@@ -28,56 +28,80 @@
2828
2929module Streamly.Internal.Data.MutArray.Type
3030 (
31− -- * Type
31+ -- ** Type
3232 -- $arrayNotes
3333 MutArray (..)
34− , MutableByteArray
34+
35+ -- ** Conversion
36+ -- *** Pinned and Unpinned
3537 , pin
3638 , unpin
3739 , isPinned
3840
39− -- * Constructing and Writing
41+ -- ** Casting
42+ , cast
43+ , castUnsafe -- XXX unsafeCast
44+ , asBytes
45+ , unsafePinnedAsPtr
46+ , unsafeAsPtr
47+
4048 -- ** Construction
41− , nil
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
4263
43− -- *** Uninitialized Arrays
44− , pinnedNew
45− , pinnedNewBytes
46− , pinnedNewAligned
47− , new
48− , newArrayWith
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
4970
50− -- *** From streams
71+ -- *** Stream Folds
5172 , ArrayUnsafe (..)
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
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
6887 , fromListN
6988 , pinnedFromListN
7089 , fromList
7190 , pinnedFromList
7291 , fromListRevN
7392 , fromListRev
74− , fromStreamDN
75− , fromStreamD
93+ , fromStreamN
94+ , fromStream
95+ , fromPureStreamN
7696 , fromPureStream
97+ , fromByteStr#
98+ , fromPtrN
99+ , fromChunksK
100+ , fromChunksRealloced -- fromSmallChunks
77101
78− -- * Random writes
102+ -- ** Random writes
79103 , putIndex
80− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
81105 , putIndices
82106 -- , putFromThenTo
83107 -- , putFrom -- start writing at the given position
@@ -85,78 +109,71 @@ module Streamly.Internal.Data.MutArray.Type
85109 -- , putFromTo
86110 -- , putFromRev
87111 -- , putUptoRev
88− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
89113 , modifyIndex
90114 , modifyIndices
91115 , modify
92116 , swapIndices
93117 , unsafeSwapIndices
94118
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.
99−
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
119+ -- ** Reading
112120
113− -- * Eliminating and Reading
114−
115− -- ** To streams
116− , reader
117− , readerRevWith
118− , readerRev
121+ -- *** Indexing
122+ , getIndex
123+ , getIndexUnsafe -- XXX unsafeGetIndex
124+ -- , getFromThenTo
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
119128
120− -- ** To containers
121− , toStreamDWith
122− , toStreamDRevWith
123− , toStreamKWith
124− , toStreamKRevWith
129+ -- *** To Streams
125130 , read
126131 , readRev
132+ , toStreamWith
133+ , toStreamRevWith
127134 , toStreamK
135+ , toStreamKWith
128136 , toStreamKRev
137+ , toStreamKRevWith
138+
139+ -- *** To Containers
129140 , toList
130141
142+ -- *** Unfolds
131143 -- experimental
132144 , producerWith
133145 , producer
134146
135− -- ** Random reads
136− , getIndex
137− , getIndexUnsafe
138− , getIndices
139− , getIndicesD
140− -- , getFromThenTo
141− , getIndexRev
142−
143− -- * Memory Management
144− , blockSize
145− , arrayChunkBytes
146− , allocBytesToElemCount
147− , realloc
148− , resize
149− , resizeExp
150− , rightSize
147+ , reader
148+ , readerRevWith
149+ , readerRev
151150
152− -- * Size
151+ -- ** Size and Capacity
152+ -- *** Size
153153 , length
154154 , byteLength
155+
156+ -- *** Capacity
155157 -- , capacity
156158 , byteCapacity
157159 , bytesFree
158160
159− -- * In-place Mutation Algorithms
161+ -- *** Capacity Management
162+ , blockSize
163+ , arrayChunkBytes
164+ , allocBytesToElemCount
165+ , realloc -- this is "resize"
166+ , grow
167+ , growExp
168+ , rightSize
169+
170+ -- ** Folding
171+ , foldl'
172+ , foldr
173+ , byteCmp
174+ , byteEq
175+
176+ -- ** In-place Mutation Algorithms
160177 , strip
161178 , reverse
162179 , permute
@@ -166,62 +183,121 @@ module Streamly.Internal.Data.MutArray.Type
166183 , mergeBy
167184 , bubble
168185
169− -- * Casting
170− , cast
171− , castUnsafe
172− , asBytes
173− , asPtrUnsafe
174−
175− -- * Folding
176− , foldl'
177− , foldr
178− , cmp
179−
180− -- * Arrays of arrays
181− -- We can add dimensionality parameter to the array type to get
182− -- multidimensional arrays. Multidimensional arrays would just be a
183− -- convenience wrapper on top of single dimensional arrays.
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.
184190
185− -- | Operations dealing with multiple arrays, streams of arrays or
186− -- multidimensional array representations.
187−
188− -- ** Construct from streams
189− , chunksOf
190− , pinnedChunksOf
191− , writeChunks
192−
193− -- ** Eliminate to streams
194− , flattenArrays
195− , flattenArraysRev
196− , fromArrayStreamK
191+ -- *** Appending elements
192+ , snocWith
193+ , snoc
194+ , snocLinear
195+ , snocMay
196+ , snocUnsafe -- XXX unsafeSnoc
197197
198− -- ** Construct from arrays
199− -- get chunks without copying
200− , getSliceUnsafe
201− , getSlice
202− -- , getSlicesFromLenN
203− , splitAt -- XXX should be able to express using getSlice
204− , breakOn
198+ -- *** Appending streams
199+ , unsafeAppendN
200+ , appendN
201+ , appendWith
202+ , append
205203
206− -- ** Cloning arrays
207− , clone
208− , pinnedClone
209−
210− -- ** Appending arrays
204+ -- *** Appending arrays
211205 , spliceCopy
212206 , spliceWith
213207 , splice
214208 , spliceExp
215− , spliceUnsafe
209+ , spliceUnsafe -- XXX unsafeSplice
216210 -- , putSlice
217211 -- , appendSlice
218212 -- , appendSliceFrom
219213
220− -- * Utilities
214+ -- ** Serialization using Unbox
215+ , pokeAppend
216+ , pokeAppendMay
217+ , pokeSkipUnsafe -- unsafePokeSkip
218+
219+ -- ** Deserialization using Unbox
220+ , peekUncons
221+ , peekUnconsUnsafe -- unsafePeekUncons
222+ , peekSkipUnsafe -- unsafePeekSkip
223+
224+ -- Arrays of arrays
225+ -- We can add dimensionality parameter to the array type to get
226+ -- multidimensional arrays. Multidimensional arrays would just be a
227+ -- convenience wrapper on top of single dimensional arrays.
228+
229+ -- ** Streams of Arrays
230+ -- *** Chunk
231+ -- | Group a stream into arrays.
232+ , chunksOf
233+ , pinnedChunksOf
234+ , buildChunks
235+
236+ -- *** Split
237+ -- | Split an array into slices.
238+
239+ -- , 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
221264 , roundUpToPower2
222265 , memcpy
223266 , memcmp
224267 , 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
225301 )
226302where
227303
@@ -232,19 +308,20 @@ where
232308
233309import Control.Monad (when, void)
234310import Control.Monad.IO.Class (MonadIO(..))
