@@ -40,22 +40,22 @@ module Streamly.Internal.Data.MutArray.Type
4040
4141 -- ** Casting
4242 , cast
43− , castUnsafe
43+ , castUnsafe -- XXX unsafeCast
4444 , asBytes
45− , asPtrUnsafe
46− , asUnpinnedPtrUnsafe
45+ , unsafePinnedAsPtr
46+ , unsafeAsPtr
4747
4848 -- ** Construction
49− , nil
49+ , empty
5050
5151 -- *** New
5252 -- | New arrays are always empty arrays with some reserve capacity to
5353 -- extend the length without reallocating.
54− , new
55− , newArrayWith
56− , pinnedNew
57− , pinnedNewBytes
58− , pinnedNewAligned
54+ , emptyOf
55+ , newArrayWith -- emptyAlignedWith
56+ , pinnedEmptyOf
57+ , pinnedNewAligned -- XXX not required
58+ -- , new -- uninitialized array of specified length
5959
6060 -- *** Cloning
6161 , clone
@@ -63,27 +63,25 @@ module Streamly.Internal.Data.MutArray.Type
6363
6464 -- *** Slicing
6565 -- | Get a subarray without copying
66− , getSliceUnsafe
66+ , getSliceUnsafe -- XXX unsafeGetSlice
6767 , getSlice
6868 , splitAt -- XXX should be able to express using getSlice
6969 , breakOn
7070
7171 -- *** Stream Folds
7272 , ArrayUnsafe (..)
73− , writeNWithUnsafe
74− , writeNWith
75− , writeNUnsafe
76− , pinnedWriteNUnsafe
77− , writeN
78− , pinnedWriteN
79− , pinnedWriteNAligned
80−
81− , writeWith
82− , write
83− , pinnedWrite
84−
85− , writeRevN
86− -- , writeRev
73+ , unsafeCreateOfWith
74+ , unsafeCreateOf
75+ , unsafePinnedCreateOf
76+ , pinnedCreateOf
77+ , createOfWith
78+ , createOf
79+ , revCreateOf
80+
81+ , pinnedCreate
82+ , createWith
83+ , create
84+ -- , revCreate
8785
8886 -- *** From containers
8987 , fromListN
@@ -99,11 +97,11 @@ module Streamly.Internal.Data.MutArray.Type
9997 , fromByteStr#
10098 , fromPtrN
10199 , fromChunksK
102− , fromChunksRealloced
100+ , fromChunksRealloced -- fromSmallChunks
103101
104102 -- ** Random writes
105103 , putIndex
106− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
107105 , putIndices
108106 -- , putFromThenTo
109107 -- , putFrom -- start writing at the given position
@@ -111,7 +109,7 @@ module Streamly.Internal.Data.MutArray.Type
111109 -- , putFromTo
112110 -- , putFromRev
113111 -- , putUptoRev
114− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
115113 , modifyIndex
116114 , modifyIndices
117115 , modify
@@ -122,11 +120,11 @@ module Streamly.Internal.Data.MutArray.Type
122120
123121 -- *** Indexing
124122 , getIndex
125− , getIndexUnsafe
126− , getIndices -- XXX indexReader
127− , getIndicesWith -- XXX indexReaderWith
123+ , getIndexUnsafe -- XXX unsafeGetIndex
128124 -- , getFromThenTo
129− , getIndexRev
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
130128
131129 -- *** To Streams
132130 , read
@@ -164,9 +162,9 @@ module Streamly.Internal.Data.MutArray.Type
164162 , blockSize
165163 , arrayChunkBytes
166164 , allocBytesToElemCount
167− , realloc
168− , resize
169− , resizeExp
165+ , realloc -- this is "resize"
166+ , grow
167+ , growExp
170168 , rightSize
171169
172170 -- ** Folding
@@ -195,24 +193,34 @@ module Streamly.Internal.Data.MutArray.Type
195193 , snoc
196194 , snocLinear
197195 , snocMay
198− , snocUnsafe
196+ , snocUnsafe -- XXX unsafeSnoc
199197
200198 -- *** Appending streams
201− , writeAppendNUnsafe
202− , writeAppendN
203− , writeAppendWith
204− , writeAppend
199+ , unsafeAppendN
200+ , appendN
201+ , appendWith
202+ , append
205203
206204 -- *** Appending arrays
207205 , spliceCopy
208206 , spliceWith
209207 , splice
210208 , spliceExp
211− , spliceUnsafe
209+ , spliceUnsafe -- XXX unsafeSplice
212210 -- , putSlice
213211 -- , appendSlice
214212 -- , appendSliceFrom
215213
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+
216224 -- Arrays of arrays
217225 -- We can add dimensionality parameter to the array type to get
218226 -- multidimensional arrays. Multidimensional arrays would just be a
@@ -234,18 +242,23 @@ module Streamly.Internal.Data.MutArray.Type
234242
235243 -- *** Concat
236244 -- | Append the arrays in a stream to form a stream of elements.
237− , concatChunks
238− , concatChunksRev
245+ , concatWith
246+ , concatRevWith
247+ , concat
248+ , concatRev
239249
240250 -- *** Compact
241251 -- | Append the arrays in a stream to form a stream of larger arrays.
242252 , SpliceState (..)
243− , pCompactChunksLE
244− , rCompactChunksLE
245− , fCompactChunksGE
246− , lCompactChunksGE
247− , compactChunksGE
248− , compactChunksEQ
253+ , pCompactLE
254+ , pPinnedCompactLE
255+ , compactLeAs
256+ , fCompactGE
257+ , fPinnedCompactGE
258+ , lCompactGE
259+ , lPinnedCompactGE
260+ , compactGE
261+ , compactEQ
249262
250263 -- ** Utilities
251264 , roundUpToPower2
@@ -254,6 +267,7 @@ module Streamly.Internal.Data.MutArray.Type
254267 , c_memchr
255268
256269 -- * Deprecated
270+ , asPtrUnsafe
257271 , writeChunks
258272 , flattenArrays
259273 , flattenArraysRev
@@ -261,6 +275,29 @@ module Streamly.Internal.Data.MutArray.Type
261275 , fromStreamDN
262276 , fromStreamD
263277 , 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
264301 )
265302where
266303
@@ -316,7 +353,7 @@ import qualified Streamly.Internal.Data.StreamK.Type as K
316353import qualified Prelude
317354
318355import Prelude hiding
319− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
320357
321358#include "DocTestDataMutArray.hs"
322359
@@ -442,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
442479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
443480-- allocatePinned in GHC source.
