cd2f03b297ae79bbd6b8df8cb5745fe6a5f7b3f9
[ghc.git] / libraries / base / GHC / PArr.hs
1 {-# OPTIONS_GHC -fparr -funbox-strict-fields #-}
2
3 -----------------------------------------------------------------------------
4 -- |
5 -- Module : GHC.PArr
6 -- Copyright : (c) 2001-2002 Manuel M T Chakravarty & Gabriele Keller
7 -- License : see libraries/base/LICENSE
8 --
9 -- Maintainer : Manuel M. T. Chakravarty <chak@cse.unsw.edu.au>
10 -- Stability : internal
11 -- Portability : non-portable (GHC Extensions)
12 --
13 -- Basic implementation of Parallel Arrays.
14 --
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.
19 --
20 --- DOCU ----------------------------------------------------------------------
21 --
22 -- Language: Haskell 98 plus unboxed values and parallel arrays
23 --
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,
34 -- such as Fortran.
35 --
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'.
41 --
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.
49 --
50 --- TODO ----------------------------------------------------------------------
51 --
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.
55 --
56 -- * Currently, functions that emphasis the constructor-based definition of
57 -- lists (such as, head, last, tail, and init) are not supported.
58 --
59 -- Is it worthwhile to support the string processing functions lines,
60 -- words, unlines, and unwords? (Currently, they are not implemented.)
61 --
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'.
64 --
65 -- * Prescans are often more useful for array programming than scans. Shall
66 -- we include them into the Prelude or the library?
67 --
68 -- * Due to the use of the iterator `loop', we could define some fusion rules
69 -- in this module.
70 --
71 -- * We might want to add bounds checks that can be deactivated.
72 --
73
74 module GHC.PArr (
75 -- [::], -- Built-in syntax
76
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:])
123
124 -- overloaded functions
125 --
126 enumFromToP, -- :: Enum a => a -> a -> [:a:]
127 enumFromThenToP, -- :: Enum a => a -> a -> a -> [:a:]
128
129 -- the following functions are not available on lists
130 --
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 dpermuteP, -- :: [:Int:] -> [:e:] -> [:e:] -> [:e:]
139 crossP, -- :: [:a:] -> [:b:] -> [:(a, b):]
140 crossMapP, -- :: [:a:] -> (a -> [:b:]) -> [:(a, b):]
141 indexOfP -- :: (a -> Bool) -> [:a:] -> [:Int:]
142 ) where
143
144 #ifndef __HADDOCK__
145
146 import Prelude
147
148 import GHC.ST ( ST(..), STRep, runST )
149 import GHC.Exts ( Int#, Array#, Int(I#), MutableArray#, newArray#,
150 unsafeFreezeArray#, indexArray#, writeArray#, (<#), (>=#) )
151
152 infixl 9 !:
153 infixr 5 +:+
154 infix 4 `elemP`, `notElemP`
155
156
157 -- representation of parallel arrays
158 -- ---------------------------------
159
160 -- this rather straight forward implementation maps parallel arrays to the
161 -- internal representation used for standard Haskell arrays in GHC's Prelude
162 -- (EXPORTED ABSTRACTLY)
163 --
164 -- * This definition *must* be kept in sync with `TysWiredIn.parrTyCon'!
