@@ -28,80 +28,56 @@
2828
2929module Streamly.Internal.Data.MutArray.Type
3030 (
31− -- ** Type
31+ -- * Type
3232 -- $arrayNotes
3333 MutArray (..)
34−
35− -- ** Conversion
36− -- *** Pinned and Unpinned
34+ , MutableByteArray
3735 , pin
3836 , unpin
3937 , isPinned
4038
41− -- ** Casting
42− , cast
43− , castUnsafe -- XXX unsafeCast
44− , asBytes
45− , unsafePinnedAsPtr
46− , unsafeAsPtr
47−
39+ -- * Constructing and Writing
4840 -- ** Construction
49− , empty
50−
51− -- *** New
52− -- | New arrays are always empty arrays with some reserve capacity to
53− -- extend the length without reallocating.
54− , emptyOf
55− , newArrayWith -- emptyAlignedWith
56− , pinnedEmptyOf
57− , pinnedNewAligned -- XXX not required
58− -- , new -- uninitialized array of specified length
59−
60− -- *** Cloning
61− , clone
62− , pinnedClone
41+ , nil
6342
64− -- *** Slicing
65− -- | Get a subarray without copying
66− , getSliceUnsafe -- XXX unsafeGetSlice
67− , getSlice
68− , splitAt -- XXX should be able to express using getSlice
69− , breakOn
43+ -- *** Uninitialized Arrays
44+ , pinnedNew
45+ , pinnedNewBytes
46+ , pinnedNewAligned
47+ , new
48+ , newArrayWith
7049
71− -- *** Stream Folds
50+ -- *** From streams
7251 , ArrayUnsafe (..)
73− , unsafeCreateOfWith
74− , unsafeCreateOf
75− , unsafePinnedCreateOf
76− , pinnedCreateOf
77− , createOfWith
78− , createOf
79− , revCreateOf
80−
81− , pinnedCreate
82− , createWith
83− , create
84− -- , revCreate
85−
86− -- *** From containers
52+ , writeNWithUnsafe
53+ , writeNWith
54+ , writeNUnsafe
55+ , pinnedWriteNUnsafe
56+ , writeN
57+ , pinnedWriteN
58+ , pinnedWriteNAligned
59+
60+ , writeWith
61+ , write
62+ , pinnedWrite
63+
64+ , writeRevN
65+ -- , writeRev
66+
67+ -- ** From containers
8768 , fromListN
8869 , pinnedFromListN
8970 , fromList
9071 , pinnedFromList
9172 , fromListRevN
9273 , fromListRev
93− , fromStreamN
94− , fromStream
95− , fromPureStreamN
74+ , fromStreamDN
75+ , fromStreamD
9676 , fromPureStream
97− , fromByteStr#
98− , fromPtrN
99− , fromChunksK
100− , fromChunksRealloced -- fromSmallChunks
10177
102− -- ** Random writes
78+ -- * Random writes
10379 , putIndex
104− , putIndexUnsafe -- XXX unsafePutIndex
80+ , putIndexUnsafe
10581 , putIndices
10682 -- , putFromThenTo
10783 -- , putFrom -- start writing at the given position
@@ -109,71 +85,78 @@ module Streamly.Internal.Data.MutArray.Type
10985 -- , putFromTo
11086 -- , putFromRev
11187 -- , putUptoRev
112− , modifyIndexUnsafe -- XXX unsafeModifyIndex
88+ , modifyIndexUnsafe
11389 , modifyIndex
11490 , modifyIndices
11591 , modify
11692 , swapIndices
11793 , unsafeSwapIndices
11894
119− -- ** Reading
95+ -- * Growing and Shrinking
96+ -- Arrays grow only at the end, though it is possible to grow on both sides
97+ -- and therefore have a cons as well as snoc. But that will require two
98+ -- bounds in the array representation.
12099
121− -- *** Indexing
122− , getIndex
123− , getIndexUnsafe -- XXX unsafeGetIndex
124− -- , getFromThenTo
125− , getIndexRev -- getRevIndex?
126− , indexReader
127− , indexReaderWith
100+ -- ** Appending elements
101+ , snocWith
102+ , snoc
103+ , snocLinear
104+ , snocMay
105+ , snocUnsafe
106+
107+ -- ** Appending streams
108+ , writeAppendNUnsafe
109+ , writeAppendN
110+ , writeAppendWith
111+ , writeAppend
128112
129− -- *** To Streams
113+ -- * Eliminating and Reading
114+
115+ -- ** To streams
116+ , reader
117+ , readerRevWith
118+ , readerRev
119+
120+ -- ** To containers
121+ , toStreamDWith
122+ , toStreamDRevWith
123+ , toStreamKWith
124+ , toStreamKRevWith
130125 , read
131126 , readRev
132− , toStreamWith
133− , toStreamRevWith
134127 , toStreamK
135− , toStreamKWith
136128 , toStreamKRev
137− , toStreamKRevWith
138−
139− -- *** To Containers
140129 , toList
141130
142− -- *** Unfolds
143131 -- experimental
144132 , producerWith
145133 , producer
146134
147− , reader
148− , readerRevWith
149− , readerRev
150−
151− -- ** Size and Capacity
152− -- *** Size
153− , length
154− , byteLength
155−
156− -- *** Capacity
157− -- , capacity
158− , byteCapacity
159− , bytesFree
135+ -- ** Random reads
136+ , getIndex
137+ , getIndexUnsafe
138+ , getIndices
139+ , getIndicesD
140+ -- , getFromThenTo
141+ , getIndexRev
160142
161− -- *** Capacity Management
143+ -- * Memory Management
162144 , blockSize
163145 , arrayChunkBytes
164146 , allocBytesToElemCount
165− , realloc -- this is "resize"
166− , grow
167− , growExp
147+ , realloc
148+ , resize
149+ , resizeExp
168150 , rightSize
169151
170− -- ** Folding
171− , foldl'
172− , foldr
173− , byteCmp
174− , byteEq
152+ -- * Size
153+ , length
154+ , byteLength
155+ -- , capacity
156+ , byteCapacity
157+ , bytesFree
175158
176− -- ** In-place Mutation Algorithms
159+ -- * In-place Mutation Algorithms
177160 , strip
178161 , reverse
179162 , permute
@@ -183,121 +166,62 @@ module Streamly.Internal.Data.MutArray.Type
183166 , mergeBy
184167 , bubble
185168
186− -- ** Growing and Shrinking
187− -- | Arrays grow only at the end, though it is possible to grow on both
188− -- sides and therefore have a cons as well as snoc. But that will require
189− -- both lower and upper bound in the array representation.
190−
191− -- *** Appending elements
192− , snocWith
193− , snoc
194− , snocLinear
195− , snocMay
196− , snocUnsafe -- XXX unsafeSnoc
197−
198− -- *** Appending streams
199− , unsafeAppendN
200− , appendN
201− , appendWith
202− , append
203−
204− -- *** Appending arrays
205− , spliceCopy
206− , spliceWith
207− , splice
208− , spliceExp
209− , spliceUnsafe -- XXX unsafeSplice
210− -- , putSlice
211− -- , appendSlice
212− -- , appendSliceFrom
213−
214− -- ** Serialization using Unbox
215− , pokeAppend
216− , pokeAppendMay
217− , pokeSkipUnsafe -- unsafePokeSkip
169+ -- * Casting
170+ , cast
171+ , castUnsafe
172+ , asBytes
173+ , asPtrUnsafe
218174
219− -- ** Deserialization using Unbox
220− , peekUncons
221− , peekUnconsUnsafe -- unsafePeekUncons
222− , peekSkipUnsafe -- unsafePeekSkip
175+ -- * Folding
176+ , foldl'
177+ , foldr
178+ , cmp
223179
224− -- Arrays of arrays
180+ -- * Arrays of arrays
225181 -- We can add dimensionality parameter to the array type to get
226182 -- multidimensional arrays. Multidimensional arrays would just be a
227183 -- convenience wrapper on top of single dimensional arrays.
228184
229− -- ** Streams of Arrays
230− -- *** Chunk
231− -- | Group a stream into arrays.
185+ -- | Operations dealing with multiple arrays, streams of arrays or
186+ -- multidimensional array representations.
187+
188+ -- ** Construct from streams
232189 , chunksOf
233190 , pinnedChunksOf
234− , buildChunks
191+ , writeChunks
235192
236− -- *** Split
237− -- | Split an array into slices.
193+ -- ** Eliminate to streams
194+ , flattenArrays
195+ , flattenArraysRev
196+ , fromArrayStreamK
238197
198+ -- ** Construct from arrays
199+ -- get chunks without copying
200+ , getSliceUnsafe
201+ , getSlice
239202 -- , getSlicesFromLenN
240− , splitOn
241− -- , slicesOf
242−
243− -- *** Concat
244− -- | Append the arrays in a stream to form a stream of elements.
245− , concatWith
246− , concatRevWith
247− , concat
248− , concatRev
249−
250− -- *** Compact
251− -- | Append the arrays in a stream to form a stream of larger arrays.
252− , SpliceState (..)
253− , pCompactLE
254− , pPinnedCompactLE
255− , compactLeAs
256− , fCompactGE
257− , fPinnedCompactGE
258− , lCompactGE
259− , lPinnedCompactGE
260− , compactGE
261− , compactEQ
262−
263− -- ** Utilities
203+ , splitAt -- XXX should be able to express using getSlice
204+ , breakOn
205+
206+ -- ** Cloning arrays
207+ , clone
208+ , pinnedClone
209+
210+ -- ** Appending arrays
211+ , spliceCopy
212+ , spliceWith
213+ , splice
214+ , spliceExp
215+ , spliceUnsafe
216+ -- , putSlice
217+ -- , appendSlice
218+ -- , appendSliceFrom
219+
220+ -- * Utilities
264221 , roundUpToPower2
265222 , memcpy
266223 , memcmp
267224 , c_memchr
268−
269− -- * Deprecated
270− , asPtrUnsafe
271− , writeChunks
272− , flattenArrays
273− , flattenArraysRev
274− , fromArrayStreamK
275− , fromStreamDN
276− , fromStreamD
277− , cmp
278− , getIndices
279− , getIndicesWith
280− , resize
281− , resizeExp
282− , nil
283− , new
284− , pinnedNew
285− , pinnedNewBytes
286− , writeAppendNUnsafe
287− , writeAppendN
288− , writeAppendWith
289− , writeAppend
290− , writeNWithUnsafe
291− , writeNWith
292− , writeNUnsafe
293− , pinnedWriteNUnsafe
294− , writeN
295− , pinnedWriteN
296− , pinnedWriteNAligned -- XXX not required
297− , writeWith
298− , write
299− , pinnedWrite
300− , writeRevN
301225 )
302226where
303227
@@ -308,20 +232,19 @@ where
308232
309233import Control.Monad (when, void)
310234import Control.Monad.IO.Class (MonadIO(..))
311−import Data.Bifunctor (first)
312235import Data.Bits (shiftR, (.|.), (.&.))
313236import Data.Functor.Identity (Identity(..))
314237import Data.Proxy (Proxy(..))
315238import Data.Word (Word8)
316239import Foreign.C.Types (CSize(..), CInt(..))
317240import Foreign.Ptr (plusPtr, minusPtr, nullPtr)
318−import Streamly.Internal.Data.MutByteArray.Type
319− ( MutByteArray(..)
241+import Streamly.Internal.Data.Unbox
242+ ( MutableByteArray(..)
243+ , Unbox(..)
320244 , PinnedState(..)
321245 , getMutableByteArray#
322246 , putSliceUnsafe
323247 )
324−import Streamly.Internal.Data.Unbox (Unbox(..))
325248import GHC.Base
326249 ( IO(..)
327250 , Int(..)
@@ -329,31 +252,27 @@ import GHC.Base
329252 , copyMutableByteArray#
330253 )
331254import GHC.Base (noinline)
332−import GHC.Exts (unsafeCoerce#, Addr#)
255+import GHC.Exts (unsafeCoerce#)
333256import GHC.Ptr (Ptr(..))
334257
335258import Streamly.Internal.Data.Fold.Type (Fold(..))
336259import Streamly.Internal.Data.Producer.Type (Producer (..))
337260import Streamly.Internal.Data.Stream.Type (Stream)
338−import Streamly.Internal.Data.Parser.Type (Parser (..))
339261import Streamly.Internal.Data.StreamK.Type (StreamK)
340262import Streamly.Internal.Data.SVar.Type (adaptState, defState)
341−import Streamly.Internal.Data.Tuple.Strict (Tuple'(..))
342263import Streamly.Internal.Data.Unfold.Type (Unfold(..))