311+import Data.Bifunctor (first)
235312import Data.Bits (shiftR, (.|.), (.&.))
236313import Data.Functor.Identity (Identity(..))
237314import Data.Proxy (Proxy(..))
238315import Data.Word (Word8)
239316import Foreign.C.Types (CSize(..), CInt(..))
240317import Foreign.Ptr (plusPtr, minusPtr, nullPtr)
241−import Streamly.Internal.Data.Unbox
242− ( MutableByteArray(..)
243− , Unbox(..)
318+import Streamly.Internal.Data.MutByteArray.Type
319+ ( MutByteArray(..)
244320 , PinnedState(..)
245321 , getMutableByteArray#
246322 , putSliceUnsafe
247323 )
324+import Streamly.Internal.Data.Unbox (Unbox(..))
248325import GHC.Base
249326 ( IO(..)
250327 , Int(..)
@@ -252,27 +329,31 @@ import GHC.Base
252329 , copyMutableByteArray#
253330 )
254331import GHC.Base (noinline)
255−import GHC.Exts (unsafeCoerce#)
332+import GHC.Exts (unsafeCoerce#, Addr#)
256333import GHC.Ptr (Ptr(..))
257334
258335import Streamly.Internal.Data.Fold.Type (Fold(..))
259336import Streamly.Internal.Data.Producer.Type (Producer (..))
260337import Streamly.Internal.Data.Stream.Type (Stream)
338+import Streamly.Internal.Data.Parser.Type (Parser (..))
261339import Streamly.Internal.Data.StreamK.Type (StreamK)
262340import Streamly.Internal.Data.SVar.Type (adaptState, defState)
341+import Streamly.Internal.Data.Tuple.Strict (Tuple'(..))
263342import Streamly.Internal.Data.Unfold.Type (Unfold(..))
264343import Streamly.Internal.System.IO (arrayPayloadSize, defaultChunkSize)
265344
266345import 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
267349import qualified Streamly.Internal.Data.Producer as Producer
268350import qualified Streamly.Internal.Data.Stream.Type as D
269351import qualified Streamly.Internal.Data.Stream.Lift as D
270352import qualified Streamly.Internal.Data.StreamK.Type as K
271−import qualified Streamly.Internal.Data.Unbox as Unboxed
272353import qualified Prelude
273354
274355import Prelude hiding
275− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
276357
277358#include "DocTestDataMutArray.hs"
278359
@@ -280,6 +361,7 @@ import Prelude hiding
280361-- Foreign helpers
281362-------------------------------------------------------------------------------
282363
364+-- NOTE: Have to be "ccall unsafe" so that we can pass unpinned memory to these
283365foreign import ccall unsafe "string.h memcpy" c_memcpy
284366 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO (Ptr Word8)
285367
@@ -289,6 +371,9 @@ foreign import ccall unsafe "string.h memchr" c_memchr
289371foreign import ccall unsafe "string.h memcmp" c_memcmp
290372 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO CInt
291373
374+foreign import ccall unsafe "string.h strlen" c_strlen
375+ :: Ptr Word8 -> IO CSize
376+
292377-- | Given an 'Unboxed' type (unused first arg) and a number of bytes, return
293378-- how many elements of that type will completely fit in those bytes.
294379--
@@ -343,7 +428,7 @@ data MutArray a =
343428 -- The array is a range into arrContents. arrContents may be a superset of
344429 -- the slice represented by the array. All offsets are in bytes.
345430 MutArray
346− { arrContents :: {-# UNPACK #-} !MutableByteArray
431+ { arrContents :: {-# UNPACK #-} !MutByteArray
347432 , arrStart :: {-# UNPACK #-} !Int -- ^ index into arrContents
348433 , arrEnd :: {-# UNPACK #-} !Int -- ^ index into arrContents
349434 -- Represents the first invalid index of
@@ -355,8 +440,8 @@ data MutArray a =
355440-- Pinning & Unpinning
356441-------------------------------------------------------------------------------
357442
358−-- | Copy the 'MutArray' to pinned memory if unpinned, else do nothing. The
359−-- overhead is a copy if the input array is unpinned.
443+-- | Return a copy of the array in pinned memory if unpinned, else return the
444+-- original array.
360445{-# INLINE pin #-}
361446pin :: MutArray a -> IO (MutArray a)
362447pin arr@MutArray{..} =
@@ -364,8 +449,8 @@ pin arr@MutArray{..} =
364449 then pure arr
365450 else pinnedClone arr
366451
367−-- | Copy the 'MutArray' to unpinned memory if pinned, else do nothing. The
368−-- overhead is a copy if the input array is pinned.
452+-- | Return a copy of the array in unpinned memory if pinned, else return the
453+-- original array.
369454{-# INLINE unpin #-}
370455unpin :: MutArray a -> IO (MutArray a)
371456unpin arr@MutArray{..} =
@@ -373,6 +458,7 @@ unpin arr@MutArray{..} =
373458 then clone arr
374459 else pure arr
375460
461+-- | Return 'True' if the array is allocated in pinned memory.
376462{-# INLINE isPinned #-}
377463isPinned :: MutArray a -> Bool
378464isPinned MutArray{..} = Unboxed.isPinned arrContents
@@ -393,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
393479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
394480-- allocatePinned in GHC source.
395481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
396484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
397485-- length and with a capacity to hold @count@ elements, using @allocator
398486-- size alignment@ as the memory allocator function.
@@ -405,7 +493,7 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
405493-- /Pre-release/
406494{-# INLINE newArrayWith #-}
407495newArrayWith :: forall m a. (MonadIO m, Unbox a)
408− => (Int -> Int -> m MutableByteArray) -> Int -> Int -> m (MutArray a)
496+ => (Int -> Int -> m MutByteArray) -> Int -> Int -> m (MutArray a)
409497newArrayWith alloc alignSize count = do
410498 let size = max (count * SIZE_OF(a)) 0
411499 contents <- alloc size alignSize
@@ -416,12 +504,25 @@ newArrayWith alloc alignSize count = do
416504 , arrBound = size
417505 }
418506
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." #-}
419520nil ::
420521#ifdef DEVBUILD
421522 Unbox a =>
422523#endif
423524 MutArray a
424−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
425526
426527{-# INLINE newBytesAs #-}
427528newBytesAs :: MonadIO m =>
@@ -438,12 +539,13 @@ newBytesAs ps bytes = do
438539 , arrBound = bytes
439540 }
440541
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.
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.
444545--
445546-- /Pre-release/
446547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
447549pinnedNewBytes :: MonadIO m =>
448550#ifdef DEVBUILD
449551 Unbox a =>
@@ -468,21 +570,31 @@ newAs ps =
468570 (error "new: alignment is not used in unpinned arrays.")
469571
470572-- XXX can unaligned allocation be more efficient when alignment is not needed?