444481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
445484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
446485-- length and with a capacity to hold @count@ elements, using @allocator
447486-- size alignment@ as the memory allocator function.
@@ -465,12 +504,25 @@ newArrayWith alloc alignSize count = do
465504 , arrBound = size
466505 }
467506
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." #-}
468520nil ::
469521#ifdef DEVBUILD
470522 Unbox a =>
471523#endif
472524 MutArray a
473−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
474526
475527{-# INLINE newBytesAs #-}
476528newBytesAs :: MonadIO m =>
@@ -487,12 +539,13 @@ newBytesAs ps bytes = do
487539 , arrBound = bytes
488540 }
489541
490−-- | Allocates a pinned empty array that can hold 'count' items. The memory of
491−-- the array is uninitialized and the allocation is aligned as per the
492−-- '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.
493545--
494546-- /Pre-release/
495547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
496549pinnedNewBytes :: MonadIO m =>
497550#ifdef DEVBUILD
498551 Unbox a =>
@@ -517,21 +570,31 @@ newAs ps =
517570 (error "new: alignment is not used in unpinned arrays.")
518571
519572-- XXX can unaligned allocation be more efficient when alignment is not needed?
520−--
521−-- | Allocates an empty pinned array that can hold 'count' items. The memory of
522−-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
523−-- instance of the type.
524−--
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." #-}
525582{-# INLINE pinnedNew #-}
526583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
527−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
528585
529−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
530−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
531588--
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." #-}
532595{-# INLINE new #-}
533596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
534−new = newAs Unpinned
597+new = emptyOf
535598
536599-------------------------------------------------------------------------------
537600-- Random writes
@@ -579,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
579642
580643 where
581644
582− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
583646
584647-- | Modify a given index of an array using a modifier function.
585648--
@@ -775,22 +838,19 @@ arrayChunkBytes = 1024
775838roundDownTo :: Int -> Int -> Int
776839roundDownTo elemSize size = size - (size `mod` elemSize)
777840
778−-- XXX See if resizing can be implemented by reading the old array as a stream
779−-- and then using writeN to the new array.
780−--
781841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
782842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
783843-- which complicates some inspection tests.
784844--
785−{-# NOINLINE reallocExplicit #-}
786−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
787−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
788848 assertM(arrEnd <= arrBound)
789849
790850 -- Allocate new array
791851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
792852 contents <-
793− if Unboxed.isPinned arrContents
853+ if ps == Pinned
794854 then Unboxed.pinnedNew newCapMaxInBytes
795855 else Unboxed.new newCapMaxInBytes
796856 let !(MutByteArray mbarrFrom#) = arrContents
@@ -815,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
815875 , arrBound = newCapInBytes
816876 }
817877
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+
818882-- | @realloc newCapacity array@ reallocates the array to the specified
819883-- capacity in bytes.
820884--
@@ -826,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
826890-- If the original array is pinned, the newly allocated array is also pinned.
827891{-# INLINABLE realloc #-}
828892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
829−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
830899
831900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
832901-- in error messages and the capSizer is used to determine the capacity of the
@@ -849,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
849918 where
850919
851920 badSize newSize =
852− concat
921+ Prelude.concat
853922 [ label
854923 , ": new array size (in bytes) is less than required size "
855924 , show newSize
856925 , ". Please check the sizing function passed."
857926 ]
858927
859−-- | @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
860929-- it is enough to hold the specified number of elements. Nothing is done if
861930-- the specified capacity is less than the length of the array.
862931--
@@ -864,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
864933-- the block size (4K).
865934--
866935-- /Pre-release/
867−{-# INLINE resize #-}
868−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
869938 Int -> MutArray a -> m (MutArray a)
870−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
871940 let req = SIZE_OF(a) * nElems
872− len = arrEnd - arrStart
873− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
874943 then return arr
875944 else realloc req arr
876945
877−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
878−-- 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.
879954--
880955-- /Pre-release/
881−{-# INLINE resizeExp #-}
882−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
883958 Int -> MutArray a -> m (MutArray a)
884−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
885960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
886961 req1 =
887962 if req > largeObjectThreshold
888963 then roundUpToPower2 req
889964 else req
890− len = arrEnd - arrStart
891− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
892967 then return arr
893968 else realloc req1 arr
894969
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+
895976-- | Resize the allocated memory to drop any reserved free space at the end of
896977-- the array and reallocate it to reduce wastage.
897978--
@@ -956,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9561037{-# INLINE snocMay #-}
9571038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9581039 MutArray a -> a -> m (Maybe (MutArray a))
959−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9601041 let newEnd = INDEX_NEXT(arrEnd,a)
9611042 if newEnd <= arrBound
9621043 then Just <$> snocNewEnd newEnd arr x
@@ -970,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9701051 -> a
9711052 -> m (MutArray a)
9721053snocWithRealloc sizer arr x = do
973− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9741055 snocUnsafe arr1 x
9751056
9761057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -992,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9921073 -> MutArray a
9931074 -> a
9941075 -> m (MutArray a)
995−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9961077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9971078 if newEnd <= arrBound arr
9981079 then snocNewEnd newEnd arr x
@@ -1036,6 +1117,126 @@ snoc = snocWith f
10361117 then oldSize * 2
10371118 else roundUpToPower2 oldSize * 2
10381119
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+
10391240-------------------------------------------------------------------------------
10401241-- Random reads
10411242-------------------------------------------------------------------------------
@@ -1078,10 +1279,10 @@ getIndexRev i MutArray{..} = do
10781279data GetIndicesState contents start end st =
10791280 GetIndicesState contents start end st
10801281
1081−{-# INLINE getIndicesWith #-}
1082−getIndicesWith :: (Monad m, Unbox a) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10831284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1084−getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10851286
10861287 where
10871288
@@ -1100,14 +1301,26 @@ getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
11001301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
11011302 D.Stop -> return D.Stop
11021303
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+
11031310-- | Given an unfold that generates array indices, read the elements on those
11041311-- indices from the supplied MutArray. An error is thrown if an index is out of
11051312-- bounds.
