# n#
205 (!:) :: [:a:] -> Int -> a
208 foldlP :: (a -> b -> a) -> a -> [:b:] -> a
209 foldlP f z = snd . loop (foldEFL (flip f)) z
211 foldl1P :: (a -> a -> a) -> [:a:] -> a
212 foldl1P f [::] = error "Prelude.foldl1P: empty array"
213 foldl1P f a = snd $ loopFromTo 1 (lengthP a - 1) (foldEFL f) (a!:0) a
215 scanlP :: (a -> b -> a) -> a -> [:b:] -> [:a:]
216 scanlP f z = fst . loop (scanEFL (flip f)) z
218 scanl1P :: (a -> a -> a) -> [:a:] -> [:a:]
219 acanl1P f [::] = error "Prelude.scanl1P: empty array"
220 scanl1P f a = fst $ loopFromTo 1 (lengthP a - 1) (scanEFL f) (a!:0) a
222 foldrP :: (a -> b -> b) -> b -> [:a:] -> b
223 foldrP = error "Prelude.foldrP: not implemented yet" -- FIXME
225 foldr1P :: (a -> a -> a) -> [:a:] -> a
226 foldr1P = error "Prelude.foldr1P: not implemented yet" -- FIXME
228 scanrP :: (a -> b -> b) -> b -> [:a:] -> [:b:]
229 scanrP = error "Prelude.scanrP: not implemented yet" -- FIXME
231 scanr1P :: (a -> a -> a) -> [:a:] -> [:a:]
232 scanr1P = error "Prelude.scanr1P: not implemented yet" -- FIXME
234 -- iterate, repeat -- parallel arrays must be finite
236 replicateP :: Int -> a -> [:a:]
237 {-# INLINE replicateP #-}
238 replicateP n e = runST (do
239 marr# <- newArray n e
242 -- cycle -- parallel arrays must be finite
244 takeP :: Int -> [:a:] -> [:a:]
245 takeP n = sliceP 0 (n - 1)
247 dropP :: Int -> [:a:] -> [:a:]
248 dropP n a = sliceP (n - 1) (lengthP a - 1) a
250 splitAtP :: Int -> [:a:] -> ([:a:],[:a:])
251 splitAtP n xs = (takeP n xs, dropP n xs)
253 takeWhileP :: (a -> Bool) -> [:a:] -> [:a:]
254 takeWhileP = error "Prelude.takeWhileP: not implemented yet" -- FIXME
256 dropWhileP :: (a -> Bool) -> [:a:] -> [:a:]
257 dropWhileP = error "Prelude.dropWhileP: not implemented yet" -- FIXME
259 spanP :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
260 spanP = error "Prelude.spanP: not implemented yet" -- FIXME
262 breakP :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
263 breakP p = spanP (not . p)
265 -- lines, words, unlines, unwords, -- is string processing really needed
267 reverseP :: [:a:] -> [:a:]
268 reverseP a = permuteP (enumFromThenToP (len - 1) (len - 2) 0) a
269 -- we can't use the [:x, y..z:] form here for tedious
270 -- reasons to do with the typechecker and the fact that
271 -- `enumFromThenToP' is defined in the same module
275 andP :: [:Bool:] -> Bool
276 andP = foldP (&&) True
278 orP :: [:Bool:] -> Bool
279 orP = foldP (||) True
281 anyP :: (a -> Bool) -> [:a:] -> Bool
282 anyP p = orP . mapP p
284 allP :: (a -> Bool) -> [:a:] -> Bool
285 allP p = andP . mapP p
287 elemP :: (Eq a) => a -> [:a:] -> Bool
288 elemP x = anyP (== x)
290 notElemP :: (Eq a) => a -> [:a:] -> Bool
291 notElemP x = allP (/= x)
293 lookupP :: (Eq a) => a -> [:(a, b):] -> Maybe b
294 lookupP = error "Prelude.lookupP: not implemented yet" -- FIXME