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−--
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." #-}
476582{-# INLINE pinnedNew #-}
477583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
478−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
479585
480−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
481−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
482588--
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." #-}
483595{-# INLINE new #-}
484596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
485−new = newAs Unpinned
597+new = emptyOf
486598
487599-------------------------------------------------------------------------------
488600-- Random writes
@@ -498,7 +610,7 @@ putIndexUnsafe :: forall m a. (MonadIO m, Unbox a)
498610putIndexUnsafe i MutArray{..} x = do
499611 let index = INDEX_OF(arrStart, i, a)
500612 assert (i >= 0 && INDEX_VALID(index, arrEnd, a)) (return ())
501− liftIO $ pokeByteIndex index arrContents x
613+ liftIO $ pokeAt index arrContents x
502614
503615invalidIndex :: String -> Int -> a
504616invalidIndex label i =
@@ -516,7 +628,7 @@ putIndex :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> a -> m ()
516628putIndex i MutArray{..} x = do
517629 let index = INDEX_OF(arrStart,i,a)
518630 if i >= 0 && INDEX_VALID(index,arrEnd,a)
519− then liftIO $ pokeByteIndex index arrContents x
631+ then liftIO $ pokeAt index arrContents x
520632 else invalidIndex "putIndex" i
521633
522634-- | Write an input stream of (index, value) pairs to an array. Throws an
@@ -530,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
530642
531643 where
532644
533− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
534646
535647-- | Modify a given index of an array using a modifier function.
536648--
@@ -542,9 +654,9 @@ modifyIndexUnsafe :: forall m a b. (MonadIO m, Unbox a) =>
542654modifyIndexUnsafe i MutArray{..} f = liftIO $ do
543655 let index = INDEX_OF(arrStart,i,a)
544656 assert (i >= 0 && INDEX_NEXT(index,a) <= arrEnd) (return ())
545− r <- peekByteIndex index arrContents
657+ r <- peekAt index arrContents
546658 let (x, res) = f r
547− pokeByteIndex index arrContents x
659+ pokeAt index arrContents x
548660 return res
549661
550662-- | Modify a given index of an array using a modifier function.
@@ -556,9 +668,9 @@ modifyIndex i MutArray{..} f = do
556668 let index = INDEX_OF(arrStart,i,a)
557669 if i >= 0 && INDEX_VALID(index,arrEnd,a)
558670 then liftIO $ do
559− r <- peekByteIndex index arrContents
671+ r <- peekAt index arrContents
560672 let (x, res) = f r
561− pokeByteIndex index arrContents x
673+ pokeAt index arrContents x
562674 return res
563675 else invalidIndex "modifyIndex" i
564676
@@ -592,8 +704,8 @@ modify MutArray{..} f = liftIO $
592704
593705 go i =
594706 when (INDEX_VALID(i,arrEnd,a)) $ do
595− r <- peekByteIndex i arrContents
596− pokeByteIndex i arrContents (f r)
707+ r <- peekAt i arrContents
708+ pokeAt i arrContents (f r)
597709 go (INDEX_NEXT(i,a))
598710
599711-- XXX We could specify the number of bytes to swap instead of Proxy. Need
@@ -602,15 +714,15 @@ modify MutArray{..} f = liftIO $
602714swapArrayByteIndices ::
603715 forall a. Unbox a
604716 => Proxy a
605− -> MutableByteArray
717+ -> MutByteArray
606718 -> Int
607719 -> Int
608720 -> IO ()
609721swapArrayByteIndices _ arrContents i1 i2 = do
610− r1 <- peekByteIndex i1 arrContents
611− r2 <- peekByteIndex i2 arrContents
612− pokeByteIndex i1 arrContents (r2 :: a)
613− pokeByteIndex i2 arrContents (r1 :: a)
722+ r1 <- peekAt i1 arrContents
723+ r2 <- peekAt i2 arrContents
724+ pokeAt i1 arrContents (r2 :: a)
725+ pokeAt i2 arrContents (r1 :: a)
614726
615727-- | Swap the elements at two indices without validating the indices.
616728--
@@ -726,26 +838,23 @@ arrayChunkBytes = 1024
726838roundDownTo :: Int -> Int -> Int
727839roundDownTo elemSize size = size - (size `mod` elemSize)
728840
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−--
732841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
733842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
734843-- which complicates some inspection tests.
735844--
736−{-# NOINLINE reallocExplicit #-}
737−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
738−reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
845+{-# NOINLINE reallocExplicitAs #-}
846+reallocExplicitAs :: PinnedState -> Int -> Int -> MutArray a -> IO (MutArray a)
847+reallocExplicitAs ps elemSize newCapacityInBytes MutArray{..} = do
739848 assertM(arrEnd <= arrBound)
740849
741850 -- Allocate new array
742851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
743852 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
853+ if ps == Pinned
854+ then Unboxed.pinnedNew newCapMaxInBytes
855+ else Unboxed.new newCapMaxInBytes
856+ let !(MutByteArray mbarrFrom#) = arrContents
857+ !(MutByteArray mbarrTo#) = contents
749858
750859 -- Copy old data
751860 let oldStart = arrStart
@@ -766,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
766875 , arrBound = newCapInBytes
767876 }
768877
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+
769882-- | @realloc newCapacity array@ reallocates the array to the specified
770883-- capacity in bytes.
771884--
@@ -777,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
777890-- If the original array is pinned, the newly allocated array is also pinned.
778891{-# INLINABLE realloc #-}
779892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
780−realloc bytes arr = liftIO $ reallocExplicit (SIZE_OF(a)) bytes arr
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
781899
782900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
783901-- in error messages and the capSizer is used to determine the capacity of the
@@ -800,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
800918 where
801919
802920 badSize newSize =
803− concat
921+ Prelude.concat
804922 [ label
805923 , ": new array size (in bytes) is less than required size "
806924 , show newSize
807925 , ". Please check the sizing function passed."
808926 ]
809927
810−-- | @resize newCapacity array@ changes the total capacity of the array so that
928+-- | @grow newCapacity array@ changes the total capacity of the array so that
811929-- it is enough to hold the specified number of elements. Nothing is done if
812930-- the specified capacity is less than the length of the array.
813931--
@@ -815,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
815933-- the block size (4K).
816934--
817935-- /Pre-release/
818−{-# INLINE resize #-}
819−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
820938 Int -> MutArray a -> m (MutArray a)
821−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
822940 let req = SIZE_OF(a) * nElems
823− len = arrEnd - arrStart
824− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
825943 then return arr
826944 else realloc req arr
827945
828−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
829−-- then it is rounded up to the closest power of 2.
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.
830954--
831955-- /Pre-release/
832−{-# INLINE resizeExp #-}
833−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
834958 Int -> MutArray a -> m (MutArray a)
835−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
836960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
837961 req1 =
838962 if req > largeObjectThreshold
839963 then roundUpToPower2 req
840964 else req
841− len = arrEnd - arrStart
842− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
843967 then return arr
844968 else realloc req1 arr
845969
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+
846976-- | Resize the allocated memory to drop any reserved free space at the end of
847977-- the array and reallocate it to reduce wastage.