343264import Streamly.Internal.System.IO (arrayPayloadSize, defaultChunkSize)
344265
345266import qualified Streamly.Internal.Data.Fold.Type as FL
346−import qualified Streamly.Internal.Data.MutByteArray.Type as Unboxed
347−import qualified Streamly.Internal.Data.Parser.Type as Parser
348−-- import qualified Streamly.Internal.Data.Fold.Type as Fold
349267import qualified Streamly.Internal.Data.Producer as Producer
350268import qualified Streamly.Internal.Data.Stream.Type as D
351269import qualified Streamly.Internal.Data.Stream.Lift as D
352270import qualified Streamly.Internal.Data.StreamK.Type as K
271+import qualified Streamly.Internal.Data.Unbox as Unboxed
353272import qualified Prelude
354273
355274import Prelude hiding
356− (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
275+ (Foldable(..), read, unlines, splitAt, reverse, truncate)
357276
358277#include "DocTestDataMutArray.hs"
359278
@@ -361,7 +280,6 @@ import Prelude hiding
361280-- Foreign helpers
362281-------------------------------------------------------------------------------
363282
364−-- NOTE: Have to be "ccall unsafe" so that we can pass unpinned memory to these
365283foreign import ccall unsafe "string.h memcpy" c_memcpy
366284 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO (Ptr Word8)
367285
@@ -371,9 +289,6 @@ foreign import ccall unsafe "string.h memchr" c_memchr
371289foreign import ccall unsafe "string.h memcmp" c_memcmp
372290 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO CInt
373291
374−foreign import ccall unsafe "string.h strlen" c_strlen
375− :: Ptr Word8 -> IO CSize
376−
377292-- | Given an 'Unboxed' type (unused first arg) and a number of bytes, return
378293-- how many elements of that type will completely fit in those bytes.
379294--
@@ -428,7 +343,7 @@ data MutArray a =
428343 -- The array is a range into arrContents. arrContents may be a superset of
429344 -- the slice represented by the array. All offsets are in bytes.
430345 MutArray
431− { arrContents :: {-# UNPACK #-} !MutByteArray
346+ { arrContents :: {-# UNPACK #-} !MutableByteArray
432347 , arrStart :: {-# UNPACK #-} !Int -- ^ index into arrContents
433348 , arrEnd :: {-# UNPACK #-} !Int -- ^ index into arrContents
434349 -- Represents the first invalid index of
@@ -440,8 +355,8 @@ data MutArray a =
440355-- Pinning & Unpinning
441356-------------------------------------------------------------------------------
442357
443−-- | Return a copy of the array in pinned memory if unpinned, else return the
444−-- original array.
358+-- | Copy the 'MutArray' to pinned memory if unpinned, else do nothing. The
359+-- overhead is a copy if the input array is unpinned.
445360{-# INLINE pin #-}
446361pin :: MutArray a -> IO (MutArray a)
447362pin arr@MutArray{..} =
@@ -449,8 +364,8 @@ pin arr@MutArray{..} =
449364 then pure arr
450365 else pinnedClone arr
451366
452−-- | Return a copy of the array in unpinned memory if pinned, else return the
453−-- original array.
367+-- | Copy the 'MutArray' to unpinned memory if pinned, else do nothing. The
368+-- overhead is a copy if the input array is pinned.
454369{-# INLINE unpin #-}
455370unpin :: MutArray a -> IO (MutArray a)
456371unpin arr@MutArray{..} =
@@ -458,7 +373,6 @@ unpin arr@MutArray{..} =
458373 then clone arr
459374 else pure arr
460375
461−-- | Return 'True' if the array is allocated in pinned memory.
462376{-# INLINE isPinned #-}
463377isPinned :: MutArray a -> Bool
464378isPinned MutArray{..} = Unboxed.isPinned arrContents
@@ -479,8 +393,6 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
479393-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
480394-- allocatePinned in GHC source.
481395
482−-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483−
484396-- | @newArrayWith allocator alignment count@ allocates a new array of zero
485397-- length and with a capacity to hold @count@ elements, using @allocator
486398-- size alignment@ as the memory allocator function.
@@ -493,7 +405,7 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
493405-- /Pre-release/
494406{-# INLINE newArrayWith #-}
495407newArrayWith :: forall m a. (MonadIO m, Unbox a)
496− => (Int -> Int -> m MutByteArray) -> Int -> Int -> m (MutArray a)
408+ => (Int -> Int -> m MutableByteArray) -> Int -> Int -> m (MutArray a)
497409newArrayWith alloc alignSize count = do
498410 let size = max (count * SIZE_OF(a)) 0
499411 contents <- alloc size alignSize
@@ -504,25 +416,12 @@ newArrayWith alloc alignSize count = do
504416 , arrBound = size
505417 }
506418
507−-- For arrays "nil" sounds a bit odd. empty is better. The only problem with
508−-- empty is that it is also used by the Alternative type class. But assuming we
509−-- will mostly import the Array module qualified this should be fine.
510−
511−-- | Create an empty array.
512−empty ::
513−#ifdef DEVBUILD
514− Unbox a =>
515−#endif
516− MutArray a
517−empty = MutArray Unboxed.empty 0 0 0
518−
519−{-# DEPRECATED nil "Please use empty instead." #-}
520419nil ::
521420#ifdef DEVBUILD
522421 Unbox a =>
523422#endif
524423 MutArray a
525−nil = empty
424+nil = MutArray Unboxed.nil 0 0 0
526425
527426{-# INLINE newBytesAs #-}
528427newBytesAs :: MonadIO m =>
@@ -539,13 +438,12 @@ newBytesAs ps bytes = do
539438 , arrBound = bytes
540439 }
541440
542−-- | Allocates a pinned empty array that with a reserved capacity of bytes.
543−-- The memory of the array is uninitialized and the allocation is aligned as
544−-- per the 'Unboxed' instance of the type.
441+-- | Allocates a pinned empty array that can hold 'count' items. The memory of
442+-- the array is uninitialized and the allocation is aligned as per the
443+-- 'Unboxed' instance of the type.
545444--
546445-- /Pre-release/
547446{-# INLINE pinnedNewBytes #-}
548−{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
549447pinnedNewBytes :: MonadIO m =>
550448#ifdef DEVBUILD
551449 Unbox a =>
@@ -570,31 +468,21 @@ newAs ps =
570468 (error "new: alignment is not used in unpinned arrays.")
571469
572470-- XXX can unaligned allocation be more efficient when alignment is not needed?
573−
574−-- | Allocates a pinned array of zero length but growable to the specified
575−-- capacity without reallocation.
576−{-# INLINE pinnedEmptyOf #-}
577−pinnedEmptyOf :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
578−pinnedEmptyOf = newAs Pinned
579−
580−-- XXX Deprecate in major
581−-- {-# DEPRECATED pinnedNew "Please use pinnedEmptyOf instead." #-}
471+--
472+-- | Allocates an empty pinned array that can hold 'count' items. The memory of
473+-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
474+-- instance of the type.
475+--
582476{-# INLINE pinnedNew #-}
583477pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
584−pinnedNew = pinnedEmptyOf
478+pinnedNew = newAs Pinned
585479
586−-- | Allocates an unpinned array of zero length but growable to the specified
587−-- capacity without reallocation.
480+-- | Allocates an empty unpinned array that can hold 'count' items. The memory
481+-- of the array is uninitialized.
588482--
589−{-# INLINE emptyOf #-}
590−emptyOf :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
591−emptyOf = newAs Unpinned
592−
593−-- XXX Deprecate in major
594−-- {-# DEPRECATED new "Please use emptyOf instead." #-}
595483{-# INLINE new #-}
596484new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
597−new = emptyOf
485+new = newAs Unpinned
598486
599487-------------------------------------------------------------------------------
600488-- Random writes
@@ -610,7 +498,7 @@ putIndexUnsafe :: forall m a. (MonadIO m, Unbox a)
610498putIndexUnsafe i MutArray{..} x = do
611499 let index = INDEX_OF(arrStart, i, a)
612500 assert (i >= 0 && INDEX_VALID(index, arrEnd, a)) (return ())
613− liftIO $ pokeAt index arrContents x
501+ liftIO $ pokeByteIndex index arrContents x
614502
615503invalidIndex :: String -> Int -> a
616504invalidIndex label i =
@@ -628,7 +516,7 @@ putIndex :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> a -> m ()
628516putIndex i MutArray{..} x = do
629517 let index = INDEX_OF(arrStart,i,a)
630518 if i >= 0 && INDEX_VALID(index,arrEnd,a)
631− then liftIO $ pokeAt index arrContents x
519+ then liftIO $ pokeByteIndex index arrContents x
632520 else invalidIndex "putIndex" i
633521
634522-- | Write an input stream of (index, value) pairs to an array. Throws an
@@ -642,7 +530,7 @@ putIndices arr = FL.foldlM' step (return ())
642530
643531 where
644532
645− step () (i, x) = putIndex i arr x
533+ step () (i, x) = liftIO (putIndex i arr x)
646534
647535-- | Modify a given index of an array using a modifier function.
648536--
@@ -654,9 +542,9 @@ modifyIndexUnsafe :: forall m a b. (MonadIO m, Unbox a) =>
654542modifyIndexUnsafe i MutArray{..} f = liftIO $ do
655543 let index = INDEX_OF(arrStart,i,a)
656544 assert (i >= 0 && INDEX_NEXT(index,a) <= arrEnd) (return ())
657− r <- peekAt index arrContents
545+ r <- peekByteIndex index arrContents
658546 let (x, res) = f r
659− pokeAt index arrContents x
547+ pokeByteIndex index arrContents x
660548 return res
661549
662550-- | Modify a given index of an array using a modifier function.
@@ -668,9 +556,9 @@ modifyIndex i MutArray{..} f = do
668556 let index = INDEX_OF(arrStart,i,a)
669557 if i >= 0 && INDEX_VALID(index,arrEnd,a)
670558 then liftIO $ do
671− r <- peekAt index arrContents
559+ r <- peekByteIndex index arrContents
672560 let (x, res) = f r
673− pokeAt index arrContents x
561+ pokeByteIndex index arrContents x
674562 return res
675563 else invalidIndex "modifyIndex" i
676564
@@ -704,8 +592,8 @@ modify MutArray{..} f = liftIO $
704592
705593 go i =
706594 when (INDEX_VALID(i,arrEnd,a)) $ do
707− r <- peekAt i arrContents
708− pokeAt i arrContents (f r)
595+ r <- peekByteIndex i arrContents
596+ pokeByteIndex i arrContents (f r)
709597 go (INDEX_NEXT(i,a))
710598
711599-- XXX We could specify the number of bytes to swap instead of Proxy. Need
@@ -714,15 +602,15 @@ modify MutArray{..} f = liftIO $
714602swapArrayByteIndices ::
715603 forall a. Unbox a
716604 => Proxy a
717− -> MutByteArray
605+ -> MutableByteArray
718606 -> Int
719607 -> Int
720608 -> IO ()
721609swapArrayByteIndices _ arrContents i1 i2 = do
722− r1 <- peekAt i1 arrContents
723− r2 <- peekAt i2 arrContents
724− pokeAt i1 arrContents (r2 :: a)
725− pokeAt i2 arrContents (r1 :: a)
610+ r1 <- peekByteIndex i1 arrContents
611+ r2 <- peekByteIndex i2 arrContents
612+ pokeByteIndex i1 arrContents (r2 :: a)
613+ pokeByteIndex i2 arrContents (r1 :: a)
726614
727615-- | Swap the elements at two indices without validating the indices.
728616--
@@ -838,23 +726,26 @@ arrayChunkBytes = 1024
838726roundDownTo :: Int -> Int -> Int
839727roundDownTo elemSize size = size - (size `mod` elemSize)
840728
729+-- XXX See if resizing can be implemented by reading the old array as a stream
730+-- and then using writeN to the new array.
731+--
841732-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
842733-- Since this is not inlined Unboxed consrraint leads to dictionary passing
843734-- which complicates some inspection tests.
844735--
845−{-# NOINLINE reallocExplicitAs #-}
846−reallocExplicitAs :: PinnedState -> Int -> Int -> MutArray a -> IO (MutArray a)
847−reallocExplicitAs ps elemSize newCapacityInBytes MutArray{..} = do
736+{-# NOINLINE reallocExplicit #-}
737+reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
738+reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
848739 assertM(arrEnd <= arrBound)
849740
850741 -- Allocate new array
851742 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
852743 contents <-
853− if ps == Pinned
854− then Unboxed.pinnedNew newCapMaxInBytes
855− else Unboxed.new newCapMaxInBytes
856− let !(MutByteArray mbarrFrom#) = arrContents
857− !(MutByteArray mbarrTo#) = contents
744+ if Unboxed.isPinned arrContents
745+ then Unboxed.pinnedNewBytes newCapMaxInBytes
746+ else Unboxed.newBytes newCapMaxInBytes
747+ let !(MutableByteArray mbarrFrom#) = arrContents
748+ !(MutableByteArray mbarrTo#) = contents
858749
859750 -- Copy old data
860751 let oldStart = arrStart
@@ -875,10 +766,6 @@ reallocExplicitAs ps elemSize newCapacityInBytes MutArray{..} = do
875766 , arrBound = newCapInBytes
876767 }
877768
878−-- XXX Should these be called reallocBytes etc?
879−-- XXX We may also need reallocAs to allocate as pinned/unpinned explicitly. In
880−-- fact clone/pinnedClone can be implemented using reallocAs.
881−
882769-- | @realloc newCapacity array@ reallocates the array to the specified
883770-- capacity in bytes.
884771--
@@ -890,12 +777,7 @@ reallocExplicitAs ps elemSize newCapacityInBytes MutArray{..} = do
890777-- If the original array is pinned, the newly allocated array is also pinned.
891778{-# INLINABLE realloc #-}
892779realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
893−realloc bytes arr =
894− let ps =
895− if isPinned arr
896− then Pinned
897− else Unpinned
898− in liftIO $ reallocExplicitAs ps (SIZE_OF(a)) bytes arr
780+realloc bytes arr = liftIO $ reallocExplicit (SIZE_OF(a)) bytes arr
899781
900782-- | @reallocWith label capSizer minIncrBytes array@. The label is used
901783-- in error messages and the capSizer is used to determine the capacity of the
@@ -918,14 +800,14 @@ reallocWith label capSizer minIncrBytes arr = do
918800 where
919801
920802 badSize newSize =
921− Prelude.concat
803+ concat
922804 [ label
923805 , ": new array size (in bytes) is less than required size "
924806 , show newSize
925807 , ". Please check the sizing function passed."