165 --
166 data [::] e = PArr Int# (Array# e)
167
168
169 -- exported operations on parallel arrays
170 -- --------------------------------------
171
172 -- operations corresponding to list operations
173 --
174
175 mapP :: (a -> b) -> [:a:] -> [:b:]
176 mapP f = fst . loop (mapEFL f) noAL
177
178 (+:+) :: [:a:] -> [:a:] -> [:a:]
179 a1 +:+ a2 = fst $ loop (mapEFL sel) noAL (enumFromToP 0 (len1 + len2 - 1))
180 -- we can't use the [:x..y:] form here for tedious
181 -- reasons to do with the typechecker and the fact that
182 -- `enumFromToP' is defined in the same module
183 where
184 len1 = lengthP a1
185 len2 = lengthP a2
186 --
187 sel i | i < len1 = a1!:i
188 | otherwise = a2!:(i - len1)
189
190 filterP :: (a -> Bool) -> [:a:] -> [:a:]
191 filterP p = fst . loop (filterEFL p) noAL
192
193 concatP :: [:[:a:]:] -> [:a:]
194 concatP xss = foldlP (+:+) [::] xss
195
196 concatMapP :: (a -> [:b:]) -> [:a:] -> [:b:]
197 concatMapP f = concatP . mapP f
198
199 -- head, last, tail, init, -- it's not wise to use them on arrays
200
201 nullP :: [:a:] -> Bool
202 nullP [::] = True
203 nullP _ = False
204
205 lengthP :: [:a:] -> Int
206 lengthP (PArr n# _) = I# n#
207
208 (!:) :: [:a:] -> Int -> a
209 (!:) = indexPArr
210
211 foldlP :: (a -> b -> a) -> a -> [:b:] -> a
212 foldlP f z = snd . loop (foldEFL (flip f)) z
213
214 foldl1P :: (a -> a -> a) -> [:a:] -> a
215 foldl1P f [::] = error "Prelude.foldl1P: empty array"
216 foldl1P f a = snd $ loopFromTo 1 (lengthP a - 1) (foldEFL f) (a!:0) a
217
218 scanlP :: (a -> b -> a) -> a -> [:b:] -> [:a:]
219 scanlP f z = fst . loop (scanEFL (flip f)) z
220
221 scanl1P :: (a -> a -> a) -> [:a:] -> [:a:]
222 scanl1P f [::] = error "Prelude.scanl1P: empty array"
223 scanl1P f a = fst $ loopFromTo 1 (lengthP a - 1) (scanEFL f) (a!:0) a
224
225 foldrP :: (a -> b -> b) -> b -> [:a:] -> b
226 foldrP = error "Prelude.foldrP: not implemented yet" -- FIXME
227
228 foldr1P :: (a -> a -> a) -> [:a:] -> a
229 foldr1P = error "Prelude.foldr1P: not implemented yet" -- FIXME
230
231 scanrP :: (a -> b -> b) -> b -> [:a:] -> [:b:]
232 scanrP = error "Prelude.scanrP: not implemented yet" -- FIXME
233
234 scanr1P :: (a -> a -> a) -> [:a:] -> [:a:]
235 scanr1P = error "Prelude.scanr1P: not implemented yet" -- FIXME
236
237 -- iterate, repeat -- parallel arrays must be finite
238
239 replicateP :: Int -> a -> [:a:]
240 {-# INLINE replicateP #-}
241 replicateP n e = runST (do
242 marr# <- newArray n e
243 mkPArr n marr#)
244
245 -- cycle -- parallel arrays must be finite
246
247 takeP :: Int -> [:a:] -> [:a:]
248 takeP n = sliceP 0 (n - 1)
249
250 dropP :: Int -> [:a:] -> [:a:]
251 dropP n a = sliceP n (lengthP a - 1) a
252
253 splitAtP :: Int -> [:a:] -> ([:a:],[:a:])
254 splitAtP n xs = (takeP n xs, dropP n xs)
255
256 takeWhileP :: (a -> Bool) -> [:a:] -> [:a:]
257 takeWhileP = error "Prelude.takeWhileP: not implemented yet" -- FIXME
258
259 dropWhileP :: (a -> Bool) -> [:a:] -> [:a:]
260 dropWhileP = error "Prelude.dropWhileP: not implemented yet" -- FIXME
261
262 spanP :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
263 spanP = error "Prelude.spanP: not implemented yet" -- FIXME
264
265 breakP :: (a -> Bool) -> [:a:] -> ([:a:], [:a:])
266 breakP p = spanP (not . p)
267
268 -- lines, words, unlines, unwords, -- is string processing really needed
269
270 reverseP :: [:a:] -> [:a:]
271 reverseP a = permuteP (enumFromThenToP (len - 1) (len - 2) 0) a
272 -- we can't use the [:x, y..z:] form here for tedious
273 -- reasons to do with the typechecker and the fact that
274 -- `enumFromThenToP' is defined in the same module
275 where
276 len = lengthP a
277
278 andP :: [:Bool:] -> Bool
279 andP = foldP (&&) True
280
281 orP :: [:Bool:] -> Bool
282 orP = foldP (||) True
283
284 anyP :: (a -> Bool) -> [:a:] -> Bool