848978--
@@ -887,7 +1017,7 @@ rightSize arr@MutArray{..} = do
8871017snocNewEnd :: (MonadIO m, Unbox a) => Int -> MutArray a -> a -> m (MutArray a)
8881018snocNewEnd newEnd arr@MutArray{..} x = liftIO $ do
8891019 assert (newEnd <= arrBound) (return ())
890− pokeByteIndex arrEnd arrContents x
1020+ pokeAt arrEnd arrContents x
8911021 return $ arr {arrEnd = newEnd}
8921022
8931023-- | Really really unsafe, appends the element into the first array, may
@@ -907,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9071037{-# INLINE snocMay #-}
9081038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9091039 MutArray a -> a -> m (Maybe (MutArray a))
910−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9111041 let newEnd = INDEX_NEXT(arrEnd,a)
9121042 if newEnd <= arrBound
9131043 then Just <$> snocNewEnd newEnd arr x
@@ -921,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9211051 -> a
9221052 -> m (MutArray a)
9231053snocWithRealloc sizer arr x = do
924− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9251055 snocUnsafe arr1 x
9261056
9271057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -943,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9431073 -> MutArray a
9441074 -> a
9451075 -> m (MutArray a)
946−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9471077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9481078 if newEnd <= arrBound arr
9491079 then snocNewEnd newEnd arr x
@@ -987,6 +1117,126 @@ snoc = snocWith f
9871117 then oldSize * 2
9881118 else roundUpToPower2 oldSize * 2
9891119
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+
9901240-------------------------------------------------------------------------------
9911241-- Random reads
9921242-------------------------------------------------------------------------------
@@ -1001,7 +1251,7 @@ getIndexUnsafe :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m a
10011251getIndexUnsafe i MutArray{..} = do
10021252 let index = INDEX_OF(arrStart,i,a)
10031253 assert (i >= 0 && INDEX_VALID(index,arrEnd,a)) (return ())
1004− liftIO $ peekByteIndex index arrContents
1254+ liftIO $ peekAt index arrContents
10051255
10061256-- | /O(1)/ Lookup the element at the given index. Index starts from 0.
10071257--
@@ -1010,7 +1260,7 @@ getIndex :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (Maybe a)
10101260getIndex i MutArray{..} = do
10111261 let index = INDEX_OF(arrStart,i,a)
10121262 if i >= 0 && INDEX_VALID(index,arrEnd,a)
1013− then liftIO $ Just <$> peekByteIndex index arrContents
1263+ then liftIO $ Just <$> peekAt index arrContents
10141264 else return Nothing
10151265
10161266-- | /O(1)/ Lookup the element at the given index from the end of the array.
@@ -1023,21 +1273,16 @@ getIndexRev :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m a
10231273getIndexRev i MutArray{..} = do
10241274 let index = RINDEX_OF(arrEnd,i,a)
10251275 if i >= 0 && index >= arrStart
1026− then liftIO $ peekByteIndex index arrContents
1276+ then liftIO $ peekAt index arrContents
10271277 else invalidIndex "getIndexRev" i
10281278
10291279data GetIndicesState contents start end st =
10301280 GetIndicesState contents start end st
10311281
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) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10391284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1040−getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10411286
10421287 where
10431288
@@ -1056,9 +1301,26 @@ getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
10561301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
10571302 D.Stop -> return D.Stop
10581303
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." #-}
10591321{-# INLINE getIndices #-}
10601322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1061−getIndices = getIndicesD liftIO
1323+getIndices = indexReader
10621324
10631325-------------------------------------------------------------------------------
10641326-- Subarrays
@@ -1163,7 +1425,7 @@ partitionBy f arr@MutArray{..} = liftIO $ do
11631425
11641426 -- Invariant low < high on entry, and on return as well
11651427 moveHigh low high = do
1166− h <- peekByteIndex high arrContents
1428+ h <- peekAt high arrContents
11671429 if f h
11681430 then
11691431 -- Correctly classified, continue the loop
@@ -1183,7 +1445,7 @@ partitionBy f arr@MutArray{..} = liftIO $ do
11831445 -- low <= high
11841446 -- Both low and high are valid locations within the array
11851447 go low high = do
1186− l <- peekByteIndex low arrContents
1448+ l <- peekAt low arrContents
11871449 if f l
11881450 then
11891451 -- low is wrongly classified
@@ -1194,8 +1456,8 @@ partitionBy f arr@MutArray{..} = liftIO $ do
11941456 case r of
11951457 Nothing -> return low
11961458 Just (high1, h) -> do -- low < high1
1197− pokeByteIndex low arrContents h
1198− pokeByteIndex high1 arrContents l
1459+ pokeAt low arrContents h
1460+ pokeAt high1 arrContents l
11991461 let low1 = INDEX_NEXT(low,a)
12001462 high2 = INDEX_PREV(high1,a)
12011463 if low1 <= high2
@@ -1329,7 +1591,7 @@ chunksOfAs ps n (D.Stream step state) =
13291591 r <- step (adaptState gst) st
13301592 case r of
13311593 D.Yield x s -> do
1332− liftIO $ pokeByteIndex end contents x
1594+ liftIO $ pokeAt end contents x
13331595 let end1 = INDEX_NEXT(end,a)
13341596 return $
13351597 if end1 >= bound
@@ -1353,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
13531615--
13541616-- Same as the following but may be more efficient:
13551617--
1356−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
13571619--
13581620-- /Pre-release/
13591621{-# INLINE_NORMAL chunksOf #-}
@@ -1362,17 +1624,29 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
13621624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
13631625-- benchmark. It seems it has difficulty producing optimized code when
13641626-- converting to StreamK. Investigate GHC optimizations.
1365−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
13661628chunksOf = chunksOfAs Unpinned
13671629
13681630-- | Like 'chunksOf' but creates pinned arrays.
13691631{-# INLINE_NORMAL pinnedChunksOf #-}
13701632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
13711633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1372−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
13731635pinnedChunksOf = chunksOfAs Pinned
13741636
1375−-- XXX This should take a PinnedState
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+
13761650-- XXX buffer to a list instead?
13771651-- | Buffer the stream into arrays in memory.
13781652{-# INLINE arrayStreamKFromStreamDAs #-}
@@ -1390,16 +1664,10 @@ data FlattenState s contents a =
13901664 OuterLoop s
13911665 | InnerLoop s contents !Int !Int
13921666
1393−-- | Use the "reader" unfold instead.
1394−--
1395−-- @flattenArrays = unfoldMany reader@
1396−--
1397−-- We can try this if there are any fusion issues in the unfold.
1398−--
1399−{-# INLINE_NORMAL flattenArrays #-}
1400−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1401− => D.Stream m (MutArray a) -> D.Stream m a
1402−flattenArrays (D.Stream step state) = D.Stream step' (OuterLoop state)
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)
14031671
14041672 where
14051673
@@ -1416,19 +1684,30 @@ flattenArrays (D.Stream step state) = D.Stream step' (OuterLoop state)
14161684 return $ D.Skip $ OuterLoop st
14171685
14181686 step' _ (InnerLoop st contents p end) = do
1419− x <- liftIO $ peekByteIndex p contents
1687+ x <- liftio $ peekAt p contents
14201688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14211689
1422−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14231691--
1424−-- @flattenArrays = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14251693--
14261694-- We can try this if there are any fusion issues in the unfold.