926808 ]
927809
928−-- | @grow newCapacity array@ changes the total capacity of the array so that
810+-- | @resize newCapacity array@ changes the total capacity of the array so that
929811-- it is enough to hold the specified number of elements. Nothing is done if
930812-- the specified capacity is less than the length of the array.
931813--
@@ -933,46 +815,34 @@ reallocWith label capSizer minIncrBytes arr = do
933815-- the block size (4K).
934816--
935817-- /Pre-release/
936−{-# INLINE grow #-}
937−grow :: forall m a. (MonadIO m, Unbox a) =>
818+{-# INLINE resize #-}
819+resize :: forall m a. (MonadIO m, Unbox a) =>
938820 Int -> MutArray a -> m (MutArray a)
939−grow nElems arr@MutArray{..} = do
821+resize nElems arr@MutArray{..} = do
940822 let req = SIZE_OF(a) * nElems
941− cap = arrBound - arrStart
942− if req < cap
823+ len = arrEnd - arrStart
824+ if req < len
943825 then return arr
944826 else realloc req arr
945827
946−{-# DEPRECATED resize "Please use grow instead." #-}
947−{-# INLINE resize #-}
948−resize :: forall m a. (MonadIO m, Unbox a) =>
949− Int -> MutArray a -> m (MutArray a)
950−resize = grow
951−
952−-- | Like 'grow' but if the requested byte capacity is more than
953−-- 'largeObjectThreshold' then it is rounded up to the closest power of 2.
828+-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
829+-- then it is rounded up to the closest power of 2.
954830--
955831-- /Pre-release/
956−{-# INLINE growExp #-}
957−growExp :: forall m a. (MonadIO m, Unbox a) =>
832+{-# INLINE resizeExp #-}
833+resizeExp :: forall m a. (MonadIO m, Unbox a) =>
958834 Int -> MutArray a -> m (MutArray a)
959−growExp nElems arr@MutArray{..} = do
835+resizeExp nElems arr@MutArray{..} = do
960836 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
961837 req1 =
962838 if req > largeObjectThreshold
963839 then roundUpToPower2 req
964840 else req
965− cap = arrBound - arrStart
966− if req1 < cap
841+ len = arrEnd - arrStart
842+ if req1 < len
967843 then return arr
968844 else realloc req1 arr
969845
970−{-# DEPRECATED resizeExp "Please use growExp instead." #-}
971−{-# INLINE resizeExp #-}
972−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
973− Int -> MutArray a -> m (MutArray a)
974−resizeExp = growExp
975−
976846-- | Resize the allocated memory to drop any reserved free space at the end of
977847-- the array and reallocate it to reduce wastage.
978848--
@@ -1017,7 +887,7 @@ rightSize arr@MutArray{..} = do
1017887snocNewEnd :: (MonadIO m, Unbox a) => Int -> MutArray a -> a -> m (MutArray a)
1018888snocNewEnd newEnd arr@MutArray{..} x = liftIO $ do
1019889 assert (newEnd <= arrBound) (return ())
1020− pokeAt arrEnd arrContents x
890+ pokeByteIndex arrEnd arrContents x
1021891 return $ arr {arrEnd = newEnd}
1022892
1023893-- | Really really unsafe, appends the element into the first array, may
@@ -1037,7 +907,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
1037907{-# INLINE snocMay #-}
1038908snocMay :: forall m a. (MonadIO m, Unbox a) =>
1039909 MutArray a -> a -> m (Maybe (MutArray a))
1040−snocMay arr@MutArray{..} x = do
910+snocMay arr@MutArray{..} x = liftIO $ do
1041911 let newEnd = INDEX_NEXT(arrEnd,a)
1042912 if newEnd <= arrBound
1043913 then Just <$> snocNewEnd newEnd arr x
@@ -1051,7 +921,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
1051921 -> a
1052922 -> m (MutArray a)
1053923snocWithRealloc sizer arr x = do
1054− arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
924+ arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1055925 snocUnsafe arr1 x
1056926
1057927-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -1073,7 +943,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
1073943 -> MutArray a
1074944 -> a
1075945 -> m (MutArray a)
1076−snocWith allocSize arr x = do
946+snocWith allocSize arr x = liftIO $ do
1077947 let newEnd = INDEX_NEXT(arrEnd arr,a)
1078948 if newEnd <= arrBound arr
1079949 then snocNewEnd newEnd arr x
@@ -1117,126 +987,6 @@ snoc = snocWith f
1117987 then oldSize * 2
1118988 else roundUpToPower2 oldSize * 2
1119989
1120−-------------------------------------------------------------------------------
1121−-- Serialization/Deserialization
1122−-------------------------------------------------------------------------------
1123−
1124−{-# INLINE pokeNewEnd #-}
1125−pokeNewEnd :: (MonadIO m, Unbox a) =>
1126− Int -> MutArray Word8 -> a -> m (MutArray Word8)
1127−pokeNewEnd newEnd arr@MutArray{..} x = liftIO $ do
1128− assert (newEnd <= arrBound) (return ())
1129− liftIO $ pokeAt arrEnd arrContents x
1130− return $ arr {arrEnd = newEnd}
1131−
1132−-- | Really really unsafe, unboxes a Haskell type and appends the resulting
1133−-- bytes to the byte array, may cause silent data corruption or if you are
1134−-- lucky a segfault if the array does not have enough space to append the
1135−-- element.
1136−--
1137−-- /Internal/
1138−{-# INLINE pokeAppendUnsafe #-}
1139−pokeAppendUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1140− MutArray Word8 -> a -> m (MutArray Word8)
1141−pokeAppendUnsafe arr@MutArray{..} = pokeNewEnd (arrEnd + SIZE_OF(a)) arr
1142−
1143−-- | Skip the specified number of bytes in the array. The data in the skipped
1144−-- region remains uninitialzed.
1145−{-# INLINE pokeSkipUnsafe #-}
1146−pokeSkipUnsafe :: Int -> MutArray Word8 -> MutArray Word8
1147−pokeSkipUnsafe n arr@MutArray{..} = do
1148− let newEnd = arrEnd + n
1149− in assert (newEnd <= arrBound) (arr {arrEnd = newEnd})
1150−
1151−-- | Like 'pokeAppend' but does not grow the array when pre-allocated array
1152−-- capacity becomes full.
1153−--
1154−-- /Internal/
1155−{-# INLINE pokeAppendMay #-}
1156−pokeAppendMay :: forall m a. (MonadIO m, Unbox a) =>
1157− MutArray Word8 -> a -> m (Maybe (MutArray Word8))
1158−pokeAppendMay arr@MutArray{..} x = liftIO $ do
1159− let newEnd = arrEnd + SIZE_OF(a)
1160− if newEnd <= arrBound
1161− then Just <$> pokeNewEnd newEnd arr x
1162− else return Nothing
1163−
1164−{-# NOINLINE pokeWithRealloc #-}
1165−pokeWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
1166− (Int -> Int)
1167− -> MutArray Word8
1168− -> a
1169− -> m (MutArray Word8)
1170−pokeWithRealloc sizer arr x = do
1171− arr1 <- liftIO $ reallocWith "pokeWithRealloc" sizer (SIZE_OF(a)) arr
1172− pokeAppendUnsafe arr1 x
1173−
1174−{-# INLINE pokeAppendWith #-}
1175−pokeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
1176− (Int -> Int)
1177− -> MutArray Word8
1178− -> a
1179− -> m (MutArray Word8)
1180−pokeAppendWith allocSize arr x = liftIO $ do
1181− let newEnd = arrEnd arr + SIZE_OF(a)
1182− if newEnd <= arrBound arr
1183− then pokeNewEnd newEnd arr x
1184− else pokeWithRealloc allocSize arr x
1185−
1186−-- | Unbox a Haskell type and append the resulting bytes to a mutable byte
1187−-- array. The array is grown exponentially when more space is needed.
1188−--
1189−-- Definition:
1190−--
1191−-- >>> pokeAppend arr x = MutArray.castUnsafe <$> MutArray.snoc (MutArray.castUnsafe arr) x
1192−--
1193−{-# INLINE pokeAppend #-}
1194−pokeAppend :: forall m a. (MonadIO m, Unbox a) =>
1195− MutArray Word8 -> a -> m (MutArray Word8)
1196−pokeAppend = pokeAppendWith f
1197−
1198− where
1199−
1200− f oldSize =
1201− if isPower2 oldSize
1202− then oldSize * 2
1203− else roundUpToPower2 oldSize * 2
1204−
1205−-- | Really really unsafe, create a Haskell value from an unboxed byte array,
1206−-- does not check if the array is big enough, may return garbage or if you are
1207−-- lucky may cause a segfault.
1208−--
1209−-- /Internal/
1210−{-# INLINE peekUnconsUnsafe #-}
1211−peekUnconsUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1212− MutArray Word8 -> m (a, MutArray Word8)
1213−peekUnconsUnsafe MutArray{..} = do
1214− let start1 = arrStart + SIZE_OF(a)
1215− assert (start1 <= arrEnd) (return ())
1216− liftIO $ do
1217− r <- peekAt arrStart arrContents
1218− return (r, MutArray arrContents start1 arrEnd arrBound)
1219−
1220−-- | Discard the specified number of bytes in the array.
1221−{-# INLINE peekSkipUnsafe #-}
1222−peekSkipUnsafe :: Int -> MutArray Word8 -> MutArray Word8
1223−peekSkipUnsafe n MutArray{..} =
1224− let start1 = arrStart + n
1225− in assert (start1 <= arrEnd) (MutArray arrContents start1 arrEnd arrBound)
1226−
1227−-- | Create a Haskell value from its unboxed representation from the head of a
1228−-- byte array, return the value and the remaining array.
1229−{-# INLINE peekUncons #-}
1230−peekUncons :: forall m a. (MonadIO m, Unbox a) =>
1231− MutArray Word8 -> m (Maybe a, MutArray Word8)
1232−peekUncons arr@MutArray{..} = do
1233− let start1 = arrStart + SIZE_OF(a)
1234− if start1 > arrEnd
1235− then return (Nothing, arr)
1236− else liftIO $ do
1237− r <- peekAt arrStart arrContents
1238− return (Just r, MutArray arrContents start1 arrEnd arrBound)
1239−
1240990-------------------------------------------------------------------------------
1241991-- Random reads
1242992-------------------------------------------------------------------------------
@@ -1251,7 +1001,7 @@ getIndexUnsafe :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m a
12511001getIndexUnsafe i MutArray{..} = do
12521002 let index = INDEX_OF(arrStart,i,a)
12531003 assert (i >= 0 && INDEX_VALID(index,arrEnd,a)) (return ())
1254− liftIO $ peekAt index arrContents
1004+ liftIO $ peekByteIndex index arrContents
12551005
12561006-- | /O(1)/ Lookup the element at the given index. Index starts from 0.
12571007--
@@ -1260,7 +1010,7 @@ getIndex :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (Maybe a)
12601010getIndex i MutArray{..} = do
12611011 let index = INDEX_OF(arrStart,i,a)
12621012 if i >= 0 && INDEX_VALID(index,arrEnd,a)
1263− then liftIO $ Just <$> peekAt index arrContents
1013+ then liftIO $ Just <$> peekByteIndex index arrContents
12641014 else return Nothing
12651015
12661016-- | /O(1)/ Lookup the element at the given index from the end of the array.
@@ -1273,16 +1023,21 @@ getIndexRev :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m a
12731023getIndexRev i MutArray{..} = do
12741024 let index = RINDEX_OF(arrEnd,i,a)
12751025 if i >= 0 && index >= arrStart
1276− then liftIO $ peekAt index arrContents
1026+ then liftIO $ peekByteIndex index arrContents
12771027 else invalidIndex "getIndexRev" i
12781028
12791029data GetIndicesState contents start end st =
12801030 GetIndicesState contents start end st
12811031
1282−{-# INLINE indexReaderWith #-}
1283−indexReaderWith :: (Monad m, Unbox a) =>
1032+-- | Given an unfold that generates array indices, read the elements on those
1033+-- indices from the supplied MutArray. An error is thrown if an index is out of
1034+-- bounds.
1035+--
1036+-- /Pre-release/
1037+{-# INLINE getIndicesD #-}
1038+getIndicesD :: (Monad m, Unbox a) =>
12841039 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1285−indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
1040+getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
12861041
12871042 where
12881043
@@ -1301,26 +1056,9 @@ indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
13011056 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
13021057 D.Stop -> return D.Stop
13031058
1304−{-# DEPRECATED getIndicesWith "Please use indexReaderWith instead." #-}
1305−{-# INLINE getIndicesWith #-}
1306−getIndicesWith :: (Monad m, Unbox a) =>
1307− (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1308−getIndicesWith = indexReaderWith
1309−
1310−-- | Given an unfold that generates array indices, read the elements on those
1311−-- indices from the supplied MutArray. An error is thrown if an index is out of
1312−-- bounds.