285 anyP p = orP . mapP p
286
287 allP :: (a -> Bool) -> [:a:] -> Bool
288 allP p = andP . mapP p
289
290 elemP :: (Eq a) => a -> [:a:] -> Bool
291 elemP x = anyP (== x)
292
293 notElemP :: (Eq a) => a -> [:a:] -> Bool
294 notElemP x = allP (/= x)
295
296 lookupP :: (Eq a) => a -> [:(a, b):] -> Maybe b
297 lookupP = error "Prelude.lookupP: not implemented yet" -- FIXME
298
299 sumP :: (Num a) => [:a:] -> a
300 sumP = foldP (+) 0
301
302 productP :: (Num a) => [:a:] -> a
303 productP = foldP (*) 1
304
305 maximumP :: (Ord a) => [:a:] -> a
306 maximumP [::] = error "Prelude.maximumP: empty parallel array"
307 maximumP xs = fold1P max xs
308
309 minimumP :: (Ord a) => [:a:] -> a
310 minimumP [::] = error "Prelude.minimumP: empty parallel array"
311 minimumP xs = fold1P min xs
312
313 zipP :: [:a:] -> [:b:] -> [:(a, b):]
314 zipP = zipWithP (,)
315
316 zip3P :: [:a:] -> [:b:] -> [:c:] -> [:(a, b, c):]
317 zip3P = zipWith3P (,,)
318
319 zipWithP :: (a -> b -> c) -> [:a:] -> [:b:] -> [:c:]
320 zipWithP f a1 a2 = let
321 len1 = lengthP a1
322 len2 = lengthP a2
323 len = len1 `min` len2
324 in
325 fst $ loopFromTo 0 (len - 1) combine 0 a1
326 where
327 combine e1 i = (Just $ f e1 (a2!:i), i + 1)
328
329 zipWith3P :: (a -> b -> c -> d) -> [:a:]->[:b:]->[:c:]->[:d:]
330 zipWith3P f a1 a2 a3 = let
331 len1 = lengthP a1
332 len2 = lengthP a2
333 len3 = lengthP a3
334 len = len1 `min` len2 `min` len3
335 in
336 fst $ loopFromTo 0 (len - 1) combine 0 a1
337 where
338 combine e1 i = (Just $ f e1 (a2!:i) (a3!:i), i + 1)
339
340 unzipP :: [:(a, b):] -> ([:a:], [:b:])
341 unzipP a = (fst $ loop (mapEFL fst) noAL a, fst $ loop (mapEFL snd) noAL a)
342 -- FIXME: these two functions should be optimised using a tupled custom loop
343 unzip3P :: [:(a, b, c):] -> ([:a:], [:b:], [:c:])
344 unzip3P a = (fst $ loop (mapEFL fst3) noAL a,
345 fst $ loop (mapEFL snd3) noAL a,
346 fst $ loop (mapEFL trd3) noAL a)
347 where
348 fst3 (a, _, _) = a
349 snd3 (_, b, _) = b
350 trd3 (_, _, c) = c
351
352 -- instances
353 --
354
355 instance Eq a => Eq [:a:] where
356 a1 == a2 | lengthP a1 == lengthP a2 = andP (zipWithP (==) a1 a2)
357 | otherwise = False
358
359 instance Ord a => Ord [:a:] where
360 compare a1 a2 = case foldlP combineOrdering EQ (zipWithP compare a1 a2) of
361 EQ | lengthP a1 == lengthP a2 -> EQ
362 | lengthP a1 < lengthP a2 -> LT
363 | otherwise -> GT
364 where
365 combineOrdering EQ EQ = EQ
366 combineOrdering EQ other = other
367 combineOrdering other _ = other
368
369 instance Functor [::] where
370 fmap = mapP
371
372 instance Monad [::] where
373 m >>= k = foldrP ((+:+) . k ) [::] m
374 m >> k = foldrP ((+:+) . const k) [::] m
375 return x = [:x:]
376 fail _ = [::]
377
378 instance Show a => Show [:a:] where
379 showsPrec _ = showPArr . fromP
380 where
381 showPArr [] s = "[::]" ++ s
382 showPArr (x:xs) s = "[:" ++ shows x (showPArr' xs s)
383
384 showPArr' [] s = ":]" ++ s
385 showPArr' (y:ys) s = ',' : shows y (showPArr' ys s)
386
387 instance Read a => Read [:a:] where
388 readsPrec _ a = [(toP v, rest) | (v, rest) <- readPArr a]
389 where
390 readPArr = readParen False (\r -> do
391 ("[:",s) <- lex r
392 readPArr1 s)
393 readPArr1 s =
394 (do { (":]", t) <- lex s; return ([], t) }) ++
395 (do { (x, t) <- reads s; (xs, u) <- readPArr2 t; return (x:xs, u) })
396
397 readPArr2 s =
398 (do { (":]", t) <- lex s; return ([], t) }) ++
399 (do { (",", t) <- lex s; (x, u) <- reads t; (xs, v) <- readPArr2 u;
400 return (x:xs, v) })
401
402 -- overloaded functions
403 --
404
405 -- Ideally, we would like `enumFromToP' and `enumFromThenToP' to be members of
406 -- `Enum'. On the other hand, we really do not want to change `Enum'. Thus,
407 -- for the moment, we hope that the compiler is sufficiently clever to
408 -- properly fuse the following definitions.