14271695--
1428−{-# INLINE_NORMAL flattenArraysRev #-}
1429−flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
14301698 => D.Stream m (MutArray a) -> D.Stream m a
1431−flattenArraysRev (D.Stream step state) = D.Stream step' (OuterLoop state)
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)
1704+ => 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)
14321711
14331712 where
14341713
@@ -1446,16 +1725,33 @@ flattenArraysRev (D.Stream step state) = D.Stream step' (OuterLoop state)
14461725 return $ D.Skip $ OuterLoop st
14471726
14481727 step' _ (InnerLoop st contents p start) = do
1449− x <- liftIO $ peekByteIndex p contents
1728+ x <- liftio $ peekAt p contents
14501729 let cur = INDEX_PREV(p,a)
14511730 return $ D.Yield x (InnerLoop st contents cur start)
14521731
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+
14531749-------------------------------------------------------------------------------
14541750-- Unfolds
14551751-------------------------------------------------------------------------------
14561752
14571753data ArrayUnsafe a = ArrayUnsafe
1458− {-# UNPACK #-} !MutableByteArray -- contents
1754+ {-# UNPACK #-} !MutByteArray -- contents
14591755 {-# UNPACK #-} !Int -- index 1
14601756 {-# UNPACK #-} !Int -- index 2
14611757
@@ -1485,7 +1781,7 @@ producerWith liftio = Producer step (return . toArrayUnsafe) extract
14851781 -- few actions for correctness and execution order sanity. We want
14861782 -- the peek to occur right here and not lazily at some later point
14871783 -- because we want the peek to be ordered with respect to the touch.
1488− !x <- liftio $ peekByteIndex cur contents
1784+ !x <- liftio $ peekAt cur contents
14891785 return $ D.Yield x (ArrayUnsafe contents (INDEX_NEXT(cur,a)) end)
14901786
14911787 extract = return . fromArrayUnsafe
@@ -1516,7 +1812,7 @@ readerRevWith liftio = Unfold step inject
15161812 {-# INLINE_LATE step #-}
15171813 step (ArrayUnsafe _ start p) | p < start = return D.Stop
15181814 step (ArrayUnsafe contents start p) = do
1519− !x <- liftio $ peekByteIndex p contents
1815+ !x <- liftio $ peekAt p contents
15201816 return $ D.Yield x (ArrayUnsafe contents start (INDEX_PREV(p,a)))
15211817
15221818-- | Unfold an array into a stream in reverse order.
@@ -1547,7 +1843,7 @@ toListFB c n MutArray{..} = go arrStart
15471843 -- evaluated/written to before we peek at them.
15481844 -- XXX
15491845 let !x = unsafeInlineIO $ do
1550− r <- peekByteIndex arrContents p
1846+ r <- peekAt arrContents p
15511847 return r
15521848 in c x (go (PTR_NEXT(p,a)))
15531849-}
@@ -1564,21 +1860,21 @@ toList MutArray{..} = liftIO $ go arrStart
15641860
15651861 go p | assert (p <= arrEnd) (p == arrEnd) = return []
15661862 go p = do
1567− x <- peekByteIndex p arrContents
1863+ x <- peekAt p arrContents
15681864 (:) x <$> go (INDEX_NEXT(p,a))
15691865
1570−{-# INLINE_NORMAL toStreamDWith #-}
1571−toStreamDWith ::
1866+{-# INLINE_NORMAL toStreamWith #-}
1867+toStreamWith ::
15721868 forall m a. (Monad m, Unbox a)
15731869 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1574−toStreamDWith liftio MutArray{..} = D.Stream step arrStart
1870+toStreamWith liftio MutArray{..} = D.Stream step arrStart
15751871
15761872 where
15771873
15781874 {-# INLINE_LATE step #-}
15791875 step _ p | assert (p <= arrEnd) (p == arrEnd) = return D.Stop
15801876 step _ p = liftio $ do
1581− r <- peekByteIndex p arrContents
1877+ r <- peekAt p arrContents
15821878 return $ D.Yield r (INDEX_NEXT(p,a))
15831879
15841880-- | Convert a 'MutArray' into a stream.
@@ -1587,7 +1883,7 @@ toStreamDWith liftio MutArray{..} = D.Stream step arrStart
15871883--
15881884{-# INLINE_NORMAL read #-}
15891885read :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1590−read = toStreamDWith liftIO
1886+read = toStreamWith liftIO
15911887
15921888{-# INLINE toStreamKWith #-}
15931889toStreamKWith ::
@@ -1599,18 +1895,18 @@ toStreamKWith liftio MutArray{..} = go arrStart
15991895
16001896 go p | assert (p <= arrEnd) (p == arrEnd) = K.nil
16011897 | otherwise =
1602− let elemM = peekByteIndex p arrContents
1898+ let elemM = peekAt p arrContents
16031899 in liftio elemM `K.consM` go (INDEX_NEXT(p,a))
16041900
16051901{-# INLINE toStreamK #-}
16061902toStreamK :: forall m a. (MonadIO m, Unbox a) => MutArray a -> StreamK m a
16071903toStreamK = toStreamKWith liftIO
16081904
1609−{-# INLINE_NORMAL toStreamDRevWith #-}
1610−toStreamDRevWith ::
1905+{-# INLINE_NORMAL toStreamRevWith #-}
1906+toStreamRevWith ::
16111907 forall m a. (Monad m, Unbox a)
16121908 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1613−toStreamDRevWith liftio MutArray{..} =
1909+toStreamRevWith liftio MutArray{..} =
16141910 let p = INDEX_PREV(arrEnd,a)
16151911 in D.Stream step p
16161912
@@ -1619,7 +1915,7 @@ toStreamDRevWith liftio MutArray{..} =
16191915 {-# INLINE_LATE step #-}
16201916 step _ p | p < arrStart = return D.Stop
16211917 step _ p = liftio $ do
1622− r <- peekByteIndex p arrContents
1918+ r <- peekAt p arrContents
16231919 return $ D.Yield r (INDEX_PREV(p,a))
16241920
16251921-- | Convert a 'MutArray' into a stream in reverse order.
@@ -1628,7 +1924,7 @@ toStreamDRevWith liftio MutArray{..} =
16281924--
16291925{-# INLINE_NORMAL readRev #-}
16301926readRev :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1631−readRev = toStreamDRevWith liftIO
1927+readRev = toStreamRevWith liftIO
16321928
16331929{-# INLINE toStreamKRevWith #-}
16341930toStreamKRevWith ::
@@ -1642,7 +1938,7 @@ toStreamKRevWith liftio MutArray {..} =
16421938
16431939 go p | p < arrStart = K.nil
16441940 | otherwise =
1645− let elemM = peekByteIndex p arrContents
1941+ let elemM = peekAt p arrContents
16461942 in liftio elemM `K.consM` go (INDEX_PREV(p,a))
16471943
16481944{-# INLINE toStreamKRev #-}
@@ -1677,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
16771973
16781974-- XXX Keep the bound intact to not lose any free space? Perf impact?
16791975
1680−-- | @writeAppendNUnsafe n arr@ appends up to @n@ input items to the supplied
1976+-- | @unsafeAppendN n arr@ appends up to @n@ input items to the supplied
16811977-- array.
16821978--
16831979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1686,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
16861982-- Any free space left in the array after appending @n@ elements is lost.
16871983--
16881984-- /Internal/
1689−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1690−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
16911987 Int
16921988 -> m (MutArray a)
16931989 -> Fold m a (MutArray a)
1694−writeAppendNUnsafe n action =
1695− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
16961991
16971992 where
16981993
@@ -1705,38 +2000,59 @@ writeAppendNUnsafe n action =
17052000 -- otherwise we lose that space.