1313−--
1314−-- /Pre-release/
1315−{-# INLINE indexReader #-}
1316−indexReader :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1317−indexReader = indexReaderWith liftIO
1318−
1319−-- XXX DO NOT REMOVE, change the signature to use Stream instead of unfold
1320−{-# DEPRECATED getIndices "Please use indexReader instead." #-}
13211059{-# INLINE getIndices #-}
13221060getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1323−getIndices = indexReader
1061+getIndices = getIndicesD liftIO
13241062
13251063-------------------------------------------------------------------------------
13261064-- Subarrays
@@ -1425,7 +1163,7 @@ partitionBy f arr@MutArray{..} = liftIO $ do
14251163
14261164 -- Invariant low < high on entry, and on return as well
14271165 moveHigh low high = do
1428− h <- peekAt high arrContents
1166+ h <- peekByteIndex high arrContents
14291167 if f h
14301168 then
14311169 -- Correctly classified, continue the loop
@@ -1445,7 +1183,7 @@ partitionBy f arr@MutArray{..} = liftIO $ do
14451183 -- low <= high
14461184 -- Both low and high are valid locations within the array
14471185 go low high = do
1448− l <- peekAt low arrContents
1186+ l <- peekByteIndex low arrContents
14491187 if f l
14501188 then
14511189 -- low is wrongly classified
@@ -1456,8 +1194,8 @@ partitionBy f arr@MutArray{..} = liftIO $ do
14561194 case r of
14571195 Nothing -> return low
14581196 Just (high1, h) -> do -- low < high1
1459− pokeAt low arrContents h
1460− pokeAt high1 arrContents l
1197+ pokeByteIndex low arrContents h
1198+ pokeByteIndex high1 arrContents l
14611199 let low1 = INDEX_NEXT(low,a)
14621200 high2 = INDEX_PREV(high1,a)
14631201 if low1 <= high2
@@ -1591,7 +1329,7 @@ chunksOfAs ps n (D.Stream step state) =
15911329 r <- step (adaptState gst) st
15921330 case r of
15931331 D.Yield x s -> do
1594− liftIO $ pokeAt end contents x
1332+ liftIO $ pokeByteIndex end contents x
15951333 let end1 = INDEX_NEXT(end,a)
15961334 return $
15971335 if end1 >= bound
@@ -1610,12 +1348,10 @@ chunksOfAs ps n (D.Stream step state) =
16101348
16111349 step' _ GroupFinish = return D.Stop
16121350
1613−-- | @chunksOf n stream@ groups the elements in the input stream into arrays of
1614−-- @n@ elements each.
1351+-- | @chunksOf n stream@ groups the input stream into a stream of
1352+-- arrays of size n.
16151353--
1616−-- Same as the following but may be more efficient:
1617−--
1618−-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
1354+-- @chunksOf n = StreamD.foldMany (MutArray.writeN n)@
16191355--
16201356-- /Pre-release/
16211357{-# INLINE_NORMAL chunksOf #-}
@@ -1624,29 +1360,17 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
16241360-- XXX the idiomatic implementation leads to large regression in the D.reverse'
16251361-- benchmark. It seems it has difficulty producing optimized code when
16261362-- converting to StreamK. Investigate GHC optimizations.
1627−-- chunksOf n = D.foldMany (createOf n)
1363+-- chunksOf n = D.foldMany (writeN n)
16281364chunksOf = chunksOfAs Unpinned
16291365
16301366-- | Like 'chunksOf' but creates pinned arrays.
16311367{-# INLINE_NORMAL pinnedChunksOf #-}
16321368pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
16331369 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1634−-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
1370+-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
16351371pinnedChunksOf = chunksOfAs Pinned
16361372
1637−-- | When we are buffering a stream of unknown size into an array we do not
1638−-- know how much space to pre-allocate. So we start with the min size and emit
1639−-- the array then keep on doubling the size every time. Thus we do not need to
1640−-- guess the optimum chunk size.
1641−--
1642−-- We can incorporate this in chunksOfAs if the additional size parameter does
1643−-- not impact perf.
1644−--
1645−{-# INLINE _chunksOfRange #-}
1646−_chunksOfRange :: -- (MonadIO m, Unbox a) =>
1647− PinnedState -> Int -> Int -> D.Stream m a -> D.Stream m (MutArray a)
1648−_chunksOfRange _ps _low _hi = undefined
1649−
1373+-- XXX This should take a PinnedState
16501374-- XXX buffer to a list instead?
16511375-- | Buffer the stream into arrays in memory.
16521376{-# INLINE arrayStreamKFromStreamDAs #-}
@@ -1664,10 +1388,16 @@ data FlattenState s contents a =
16641388 OuterLoop s
16651389 | InnerLoop s contents !Int !Int
16661390
1667−{-# INLINE_NORMAL concatWith #-}
1668−concatWith :: forall m a. (Monad m, Unbox a)
1669− => (forall b. IO b -> m b) -> D.Stream m (MutArray a) -> D.Stream m a
1670−concatWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
1391+-- | Use the "reader" unfold instead.
1392+--
1393+-- @flattenArrays = unfoldMany reader@
1394+--
1395+-- We can try this if there are any fusion issues in the unfold.
1396+--
1397+{-# INLINE_NORMAL flattenArrays #-}
1398+flattenArrays :: forall m a. (MonadIO m, Unbox a)
1399+ => D.Stream m (MutArray a) -> D.Stream m a
1400+flattenArrays (D.Stream step state) = D.Stream step' (OuterLoop state)
16711401
16721402 where
16731403
@@ -1684,30 +1414,19 @@ concatWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
16841414 return $ D.Skip $ OuterLoop st
16851415
16861416 step' _ (InnerLoop st contents p end) = do
1687− x <- liftio $ peekAt p contents
1417+ x <- liftIO $ peekByteIndex p contents
16881418 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
16891419
1690−-- | Use the "reader" unfold instead.
1420+-- | Use the "readerRev" unfold instead.
16911421--
1692−-- @concat = unfoldMany reader@
1422+-- @flattenArrays = unfoldMany readerRev@
16931423--
16941424-- We can try this if there are any fusion issues in the unfold.
16951425--
1696−{-# INLINE_NORMAL concat #-}
1697−concat :: forall m a. (MonadIO m, Unbox a)
1698− => D.Stream m (MutArray a) -> D.Stream m a
1699−concat = concatWith liftIO
1700−
1701−{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1702−{-# INLINE flattenArrays #-}
1703−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1426+{-# INLINE_NORMAL flattenArraysRev #-}
1427+flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
17041428 => D.Stream m (MutArray a) -> D.Stream m a
1705−flattenArrays = concat
1706−
1707−{-# INLINE_NORMAL concatRevWith #-}
1708−concatRevWith :: forall m a. (Monad m, Unbox a)
1709− => (forall b. IO b -> m b) -> D.Stream m (MutArray a) -> D.Stream m a
1710−concatRevWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
1429+flattenArraysRev (D.Stream step state) = D.Stream step' (OuterLoop state)
17111430
17121431 where
17131432
@@ -1725,33 +1444,16 @@ concatRevWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
17251444 return $ D.Skip $ OuterLoop st
17261445
17271446 step' _ (InnerLoop st contents p start) = do
1728− x <- liftio $ peekAt p contents
1447+ x <- liftIO $ peekByteIndex p contents
17291448 let cur = INDEX_PREV(p,a)
17301449 return $ D.Yield x (InnerLoop st contents cur start)
17311450
1732−-- | Use the "readerRev" unfold instead.
1733−--
1734−-- @concat = unfoldMany readerRev@
1735−--
1736−-- We can try this if there are any fusion issues in the unfold.
1737−--
1738−{-# INLINE_NORMAL concatRev #-}
1739−concatRev :: forall m a. (MonadIO m, Unbox a)
1740− => D.Stream m (MutArray a) -> D.Stream m a
1741−concatRev = concatRevWith liftIO
1742−
1743−{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744−{-# INLINE flattenArraysRev #-}
1745−flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
1746− => D.Stream m (MutArray a) -> D.Stream m a
1747−flattenArraysRev = concatRev
1748−
17491451-------------------------------------------------------------------------------
17501452-- Unfolds
17511453-------------------------------------------------------------------------------
17521454
17531455data ArrayUnsafe a = ArrayUnsafe
1754− {-# UNPACK #-} !MutByteArray -- contents
1456+ {-# UNPACK #-} !MutableByteArray -- contents
17551457 {-# UNPACK #-} !Int -- index 1
17561458 {-# UNPACK #-} !Int -- index 2
17571459
@@ -1781,7 +1483,7 @@ producerWith liftio = Producer step (return . toArrayUnsafe) extract
17811483 -- few actions for correctness and execution order sanity. We want
17821484 -- the peek to occur right here and not lazily at some later point
17831485 -- because we want the peek to be ordered with respect to the touch.
1784− !x <- liftio $ peekAt cur contents
1486+ !x <- liftio $ peekByteIndex cur contents
17851487 return $ D.Yield x (ArrayUnsafe contents (INDEX_NEXT(cur,a)) end)
17861488
17871489 extract = return . fromArrayUnsafe
@@ -1812,7 +1514,7 @@ readerRevWith liftio = Unfold step inject
18121514 {-# INLINE_LATE step #-}
18131515 step (ArrayUnsafe _ start p) | p < start = return D.Stop
18141516 step (ArrayUnsafe contents start p) = do
1815− !x <- liftio $ peekAt p contents
1517+ !x <- liftio $ peekByteIndex p contents
18161518 return $ D.Yield x (ArrayUnsafe contents start (INDEX_PREV(p,a)))
18171519
18181520-- | Unfold an array into a stream in reverse order.
@@ -1843,7 +1545,7 @@ toListFB c n MutArray{..} = go arrStart
18431545 -- evaluated/written to before we peek at them.
18441546 -- XXX
18451547 let !x = unsafeInlineIO $ do
1846− r <- peekAt arrContents p
1548+ r <- peekByteIndex arrContents p
18471549 return r
18481550 in c x (go (PTR_NEXT(p,a)))
18491551-}
@@ -1860,21 +1562,21 @@ toList MutArray{..} = liftIO $ go arrStart
18601562
18611563 go p | assert (p <= arrEnd) (p == arrEnd) = return []
18621564 go p = do
1863− x <- peekAt p arrContents
1565+ x <- peekByteIndex p arrContents
18641566 (:) x <$> go (INDEX_NEXT(p,a))
18651567
1866−{-# INLINE_NORMAL toStreamWith #-}
1867−toStreamWith ::
1568+{-# INLINE_NORMAL toStreamDWith #-}
1569+toStreamDWith ::
18681570 forall m a. (Monad m, Unbox a)
18691571 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1870−toStreamWith liftio MutArray{..} = D.Stream step arrStart
1572+toStreamDWith liftio MutArray{..} = D.Stream step arrStart
18711573
18721574 where
18731575
18741576 {-# INLINE_LATE step #-}
18751577 step _ p | assert (p <= arrEnd) (p == arrEnd) = return D.Stop
18761578 step _ p = liftio $ do
1877− r <- peekAt p arrContents
1579+ r <- peekByteIndex p arrContents
18781580 return $ D.Yield r (INDEX_NEXT(p,a))
18791581
18801582-- | Convert a 'MutArray' into a stream.
@@ -1883,7 +1585,7 @@ toStreamWith liftio MutArray{..} = D.Stream step arrStart
18831585--
18841586{-# INLINE_NORMAL read #-}
18851587read :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1886−read = toStreamWith liftIO
1588+read = toStreamDWith liftIO
18871589
18881590{-# INLINE toStreamKWith #-}
18891591toStreamKWith ::
@@ -1895,18 +1597,18 @@ toStreamKWith liftio MutArray{..} = go arrStart
18951597
18961598 go p | assert (p <= arrEnd) (p == arrEnd) = K.nil
18971599 | otherwise =
1898− let elemM = peekAt p arrContents
1600+ let elemM = peekByteIndex p arrContents
18991601 in liftio elemM `K.consM` go (INDEX_NEXT(p,a))
19001602
19011603{-# INLINE toStreamK #-}
19021604toStreamK :: forall m a. (MonadIO m, Unbox a) => MutArray a -> StreamK m a
19031605toStreamK = toStreamKWith liftIO
19041606
1905−{-# INLINE_NORMAL toStreamRevWith #-}
1906−toStreamRevWith ::
1607+{-# INLINE_NORMAL toStreamDRevWith #-}
1608+toStreamDRevWith ::
19071609 forall m a. (Monad m, Unbox a)
19081610 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1909−toStreamRevWith liftio MutArray{..} =
1611+toStreamDRevWith liftio MutArray{..} =
19101612 let p = INDEX_PREV(arrEnd,a)
19111613 in D.Stream step p
19121614
@@ -1915,7 +1617,7 @@ toStreamRevWith liftio MutArray{..} =
19151617 {-# INLINE_LATE step #-}
19161618 step _ p | p < arrStart = return D.Stop
19171619 step _ p = liftio $ do
1918− r <- peekAt p arrContents
1620+ r <- peekByteIndex p arrContents
19191621 return $ D.Yield r (INDEX_PREV(p,a))
19201622
19211623-- | Convert a 'MutArray' into a stream in reverse order.
@@ -1924,7 +1626,7 @@ toStreamRevWith liftio MutArray{..} =
19241626--
19251627{-# INLINE_NORMAL readRev #-}
19261628readRev :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1927−readRev = toStreamRevWith liftIO
1629+readRev = toStreamDRevWith liftIO
19281630
19291631{-# INLINE toStreamKRevWith #-}
19301632toStreamKRevWith ::
@@ -1938,7 +1640,7 @@ toStreamKRevWith liftio MutArray {..} =
19381640
19391641 go p | p < arrStart = K.nil
19401642 | otherwise =
1941− let elemM = peekAt p arrContents
1643+ let elemM = peekByteIndex p arrContents
19421644 in liftio elemM `K.consM` go (INDEX_PREV(p,a))
19431645
19441646{-# INLINE toStreamKRev #-}
@@ -1973,7 +1675,7 @@ foldr f z arr = D.foldr f z $ read arr
19731675
19741676-- XXX Keep the bound intact to not lose any free space? Perf impact?