11061313--
11071314-- /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." #-}
11081321{-# INLINE getIndices #-}
11091322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1110−getIndices = getIndicesWith liftIO
1323+getIndices = indexReader
11111324
11121325-------------------------------------------------------------------------------
11131326-- Subarrays
@@ -1402,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
14021615--
14031616-- Same as the following but may be more efficient:
14041617--
1405−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
14061619--
14071620-- /Pre-release/
14081621{-# INLINE_NORMAL chunksOf #-}
@@ -1411,14 +1624,14 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
14111624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
14121625-- benchmark. It seems it has difficulty producing optimized code when
14131626-- converting to StreamK. Investigate GHC optimizations.
1414−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
14151628chunksOf = chunksOfAs Unpinned
14161629
14171630-- | Like 'chunksOf' but creates pinned arrays.
14181631{-# INLINE_NORMAL pinnedChunksOf #-}
14191632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
14201633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1421−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
14221635pinnedChunksOf = chunksOfAs Pinned
14231636
14241637-- | When we are buffering a stream of unknown size into an array we do not
@@ -1451,16 +1664,10 @@ data FlattenState s contents a =
14511664 OuterLoop s
14521665 | InnerLoop s contents !Int !Int
14531666
1454−-- | Use the "reader" unfold instead.
1455−--
1456−-- @concatChunks = unfoldMany reader@
1457−--
1458−-- We can try this if there are any fusion issues in the unfold.
1459−--
1460−{-# INLINE_NORMAL concatChunks #-}
1461−concatChunks :: forall m a. (MonadIO m, Unbox a)
1462− => D.Stream m (MutArray a) -> D.Stream m a
1463−concatChunks (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)
14641671
14651672 where
14661673
@@ -1477,24 +1684,30 @@ concatChunks (D.Stream step state) = D.Stream step' (OuterLoop state)
14771684 return $ D.Skip $ OuterLoop st
14781685
14791686 step' _ (InnerLoop st contents p end) = do
1480− x <- liftIO $ peekAt p contents
1687+ x <- liftio $ peekAt p contents
14811688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14821689
1483−{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1484−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1485− => D.Stream m (MutArray a) -> D.Stream m a
1486−flattenArrays = concatChunks
1487−
1488−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14891691--
1490−-- @concatChunks = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14911693--
14921694-- We can try this if there are any fusion issues in the unfold.
14931695--
1494−{-# INLINE_NORMAL concatChunksRev #-}
1495−concatChunksRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
14961698 => D.Stream m (MutArray a) -> D.Stream m a
1497−concatChunksRev (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)
14981711
14991712 where
15001713
@@ -1512,14 +1725,26 @@ concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
15121725 return $ D.Skip $ OuterLoop st
15131726
15141727 step' _ (InnerLoop st contents p start) = do
1515− x <- liftIO $ peekAt p contents
1728+ x <- liftio $ peekAt p contents
15161729 let cur = INDEX_PREV(p,a)
15171730 return $ D.Yield x (InnerLoop st contents cur start)
15181731
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+
15191743{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744+{-# INLINE flattenArraysRev #-}
15201745flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
15211746 => D.Stream m (MutArray a) -> D.Stream m a
1522−flattenArraysRev = concatChunksRev
1747+flattenArraysRev = concatRev
15231748
15241749-------------------------------------------------------------------------------
15251750-- Unfolds
@@ -1748,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
17481973
17491974-- XXX Keep the bound intact to not lose any free space? Perf impact?
17501975
1751−-- | @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
17521977-- array.
17531978--
17541979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1757,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
17571982-- Any free space left in the array after appending @n@ elements is lost.
17581983--
17591984-- /Internal/
1760−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1761−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17621987 Int
17631988 -> m (MutArray a)
17641989 -> Fold m a (MutArray a)
1765−writeAppendNUnsafe n action =
1766− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
17671991
17681992 where
17691993
@@ -1776,7 +2000,7 @@ writeAppendNUnsafe n action =
17762000 -- otherwise we lose that space.
17772001 arr1 <-
17782002 if free < needed
1779− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17802004 else return arr
17812005 return $ toArrayUnsafe arr1
17822006
@@ -1784,30 +2008,51 @@ writeAppendNUnsafe n action =
17842008 liftIO $ pokeAt end contents x
17852009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17862010
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+
17872019-- | Append @n@ elements to an existing array. Any free space left in the array
17882020-- after appending @n@ elements is lost.
17892021--
1790−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17912023--
1792−{-# 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 #-}
17932032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17942033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1795−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17962035
1797−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17982037-- append the input stream. If there is no reserved space available in the
17992038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
18002039-- where @oldSize@ is the current size of the array in bytes.
18012040--
18022041-- Note that the returned array may be a mutated version of original array.
18032042--
1804−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
18052044--
18062045-- /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." #-}
18072052{-# INLINE writeAppendWith #-}
18082053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
18092054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1810−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
18112056
18122057-- | @append action@ mutates the array generated by @action@ to append the
18132058-- input stream. If there is no reserved space available in the array it is
@@ -1815,26 +2060,33 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
18152060--
18162061-- Note that the returned array may be a mutated version of original array.
18172062--
1818−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
18192064--
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." #-}
18202072{-# INLINE writeAppend #-}
18212073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
18222074 m (MutArray a) -> Fold m a (MutArray a)
1823−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
18242076
18252077-- XXX We can carry bound as well in the state to make sure we do not lose the
18262078-- remaining capacity. Need to check perf impact.
18272079--
1828−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1829−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
18302082--
1831−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
18322084--
18332085-- /Pre-release/
1834−{-# INLINE_NORMAL writeNWithUnsafe #-}
1835−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
18362088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1837−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18382090
18392091 where
18402092
@@ -1845,40 +2097,64 @@ writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18452097 return
18462098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
18472099
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+
18482106{-# INLINE_NORMAL writeNUnsafeAs #-}
18492107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
18502108 => PinnedState -> Int -> Fold m a (MutArray a)
1851−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
18522110
1853−-- | 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
18542112-- driver must not call the step function more than 'n' times otherwise it will
18552113-- corrupt the memory and crash. This function exists mainly because any
18562114-- conditional in the step function blocks fusion causing 10x performance
18572115-- slowdown.