296 sumP :: (Num a) => [:a:] -> a
299 productP :: (Num a) => [:a:] -> a
300 productP = foldP (*) 0
302 maximumP :: (Ord a) => [:a:] -> a
303 maximumP [::] = error "Prelude.maximumP: empty parallel array"
304 maximumP xs = fold1P max xs
306 minimumP :: (Ord a) => [:a:] -> a
307 minimumP [::] = error "Prelude.minimumP: empty parallel array"
308 minimumP xs = fold1P min xs
310 zipP :: [:a:] -> [:b:] -> [:(a, b):]
313 zip3P :: [:a:] -> [:b:] -> [:c:] -> [:(a, b, c):]
314 zip3P = zipWith3P (,,)
316 zipWithP :: (a -> b -> c) -> [:a:] -> [:b:] -> [:c:]
317 zipWithP f a1 a2 = let
320 len = len1 `min` len2
322 fst $ loopFromTo 0 (len - 1) combine 0 a1
324 combine e1 i = (Just $ f e1 (a2!:i), i + 1)
326 zipWith3P :: (a -> b -> c -> d) -> [:a:]->[:b:]->[:c:]->[:d:]
327 zipWith3P f a1 a2 a3 = let
331 len = len1 `min` len2 `min` len3
333 fst $ loopFromTo 0 (len - 1) combine 0 a1
335 combine e1 i = (Just $ f e1 (a2!:i) (a3!:i), i + 1)
337 unzipP :: [:(a, b):] -> ([:a:], [:b:])
338 unzipP a = (fst $ loop (mapEFL fst) noAL a, fst $ loop (mapEFL snd) noAL a)
339 -- FIXME: these two functions should be optimised using a tupled custom loop
340 unzip3P :: [:(a, b, c):] -> ([:a:], [:b:], [:c:])
341 unzip3P a = (fst $ loop (mapEFL fst3) noAL a,
342 fst $ loop (mapEFL snd3) noAL a,
343 fst $ loop (mapEFL trd3) noAL a)
352 instance Eq a => Eq [:a:] where
353 a1 == a2 | lengthP a1 == lengthP a2 = andP (zipWithP (==) a1 a2)
356 instance Ord a => Ord [:a:] where
357 compare a1 a2 = case foldlP combineOrdering EQ (zipWithP compare a1 a2) of
358 EQ | lengthP a1 == lengthP a2 -> EQ
359 | lengthP a1 < lengthP a2 -> LT
362 combineOrdering EQ EQ = EQ
363 combineOrdering EQ other = other
364 combineOrdering other _ = other
366 instance Functor [::] where
369 instance Monad [::] where
370 m >>= k = foldrP ((+:+) . k ) [::] m
371 m >> k = foldrP ((+:+) . const k) [::] m
375 instance Show a => Show [:a:] where
376 showsPrec _ = showPArr . fromP
378 showPArr [] s = "[::]" ++ s
379 showPArr (x:xs) s = "[:" ++ shows x (showPArr' xs s)
381 showPArr' [] s = ":]" ++ s
382 showPArr' (y:ys) s = ',' : shows y (showPArr' ys s)
384 instance Read a => Read [:a:] where
385 readsPrec _ a = [(toP v, rest) | (v, rest) <- readPArr a]
387 readPArr = readParen False (\r -> do
391 (do { (":]", t) <- lex s; return ([], t) }) ++
392 (do { (x, t) <- reads s; (xs, u) <- readPArr2 t; return (x:xs, u) })
395 (do { (":]", t) <- lex s; return ([], t) }) ++
396 (do { (",", t) <- lex s; (x, u) <- reads t; (xs, v) <- readPArr2 u;
399 -- overloaded functions
402 -- Ideally, we would like `enumFromToP' and `enumFromThenToP' to be members of
403 -- `Enum'. On the other hand, we really do not want to change `Enum'. Thus,
404 -- for the moment, we hope that the compiler is sufficiently clever to
405 -- properly fuse the following definition.