17062001 arr1 <-
17072002 if free < needed
1708− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17092004 else return arr
17102005 return $ toArrayUnsafe arr1
17112006
17122007 step (ArrayUnsafe contents start end) x = do
1713− liftIO $ pokeByteIndex end contents x
2008+ liftIO $ pokeAt end contents x
17142009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17152010
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+
17162019-- | Append @n@ elements to an existing array. Any free space left in the array
17172020-- after appending @n@ elements is lost.
17182021--
1719−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17202023--
1721−{-# INLINE_NORMAL writeAppendN #-}
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 #-}
17222032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17232033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1724−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17252035
1726−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17272037-- append the input stream. If there is no reserved space available in the
17282038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
17292039-- where @oldSize@ is the current size of the array in bytes.
17302040--
17312041-- Note that the returned array may be a mutated version of original array.
17322042--
1733−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
17342044--
17352045-- /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." #-}
17362052{-# INLINE writeAppendWith #-}
17372053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
17382054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1739−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
17402056
17412057-- | @append action@ mutates the array generated by @action@ to append the
17422058-- input stream. If there is no reserved space available in the array it is
@@ -1744,70 +2060,101 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
17442060--
17452061-- Note that the returned array may be a mutated version of original array.
17462062--
1747−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
17482064--
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." #-}
17492072{-# INLINE writeAppend #-}
17502073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
17512074 m (MutArray a) -> Fold m a (MutArray a)
1752−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
17532076
17542077-- XXX We can carry bound as well in the state to make sure we do not lose the
17552078-- remaining capacity. Need to check perf impact.
17562079--
1757−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1758−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
17592082--
1760−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
17612084--
17622085-- /Pre-release/
1763−{-# INLINE_NORMAL writeNWithUnsafe #-}
1764−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
17652088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1766−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
17672090
17682091 where
17692092
17702093 initial = toArrayUnsafe <$> alloc (max n 0)
17712094
17722095 step (ArrayUnsafe contents start end) x = do
1773− liftIO $ pokeByteIndex end contents x
2096+ liftIO $ pokeAt end contents x
17742097 return
17752098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17762099
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+
17772106{-# INLINE_NORMAL writeNUnsafeAs #-}
17782107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
17792108 => PinnedState -> Int -> Fold m a (MutArray a)
1780−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
17812110
1782−-- | Like 'writeN' but does not check the array bounds when writing. The fold
2111+-- | Like 'createOf' but does not check the array bounds when writing. The fold
17832112-- driver must not call the step function more than 'n' times otherwise it will
17842113-- corrupt the memory and crash. This function exists mainly because any
17852114-- conditional in the step function blocks fusion causing 10x performance
17862115-- slowdown.
17872116--
1788−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
17892118--
1790−{-# INLINE_NORMAL writeNUnsafe #-}
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 #-}
17912126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
17922127 => Int -> Fold m a (MutArray a)
1793−writeNUnsafe = writeNUnsafeAs Unpinned
2128+writeNUnsafe = unsafeCreateOf
17942129
1795−-- | Like 'writeNUnsafe' but creates a pinned array.
1796−{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
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 #-}
17972138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
17982139 => Int -> Fold m a (MutArray a)
1799−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18002141
1801−-- | @writeNWith alloc n@ folds a maximum of @n@ elements into an array
2142+-- | @createOfWith alloc n@ folds a maximum of @n@ elements into an array
18022143-- allocated using the @alloc@ function.
18032144--
1804−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1805−-- >>> writeNWith alloc n = MutArray.writeAppendN (alloc n) n
2145+-- >>> createOfWith alloc n = Fold.take n (MutArray.unsafeCreateOfWith alloc n)
2146+-- >>> createOfWith alloc n = MutArray.appendN (alloc n) n
18062147--
1807−{-# INLINE_NORMAL writeNWith #-}
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 #-}
18082155writeNWith :: forall m a. (MonadIO m, Unbox a)
18092156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1810−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18112158
18122159{-# INLINE_NORMAL writeNAs #-}
18132160writeNAs ::
@@ -1815,28 +2162,42 @@ writeNAs ::
18152162 => PinnedState
18162163 -> Int
18172164 -> Fold m a (MutArray a)
1818−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18192166
1820−-- | @writeN n@ folds a maximum of @n@ elements from the input stream to an
2167+-- | @createOf n@ folds a maximum of @n@ elements from the input stream to an
18212168-- 'MutArray'.
18222169--
1823−-- >>> writeN = MutArray.writeNWith MutArray.new
1824−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1825−-- >>> writeN n = MutArray.writeAppendN n (MutArray.new n)
2170+-- >>> createOf = MutArray.createOfWith MutArray.new
2171+-- >>> createOf n = Fold.take n (MutArray.unsafeCreateOf n)
2172+-- >>> createOf n = MutArray.appendN n (MutArray.emptyOf n)
18262173--
1827−{-# INLINE_NORMAL writeN #-}
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 #-}
18282181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1829−writeN = writeNAs Unpinned
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
18302191
1831−-- | Like 'writeN' but creates a pinned array.
1832−{-# INLINE_NORMAL pinnedWriteN #-}
2192+{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193+{-# INLINE pinnedWriteN #-}
18332194pinnedWriteN ::
18342195 forall m a. (MonadIO m, Unbox a)
18352196 => Int
18362197 -> Fold m a (MutArray a)
1837−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
18382199
1839−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
18402201--
18412202-- /Internal/
18422203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1853,11 +2214,11 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18532214
18542215 step (ArrayUnsafe contents start end) x = do
18552216 let ptr = INDEX_PREV(start,a)
1856− liftIO $ pokeByteIndex ptr contents x
2217+ liftIO $ pokeAt ptr contents x
18572218 return
18582219 $ ArrayUnsafe contents ptr end
18592220
1860−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
18612222--
18622223-- /Internal/
18632224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1865,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
18652226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
18662227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
18672228
1868−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
18692230--
18702231-- /Pre-release/
1871−{-# INLINE_NORMAL writeRevN #-}
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 #-}
18722238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1873−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
18742240
18752241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
18762242-- input stream to a 'MutArray' aligned to the given size.
18772243--
1878−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1879−-- >>> pinnedWriteNAligned align n = MutArray.writeAppendN n (MutArray.pinnedNewAligned align n)
2244+-- >>> pinnedWriteNAligned align = MutArray.createOfWith (MutArray.pinnedNewAligned align)
2245+-- >>> pinnedWriteNAligned align n = MutArray.appendN n (MutArray.pinnedNewAligned align n)
18802246--
18812247-- /Pre-release/
18822248--
18832249{-# INLINE_NORMAL pinnedWriteNAligned #-}
18842250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
18852251 => Int -> Int -> Fold m a (MutArray a)
1886−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
18872253
18882254-- XXX Buffer to a list instead?
18892255
18902256-- | Buffer a stream into a stream of arrays.