18582116--
1859−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
18602118--
1861−{-# 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 #-}
18622126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
18632127 => Int -> Fold m a (MutArray a)
1864−writeNUnsafe = writeNUnsafeAs Unpinned
2128+writeNUnsafe = unsafeCreateOf
18652129
1866−-- | Like 'writeNUnsafe' but creates a pinned array.
1867−{-# 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 #-}
18682138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
18692139 => Int -> Fold m a (MutArray a)
1870−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18712141
1872−-- | @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
18732143-- allocated using the @alloc@ function.
18742144--
1875−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1876−-- >>> 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
18772147--
1878−{-# 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 #-}
18792155writeNWith :: forall m a. (MonadIO m, Unbox a)
18802156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1881−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18822158
18832159{-# INLINE_NORMAL writeNAs #-}
18842160writeNAs ::
@@ -1886,28 +2162,42 @@ writeNAs ::
18862162 => PinnedState
18872163 -> Int
18882164 -> Fold m a (MutArray a)
1889−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18902166
1891−-- | @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
18922168-- 'MutArray'.
18932169--
1894−-- >>> writeN = MutArray.writeNWith MutArray.new
1895−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1896−-- >>> 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)
18972173--
1898−{-# 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 #-}
18992181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1900−writeN = writeNAs Unpinned
2182+writeN = createOf
19012183
1902−-- | Like 'writeN' but creates a pinned array.
1903−{-# INLINE_NORMAL pinnedWriteN #-}
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
2191+
2192+{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193+{-# INLINE pinnedWriteN #-}
19042194pinnedWriteN ::
19052195 forall m a. (MonadIO m, Unbox a)
19062196 => Int
19072197 -> Fold m a (MutArray a)
1908−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
19092199
1910−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
19112201--
19122202-- /Internal/
19132203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1928,7 +2218,7 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
19282218 return
19292219 $ ArrayUnsafe contents ptr end
19302220
1931−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
19322222--
19332223-- /Internal/
19342224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1936,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
19362226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
19372227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
19382228
1939−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
19402230--
19412231-- /Pre-release/
1942−{-# 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 #-}
19432238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1944−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
19452240
19462241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
19472242-- input stream to a 'MutArray' aligned to the given size.
19482243--
1949−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1950−-- >>> 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)
19512246--
19522247-- /Pre-release/
19532248--
19542249{-# INLINE_NORMAL pinnedWriteNAligned #-}
19552250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
19562251 => Int -> Int -> Fold m a (MutArray a)
1957−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
19582253
19592254-- XXX Buffer to a list instead?
19602255
19612256-- | Buffer a stream into a stream of arrays.
19622257--
1963−-- >>> buildChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
19642259--
19652260-- Breaking an array into an array stream can be useful to consume a large
19662261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1972,9 +2267,10 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
19722267{-# INLINE_NORMAL buildChunks #-}
19732268buildChunks :: (MonadIO m, Unbox a) =>
19742269 Int -> Fold m a (StreamK n (MutArray a))
1975−buildChunks n = FL.many (writeN n) FL.toStreamK
2270+buildChunks n = FL.many (createOf n) FL.toStreamK
19762271
19772272{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273+{-# INLINE writeChunks #-}
19782274writeChunks :: (MonadIO m, Unbox a) =>
19792275 Int -> Fold m a (StreamK n (MutArray a))
19802276writeChunks = buildChunks
@@ -1982,29 +2278,29 @@ writeChunks = buildChunks
19822278{-# INLINE_NORMAL writeWithAs #-}
19832279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19842280 => PinnedState -> Int -> Fold m a (MutArray a)
1985−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19862282writeWithAs ps elemCount =
19872283 FL.rmapM extract $ FL.foldlM' step initial
19882284
19892285 where
19902286
19912287 initial = do
1992− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1993− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19942290
19952291 step arr@(MutArray _ start end bound) x
19962292 | INDEX_NEXT(end,a) > bound = do
19972293 let oldSize = end - start
19982294 newSize = max (oldSize * 2) 1
1999− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
20002296 snocUnsafe arr1 x
20012297 step arr x = snocUnsafe arr x
20022298
20032299 extract = liftIO . rightSize
20042300
2005−-- XXX Compare writeWith with fromStreamD which uses an array of streams
2301+-- XXX Compare createWith with fromStreamD which uses an array of streams
20062302-- implementation. We can write this using buildChunks above if that is faster.
2007−-- If writeWith is faster then we should use that to implement
2303+-- If createWith is faster then we should use that to implement
20082304-- fromStreamD.
20092305--
20102306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -2019,39 +2315,56 @@ writeWithAs ps elemCount =
20192315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
20202316-- malloc to compare against.
20212317
2022−-- | @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
20232319-- starts at a size big enough to hold minCount elements, the size is doubled
20242320-- every time the array needs to be grown.
20252321--
20262322-- /Caution! Do not use this on infinite streams./
20272323--
2028−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
2029−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
2030−-- >>> writeWith n = Fold.rmapM MutArray.fromChunksK (MutArray.buildChunks 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)
20312327--
20322328-- /Pre-release/
2033−{-# 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 #-}
20342337writeWith :: forall m a. (MonadIO m, Unbox a)
20352338 => Int -> Fold m a (MutArray a)
2036−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
2037−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
20382340
20392341-- | Fold the whole input to a single array.
20402342--
2041−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
20422344-- rounded up to the element size.
20432345--
20442346-- /Caution! Do not use this on infinite streams./
20452347--
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." #-}
20462354{-# INLINE write #-}
20472355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2048−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
2357+
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)
20492363
2050−-- | Like 'write' but creates a pinned array.