409
410 enumFromToP :: Enum a => a -> a -> [:a:]
411 enumFromToP x y = mapP toEnum (eftInt (fromEnum x) (fromEnum y))
412 where
413 eftInt x y = scanlP (+) x $ replicateP (y - x + 1) 1
414
415 enumFromThenToP :: Enum a => a -> a -> a -> [:a:]
416 enumFromThenToP x y z =
417 mapP toEnum (efttInt (fromEnum x) (fromEnum y) (fromEnum z))
418 where
419 efttInt x y z = scanlP (+) x $
420 replicateP (abs (z - x) `div` abs delta + 1) delta
421 where
422 delta = y - x
423
424 -- the following functions are not available on lists
425 --
426
427 -- create an array from a list (EXPORTED)
428 --
429 toP :: [a] -> [:a:]
430 toP l = fst $ loop store l (replicateP (length l) ())
431 where
432 store _ (x:xs) = (Just x, xs)
433
434 -- convert an array to a list (EXPORTED)
435 --
436 fromP :: [:a:] -> [a]
437 fromP a = [a!:i | i <- [0..lengthP a - 1]]
438
439 -- cut a subarray out of an array (EXPORTED)
440 --
441 sliceP :: Int -> Int -> [:e:] -> [:e:]
442 sliceP from to a =
443 fst $ loopFromTo (0 `max` from) (to `min` (lengthP a - 1)) (mapEFL id) noAL a
444
445 -- parallel folding (EXPORTED)
446 --
447 -- * the first argument must be associative; otherwise, the result is undefined
448 --
449 foldP :: (e -> e -> e) -> e -> [:e:] -> e
450 foldP = foldlP
451
452 -- parallel folding without explicit neutral (EXPORTED)
453 --
454 -- * the first argument must be associative; otherwise, the result is undefined
455 --
456 fold1P :: (e -> e -> e) -> [:e:] -> e
457 fold1P = foldl1P
458
459 -- permute an array according to the permutation vector in the first argument
460 -- (EXPORTED)
461 --
462 permuteP :: [:Int:] -> [:e:] -> [:e:]
463 permuteP is es
464 | isLen /= esLen = error "GHC.PArr: arguments must be of the same length"
465 | otherwise = runST (do
466 marr <- newArray isLen noElem
467 permute marr is es
468 mkPArr isLen marr)
469 where
470 noElem = error "GHC.PArr.permuteP: I do not exist!"
471 -- unlike standard Haskell arrays, this value represents an
472 -- internal error
473 isLen = lengthP is
474 esLen = lengthP es
475
476 -- permute an array according to the back-permutation vector in the first
477 -- argument (EXPORTED)
478 --
479 -- * the permutation vector must represent a surjective function; otherwise,
480 -- the result is undefined
481 --
482 bpermuteP :: [:Int:] -> [:e:] -> [:e:]
483 bpermuteP is es = fst $ loop (mapEFL (es!:)) noAL is
484
485 -- permute an array according to the permutation vector in the first
486 -- argument, which need not be surjective (EXPORTED)
487 --
488 -- * any elements in the result that are not covered by the permutation
489 -- vector assume the value of the corresponding position of the third
490 -- argument
491 --
492 dpermuteP :: [:Int:] -> [:e:] -> [:e:] -> [:e:]
493 dpermuteP is es dft
494 | isLen /= esLen = error "GHC.PArr: arguments must be of the same length"
495 | otherwise = runST (do
496 marr <- newArray dftLen noElem
497 trans 0 (isLen - 1) marr dft copyOne noAL
498 permute marr is es
499 mkPArr dftLen marr)
500 where
501 noElem = error "GHC.PArr.permuteP: I do not exist!"