19751677
1976−-- | @unsafeAppendN n arr@ appends up to @n@ input items to the supplied
1678+-- | @writeAppendNUnsafe n arr@ appends up to @n@ input items to the supplied
19771679-- array.
19781680--
19791681-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1982,12 +1684,13 @@ foldr f z arr = D.foldr f z $ read arr
19821684-- Any free space left in the array after appending @n@ elements is lost.
19831685--
19841686-- /Internal/
1985−{-# INLINE_NORMAL unsafeAppendN #-}
1986−unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
1687+{-# INLINE_NORMAL writeAppendNUnsafe #-}
1688+writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
19871689 Int
19881690 -> m (MutArray a)
19891691 -> Fold m a (MutArray a)
1990−unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
1692+writeAppendNUnsafe n action =
1693+ fmap fromArrayUnsafe $ FL.foldlM' step initial
19911694
19921695 where
19931696
@@ -2000,59 +1703,38 @@ unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
20001703 -- otherwise we lose that space.
20011704 arr1 <-
20021705 if free < needed
2003− then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
1706+ then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
20041707 else return arr
20051708 return $ toArrayUnsafe arr1
20061709
20071710 step (ArrayUnsafe contents start end) x = do
2008− liftIO $ pokeAt end contents x
1711+ liftIO $ pokeByteIndex end contents x
20091712 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
20101713
2011−{-# DEPRECATED writeAppendNUnsafe "Please use unsafeAppendN instead." #-}
2012−{-# INLINE writeAppendNUnsafe #-}
2013−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
2014− Int
2015− -> m (MutArray a)
2016− -> Fold m a (MutArray a)
2017−writeAppendNUnsafe = unsafeAppendN
2018−
20191714-- | Append @n@ elements to an existing array. Any free space left in the array
20201715-- after appending @n@ elements is lost.
20211716--
2022−-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
1717+-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
20231718--
2024−{-# INLINE_NORMAL appendN #-}
2025−appendN :: forall m a. (MonadIO m, Unbox a) =>
2026− Int -> m (MutArray a) -> Fold m a (MutArray a)
2027−appendN n initial = FL.take n (unsafeAppendN n initial)
2028−
2029−-- XXX Deprecate in major
2030−-- {-# DEPRECATED writeAppendN "Please use appendN instead." #-}
2031−{-# INLINE writeAppendN #-}
1719+{-# INLINE_NORMAL writeAppendN #-}
20321720writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
20331721 Int -> m (MutArray a) -> Fold m a (MutArray a)
2034−writeAppendN = appendN
1722+writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
20351723
2036−-- | @appendWith realloc action@ mutates the array generated by @action@ to
1724+-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
20371725-- append the input stream. If there is no reserved space available in the
20381726-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
20391727-- where @oldSize@ is the current size of the array in bytes.
20401728--
20411729-- Note that the returned array may be a mutated version of original array.
20421730--
2043−-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
1731+-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
20441732--
20451733-- /Pre-release/
2046−{-# INLINE appendWith #-}
2047−appendWith :: forall m a. (MonadIO m, Unbox a) =>
2048− (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
2049−appendWith sizer = FL.foldlM' (snocWith sizer)
2050−
2051−{-# DEPRECATED writeAppendWith "Please use appendWith instead." #-}
20521734{-# INLINE writeAppendWith #-}
20531735writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
20541736 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
2055−writeAppendWith = appendWith
1737+writeAppendWith sizer = FL.foldlM' (snocWith sizer)
20561738
20571739-- | @append action@ mutates the array generated by @action@ to append the
20581740-- input stream. If there is no reserved space available in the array it is
@@ -2060,101 +1742,70 @@ writeAppendWith = appendWith
20601742--
20611743-- Note that the returned array may be a mutated version of original array.
20621744--
2063−-- >>> append = MutArray.appendWith (* 2)
1745+-- >>> writeAppend = MutArray.writeAppendWith (* 2)
20641746--
2065−{-# INLINE append #-}
2066−append :: forall m a. (MonadIO m, Unbox a) =>
2067− m (MutArray a) -> Fold m a (MutArray a)
2068−append = appendWith (* 2)
2069−
2070−-- XXX Deprecate in major
2071−-- {-# DEPRECATED writeAppend "Please use append instead." #-}
20721747{-# INLINE writeAppend #-}
20731748writeAppend :: forall m a. (MonadIO m, Unbox a) =>
20741749 m (MutArray a) -> Fold m a (MutArray a)
2075−writeAppend = append
1750+writeAppend = writeAppendWith (* 2)
20761751
20771752-- XXX We can carry bound as well in the state to make sure we do not lose the
20781753-- remaining capacity. Need to check perf impact.
20791754--
2080−-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081−-- function as argument.
1755+-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1756+-- as argument.
20821757--
2083−-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
1758+-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
20841759--
20851760-- /Pre-release/
2086−{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087−unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
1761+{-# INLINE_NORMAL writeNWithUnsafe #-}
1762+writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
20881763 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2089−unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
1764+writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
20901765
20911766 where
20921767
20931768 initial = toArrayUnsafe <$> alloc (max n 0)
20941769
20951770 step (ArrayUnsafe contents start end) x = do
2096− liftIO $ pokeAt end contents x
1771+ liftIO $ pokeByteIndex end contents x
20971772 return
20981773 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
20991774
2100−{-# DEPRECATED writeNWithUnsafe "Please use unsafeCreateOfWith instead." #-}
2101−{-# INLINE writeNWithUnsafe #-}
2102−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2103− => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2104−writeNWithUnsafe = unsafeCreateOfWith
2105−
21061775{-# INLINE_NORMAL writeNUnsafeAs #-}
21071776writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
21081777 => PinnedState -> Int -> Fold m a (MutArray a)
2109−writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
1778+writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
21101779
2111−-- | Like 'createOf' but does not check the array bounds when writing. The fold
1780+-- | Like 'writeN' but does not check the array bounds when writing. The fold
21121781-- driver must not call the step function more than 'n' times otherwise it will
21131782-- corrupt the memory and crash. This function exists mainly because any
21141783-- conditional in the step function blocks fusion causing 10x performance
21151784-- slowdown.
21161785--
2117−-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
1786+-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
21181787--
2119−{-# INLINE_NORMAL unsafeCreateOf #-}
2120−unsafeCreateOf :: forall m a. (MonadIO m, Unbox a)
2121− => Int -> Fold m a (MutArray a)
2122−unsafeCreateOf = writeNUnsafeAs Unpinned
2123−
2124−{-# DEPRECATED writeNUnsafe "Please use unsafeCreateOf instead." #-}
2125−{-# INLINE writeNUnsafe #-}
1788+{-# INLINE_NORMAL writeNUnsafe #-}
21261789writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
21271790 => Int -> Fold m a (MutArray a)
2128−writeNUnsafe = unsafeCreateOf
1791+writeNUnsafe = writeNUnsafeAs Unpinned
21291792
2130−-- | Like 'unsafeCreateOf' but creates a pinned array.
2131−{-# INLINE_NORMAL unsafePinnedCreateOf #-}
2132−unsafePinnedCreateOf :: forall m a. (MonadIO m, Unbox a)
2133− => Int -> Fold m a (MutArray a)
2134−unsafePinnedCreateOf = writeNUnsafeAs Pinned
2135−
2136−{-# DEPRECATED pinnedWriteNUnsafe "Please use unsafePinnedCreateOf instead." #-}
2137−{-# INLINE pinnedWriteNUnsafe #-}
1793+-- | Like 'writeNUnsafe' but creates a pinned array.
1794+{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
21381795pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
21391796 => Int -> Fold m a (MutArray a)
2140−pinnedWriteNUnsafe = unsafePinnedCreateOf
1797+pinnedWriteNUnsafe = writeNUnsafeAs Pinned
21411798
2142−-- | @createOfWith alloc n@ folds a maximum of @n@ elements into an array
1799+-- | @writeNWith alloc n@ folds a maximum of @n@ elements into an array
21431800-- allocated using the @alloc@ function.
21441801--
2145−-- >>> createOfWith alloc n = Fold.take n (MutArray.unsafeCreateOfWith alloc n)
2146−-- >>> createOfWith alloc n = MutArray.appendN (alloc n) n
1802+-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1803+-- >>> writeNWith alloc n = MutArray.writeAppendN (alloc n) n
21471804--
2148−{-# INLINE_NORMAL createOfWith #-}
2149−createOfWith :: forall m a. (MonadIO m, Unbox a)
2150− => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2151−createOfWith alloc n = FL.take n (unsafeCreateOfWith alloc n)
2152−
2153−-- {-# DEPRECATED writeNWith "Please use createOfWith instead." #-}
2154−{-# INLINE writeNWith #-}
1805+{-# INLINE_NORMAL writeNWith #-}
21551806writeNWith :: forall m a. (MonadIO m, Unbox a)
21561807 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2157−writeNWith = createOfWith
1808+writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
21581809
21591810{-# INLINE_NORMAL writeNAs #-}
21601811writeNAs ::
@@ -2162,42 +1813,28 @@ writeNAs ::
21621813 => PinnedState
21631814 -> Int
21641815 -> Fold m a (MutArray a)
2165−writeNAs ps = createOfWith (newAs ps)
1816+writeNAs ps = writeNWith (newAs ps)
21661817
2167−-- | @createOf n@ folds a maximum of @n@ elements from the input stream to an
1818+-- | @writeN n@ folds a maximum of @n@ elements from the input stream to an
21681819-- 'MutArray'.
21691820--
2170−-- >>> createOf = MutArray.createOfWith MutArray.new
2171−-- >>> createOf n = Fold.take n (MutArray.unsafeCreateOf n)
2172−-- >>> createOf n = MutArray.appendN n (MutArray.emptyOf n)
1821+-- >>> writeN = MutArray.writeNWith MutArray.new
1822+-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1823+-- >>> writeN n = MutArray.writeAppendN n (MutArray.new n)
21731824--
2174−{-# INLINE_NORMAL createOf #-}
2175−createOf :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2176−createOf = writeNAs Unpinned
2177−
2178−-- XXX Deprecate in major
2179−-- {-# DEPRECATED writeN "Please use createOf instead." #-}
2180−{-# INLINE writeN #-}
1825+{-# INLINE_NORMAL writeN #-}
21811826writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2182−writeN = createOf
2183−
2184−-- | Like 'createOf' but creates a pinned array.
2185−{-# INLINE_NORMAL pinnedCreateOf #-}
2186−pinnedCreateOf ::
2187− forall m a. (MonadIO m, Unbox a)
2188− => Int
2189− -> Fold m a (MutArray a)
2190−pinnedCreateOf = writeNAs Pinned
1827+writeN = writeNAs Unpinned
21911828
2192−{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193−{-# INLINE pinnedWriteN #-}
1829+-- | Like 'writeN' but creates a pinned array.
1830+{-# INLINE_NORMAL pinnedWriteN #-}
21941831pinnedWriteN ::
21951832 forall m a. (MonadIO m, Unbox a)
21961833 => Int
21971834 -> Fold m a (MutArray a)
2198−pinnedWriteN = pinnedCreateOf
1835+pinnedWriteN = writeNAs Pinned
21991836
2200−-- | Like unsafeCreateOfWith but writes the array in reverse order.
1837+-- | Like writeNWithUnsafe but writes the array in reverse order.
22011838--
22021839-- /Internal/
22031840{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -2214,11 +1851,11 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
22141851
22151852 step (ArrayUnsafe contents start end) x = do
22161853 let ptr = INDEX_PREV(start,a)
2217− liftIO $ pokeAt ptr contents x
1854+ liftIO $ pokeByteIndex ptr contents x
22181855 return
22191856 $ ArrayUnsafe contents ptr end
22201857
2221−-- | Like createOfWith but writes the array in reverse order.
1858+-- | Like writeNWith but writes the array in reverse order.
22221859--
22231860-- /Internal/
22241861{-# INLINE_NORMAL writeRevNWith #-}
@@ -2226,36 +1863,31 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
22261863 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
22271864writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
22281865
2229−-- | Like 'createOf' but writes the array in reverse order.
1866+-- | Like writeN but writes the array in reverse order.
22301867--
22311868-- /Pre-release/
2232−{-# INLINE_NORMAL revCreateOf #-}
2233−revCreateOf :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2234−revCreateOf = writeRevNWith new
2235−
2236−{-# DEPRECATED writeRevN "Please use revCreateOf instead." #-}
2237−{-# INLINE writeRevN #-}
1869+{-# INLINE_NORMAL writeRevN #-}
22381870writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2239−writeRevN = revCreateOf
1871+writeRevN = writeRevNWith new
22401872
22411873-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
22421874-- input stream to a 'MutArray' aligned to the given size.