407 enumFromToP :: Enum a => a -> a -> [:a:]
408 enumFromToP x y = mapP toEnum (eftInt (fromEnum x) (fromEnum y))
410 eftInt x y = scanlP (+) x $ replicateP (y - x + 1) 1
412 enumFromThenToP :: Enum a => a -> a -> a -> [:a:]
413 enumFromThenToP x y z =
414 mapP toEnum (efttInt (fromEnum x) (fromEnum y) (fromEnum z))
416 efttInt x y z = scanlP (+) x $
417 replicateP ((z - x + 1) `div` delta - 1) delta
421 -- the following functions are not available on lists
424 -- create an array from a list (EXPORTED)
427 toP l = fst $ loop store l (replicateP (length l) ())
429 store _ (x:xs) = (Just x, xs)
431 -- convert an array to a list (EXPORTED)
433 fromP :: [:a:] -> [a]
434 fromP a = [a!:i | i <- [0..lengthP a - 1]]
436 -- cut a subarray out of an array (EXPORTED)
438 sliceP :: Int -> Int -> [:e:] -> [:e:]
440 fst $ loopFromTo (0 `max` from) (to `min` (lengthP a - 1)) (mapEFL id) noAL a
442 -- parallel folding (EXPORTED)
444 -- * the first argument must be associative; otherwise, the result is undefined
446 foldP :: (e -> e -> e) -> e -> [:e:] -> e
449 -- parallel folding without explicit neutral (EXPORTED)
451 -- * the first argument must be associative; otherwise, the result is undefined
453 fold1P :: (e -> e -> e) -> [:e:] -> e
456 -- permute an array according to the permutation vector in the first argument
459 permuteP :: [:Int:] -> [:e:] -> [:e:]
460 permuteP is es = fst $ loop (mapEFL (es!:)) noAL is
462 -- permute an array according to the back-permutation vector in the first
463 -- argument (EXPORTED)
465 -- * the permutation vector must represent a surjective function; otherwise,
466 -- the result is undefined
468 bpermuteP :: [:Int:] -> [:e:] -> [:e:]
469 bpermuteP is es = error "Prelude.bpermuteP: not implemented yet" -- FIXME
471 -- permute an array according to the back-permutation vector in the first
472 -- argument, which need not be surjective (EXPORTED)
474 -- * any elements in the result that are not covered by the back-permutation
475 -- vector assume the value of the corresponding position of the third
478 bpermuteDftP :: [:Int:] -> [:e:] -> [:e:] -> [:e:]
479 bpermuteDftP is es = error "Prelude.bpermuteDftP: not implemented yet"-- FIXME
481 -- computes the cross combination of two arrays (EXPORTED)
483 crossP :: [:a:] -> [:b:] -> [:(a, b):]
484 crossP a1 a2 = fst $ loop combine (0, 0) $ replicateP len ()
490 combine _ (i, j) = (Just $ (a1!:i, a2!:j), next)
492 next | (i + 1) == len1 = (0 , j + 1)
493 | otherwise = (i + 1, j)
495 {- An alternative implementation
496 * The one above is certainly better for flattened code, but here where we
497 are handling boxed arrays, the trade off is less clear. However, I
498 think, the above one is still better.