502 -- unlike standard Haskell arrays, this value represents an
503 -- internal error
504 isLen = lengthP is
505 esLen = lengthP es
506 dftLen = lengthP dft
507
508 copyOne e _ = (Just e, noAL)
509
510 -- computes the cross combination of two arrays (EXPORTED)
511 --
512 crossP :: [:a:] -> [:b:] -> [:(a, b):]
513 crossP a1 a2 = fst $ loop combine (0, 0) $ replicateP len ()
514 where
515 len1 = lengthP a1
516 len2 = lengthP a2
517 len = len1 * len2
518 --
519 combine _ (i, j) = (Just $ (a1!:i, a2!:j), next)
520 where
521 next | (i + 1) == len1 = (0 , j + 1)
522 | otherwise = (i + 1, j)
523
524 {- An alternative implementation
525 * The one above is certainly better for flattened code, but here where we
526 are handling boxed arrays, the trade off is less clear. However, I
527 think, the above one is still better.
528
529 crossP a1 a2 = let
530 len1 = lengthP a1
531 len2 = lengthP a2
532 x1 = concatP $ mapP (replicateP len2) a1
533 x2 = concatP $ replicateP len1 a2
534 in
535 zipP x1 x2
536 -}
537
538 -- |Compute a cross of an array and the arrays produced by the given function
539 -- for the elements of the first array.
540 --
541 crossMapP :: [:a:] -> (a -> [:b:]) -> [:(a, b):]
542 crossMapP a f = let
543 bs = mapP f a
544 segd = mapP lengthP bs
545 as = zipWithP replicateP segd a
546 in
547 zipP (concatP as) (concatP bs)
548
549 {- The following may seem more straight forward, but the above is very cheap
550 with segmented arrays, as `mapP lengthP', `zipP', and `concatP' are
551 constant time, and `map f' uses the lifted version of `f'.
552
553 crossMapP a f = concatP $ mapP (\x -> mapP ((,) x) (f x)) a
554
555 -}
556
557 -- computes an index array for all elements of the second argument for which
558 -- the predicate yields `True' (EXPORTED)
559 --
560 indexOfP :: (a -> Bool) -> [:a:] -> [:Int:]
561 indexOfP p a = fst $ loop calcIdx 0 a
562 where
563 calcIdx e idx | p e = (Just idx, idx + 1)
564 | otherwise = (Nothing , idx )
565
566
567 -- auxiliary functions
568 -- -------------------
569
570 -- internally used mutable boxed arrays
571 --
572 data MPArr s e = MPArr Int# (MutableArray# s e)
573
574 -- allocate a new mutable array that is pre-initialised with a given value
575 --
576 newArray :: Int -> e -> ST s (MPArr s e)
577 {-# INLINE newArray #-}
578 newArray (I# n#) e = ST $ \s1# ->
579 case newArray# n# e s1# of { (# s2#, marr# #) ->
580 (# s2#, MPArr n# marr# #)}
581
582 -- convert a mutable array into the external parallel array representation
583 --
584 mkPArr :: Int -> MPArr s e -> ST s [:e:]
585 {-# INLINE mkPArr #-}
586 mkPArr (I# n#) (MPArr _ marr#) = ST $ \s1# ->
587 case unsafeFreezeArray# marr# s1# of { (# s2#, arr# #) ->
588 (# s2#, PArr n# arr# #) }
589
590 -- general array iterator
591 --
592 -- * corresponds to `loopA' from ``Functional Array Fusion'', Chakravarty &
593 -- Keller, ICFP 2001
594 --
595 loop :: (e -> acc -> (Maybe e', acc)) -- mapping & folding, once per element
596 -> acc -- initial acc value
597 -> [:e:] -- input array
598 -> ([:e':], acc)
599 {-# INLINE loop #-}
600 loop mf acc arr = loopFromTo 0 (lengthP arr - 1) mf acc arr
601
602 -- general array iterator with bounds
603 --
604 loopFromTo :: Int -- from index
605 -> Int -- to index
606 -> (e -> acc -> (Maybe e', acc))
607 -> acc
608 -> [:e:]
609 -> ([:e':], acc)
610 {-# INLINE loopFromTo #-}
611 loopFromTo from to mf start arr = runST (do
612 marr <- newArray (to - from + 1) noElem
613 (n', acc) <- trans from to marr arr mf start
614 arr <- mkPArr n' marr
615 return (arr, acc))
616 where
617 noElem = error "GHC.PArr.loopFromTo: I do not exist!"