22431875--
2244−-- >>> pinnedWriteNAligned align = MutArray.createOfWith (MutArray.pinnedNewAligned align)
2245−-- >>> pinnedWriteNAligned align n = MutArray.appendN n (MutArray.pinnedNewAligned align n)
1876+-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1877+-- >>> pinnedWriteNAligned align n = MutArray.writeAppendN n (MutArray.pinnedNewAligned align n)
22461878--
22471879-- /Pre-release/
22481880--
22491881{-# INLINE_NORMAL pinnedWriteNAligned #-}
22501882pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
22511883 => Int -> Int -> Fold m a (MutArray a)
2252−pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
1884+pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
22531885
22541886-- XXX Buffer to a list instead?
2255−
1887+--
22561888-- | Buffer a stream into a stream of arrays.
22571889--
2258−-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
1890+-- >>> writeChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
22591891--
22601892-- Breaking an array into an array stream can be useful to consume a large
22611893-- array sequentially such that memory of the array is released incrementatlly.
@@ -2264,43 +1896,37 @@ pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
22641896--
22651897-- /Unimplemented/
22661898--
2267−{-# INLINE_NORMAL buildChunks #-}
2268−buildChunks :: (MonadIO m, Unbox a) =>
2269− Int -> Fold m a (StreamK n (MutArray a))
2270−buildChunks n = FL.many (createOf n) FL.toStreamK
2271−
2272−{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273−{-# INLINE writeChunks #-}
1899+{-# INLINE_NORMAL writeChunks #-}
22741900writeChunks :: (MonadIO m, Unbox a) =>
22751901 Int -> Fold m a (StreamK n (MutArray a))
2276−writeChunks = buildChunks
1902+writeChunks n = FL.many (writeN n) FL.toStreamK
22771903
22781904{-# INLINE_NORMAL writeWithAs #-}
22791905writeWithAs :: forall m a. (MonadIO m, Unbox a)
22801906 => PinnedState -> Int -> Fold m a (MutArray a)
2281−-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
1907+-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
22821908writeWithAs ps elemCount =
22831909 FL.rmapM extract $ FL.foldlM' step initial
22841910
22851911 where
22861912
22871913 initial = do
2288− when (elemCount < 0) $ error "createWith: elemCount is negative"
2289− newAs ps elemCount
1914+ when (elemCount < 0) $ error "writeWith: elemCount is negative"
1915+ liftIO $ newAs ps elemCount
22901916
22911917 step arr@(MutArray _ start end bound) x
22921918 | INDEX_NEXT(end,a) > bound = do
22931919 let oldSize = end - start
22941920 newSize = max (oldSize * 2) 1
2295− arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
1921+ arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
22961922 snocUnsafe arr1 x
22971923 step arr x = snocUnsafe arr x
22981924
22991925 extract = liftIO . rightSize
23001926
2301−-- XXX Compare createWith with fromStreamD which uses an array of streams
2302−-- implementation. We can write this using buildChunks above if that is faster.
2303−-- If createWith is faster then we should use that to implement
1927+-- XXX Compare writeWith with fromStreamD which uses an array of streams
1928+-- implementation. We can write this using writeChunks above if that is faster.
1929+-- If writeWith is faster then we should use that to implement
23041930-- fromStreamD.
23051931--
23061932-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -2315,56 +1941,39 @@ writeWithAs ps elemCount =
23151941-- XXX check if GHC's memory allocator is efficient enough. We can try the C
23161942-- malloc to compare against.
23171943
2318−-- | @createWith minCount@ folds the whole input to a single array. The array
1944+-- | @writeWith minCount@ folds the whole input to a single array. The array
23191945-- starts at a size big enough to hold minCount elements, the size is doubled
23201946-- every time the array needs to be grown.
23211947--
23221948-- /Caution! Do not use this on infinite streams./
23231949--
2324−-- >>> f n = MutArray.appendWith (* 2) (MutArray.emptyOf n)
2325−-- >>> createWith n = Fold.rmapM MutArray.rightSize (f n)
2326−-- >>> createWith n = Fold.rmapM MutArray.fromChunksK (MutArray.buildChunks n)
1950+-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
1951+-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
1952+-- >>> writeWith n = Fold.rmapM MutArray.fromArrayStreamK (MutArray.writeChunks n)
23271953--
23281954-- /Pre-release/
2329−{-# INLINE_NORMAL createWith #-}
2330−createWith :: forall m a. (MonadIO m, Unbox a)
2331− => Int -> Fold m a (MutArray a)
2332−-- createWith n = FL.rmapM rightSize $ appendWith (* 2) (emptyOf n)
2333−createWith = writeWithAs Unpinned
2334−
2335−{-# DEPRECATED writeWith "Please use createWith instead." #-}
2336−{-# INLINE writeWith #-}
1955+{-# INLINE_NORMAL writeWith #-}
23371956writeWith :: forall m a. (MonadIO m, Unbox a)
23381957 => Int -> Fold m a (MutArray a)
2339−writeWith = createWith
1958+-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
1959+writeWith = writeWithAs Unpinned
23401960
23411961-- | Fold the whole input to a single array.
23421962--
2343−-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
1963+-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
23441964-- rounded up to the element size.
23451965--
23461966-- /Caution! Do not use this on infinite streams./
23471967--
2348−{-# INLINE create #-}
2349−create :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2350−create = createWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2351−
2352−-- XXX Deprecate in major
2353−-- {-# DEPRECATED write "Please use create instead." #-}
23541968{-# INLINE write #-}
23551969write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2356−write = create
1970+write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
23571971
2358−-- | Like 'create' but creates a pinned array.
2359−{-# INLINE pinnedCreate #-}
2360−pinnedCreate :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2361−pinnedCreate =
2362− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2363−
2364−{-# DEPRECATED pinnedWrite "Please use pinnedCreate instead." #-}
1972+-- | Like 'write' but creates a pinned array.
23651973{-# INLINE pinnedWrite #-}
23661974pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2367−pinnedWrite = pinnedCreate
1975+pinnedWrite =
1976+ writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
23681977
23691978-------------------------------------------------------------------------------
23701979-- construct from streams, known size
@@ -2374,31 +1983,25 @@ pinnedWrite = pinnedCreate
23741983fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
23751984 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
23761985fromStreamDNAs ps limit str = do
2377− (arr :: MutArray a) <- newAs ps limit
1986+ (arr :: MutArray a) <- liftIO $ newAs ps limit
23781987 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
23791988 return $ arr {arrEnd = end}
23801989
23811990 where
23821991
23831992 fwrite arrContents ptr x = do
2384− liftIO $ pokeAt ptr arrContents x
1993+ liftIO $ pokeByteIndex ptr arrContents x
23851994 return $ INDEX_NEXT(ptr,a)
23861995
2387−-- | Use the 'createOf' fold instead.
1996+-- | Use the 'writeN' fold instead.
23881997--
2389−-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
1998+-- >>> fromStreamDN n = Stream.fold (MutArray.writeN n)
23901999--
2391−{-# INLINE_NORMAL fromStreamN #-}
2392−fromStreamN :: forall m a. (MonadIO m, Unbox a)
2393− => Int -> D.Stream m a -> m (MutArray a)
2394−-- fromStreamDN n = D.fold (createOf n)
2395−fromStreamN = fromStreamDNAs Unpinned
2396−
2397−{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398−{-# INLINE fromStreamDN #-}
2000+{-# INLINE_NORMAL fromStreamDN #-}
23992001fromStreamDN :: forall m a. (MonadIO m, Unbox a)
24002002 => Int -> D.Stream m a -> m (MutArray a)
2401−fromStreamDN = fromStreamN
2003+-- fromStreamDN n = D.fold (writeN n)
2004+fromStreamDN = fromStreamDNAs Unpinned
24022005
24032006-- | Create a 'MutArray' from the first N elements of a list. The array is
24042007-- allocated to size N, if the list terminates before N elements then the
@@ -2418,70 +2021,16 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
24182021-- /Pre-release/
24192022{-# INLINE fromListRevN #-}
24202023fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2421−fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
2422−
2423−-- | Convert a pure stream in Identity monad to a mutable array.
2424−{-# INLINABLE fromPureStreamN #-}
2425−fromPureStreamN :: (MonadIO m, Unbox a) =>
2426− Int -> Stream Identity a -> m (MutArray a)
2427−fromPureStreamN n xs =
2428− D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
2024+fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
24292025
24302026-- | Convert a pure stream in Identity monad to a mutable array.
24312027{-# INLINABLE fromPureStream #-}
24322028fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
24332029fromPureStream xs =
2434− D.fold create $ D.morphInner (return . runIdentity) xs
2435−
2436−{-# INLINABLE fromPtrN #-}
2437−fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
2438−fromPtrN len addr = do
2439− -- memcpy is better than stream copy when the size is known.
2440− -- XXX We can implement a stream copy in a similar way by streaming Word64
2441− -- first and then remaining Word8.
2442− arr <- new len
2443− _ <- unsafeAsPtr arr
2444− (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
2445− return (arr {arrEnd = len})
2446−
2447−{-# INLINABLE fromByteStr# #-}
2448−fromByteStr# :: MonadIO m => Addr# -> m (MutArray Word8)
2449−fromByteStr# addr = do
2450− -- It is better to count the size first and allocate exact space.
2451− -- Also, memcpy is better than stream copy when the size is known.
2452− -- C strlen compares 4 bytes at a time, so is better than the stream
2453− -- version. https://github.com/bminor/glibc/blob/master/string/strlen.c
2454− -- XXX We can possibly use a stream of Word64 to do the same.
2455− -- fromByteStr# addr = fromPureStream (D.fromByteStr# addr)
2456− len <- liftIO $ c_strlen (Ptr addr)
2457− let lenInt = fromIntegral len
2458− arr <- new lenInt
2459− _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2460− return (arr {arrEnd = lenInt})
2030+ liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
24612031
24622032-------------------------------------------------------------------------------
2463−-- convert a stream of arrays to a single array by reallocating and copying
2464−-------------------------------------------------------------------------------
2465−
2466−-- XXX Both of these implementations of splicing seem to perform equally well.
2467−-- We need to perform benchmarks over a range of sizes though.
2468−
2469−-- | Also see 'fromChunksK'.
2470−{-# INLINE fromChunksRealloced #-}
2471−fromChunksRealloced :: forall m a. (MonadIO m, Unbox a)
2472− => Stream m (MutArray a) -> m (MutArray a)
2473−fromChunksRealloced s = do
2474− res <- D.uncons s
2475− case res of
2476− Just (a, strm) -> do
2477− arr <- D.foldlM' spliceExp (pure a) strm
2478− -- Reallocation is exponential so there may be 50% empty space in
2479− -- worst case. One more reallocation to reclaim the space.
2480− rightSize arr
2481− Nothing -> pure nil
2482−
2483−-------------------------------------------------------------------------------
2484−-- convert a stream of arrays to a single array by buffering arrays first
2033+-- convert stream to a single array
24852034-------------------------------------------------------------------------------
24862035
24872036{-# INLINE arrayStreamKLength #-}
@@ -2491,72 +2040,34 @@ arrayStreamKLength as = K.foldl' (+) 0 (K.map length as)
24912040-- | Convert an array stream to an array. Note that this requires peak memory
24922041-- that is double the size of the array stream.
24932042--
2494−{-# INLINE fromChunkskAs #-}
2495−fromChunkskAs :: (Unbox a, MonadIO m) =>
2496− PinnedState -> StreamK m (MutArray a) -> m (MutArray a)
2497−fromChunkskAs ps as = do
2498− len <- arrayStreamKLength as
2499− arr <- newAs ps len
2500− -- XXX is StreamK fold faster or StreamD fold?
2501− K.foldlM' spliceUnsafe (pure arr) as
2502− -- fromStreamDN len $ D.unfoldMany reader $ D.fromStreamK as
2503−
2504−-- XXX Need to compare this with fromChunks and fromChunkList and keep the
2505−-- fastest or simplest one if all are equally fast.
2506−
2507−-- | Convert an array stream to an array. Note that this requires peak memory
2508−-- that is double the size of the array stream.
2509−--
2510−-- Also see 'fromChunksRealloced'.
2511−--
2512−{-# INLINE fromChunksK #-}
2513−fromChunksK :: (Unbox a, MonadIO m) =>
2514− StreamK m (MutArray a) -> m (MutArray a)
2515−fromChunksK = fromChunkskAs Unpinned
2516−
2517−{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
25182043{-# INLINE fromArrayStreamK #-}
25192044fromArrayStreamK :: (Unbox a, MonadIO m) =>
25202045 StreamK m (MutArray a) -> m (MutArray a)
2521−fromArrayStreamK = fromChunksK
2046+fromArrayStreamK as = do
2047+ len <- arrayStreamKLength as
2048+ fromStreamDN len $ D.unfoldMany reader $ D.fromStreamK as
25222049
25232050{-# INLINE fromStreamDAs #-}
25242051fromStreamDAs ::
25252052 (MonadIO m, Unbox a) => PinnedState -> D.Stream m a -> m (MutArray a)
2526−fromStreamDAs ps m =
2527− arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
2053+fromStreamDAs ps m = arrayStreamKFromStreamDAs ps m >>= fromArrayStreamK
25282054
2529−-- | Create an 'Array' from a stream. This is useful when we want to create a
2530−-- single array from a stream of unknown size. 'createOf' is at least twice
2531−-- as efficient when the size is already known.
2532−--
2533−-- Note that if the input stream is too large memory allocation for the array
2534−-- may fail. When the stream size is not known, `chunksOf` followed by
2535−-- processing of indvidual arrays in the resulting stream should be preferred.