2364+{-# DEPRECATED pinnedWrite "Please use pinnedCreate instead." #-}
20512365{-# INLINE pinnedWrite #-}
20522366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2053−pinnedWrite =
2054− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
20552368
20562369-------------------------------------------------------------------------------
20572370-- construct from streams, known size
@@ -2061,7 +2374,7 @@ pinnedWrite =
20612374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
20622375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
20632376fromStreamDNAs ps limit str = do
2064− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
20652378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
20662379 return $ arr {arrEnd = end}
20672380
@@ -2071,17 +2384,18 @@ fromStreamDNAs ps limit str = do
20712384 liftIO $ pokeAt ptr arrContents x
20722385 return $ INDEX_NEXT(ptr,a)
20732386
2074−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
20752388--
2076−-- >>> fromStreamN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20772390--
20782391{-# INLINE_NORMAL fromStreamN #-}
20792392fromStreamN :: forall m a. (MonadIO m, Unbox a)
20802393 => Int -> D.Stream m a -> m (MutArray a)
2081−-- fromStreamDN n = D.fold (writeN n)
2394+-- fromStreamDN n = D.fold (createOf n)
20822395fromStreamN = fromStreamDNAs Unpinned
20832396
20842397{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398+{-# INLINE fromStreamDN #-}
20852399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20862400 => Int -> D.Stream m a -> m (MutArray a)
20872401fromStreamDN = fromStreamN
@@ -2104,20 +2418,20 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
21042418-- /Pre-release/
21052419{-# INLINE fromListRevN #-}
21062420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2107−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
2421+fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
21082422
21092423-- | Convert a pure stream in Identity monad to a mutable array.
21102424{-# INLINABLE fromPureStreamN #-}
21112425fromPureStreamN :: (MonadIO m, Unbox a) =>
21122426 Int -> Stream Identity a -> m (MutArray a)
21132427fromPureStreamN n xs =
2114− liftIO $ D.fold (writeN n) $ D.morphInner (return . runIdentity) xs
2428+ D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
21152429
21162430-- | Convert a pure stream in Identity monad to a mutable array.
21172431{-# INLINABLE fromPureStream #-}
21182432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
21192433fromPureStream xs =
2120− liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
2434+ D.fold create $ D.morphInner (return . runIdentity) xs
21212435
21222436{-# INLINABLE fromPtrN #-}
21232437fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
@@ -2126,7 +2440,7 @@ fromPtrN len addr = do
21262440 -- XXX We can implement a stream copy in a similar way by streaming Word64
21272441 -- first and then remaining Word8.
21282442 arr <- new len
2129− _ <- asUnpinnedPtrUnsafe arr
2443+ _ <- unsafeAsPtr arr
21302444 (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
21312445 return (arr {arrEnd = len})
21322446
@@ -2142,7 +2456,7 @@ fromByteStr# addr = do
21422456 len <- liftIO $ c_strlen (Ptr addr)
21432457 let lenInt = fromIntegral len
21442458 arr <- new lenInt
2145− _ <- asUnpinnedPtrUnsafe arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2459+ _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
21462460 return (arr {arrEnd = lenInt})
21472461
21482462-------------------------------------------------------------------------------
@@ -2163,7 +2477,7 @@ fromChunksRealloced s = do
21632477 arr <- D.foldlM' spliceExp (pure a) strm
21642478 -- Reallocation is exponential so there may be 50% empty space in
21652479 -- worst case. One more reallocation to reclaim the space.
2166− liftIO $ rightSize arr
2480+ rightSize arr
21672481 Nothing -> pure nil
21682482
21692483-------------------------------------------------------------------------------
@@ -2201,6 +2515,7 @@ fromChunksK :: (Unbox a, MonadIO m) =>
22012515fromChunksK = fromChunkskAs Unpinned
22022516
22032517{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
2518+{-# INLINE fromArrayStreamK #-}
22042519fromArrayStreamK :: (Unbox a, MonadIO m) =>
22052520 StreamK m (MutArray a) -> m (MutArray a)
22062521fromArrayStreamK = fromChunksK
@@ -2212,7 +2527,7 @@ fromStreamDAs ps m =
22122527 arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
22132528
22142529-- | Create an 'Array' from a stream. This is useful when we want to create a
2215−-- single array from a stream of unknown size. 'writeN' is at least twice
2530+-- single array from a stream of unknown size. 'createOf' is at least twice
22162531-- as efficient when the size is already known.
22172532--
22182533-- Note that if the input stream is too large memory allocation for the array
@@ -2224,19 +2539,19 @@ fromStreamDAs ps m =
22242539fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
22252540fromStream = fromStreamDAs Unpinned
22262541
2227−-- fromStream (Stream m) = P.fold write m
2542+-- fromStream (Stream m) = P.fold create m
22282543-- CAUTION: a very large number (millions) of arrays can degrade performance
22292544-- due to GC overhead because we need to buffer the arrays before we flatten
22302545-- all the arrays.
22312546--
2232−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
22332548--
22342549-- | We could take the approach of doubling the memory allocation on each
22352550-- overflow. This would result in more or less the same amount of copying as in
22362551-- the chunking approach. However, if we have to shrink in the end then it may
22372552-- result in an extra copy of the entire data.
22382553--
2239−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
22402555--
22412556{-# INLINE fromStreamD #-}
22422557{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
@@ -2266,6 +2581,10 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
22662581-- Cloning
22672582-------------------------------------------------------------------------------
22682583
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+
22692588{-# INLINE cloneAs #-}
22702589cloneAs ::
22712590 ( MonadIO m
@@ -2275,7 +2594,7 @@ cloneAs ::
22752594 )
22762595 => PinnedState -> MutArray a -> m (MutArray a)
22772596cloneAs ps src =
2278− liftIO $ do
2597+ do
22792598 let startSrc = arrStart src
22802599 srcLen = arrEnd src - startSrc
22812600 newArrContents <-
@@ -2314,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
23142633 Unbox a =>
23152634#endif
23162635 MutArray a -> MutArray a -> m (MutArray a)
2317−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
23182637 let start1 = arrStart arr1
23192638 start2 = arrStart arr2
23202639 len1 = arrEnd arr1 - start1
23212640 len2 = arrEnd arr2 - start2
2322− let newLen = len1 + len2
2641+ let len = len1 + len2
23232642 newArrContents <-
23242643 if Unboxed.isPinned (arrContents arr1)
2325− then Unboxed.pinnedNew newLen
2326− else Unboxed.new newLen
2327− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
23282646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
23292647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
23302648 return $ MutArray newArrContents 0 len len
@@ -2336,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
23362654spliceUnsafe :: MonadIO m =>
23372655 MutArray a -> MutArray a -> m (MutArray a)
23382656spliceUnsafe dst src =
2339− liftIO $ do
2657+ do
23402658 let startSrc = arrStart src
23412659 srcLen = arrEnd src - startSrc
23422660 endDst = arrEnd dst
@@ -2358,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
23582676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
23592677spliceWith sizer dst@(MutArray _ start end bound) src = do
23602678{-
2361− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
23622680 in D.fold f (toStreamD src)
23632681-}
23642682 assert (end <= bound) (return ())
@@ -2374,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
23742692 $ "splice: newSize is less than the total size "
23752693 ++ "of arrays being appended. Please check the "
23762694 ++ "sizer function passed."