503 x1 = concatP $ mapP (replicateP len2) a1
504 x2 = concatP $ replicateP len1 a2
509 -- computes an index array for all elements of the second argument for which
510 -- the predicate yields `True' (EXPORTED)
512 indexOfP :: (a -> Bool) -> [:a:] -> [:Int:]
513 indexOfP p a = fst $ loop calcIdx 0 a
515 calcIdx e idx | p e = (Just idx, idx + 1)
516 | otherwise = (Nothing , idx )
519 -- auxiliary functions
520 -- -------------------
522 -- internally used mutable boxed arrays
524 data MPArr s e = MPArr Int# (MutableArray# s e)
526 -- allocate a new mutable array that is pre-initialised with a given value
528 newArray :: Int -> e -> ST s (MPArr s e)
529 {-# INLINE newArray #-}
530 newArray (I# n#) e = ST $ \s1# ->
531 case newArray# n# e s1# of { (# s2#, marr# #) ->
532 (# s2#, MPArr n# marr# #)}
534 -- convert a mutable array into the external parallel array representation
536 mkPArr :: Int -> MPArr s e -> ST s [:e:]
537 {-# INLINE mkPArr #-}
538 mkPArr (I# n#) (MPArr _ marr#) = ST $ \s1# ->
539 case unsafeFreezeArray# marr# s1# of { (# s2#, arr# #) ->
540 (# s2#, PArr n# arr# #) }
542 -- general array iterator
544 -- * corresponds to `loopA' from ``Functional Array Fusion'', Chakravarty &
547 loop :: (e -> acc -> (Maybe e', acc)) -- mapping & folding, once per element
548 -> acc -- initial acc value
549 -> [:e:] -- input array
552 loop mf acc arr = loopFromTo 0 (lengthP arr - 1) mf acc arr
554 -- general array iterator with bounds
556 loopFromTo :: Int -- from index
558 -> (e -> acc -> (Maybe e', acc))
562 {-# INLINE loopFromTo #-}
563 loopFromTo from to mf start arr = runST (do
564 marr <- newArray (to - from + 1) noElem
565 (n', acc) <- trans from to marr arr mf start
566 arr <- mkPArr n' marr
569 noElem = error "PrelPArr.loopFromTo: I do not exist!"
570 -- unlike standard Haskell arrays, this value represents an
573 -- actually loop body of `loop'
575 -- * for this to be really efficient, it has to be translated with the
576 -- constructor specialisation phase "SpecConstr" switched on; as of GHC 5.03
577 -- this requires an optimisation level of at least -O2
579 trans :: Int -- index of first elem to process
580 -> Int -- index of last elem to process
581 -> MPArr s e' -- destination array
582 -> [:e:] -- source array
583 -> (e -> acc -> (Maybe e', acc)) -- mutator
584 -> acc -- initial accumulator
585 -> ST s (Int, acc) -- final destination length/final acc
587 trans from to marr arr mf start = trans' from 0 start
589 trans' arrOff marrOff acc
590 | arrOff > to = return (marrOff, acc)
592 let (oe', acc') = mf (arr `indexPArr` arrOff) acc
593 marrOff' <- case oe' of
594 Nothing -> return marrOff
596 writeMPArr marr marrOff e'
598 trans' (arrOff + 1) marrOff' acc'
601 -- common patterns for using `loop'
604 -- initial value for the accumulator when the accumulator is not needed
609 -- `loop' mutator maps a function over array elements
611 mapEFL :: (e -> e') -> (e -> () -> (Maybe e', ()))
612 {-# INLINE mapEFL #-}
613 mapEFL f = \e a -> (Just $ f e, ())
615 -- `loop' mutator that filter elements according to a predicate
617 filterEFL :: (e -> Bool) -> (e -> () -> (Maybe e, ()))
618 {-# INLINE filterEFL #-}
619 filterEFL p = \e a -> if p e then (Just e, ()) else (Nothing, ())
621 -- `loop' mutator for array folding
623 foldEFL :: (e -> acc -> acc) -> (e -> acc -> (Maybe (), acc))
624 {-# INLINE foldEFL #-}
625 foldEFL f = \e a -> (Nothing, f e a)
627 -- `loop' mutator for array scanning
629 scanEFL :: (e -> acc -> acc) -> (e -> acc -> (Maybe acc, acc))
630 {-# INLINE scanEFL #-}
631 scanEFL f = \e a -> (Just a, f e a)
633 -- elementary array operations
636 -- unlifted array indexing
638 indexPArr :: [:e:] -> Int -> e
639 {-# INLINE indexPArr #-}
640 indexPArr (PArr _ arr#) (I# i#) =
641 case indexArray# arr# i# of (# e #) -> e
643 -- encapsulate writing into a mutable array into the `ST' monad
645 writeMPArr :: MPArr s e -> Int -> e -> ST s ()
646 {-# INLINE writeMPArr #-}
647 writeMPArr (MPArr _ marr#) (I# i#) e = ST $ \s# ->
648 case writeArray# marr# i# e s# of s'# -> (# s'#, () #)