2536−--
2537−-- /Pre-release/
2538−{-# INLINE fromStream #-}
2539−fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
2540−fromStream = fromStreamDAs Unpinned
2541−
2542−-- fromStream (Stream m) = P.fold create m
25432055-- CAUTION: a very large number (millions) of arrays can degrade performance
25442056-- due to GC overhead because we need to buffer the arrays before we flatten
25452057-- all the arrays.
25462058--
2547−-- XXX Compare if this is faster or "fold create".
2059+-- XXX Compare if this is faster or "fold write".
25482060--
25492061-- | We could take the approach of doubling the memory allocation on each
25502062-- overflow. This would result in more or less the same amount of copying as in
25512063-- the chunking approach. However, if we have to shrink in the end then it may
25522064-- result in an extra copy of the entire data.
25532065--
2554−-- >>> fromStreamD = StreamD.fold MutArray.create
2066+-- >>> fromStreamD = StreamD.fold MutArray.write
25552067--
25562068{-# INLINE fromStreamD #-}
2557−{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
25582069fromStreamD :: (MonadIO m, Unbox a) => D.Stream m a -> m (MutArray a)
2559−fromStreamD = fromStream
2070+fromStreamD = fromStreamDAs Unpinned
25602071
25612072-- | Create a 'MutArray' from a list. The list must be of finite size.
25622073--
@@ -2581,44 +2092,22 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
25812092-- Cloning
25822093-------------------------------------------------------------------------------
25832094
2584−-- Arrays are aligned on 64-bit boundaries. The fastest way to copy an array is
2585−-- to unsafeCast it to Word64, read it, write it to Word64 array and unsafeCast
2586−-- it again. We can use SIMD read/write as well.
2587−
25882095{-# INLINE cloneAs #-}
2589−cloneAs ::
2590− ( MonadIO m
2591−#ifdef DEVBUILD
2592− , Unbox a
2593−#endif
2594− )
2595− => PinnedState -> MutArray a -> m (MutArray a)
2096+cloneAs :: MonadIO m => PinnedState -> MutArray a -> m (MutArray a)
25962097cloneAs ps src =
2597− do
2098+ liftIO $ do
25982099 let startSrc = arrStart src
25992100 srcLen = arrEnd src - startSrc
2600− newArrContents <-
2601− Unboxed.cloneSliceUnsafeAs ps startSrc srcLen (arrContents src)
2101+ newArrContents <- Unboxed.newBytesAs ps srcLen
2102+ putSliceUnsafe (arrContents src) startSrc newArrContents 0 srcLen
26022103 return $ MutArray newArrContents 0 srcLen srcLen
26032104
26042105{-# INLINE clone #-}
2605−clone ::
2606− ( MonadIO m
2607−#ifdef DEVBUILD
2608− , Unbox a
2609−#endif
2610− )
2611− => MutArray a -> m (MutArray a)
2106+clone :: MonadIO m => MutArray a -> m (MutArray a)
26122107clone = cloneAs Unpinned
26132108
26142109{-# INLINE pinnedClone #-}
2615−pinnedClone ::
2616− ( MonadIO m
2617−#ifdef DEVBUILD
2618− , Unbox a
2619−#endif
2620− )
2621− => MutArray a -> m (MutArray a)
2110+pinnedClone :: MonadIO m => MutArray a -> m (MutArray a)
26222111pinnedClone = cloneAs Pinned
26232112
26242113-------------------------------------------------------------------------------
@@ -2633,16 +2122,17 @@ spliceCopy :: forall m a. MonadIO m =>
26332122 Unbox a =>
26342123#endif
26352124 MutArray a -> MutArray a -> m (MutArray a)
2636−spliceCopy arr1 arr2 = do
2125+spliceCopy arr1 arr2 = liftIO $ do
26372126 let start1 = arrStart arr1
26382127 start2 = arrStart arr2
26392128 len1 = arrEnd arr1 - start1
26402129 len2 = arrEnd arr2 - start2
2641− let len = len1 + len2
2130+ let newLen = len1 + len2
26422131 newArrContents <-
26432132 if Unboxed.isPinned (arrContents arr1)
2644− then liftIO $ Unboxed.pinnedNew len
2645− else liftIO $ Unboxed.new len
2133+ then Unboxed.pinnedNewBytes newLen
2134+ else Unboxed.newBytes newLen
2135+ let len = len1 + len2
26462136 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
26472137 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
26482138 return $ MutArray newArrContents 0 len len
@@ -2654,7 +2144,7 @@ spliceCopy arr1 arr2 = do
26542144spliceUnsafe :: MonadIO m =>
26552145 MutArray a -> MutArray a -> m (MutArray a)
26562146spliceUnsafe dst src =
2657− do
2147+ liftIO $ do
26582148 let startSrc = arrStart src
26592149 srcLen = arrEnd src - startSrc
26602150 endDst = arrEnd dst
@@ -2676,7 +2166,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
26762166 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
26772167spliceWith sizer dst@(MutArray _ start end bound) src = do
26782168{-
2679− let f = appendWith (`sizer` byteLength src) (return dst)
2169+ let f = writeAppendWith (`sizer` byteLength src) (return dst)
26802170 in D.fold f (toStreamD src)
26812171-}
26822172 assert (end <= bound) (return ())
@@ -2692,7 +2182,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
26922182 $ "splice: newSize is less than the total size "
26932183 ++ "of arrays being appended. Please check the "
26942184 ++ "sizer function passed."
2695− realloc newSizeInBytes dst
2185+ liftIO $ realloc newSizeInBytes dst
26962186 else return dst
26972187 spliceUnsafe dst1 src
26982188
@@ -2729,24 +2219,14 @@ spliceExp = spliceWith (\l1 l2 -> max (l1 * 2) (l1 + l2))
27292219-- Splitting
27302220-------------------------------------------------------------------------------
27312221
2732−-- | Generate a stream of array slices using a predicate. The array element
2733−-- matching the predicate is dropped.
2734−--
2735−-- /Pre-release/
2736−{-# INLINE splitOn #-}
2737−splitOn :: (MonadIO m, Unbox a) =>
2738− (a -> Bool) -> MutArray a -> Stream m (MutArray a)
2739−splitOn predicate arr =
2740− fmap (\(i, len) -> getSliceUnsafe i len arr)
2741− $ D.indexOnSuffix predicate (read arr)
2742−
27432222-- | Drops the separator byte
27442223{-# INLINE breakOn #-}
27452224breakOn :: MonadIO m
27462225 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2747−breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
2226+breakOn sep arr@MutArray{..} = asPtrUnsafe arr $ \p -> liftIO $ do
2227+ -- XXX Instead of using asPtrUnsafe (pinning memory) we can pass unlifted
2228+ -- Addr# to memchr and it should be safe (from ghc 8.4).
27482229 -- XXX We do not need memchr here, we can use a Haskell equivalent.
2749− -- Need efficient stream based primitives that work on Word64.
27502230 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
27512231 let sepIndex = loc `minusPtr` p
27522232 return $
@@ -2767,41 +2247,32 @@ breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
27672247 }
27682248 )
27692249
2770−-- | Like 'splitAt' but does not check whether the index is valid.
2771−--
2772−{-# INLINE unsafeSplitAt #-}
2773−unsafeSplitAt :: forall a. Unbox a =>
2774− Int -> MutArray a -> (MutArray a, MutArray a)
2775−unsafeSplitAt i MutArray{..} =
2776− let off = i * SIZE_OF(a)
2777− p = arrStart + off
2778− in ( MutArray
2779− { arrContents = arrContents
2780− , arrStart = arrStart
2781− , arrEnd = p
2782− , arrBound = p
2783− }
2784− , MutArray
2785− { arrContents = arrContents
2786− , arrStart = p
2787− , arrEnd = arrEnd
2788− , arrBound = arrBound
2789− }
2790− )
2791−
27922250-- | Create two slices of an array without copying the original array. The
27932251-- specified index @i@ is the first index of the second slice.
27942252--
2795−{-# INLINE splitAt #-}
27962253splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2797−splitAt i arr =
2254+splitAt i arr@MutArray{..} =
27982255 let maxIndex = length arr - 1
27992256 in if i < 0
28002257 then error "sliceAt: negative array index"
28012258 else if i > maxIndex
28022259 then error $ "sliceAt: specified array index " ++ show i
28032260 ++ " is beyond the maximum index " ++ show maxIndex
2804− else unsafeSplitAt i arr
2261+ else let off = i * SIZE_OF(a)
2262+ p = arrStart + off
2263+ in ( MutArray
2264+ { arrContents = arrContents
2265+ , arrStart = arrStart
2266+ , arrEnd = p
2267+ , arrBound = p
2268+ }
2269+ , MutArray
2270+ { arrContents = arrContents
2271+ , arrStart = p
2272+ , arrEnd = arrEnd
2273+ , arrBound = arrBound
2274+ }
2275+ )
28052276
28062277-------------------------------------------------------------------------------
28072278-- Casting
@@ -2851,13 +2322,9 @@ cast arr =
28512322-- guarantees that unsafe calls will be performed in the calling thread. Making
28522323-- it safe to pass heap-allocated objects to unsafe functions.
28532324
2854−-- XXX Should we just name it asPtr, the unsafety is implicit for any pointer
2325+-- Should we just name it asPtr, the unsafety is implicit for any pointer
28552326-- operations. And we are safe from Haskell perspective because we will be
28562327-- pinning the memory.
2857−--
2858−-- XXX we cannot pass the length of the ptr here as in some cases it may not be
2859−-- available e.g. a null terminated C string. However, we can create another
2860−-- flavor of the API e.g. asPtrN.
28612328
28622329-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
28632330-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2875,377 +2342,48 @@ cast arr =
28752342--
28762343-- /Pre-release/
28772344--
2878−{-# INLINE unsafePinnedAsPtr #-}
2879−unsafePinnedAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2880−unsafePinnedAsPtr arr f =
2881− Unboxed.unsafePinnedAsPtr
2882− (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2883−
2884−{-# DEPRECATED asPtrUnsafe "Please use unsafePinnedAsPtr instead." #-}
28852345{-# INLINE asPtrUnsafe #-}
28862346asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2887−asPtrUnsafe = unsafePinnedAsPtr
2888−
2889−{-# INLINE unsafeAsPtr #-}
2890−unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891−unsafeAsPtr arr f =
2892− Unboxed.unsafeAsPtr
2347+asPtrUnsafe arr f =
2348+ Unboxed.asPtrUnsafe
28932349 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
28942350
28952351-------------------------------------------------------------------------------
28962352-- Equality
28972353-------------------------------------------------------------------------------
28982354
2899−-- | Byte compare two arrays. Compare the length of the arrays. If the length
2900−-- is equal, compare the lexicographical ordering of two underlying byte arrays
2901−-- otherwise return the result of length comparison.
2902−--
2903−-- /Unsafe/: Note that the 'Unbox' instance of sum types with constructors of
2904−-- different sizes may leave some memory uninitialized which can make byte
2905−-- comparison unreliable.
2355+-- | Compare the length of the arrays. If the length is equal, compare the
2356+-- lexicographical ordering of two underlying byte arrays otherwise return the
2357+-- result of length comparison.
29062358--
29072359-- /Pre-release/
2908−{-# INLINE byteCmp #-}
2909−byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2910−byteCmp arr1 arr2 = do
2911− let marr1 = getMutableByteArray# (arrContents arr1)
2912− marr2 = getMutableByteArray# (arrContents arr2)
2913− !(I# st1#) = arrStart arr1
2914− !(I# st2#) = arrStart arr2
2915− !(I# len#) = byteLength arr1
2916− case compare (byteLength arr1) (byteLength arr2) of
2917− EQ -> do
2918− r <- liftIO $ IO $ \s# ->
2919− let res =
2920− I#
2921− (compareByteArrays#
2922− (unsafeCoerce# marr1)
2923− st1#
2924− (unsafeCoerce# marr2)
2925− st2#
2926− len#)
2927− in (# s#, res #)
2928− return $ compare r 0
2929− x -> return x
2930−
29312360{-# INLINE cmp #-}
2932−{-# DEPRECATED cmp "Please use byteCmp instead." #-}
29332361cmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2934−cmp = byteCmp
2935−
2936−-- | Byte equality of two arrays.
2937−--
2938−-- >>> byteEq arr1 arr2 = (==) EQ $ MArray.byteCmp arr1 arr2
2939−--
2940−-- /Unsafe/: See 'byteCmp'.
2941−{-# INLINE byteEq #-}
2942−byteEq :: MonadIO m => MutArray a -> MutArray a -> m Bool
2943−byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
2944−
2945−-------------------------------------------------------------------------------
2946−-- Compact
2947−-------------------------------------------------------------------------------
2948−
2949−-- Note: LE versions avoid an extra copy compared to GE. LE parser trades
2950−-- backtracking one array in lieu of avoiding a copy. However, LE and GE both
2951−-- can leave some memory unused. They can split the last array to fit it
2952−-- exactly in the space.
2953−
2954−{-# INLINE_NORMAL pCompactLeAs #-}
2955−pCompactLeAs ::
2956− forall m a. (MonadIO m, Unbox a)
2957− => PinnedState -> Int -> Parser (MutArray a) m (MutArray a)
2958−pCompactLeAs ps maxElems = Parser step initial extract
2959−
2960− where
2961−
2962− maxBytes = maxElems * SIZE_OF(a)
2963−
2964− functionName = "Streamly.Internal.Data.MutArray.pCompactLE"
2965−
2966− initial =
2967− return
2968− $ if maxElems <= 0
2969− then error
2970− $ functionName
2971− ++ ": the size of arrays ["
2972− ++ show maxElems ++ "] must be a natural number"
2973− else Parser.IPartial Nothing
2974−
2975− step Nothing arr =
2976− return
2977− $ let len = byteLength arr
2978− in if len >= maxBytes
2979− then Parser.Done 0 arr
2980− else Parser.Partial 0 (Just arr)
2981− -- XXX Split the last array to use the space more compactly.