618 -- unlike standard Haskell arrays, this value represents an
619 -- internal error
620
621 -- actual loop body of `loop'
622 --
623 -- * for this to be really efficient, it has to be translated with the
624 -- constructor specialisation phase "SpecConstr" switched on; as of GHC 5.03
625 -- this requires an optimisation level of at least -O2
626 --
627 trans :: Int -- index of first elem to process
628 -> Int -- index of last elem to process
629 -> MPArr s e' -- destination array
630 -> [:e:] -- source array
631 -> (e -> acc -> (Maybe e', acc)) -- mutator
632 -> acc -- initial accumulator
633 -> ST s (Int, acc) -- final destination length/final acc
634 {-# INLINE trans #-}
635 trans from to marr arr mf start = trans' from 0 start
636 where
637 trans' arrOff marrOff acc
638 | arrOff > to = return (marrOff, acc)
639 | otherwise = do
640 let (oe', acc') = mf (arr `indexPArr` arrOff) acc
641 marrOff' <- case oe' of
642 Nothing -> return marrOff
643 Just e' -> do
644 writeMPArr marr marrOff e'
645 return $ marrOff + 1
646 trans' (arrOff + 1) marrOff' acc'
647
648 -- Permute the given elements into the mutable array.
649 --
650 permute :: MPArr s e -> [:Int:] -> [:e:] -> ST s ()
651 permute marr is es = perm 0
652 where
653 perm i
654 | i == n = return ()
655 | otherwise = writeMPArr marr (is!:i) (es!:i) >> perm (i + 1)
656 where
657 n = lengthP is
658
659
660 -- common patterns for using `loop'
661 --
662
663 -- initial value for the accumulator when the accumulator is not needed
664 --
665 noAL :: ()
666 noAL = ()
667
668 -- `loop' mutator maps a function over array elements
669 --
670 mapEFL :: (e -> e') -> (e -> () -> (Maybe e', ()))
671 {-# INLINE mapEFL #-}
672 mapEFL f = \e a -> (Just $ f e, ())
673
674 -- `loop' mutator that filter elements according to a predicate
675 --
676 filterEFL :: (e -> Bool) -> (e -> () -> (Maybe e, ()))
677 {-# INLINE filterEFL #-}
678 filterEFL p = \e a -> if p e then (Just e, ()) else (Nothing, ())
679
680 -- `loop' mutator for array folding
681 --
682 foldEFL :: (e -> acc -> acc) -> (e -> acc -> (Maybe (), acc))
683 {-# INLINE foldEFL #-}
684 foldEFL f = \e a -> (Nothing, f e a)
685
686 -- `loop' mutator for array scanning
687 --
688 scanEFL :: (e -> acc -> acc) -> (e -> acc -> (Maybe acc, acc))
689 {-# INLINE scanEFL #-}
690 scanEFL f = \e a -> (Just a, f e a)
691
692 -- elementary array operations
693 --
694
695 -- unlifted array indexing
696 --
697 indexPArr :: [:e:] -> Int -> e
698 {-# INLINE indexPArr #-}
699 indexPArr (PArr n# arr#) (I# i#)
700 | i# >=# 0# && i# <# n# =
701 case indexArray# arr# i# of (# e #) -> e
702 | otherwise = error $ "indexPArr: out of bounds parallel array index; " ++
703 "idx = " ++ show (I# i#) ++ ", arr len = "
704 ++ show (I# n#)
705
706 -- encapsulate writing into a mutable array into the `ST' monad
707 --
708 writeMPArr :: MPArr s e -> Int -> e -> ST s ()
709 {-# INLINE writeMPArr #-}
710 writeMPArr (MPArr n# marr#) (I# i#) e
711 | i# >=# 0# && i# <# n# =
712 ST $ \s# ->
713 case writeArray# marr# i# e s# of s'# -> (# s'#, () #)
714 | otherwise = error $ "writeMPArr: out of bounds parallel array index; " ++
715 "idx = " ++ show (I# i#) ++ ", arr len = "
716 ++ show (I# n#)
717
718 #endif /* __HADDOCK__ */
719