2377− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
23782696 else return dst
23792697 spliceUnsafe dst1 src
23802698
@@ -2426,7 +2744,7 @@ splitOn predicate arr =
24262744{-# INLINE breakOn #-}
24272745breakOn :: MonadIO m
24282746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2429−breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
24302748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
24312749 -- Need efficient stream based primitives that work on Word64.
24322750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
@@ -2449,33 +2767,41 @@ breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
24492767 }
24502768 )
24512769
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+
24522792-- | Create two slices of an array without copying the original array. The
24532793-- specified index @i@ is the first index of the second slice.
24542794--
24552795{-# INLINE splitAt #-}
24562796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2457−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
24582798 let maxIndex = length arr - 1
24592799 in if i < 0
24602800 then error "sliceAt: negative array index"
24612801 else if i > maxIndex
24622802 then error $ "sliceAt: specified array index " ++ show i
24632803 ++ " is beyond the maximum index " ++ show maxIndex
2464− else let off = i * SIZE_OF(a)
2465− p = arrStart + off
2466− in ( MutArray
2467− { arrContents = arrContents
2468− , arrStart = arrStart
2469− , arrEnd = p
2470− , arrBound = p
2471− }
2472− , MutArray
2473− { arrContents = arrContents
2474− , arrStart = p
2475− , arrEnd = arrEnd
2476− , arrBound = arrBound
2477− }
2478− )
2804+ else unsafeSplitAt i arr
24792805
24802806-------------------------------------------------------------------------------
24812807-- Casting
@@ -2525,9 +2851,13 @@ cast arr =
25252851-- guarantees that unsafe calls will be performed in the calling thread. Making
25262852-- it safe to pass heap-allocated objects to unsafe functions.
25272853
2528−-- 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
25292855-- operations. And we are safe from Haskell perspective because we will be
25302856-- 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.
25312861
25322862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
25332863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2545,16 +2875,21 @@ cast arr =
25452875--
25462876-- /Pre-release/
25472877--
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." #-}
25482885{-# INLINE asPtrUnsafe #-}
25492886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2550−asPtrUnsafe arr f =
2551− Unboxed.asPtrUnsafe
2552− (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2887+asPtrUnsafe = unsafePinnedAsPtr
25532888
2554−{-# INLINE asUnpinnedPtrUnsafe #-}
2555−asUnpinnedPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2556−asUnpinnedPtrUnsafe arr f =
2557− Unboxed.asUnpinnedPtrUnsafe
2889+{-# INLINE unsafeAsPtr #-}
2890+unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891+unsafeAsPtr arr f =
2892+ Unboxed.unsafeAsPtr
25582893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
25592894
25602895-------------------------------------------------------------------------------
@@ -2572,28 +2907,26 @@ asUnpinnedPtrUnsafe arr f =
25722907-- /Pre-release/
25732908{-# INLINE byteCmp #-}
25742909byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2575−byteCmp arr1 arr2 =
2576− liftIO
2577− $ do
2578− let marr1 = getMutableByteArray# (arrContents arr1)
2579− marr2 = getMutableByteArray# (arrContents arr2)
2580− !(I# st1#) = arrStart arr1
2581− !(I# st2#) = arrStart arr2
2582− !(I# len#) = byteLength arr1
2583− case compare (byteLength arr1) (byteLength arr2) of
2584− EQ -> do
2585− r <- liftIO $ IO $ \s# ->
2586− let res =
2587− I#
2588− (compareByteArrays#
2589− (unsafeCoerce# marr1)
2590− st1#
2591− (unsafeCoerce# marr2)
2592− st2#
2593− len#)
2594− in (# s#, res #)
2595− return $ compare r 0
2596− x -> return x
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
25972930
25982931{-# INLINE cmp #-}
25992932{-# DEPRECATED cmp "Please use byteCmp instead." #-}
@@ -2613,46 +2946,48 @@ byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
26132946-- Compact
26142947-------------------------------------------------------------------------------
26152948
2616−-- | Parser @pCompactChunksLE n@ coalesces adjacent arrays in the input stream
2617−-- only if the combined size would be less than or equal to n.
2618−--
2619−-- /Internal/
2620−{-# INLINE_NORMAL pCompactChunksLE #-}
2621−pCompactChunksLE ::
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 ::
26222956 forall m a. (MonadIO m, Unbox a)
2623− => Int -> Parser (MutArray a) m (MutArray a)
2624−pCompactChunksLE n = Parser step initial extract
2957+ => PinnedState -> Int -> Parser (MutArray a) m (MutArray a)
2958+pCompactLeAs ps maxElems = Parser step initial extract
26252959
26262960 where
26272961
2628− nBytes = n * SIZE_OF(a)
2962+ maxBytes = maxElems * SIZE_OF(a)
26292963
2630− functionName = "Streamly.Internal.Data.MutArray.pCompactChunksLE"
2964+ functionName = "Streamly.Internal.Data.MutArray.pCompactLE"
26312965
26322966 initial =
26332967 return
2634− $ if n <= 0
2968+ $ if maxElems <= 0
26352969 then error
26362970 $ functionName
26372971 ++ ": the size of arrays ["
2638− ++ show n ++ "] must be a natural number"
2972+ ++ show maxElems ++ "] must be a natural number"
26392973 else Parser.IPartial Nothing
26402974
26412975 step Nothing arr =
26422976 return
26432977 $ let len = byteLength arr
2644− in if len >= nBytes
2978+ in if len >= maxBytes
26452979 then Parser.Done 0 arr
26462980 else Parser.Partial 0 (Just arr)
2981+ -- XXX Split the last array to use the space more compactly.