2982− step (Just buf) arr =
2983− let len = byteLength buf + byteLength arr
2984− in if len > maxBytes
2985− then return $ Parser.Done 1 buf
2986− else do
2987− buf1 <-
2988− if byteCapacity buf < maxBytes
2989− then liftIO $ reallocExplicitAs
2990− ps (SIZE_OF(a)) maxBytes buf
2991− else return buf
2992− buf2 <- spliceUnsafe buf1 arr
2993− return $ Parser.Partial 0 (Just buf2)
2994−
2995− extract Nothing = return $ Parser.Done 0 nil
2996− extract (Just buf) = return $ Parser.Done 0 buf
2997−
2998−-- | Parser @pCompactLE maxElems@ coalesces adjacent arrays in the input stream
2999−-- only if the combined size would be less than or equal to @maxElems@
3000−-- elements. Note that it won't split an array if the original array is already
3001−-- larger than maxElems.
3002−--
3003−-- @maxElems@ must be greater than 0.
3004−--
3005−-- Generates unpinned arrays irrespective of the pinning status of input
3006−-- arrays.
3007−--
3008−-- /Internal/
3009−{-# INLINE pCompactLE #-}
3010−pCompactLE ::
3011− forall m a. (MonadIO m, Unbox a)
3012− => Int -> Parser (MutArray a) m (MutArray a)
3013−pCompactLE = pCompactLeAs Unpinned
3014−
3015−-- | Pinned version of 'pCompactLE'.
3016−{-# INLINE pPinnedCompactLE #-}
3017−pPinnedCompactLE ::
3018− forall m a. (MonadIO m, Unbox a)
3019− => Int -> Parser (MutArray a) m (MutArray a)
3020−pPinnedCompactLE = pCompactLeAs Pinned
3021−
3022−data SpliceState s arr
3023− = SpliceInitial s
3024− | SpliceBuffering s arr
3025− | SpliceYielding arr (SpliceState s arr)
3026− | SpliceFinish
3027−
3028−-- This mutates the first array (if it has space) to append values from the
3029−-- second one. This would work for immutable arrays as well because an
3030−-- immutable array never has additional space so a new array is allocated
3031−-- instead of mutating it.
3032−
3033−{-# INLINE_NORMAL compactLeAs #-}
3034−compactLeAs :: forall m a. (MonadIO m, Unbox a)
3035− => PinnedState -> Int -> D.Stream m (MutArray a) -> D.Stream m (MutArray a)
3036−compactLeAs ps maxElems (D.Stream step state) =
3037− D.Stream step' (SpliceInitial state)
3038−
3039− where
3040−
3041− maxBytes = maxElems * SIZE_OF(a)
3042−
3043− functionName = "Streamly.Internal.Data.MutArray.rCompactLE"
3044−
3045− {-# INLINE_LATE step' #-}
3046− step' gst (SpliceInitial st) = do
3047− when (maxElems <= 0) $
3048− -- XXX we can pass the module string from the higher level API
3049− error $ functionName ++ ": the size of arrays [" ++ show maxElems
3050− ++ "] must be a natural number"
3051− r <- step gst st
3052− case r of
3053− D.Yield arr s -> return $
3054− let len = byteLength arr
3055− in if len >= maxBytes
3056− then D.Skip (SpliceYielding arr (SpliceInitial s))
3057− else D.Skip (SpliceBuffering s arr)
3058− D.Skip s -> return $ D.Skip (SpliceInitial s)
3059− D.Stop -> return D.Stop
3060−
3061− -- XXX Split the last array to use the space more compactly.
3062− step' gst (SpliceBuffering st buf) = do
3063− r <- step gst st
3064− case r of
3065− D.Yield arr s -> do
3066− let len = byteLength buf + byteLength arr
3067− if len > maxBytes
3068− then return $
3069− D.Skip (SpliceYielding buf (SpliceBuffering s arr))
3070− else do
3071− buf1 <- if byteCapacity buf < maxBytes
3072− then liftIO $ reallocExplicitAs
3073− ps (SIZE_OF(a)) maxBytes buf
3074− else return buf
3075− buf2 <- spliceUnsafe buf1 arr
3076− return $ D.Skip (SpliceBuffering s buf2)
3077− D.Skip s -> return $ D.Skip (SpliceBuffering s buf)
3078− D.Stop -> return $ D.Skip (SpliceYielding buf SpliceFinish)
3079−
3080− step' _ SpliceFinish = return D.Stop
3081−
3082− step' _ (SpliceYielding arr next) = return $ D.Yield arr next
3083−
3084−
3085−{-# INLINE_NORMAL fCompactGeAs #-}
3086−fCompactGeAs ::
3087− forall m a. (MonadIO m, Unbox a)
3088− => PinnedState -> Int -> FL.Fold m (MutArray a) (MutArray a)
3089−fCompactGeAs ps minElems = Fold step initial extract extract
3090−
3091− where
3092−
3093− minBytes = minElems * SIZE_OF(a)
3094−
3095− functionName = "Streamly.Internal.Data.MutArray.fCompactGE"
3096−
3097− initial =
3098− return
3099− $ if minElems < 0
3100− then error
3101− $ functionName
3102− ++ ": the size of arrays ["
3103− ++ show minElems ++ "] must be a natural number"
3104− else FL.Partial Nothing
3105−
3106− step Nothing arr =
3107− return
3108− $ let len = byteLength arr
3109− in if len >= minBytes
3110− then FL.Done arr
3111− else FL.Partial (Just arr)
3112− -- XXX Buffer arrays as a list to avoid copy and reallocations
3113− step (Just buf) arr = do
3114− let len = byteLength buf + byteLength arr
3115− buf1 <-
3116− if byteCapacity buf < len
3117− then liftIO $ reallocExplicitAs
3118− ps (SIZE_OF(a)) (max minBytes len) buf
3119− else return buf
3120− buf2 <- spliceUnsafe buf1 arr
3121− if len >= minBytes
3122− then return $ FL.Done buf2
3123− else return $ FL.Partial (Just buf2)
3124−
3125− extract Nothing = return nil
3126− extract (Just buf) = return buf
3127−
3128−-- | Fold @fCompactGE minElems@ coalesces adjacent arrays in the input stream
3129−-- until the size becomes greater than or equal to @minElems@.
3130−--
3131−-- Generates unpinned arrays irrespective of the pinning status of input
3132−-- arrays.
3133−{-# INLINE fCompactGE #-}
3134−fCompactGE ::
3135− forall m a. (MonadIO m, Unbox a)
3136− => Int -> FL.Fold m (MutArray a) (MutArray a)
3137−fCompactGE = fCompactGeAs Unpinned
3138−
3139−-- | Pinned version of 'fCompactGE'.
3140−{-# INLINE fPinnedCompactGE #-}
3141−fPinnedCompactGE ::
3142− forall m a. (MonadIO m, Unbox a)
3143− => Int -> FL.Fold m (MutArray a) (MutArray a)
3144−fPinnedCompactGE = fCompactGeAs Pinned
3145−
3146−{-# INLINE_NORMAL lCompactGeAs #-}
3147−lCompactGeAs :: forall m a. (MonadIO m, Unbox a)
3148− => PinnedState -> Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3149−-- The fold version turns out to be a little bit slower.
3150−-- lCompactGeAs ps n = FL.many (fCompactGeAs ps n)
3151−lCompactGeAs ps minElems (Fold step1 initial1 _ final1) =
3152− Fold step initial extract final
3153−
3154− where
3155−
3156− minBytes = minElems * SIZE_OF(a)
3157−
3158− functionName = "Streamly.Internal.Data.MutArray.lCompactGE"
3159−
3160− initial = do
3161− when (minElems <= 0) $
3162− -- XXX we can pass the module string from the higher level API
3163− error $ functionName ++ ": the size of arrays ["
3164− ++ show minElems ++ "] must be a natural number"
3165−
3166− r <- initial1
3167− return $ first (Tuple' Nothing) r
3168−
3169− {-# INLINE runInner #-}
3170− runInner len acc buf =
3171− if len >= minBytes
3172− then do
3173− r <- step1 acc buf
3174− case r of
3175− FL.Done _ -> return $ FL.Done ()
3176− FL.Partial s -> do
3177− _ <- final1 s
3178− res <- initial1
3179− return $ first (Tuple' Nothing) res
3180− else return $ FL.Partial $ Tuple' (Just buf) acc
3181−
3182− step (Tuple' Nothing r1) arr =
3183− runInner (byteLength arr) r1 arr
3184−
3185− -- XXX Buffer arrays as a list to avoid copy and reallocations
3186− step (Tuple' (Just buf) r1) arr = do
3187− let len = byteLength buf + byteLength arr
3188− buf1 <- if byteCapacity buf < len
3189− then liftIO $ reallocExplicitAs
3190− ps (SIZE_OF(a)) (max minBytes len) buf
3191− else return buf
3192− buf2 <- spliceUnsafe buf1 arr
3193− runInner len r1 buf2
3194−
3195− -- XXX Several folds do extract >=> final, therefore, we need to make final
3196− -- return "m b" rather than using extract post it if we want extract to be
3197− -- partial.
3198− --
3199− -- extract forces the pending buffer to be sent to the fold which is not
3200− -- what we want.
3201− extract _ = error "lCompactGE: not designed for scanning"
3202−
3203− final (Tuple' Nothing r1) = final1 r1
3204− final (Tuple' (Just buf) r1) = do
3205− r <- step1 r1 buf
3206− case r of
3207− FL.Partial rr -> final1 rr
3208− FL.Done _ -> return ()
3209−
3210−-- | Like 'compactGE' but for transforming folds instead of stream.
3211−--
3212−-- >>> lCompactGE n = Fold.many (MutArray.fCompactGE n)
3213−--
3214−-- Generates unpinned arrays irrespective of the pinning status of input
3215−-- arrays.
3216−{-# INLINE lCompactGE #-}
3217−lCompactGE :: forall m a. (MonadIO m, Unbox a)
3218− => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3219−lCompactGE = lCompactGeAs Unpinned
3220−
3221−-- | Pinned version of 'lCompactGE'.
3222−{-# INLINE lPinnedCompactGE #-}
3223−lPinnedCompactGE :: forall m a. (MonadIO m, Unbox a)
3224− => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3225−lPinnedCompactGE = lCompactGeAs Pinned
3226−
3227−-- | @compactGE n stream@ coalesces adjacent arrays in the @stream@ until
3228−-- the size becomes greater than or equal to @n@.
3229−--
3230−-- >>> compactGE n = Stream.foldMany (MutArray.fCompactGE n)
3231−--
3232−{-# INLINE compactGE #-}
3233−compactGE ::
3234− (MonadIO m, Unbox a)
3235− => Int -> Stream m (MutArray a) -> Stream m (MutArray a)
3236−compactGE n = D.foldMany (fCompactGE n)
3237−
3238−-- | 'compactEQ n' coalesces adajacent arrays in the input stream to
3239−-- arrays of exact size @n@.
3240−--
3241−-- /Unimplemented/
3242−{-# INLINE compactEQ #-}
3243−compactEQ :: -- (MonadIO m, Unbox a) =>
3244− Int -> Stream m (MutArray a) -> Stream m (MutArray a)
3245−compactEQ _n = undefined -- D.parseManyD (pCompactEQ n)
2362+cmp arr1 arr2 =
2363+ liftIO
2364+ $ do
2365+ let marr1 = getMutableByteArray# (arrContents arr1)
2366+ marr2 = getMutableByteArray# (arrContents arr2)
2367+ !(I# st1#) = arrStart arr1
2368+ !(I# st2#) = arrStart arr2
2369+ !(I# len#) = byteLength arr1
2370+ case compare (byteLength arr1) (byteLength arr2) of
2371+ EQ -> do
2372+ r <- liftIO $ IO $ \s# ->
2373+ let res =
2374+ I#
2375+ (compareByteArrays#
2376+ (unsafeCoerce# marr1)
2377+ st1#
2378+ (unsafeCoerce# marr2)
2379+ st2#
2380+ len#)
2381+ in (# s#, res #)
2382+ return $ compare r 0
2383+ x -> return x
32462384
32472385-------------------------------------------------------------------------------
3248−-- In-place mutation algorithms
2386+-- NFData
32492387-------------------------------------------------------------------------------
32502388
32512389-- | Strip elements which match with predicate from both ends.
@@ -3274,7 +2412,7 @@ strip eq arr@MutArray{..} = liftIO $ do
32742412 getStart cur = do
32752413 if cur < arrEnd
32762414 then do
3277− r <- peekAt cur arrContents
2415+ r <- peekByteIndex cur arrContents
32782416 if eq r
32792417 then getStart (INDEX_NEXT(cur,a))
32802418 else return cur
@@ -3284,7 +2422,7 @@ strip eq arr@MutArray{..} = liftIO $ do
32842422 if cur > low
32852423 then do
32862424 let prev = INDEX_PREV(cur,a)
3287− r <- peekAt prev arrContents
2425+ r <- peekByteIndex prev arrContents
32882426 if eq r
32892427 then getLast prev low
32902428 else return cur