18912257--
1892−-- >>> writeChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
18932259--
18942260-- Breaking an array into an array stream can be useful to consume a large
18952261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1898,37 +2264,43 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
18982264--
18992265-- /Unimplemented/
19002266--
1901−{-# INLINE_NORMAL writeChunks #-}
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 #-}
19022274writeChunks :: (MonadIO m, Unbox a) =>
19032275 Int -> Fold m a (StreamK n (MutArray a))
1904−writeChunks n = FL.many (writeN n) FL.toStreamK
2276+writeChunks = buildChunks
19052277
19062278{-# INLINE_NORMAL writeWithAs #-}
19072279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19082280 => PinnedState -> Int -> Fold m a (MutArray a)
1909−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19102282writeWithAs ps elemCount =
19112283 FL.rmapM extract $ FL.foldlM' step initial
19122284
19132285 where
19142286
19152287 initial = do
1916− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1917− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19182290
19192291 step arr@(MutArray _ start end bound) x
19202292 | INDEX_NEXT(end,a) > bound = do
19212293 let oldSize = end - start
19222294 newSize = max (oldSize * 2) 1
1923− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
19242296 snocUnsafe arr1 x
19252297 step arr x = snocUnsafe arr x
19262298
19272299 extract = liftIO . rightSize
19282300
1929−-- XXX Compare writeWith with fromStreamD which uses an array of streams
1930−-- implementation. We can write this using writeChunks above if that is faster.
1931−-- If writeWith is faster then we should use that to implement
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
19322304-- fromStreamD.
19332305--
19342306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -1943,39 +2315,56 @@ writeWithAs ps elemCount =
19432315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
19442316-- malloc to compare against.
19452317
1946−-- | @writeWith minCount@ folds the whole input to a single array. The array
2318+-- | @createWith minCount@ folds the whole input to a single array. The array
19472319-- starts at a size big enough to hold minCount elements, the size is doubled
19482320-- every time the array needs to be grown.
19492321--
19502322-- /Caution! Do not use this on infinite streams./
19512323--
1952−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
1953−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
1954−-- >>> writeWith n = Fold.rmapM MutArray.fromArrayStreamK (MutArray.writeChunks n)
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)
19552327--
19562328-- /Pre-release/
1957−{-# INLINE_NORMAL writeWith #-}
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 #-}
19582337writeWith :: forall m a. (MonadIO m, Unbox a)
19592338 => Int -> Fold m a (MutArray a)
1960−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
1961−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
19622340
19632341-- | Fold the whole input to a single array.
19642342--
1965−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
19662344-- rounded up to the element size.
19672345--
19682346-- /Caution! Do not use this on infinite streams./
19692347--
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." #-}
19702354{-# INLINE write #-}
19712355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
1972−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
19732357
1974−-- | Like 'write' but creates a pinned array.
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." #-}
19752365{-# INLINE pinnedWrite #-}
19762366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
1977−pinnedWrite =
1978− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
19792368
19802369-------------------------------------------------------------------------------
19812370-- construct from streams, known size
@@ -1985,25 +2374,31 @@ pinnedWrite =
19852374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
19862375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
19872376fromStreamDNAs ps limit str = do
1988− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
19892378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
19902379 return $ arr {arrEnd = end}
19912380
19922381 where
19932382
19942383 fwrite arrContents ptr x = do
1995− liftIO $ pokeByteIndex ptr arrContents x
2384+ liftIO $ pokeAt ptr arrContents x
19962385 return $ INDEX_NEXT(ptr,a)
19972386
1998−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
19992388--
2000−-- >>> fromStreamDN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20012390--
2002−{-# INLINE_NORMAL fromStreamDN #-}
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 #-}
20032399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20042400 => Int -> D.Stream m a -> m (MutArray a)
2005−-- fromStreamDN n = D.fold (writeN n)
2006−fromStreamDN = fromStreamDNAs Unpinned
2401+fromStreamDN = fromStreamN
20072402
20082403-- | Create a 'MutArray' from the first N elements of a list. The array is
20092404-- allocated to size N, if the list terminates before N elements then the
@@ -2023,16 +2418,70 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
20232418-- /Pre-release/
20242419{-# INLINE fromListRevN #-}
20252420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2026−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
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
20272429
20282430-- | Convert a pure stream in Identity monad to a mutable array.
20292431{-# INLINABLE fromPureStream #-}
20302432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
20312433fromPureStream xs =
2032− liftIO $ D.fold write $ D.morphInner (return . runIdentity) 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})
20332461
20342462-------------------------------------------------------------------------------
2035−-- convert stream to a single array
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
20362485-------------------------------------------------------------------------------
20372486
20382487{-# INLINE arrayStreamKLength #-}
@@ -2042,34 +2491,72 @@ arrayStreamKLength as = K.foldl' (+) 0 (K.map length as)
20422491-- | Convert an array stream to an array. Note that this requires peak memory
20432492-- that is double the size of the array stream.
20442493--
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." #-}
20452518{-# INLINE fromArrayStreamK #-}
20462519fromArrayStreamK :: (Unbox a, MonadIO m) =>
20472520 StreamK m (MutArray a) -> m (MutArray a)
2048−fromArrayStreamK as = do
2049− len <- arrayStreamKLength as
2050− fromStreamDN len $ D.unfoldMany reader $ D.fromStreamK as
2521+fromArrayStreamK = fromChunksK
20512522
20522523{-# INLINE fromStreamDAs #-}
20532524fromStreamDAs ::
20542525 (MonadIO m, Unbox a) => PinnedState -> D.Stream m a -> m (MutArray a)
2055−fromStreamDAs ps m = arrayStreamKFromStreamDAs ps m >>= fromArrayStreamK
2526+fromStreamDAs ps m =
2527+ arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
20562528
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
20572543-- CAUTION: a very large number (millions) of arrays can degrade performance
20582544-- due to GC overhead because we need to buffer the arrays before we flatten
20592545-- all the arrays.
20602546--
2061−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
20622548--
20632549-- | We could take the approach of doubling the memory allocation on each
20642550-- overflow. This would result in more or less the same amount of copying as in
20652551-- the chunking approach. However, if we have to shrink in the end then it may
20662552-- result in an extra copy of the entire data.
20672553--
2068−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
20692555--
20702556{-# INLINE fromStreamD #-}
2557+{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
20712558fromStreamD :: (MonadIO m, Unbox a) => D.Stream m a -> m (MutArray a)
2072−fromStreamD = fromStreamDAs Unpinned
2559+fromStreamD = fromStream
20732560
20742561-- | Create a 'MutArray' from a list. The list must be of finite size.