26472982 step (Just buf) arr =
26482983 let len = byteLength buf + byteLength arr
2649− in if len > nBytes
2984+ in if len > maxBytes
26502985 then return $ Parser.Done 1 buf
26512986 else do
2652− -- XXX check this in the Nothing case
26532987 buf1 <-
2654− if byteCapacity buf < nBytes
2655− then liftIO $ realloc nBytes buf
2988+ if byteCapacity buf < maxBytes
2989+ then liftIO $ reallocExplicitAs
2990+ ps (SIZE_OF(a)) maxBytes buf
26562991 else return buf
26572992 buf2 <- spliceUnsafe buf1 arr
26582993 return $ Parser.Partial 0 (Just buf2)
@@ -2660,6 +2995,30 @@ pCompactChunksLE n = Parser step initial extract
26602995 extract Nothing = return $ Parser.Done 0 nil
26612996 extract (Just buf) = return $ Parser.Done 0 buf
26622997
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+
26633022data SpliceState s arr
26643023 = SpliceInitial s
26653024 | SpliceBuffering s arr
@@ -2671,48 +3030,47 @@ data SpliceState s arr
26713030-- immutable array never has additional space so a new array is allocated
26723031-- instead of mutating it.
26733032
2674−-- | Scan @rCompactChunksLE n@ coalesces adjacent arrays in the input stream
2675−-- only if the combined size would be less than or equal to n.
2676−--
2677−-- /Internal/
2678−{-# INLINE_NORMAL rCompactChunksLE #-}
2679−rCompactChunksLE :: (MonadIO m, Unbox a)
2680− => Int -> D.Stream m (MutArray a) -> D.Stream m (MutArray a)
2681−rCompactChunksLE n (D.Stream step state) =
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) =
26823037 D.Stream step' (SpliceInitial state)
26833038
26843039 where
26853040
2686− functionName = "Streamly.Internal.Data.MutArray.rCompactChunksLE"
3041+ maxBytes = maxElems * SIZE_OF(a)
3042+
3043+ functionName = "Streamly.Internal.Data.MutArray.rCompactLE"
26873044
26883045 {-# INLINE_LATE step' #-}
26893046 step' gst (SpliceInitial st) = do
2690− when (n <= 0) $
3047+ when (maxElems <= 0) $
26913048 -- XXX we can pass the module string from the higher level API
2692− error $ functionName ++ ": the size of arrays [" ++ show n
3049+ error $ functionName ++ ": the size of arrays [" ++ show maxElems
26933050 ++ "] must be a natural number"
26943051 r <- step gst st
26953052 case r of
26963053 D.Yield arr s -> return $
26973054 let len = byteLength arr
2698− in if len >= n
3055+ in if len >= maxBytes
26993056 then D.Skip (SpliceYielding arr (SpliceInitial s))
27003057 else D.Skip (SpliceBuffering s arr)
27013058 D.Skip s -> return $ D.Skip (SpliceInitial s)
27023059 D.Stop -> return D.Stop
27033060
3061+ -- XXX Split the last array to use the space more compactly.
27043062 step' gst (SpliceBuffering st buf) = do
27053063 r <- step gst st
27063064 case r of
27073065 D.Yield arr s -> do
27083066 let len = byteLength buf + byteLength arr
2709− if len > n
3067+ if len > maxBytes
27103068 then return $
27113069 D.Skip (SpliceYielding buf (SpliceBuffering s arr))
27123070 else do
2713− -- XXX check this in the SpliceInitial case
2714− buf1 <- if byteCapacity buf < n
2715− then liftIO $ realloc n buf
3071+ buf1 <- if byteCapacity buf < maxBytes
3072+ then liftIO $ reallocExplicitAs
3073+ ps (SIZE_OF(a)) maxBytes buf
27163074 else return buf
27173075 buf2 <- spliceUnsafe buf1 arr
27183076 return $ D.Skip (SpliceBuffering s buf2)
@@ -2723,113 +3081,124 @@ rCompactChunksLE n (D.Stream step state) =
27233081
27243082 step' _ (SpliceYielding arr next) = return $ D.Yield arr next
27253083
2726−-- | Fold @fCompactChunksGE n@ coalesces adjacent arrays in the input stream
2727−-- until the size becomes greater than or equal to n.
2728−--
2729−{-# INLINE_NORMAL fCompactChunksGE #-}
2730−fCompactChunksGE ::
3084+
3085+{-# INLINE_NORMAL fCompactGeAs #-}
3086+fCompactGeAs ::
27313087 forall m a. (MonadIO m, Unbox a)
2732− => Int -> FL.Fold m (MutArray a) (MutArray a)
2733−fCompactChunksGE n = Fold step initial extract extract
3088+ => PinnedState -> Int -> FL.Fold m (MutArray a) (MutArray a)
3089+fCompactGeAs ps minElems = Fold step initial extract extract
27343090
27353091 where
27363092
2737− nBytes = n * SIZE_OF(a)
3093+ minBytes = minElems * SIZE_OF(a)
27383094
2739− functionName = "Streamly.Internal.Data.MutArray.fCompactChunksGE"
3095+ functionName = "Streamly.Internal.Data.MutArray.fCompactGE"
27403096
27413097 initial =
27423098 return
2743− $ if n < 0
3099+ $ if minElems < 0
27443100 then error
27453101 $ functionName
27463102 ++ ": the size of arrays ["
2747− ++ show n ++ "] must be a natural number"
3103+ ++ show minElems ++ "] must be a natural number"
27483104 else FL.Partial Nothing
27493105
27503106 step Nothing arr =
27513107 return
27523108 $ let len = byteLength arr
2753− in if len >= nBytes
3109+ in if len >= minBytes
27543110 then FL.Done arr
27553111 else FL.Partial (Just arr)
3112+ -- XXX Buffer arrays as a list to avoid copy and reallocations
27563113 step (Just buf) arr = do
27573114 let len = byteLength buf + byteLength arr
27583115 buf1 <-
27593116 if byteCapacity buf < len
2760− then liftIO $ realloc (max len nBytes) buf
3117+ then liftIO $ reallocExplicitAs
3118+ ps (SIZE_OF(a)) (max minBytes len) buf
27613119 else return buf
27623120 buf2 <- spliceUnsafe buf1 arr
2763− if len >= n
3121+ if len >= minBytes
27643122 then return $ FL.Done buf2
27653123 else return $ FL.Partial (Just buf2)
27663124
27673125 extract Nothing = return nil
27683126 extract (Just buf) = return buf
27693127
2770−-- | Like 'compactChunksGE' but for transforming folds instead of stream.
2771−--
2772−-- >>> lCompactChunksGE n = Fold.many (MutArray.fCompactChunksGE n)
3128+-- | Fold @fCompactGE minElems@ coalesces adjacent arrays in the input stream
3129+-- until the size becomes greater than or equal to @minElems@.
27733130--
2774−{-# INLINE_NORMAL lCompactChunksGE #-}
2775−lCompactChunksGE :: (MonadIO m, Unbox a)
2776− => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
2777−-- lCompactChunksGE n = Fold.many (fCompactChunksGE n)
2778−lCompactChunksGE n (Fold step1 initial1 _ final1) =
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) =
27793152 Fold step initial extract final
27803153
27813154 where
27823155
2783− functionName = "Streamly.Internal.Data.MutArray.lCompactChunksGE"
3156+ minBytes = minElems * SIZE_OF(a)
3157+
3158+ functionName = "Streamly.Internal.Data.MutArray.lCompactGE"
27843159
27853160 initial = do
2786− when (n <= 0) $
3161+ when (minElems <= 0) $
27873162 -- XXX we can pass the module string from the higher level API
27883163 error $ functionName ++ ": the size of arrays ["
2789− ++ show n ++ "] must be a natural number"
3164+ ++ show minElems ++ "] must be a natural number"
27903165
27913166 r <- initial1
27923167 return $ first (Tuple' Nothing) r
27933168
2794− step (Tuple' Nothing r1) arr =
2795− let len = byteLength arr
2796− in if len >= n
2797− then do
2798− r <- step1 r1 arr
2799− case r of
2800− FL.Done _ -> return $ FL.Done ()
2801− FL.Partial s -> do
2802− _ <- final1 s
2803− res <- initial1
2804− return $ first (Tuple' Nothing) res
2805− else return $ FL.Partial $ Tuple' (Just arr) r1
2806−
2807− step (Tuple' (Just buf) r1) arr = do
2808− let len = byteLength buf + byteLength arr
2809− buf1 <- if byteCapacity buf < len
2810− then liftIO $ realloc (max n len) buf
2811− else return buf
2812− buf2 <- spliceUnsafe buf1 arr
2813−
2814− -- XXX this is common in both the equations of step
2815− if len >= n
3169+ {-# INLINE runInner #-}
3170+ runInner len acc buf =
3171+ if len >= minBytes
28163172 then do
2817− r <- step1 r1 buf2
3173+ r <- step1 acc buf
28183174 case r of
28193175 FL.Done _ -> return $ FL.Done ()
28203176 FL.Partial s -> do
28213177 _ <- final1 s
28223178 res <- initial1
28233179 return $ first (Tuple' Nothing) res
2824− else return $ FL.Partial $ Tuple' (Just buf2) r1
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
28253194
28263195 -- XXX Several folds do extract >=> final, therefore, we need to make final
2827− -- return "m b" rather than using extract post it if we want extract to be
3196+ -- return "m b" rather than using extract post it if we want extract to be
28283197 -- partial.
28293198 --
28303199 -- extract forces the pending buffer to be sent to the fold which is not
28313200 -- what we want.
2832− extract _ = error "lCompactChunksGE: not designed for scanning"
3201+ extract _ = error "lCompactGE: not designed for scanning"
28333202
28343203 final (Tuple' Nothing r1) = final1 r1
28353204 final (Tuple' (Just buf) r1) = do
@@ -2838,25 +3207,42 @@ lCompactChunksGE n (Fold step1 initial1 _ final1) =
28383207 FL.Partial rr -> final1 rr
28393208 FL.Done _ -> return ()
28403209
2841−-- | @compactChunksGE n stream@ coalesces adjacent arrays in the @stream@ until
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
28423228-- the size becomes greater than or equal to @n@.
28433229--
2844−-- >>> compactChunksGE n = Stream.foldMany (MutArray.fCompactChunksGE n)
3230+-- >>> compactGE n = Stream.foldMany (MutArray.fCompactGE n)
28453231--
2846−{-# INLINE compactChunksGE #-}
2847−compactChunksGE ::
3232+{-# INLINE compactGE #-}
3233+compactGE ::
28483234 (MonadIO m, Unbox a)
28493235 => Int -> Stream m (MutArray a) -> Stream m (MutArray a)
2850−compactChunksGE n = D.foldMany (fCompactChunksGE n)
3236+compactGE n = D.foldMany (fCompactGE n)
28513237
2852−-- | 'compactChunksEQ n' coalesces adajacent arrays in the input stream to
3238+-- | 'compactEQ n' coalesces adajacent arrays in the input stream to
28533239-- arrays of exact size @n@.
28543240--
28553241-- /Unimplemented/
2856−{-# INLINE compactChunksEQ #-}
2857−compactChunksEQ :: -- (MonadIO m, Unbox a) =>
3242+{-# INLINE compactEQ #-}
3243+compactEQ :: -- (MonadIO m, Unbox a) =>
28583244 Int -> Stream m (MutArray a) -> Stream m (MutArray a)
2859−compactChunksEQ _n = undefined -- D.parseManyD (pCompactChunksEQ n)
3245+compactEQ _n = undefined -- D.parseManyD (pCompactEQ n)
28603246
28613247-------------------------------------------------------------------------------
28623248-- In-place mutation algorithms