20752562--
@@ -2094,22 +2581,44 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
20942581-- Cloning
20952582-------------------------------------------------------------------------------
20962583
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+
20972588{-# INLINE cloneAs #-}
2098−cloneAs :: MonadIO m => PinnedState -> MutArray a -> m (MutArray a)
2589+cloneAs ::
2590+ ( MonadIO m
2591+#ifdef DEVBUILD
2592+ , Unbox a
2593+#endif
2594+ )
2595+ => PinnedState -> MutArray a -> m (MutArray a)
20992596cloneAs ps src =
2100− liftIO $ do
2597+ do
21012598 let startSrc = arrStart src
21022599 srcLen = arrEnd src - startSrc
2103− newArrContents <- Unboxed.newBytesAs ps srcLen
2104− putSliceUnsafe (arrContents src) startSrc newArrContents 0 srcLen
2600+ newArrContents <-
2601+ Unboxed.cloneSliceUnsafeAs ps startSrc srcLen (arrContents src)
21052602 return $ MutArray newArrContents 0 srcLen srcLen
21062603
21072604{-# INLINE clone #-}
2108−clone :: MonadIO m => MutArray a -> m (MutArray a)
2605+clone ::
2606+ ( MonadIO m
2607+#ifdef DEVBUILD
2608+ , Unbox a
2609+#endif
2610+ )
2611+ => MutArray a -> m (MutArray a)
21092612clone = cloneAs Unpinned
21102613
21112614{-# INLINE pinnedClone #-}
2112−pinnedClone :: MonadIO m => MutArray a -> m (MutArray a)
2615+pinnedClone ::
2616+ ( MonadIO m
2617+#ifdef DEVBUILD
2618+ , Unbox a
2619+#endif
2620+ )
2621+ => MutArray a -> m (MutArray a)
21132622pinnedClone = cloneAs Pinned
21142623
21152624-------------------------------------------------------------------------------
@@ -2124,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
21242633 Unbox a =>
21252634#endif
21262635 MutArray a -> MutArray a -> m (MutArray a)
2127−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
21282637 let start1 = arrStart arr1
21292638 start2 = arrStart arr2
21302639 len1 = arrEnd arr1 - start1
21312640 len2 = arrEnd arr2 - start2
2132− let newLen = len1 + len2
2641+ let len = len1 + len2
21332642 newArrContents <-
21342643 if Unboxed.isPinned (arrContents arr1)
2135− then Unboxed.pinnedNewBytes newLen
2136− else Unboxed.newBytes newLen
2137− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
21382646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
21392647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
21402648 return $ MutArray newArrContents 0 len len
@@ -2146,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
21462654spliceUnsafe :: MonadIO m =>
21472655 MutArray a -> MutArray a -> m (MutArray a)
21482656spliceUnsafe dst src =
2149− liftIO $ do
2657+ do
21502658 let startSrc = arrStart src
21512659 srcLen = arrEnd src - startSrc
21522660 endDst = arrEnd dst
@@ -2168,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
21682676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
21692677spliceWith sizer dst@(MutArray _ start end bound) src = do
21702678{-
2171− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
21722680 in D.fold f (toStreamD src)
21732681-}
21742682 assert (end <= bound) (return ())
@@ -2184,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
21842692 $ "splice: newSize is less than the total size "
21852693 ++ "of arrays being appended. Please check the "
21862694 ++ "sizer function passed."
2187− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
21882696 else return dst
21892697 spliceUnsafe dst1 src
21902698
@@ -2221,14 +2729,24 @@ spliceExp = spliceWith (\l1 l2 -> max (l1 * 2) (l1 + l2))
22212729-- Splitting
22222730-------------------------------------------------------------------------------
22232731
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+
22242743-- | Drops the separator byte
22252744{-# INLINE breakOn #-}
22262745breakOn :: MonadIO m
22272746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2228−breakOn sep arr@MutArray{..} = asPtrUnsafe arr $ \p -> liftIO $ do
2229− -- XXX Instead of using asPtrUnsafe (pinning memory) we can pass unlifted
2230− -- Addr# to memchr and it should be safe (from ghc 8.4).
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
22312748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
2749+ -- Need efficient stream based primitives that work on Word64.
22322750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
22332751 let sepIndex = loc `minusPtr` p
22342752 return $
@@ -2249,32 +2767,41 @@ breakOn sep arr@MutArray{..} = asPtrUnsafe arr $ \p -> liftIO $ do
22492767 }
22502768 )
22512769
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+
22522792-- | Create two slices of an array without copying the original array. The
22532793-- specified index @i@ is the first index of the second slice.
22542794--
2795+{-# INLINE splitAt #-}
22552796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2256−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
22572798 let maxIndex = length arr - 1
22582799 in if i < 0
22592800 then error "sliceAt: negative array index"
22602801 else if i > maxIndex
22612802 then error $ "sliceAt: specified array index " ++ show i
22622803 ++ " is beyond the maximum index " ++ show maxIndex
2263− else let off = i * SIZE_OF(a)
2264− p = arrStart + off
2265− in ( MutArray
2266− { arrContents = arrContents
2267− , arrStart = arrStart
2268− , arrEnd = p
2269− , arrBound = p
2270− }
2271− , MutArray
2272− { arrContents = arrContents
2273− , arrStart = p
2274− , arrEnd = arrEnd
2275− , arrBound = arrBound
2276− }
2277− )
2804+ else unsafeSplitAt i arr
22782805
22792806-------------------------------------------------------------------------------
22802807-- Casting
@@ -2324,9 +2851,13 @@ cast arr =
23242851-- guarantees that unsafe calls will be performed in the calling thread. Making
23252852-- it safe to pass heap-allocated objects to unsafe functions.
23262853
2327−-- Should we just name it asPtr, the unsafety is implicit for any pointer
2854+-- XXX Should we just name it asPtr, the unsafety is implicit for any pointer
23282855-- operations. And we are safe from Haskell perspective because we will be
23292856-- 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.
23302861
23312862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
23322863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2344,48 +2875,377 @@ cast arr =
23442875--
23452876-- /Pre-release/
23462877--
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." #-}
23472885{-# INLINE asPtrUnsafe #-}
23482886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2349−asPtrUnsafe arr f =
2350− Unboxed.asPtrUnsafe
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
23512893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
23522894
23532895-------------------------------------------------------------------------------
23542896-- Equality
23552897-------------------------------------------------------------------------------
23562898
2357−-- | Compare the length of the arrays. If the length is equal, compare the
2358−-- lexicographical ordering of two underlying byte arrays otherwise return the
2359−-- result of length comparison.
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.
23602906--
23612907-- /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+
23622931{-# INLINE cmp #-}
2932+{-# DEPRECATED cmp "Please use byteCmp instead." #-}
23632933cmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2364−cmp arr1 arr2 =
2365− liftIO
2366− $ do
2367− let marr1 = getMutableByteArray# (arrContents arr1)
2368− marr2 = getMutableByteArray# (arrContents arr2)
2369− !(I# st1#) = arrStart arr1
2370− !(I# st2#) = arrStart arr2
2371− !(I# len#) = byteLength arr1
2372− case compare (byteLength arr1) (byteLength arr2) of
2373− EQ -> do
2374− r <- liftIO $ IO $ \s# ->
2375− let res =
2376− I#
2377− (compareByteArrays#
2378− (unsafeCoerce# marr1)
2379− st1#
2380− (unsafeCoerce# marr2)
2381− st2#
2382− len#)
2383− in (# s#, res #)
2384− return $ compare r 0
2385− x -> return x
2934+cmp = byteCmp
23862935
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
23872947-------------------------------------------------------------------------------
2388−-- NFData
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)
3246+
3247+-------------------------------------------------------------------------------
3248+-- In-place mutation algorithms
23893249-------------------------------------------------------------------------------
23903250
23913251-- | Strip elements which match with predicate from both ends.
@@ -2414,7 +3274,7 @@ strip eq arr@MutArray{..} = liftIO $ do
24143274 getStart cur = do
24153275 if cur < arrEnd
24163276 then do
2417− r <- peekByteIndex cur arrContents
3277+ r <- peekAt cur arrContents
24183278 if eq r
24193279 then getStart (INDEX_NEXT(cur,a))
24203280 else return cur
@@ -2424,7 +3284,7 @@ strip eq arr@MutArray{..} = liftIO $ do
24243284 if cur > low
24253285 then do
24263286 let prev = INDEX_PREV(cur,a)
2427− r <- peekByteIndex prev arrContents
3287+ r <- peekAt prev arrContents
24283288 if eq r
24293289 then getLast prev low
24303290 else return cur