synthesizer-core-0.4: src/Synthesizer/Storable/Cut.hs
module Synthesizer.Storable.Cut where
import qualified Synthesizer.Storable.Signal as Sig
import qualified Data.StorableVector as SV
import qualified Data.StorableVector.Lazy as SVL
import qualified Data.StorableVector.ST.Strict as SVST
import Control.Monad.ST.Strict (ST, runST, )
import qualified Data.EventList.Relative.TimeBody as EventList
import qualified Data.EventList.Relative.TimeMixed as EventListTM
import qualified Data.EventList.Absolute.TimeBody as AbsEventList
import Control.Monad.Trans.State (runState, modify, gets, put, )
-- import Control.Monad (mapM, )
import Data.Tuple.HT (mapSnd, )
-- import qualified Algebra.RealRing as RealRing
import qualified Algebra.Additive as Additive
import qualified Number.NonNegative as NonNeg
import Foreign.Storable (Storable)
import NumericPrelude.Base
import NumericPrelude.Numeric
{-# INLINE arrange #-}
arrange :: (Storable v, Additive.C v) =>
Sig.ChunkSize
-> EventList.T NonNeg.Int (Sig.T v)
{-^ A list of pairs: (relative start time, signal part),
The start time is relative to the start time
of the previous event. -}
-> Sig.T v
{-^ The mixed signal. -}
arrange =
arrangeEquidist
{- |
Chunk sizes are adapted to the time differences.
Explicit ChunkSize parameter is only required for zero padding.
Since no ST monad is needed, this can be generalized to Generic.Signal.Transform class.
-}
arrangeAdaptive :: (Storable v, Additive.C v) =>
Sig.ChunkSize
-> EventList.T NonNeg.Int (Sig.T v)
{-^ A list of pairs: (relative start time, signal part),
The start time is relative to the start time
of the previous event. -}
-> Sig.T v
{-^ The mixed signal. -}
arrangeAdaptive size =
uncurry Sig.append .
flip runState Sig.empty .
fmap (Sig.concat . EventList.getTimes) .
EventList.mapM
(\timeNN ->
let time = NonNeg.toNumber timeNN
in do (prefix,suffix) <- gets (Sig.splitAtPad size time)
put suffix
return prefix)
(\body ->
modify (Sig.mixSndPattern body))
{- |
This function also uses the time differences as chunk sizes,
but may occasionally use smaller chunk sizes due to the chunk structure
of an input signal until the next signal starts.
-}
arrangeList :: (Storable v, Additive.C v) =>
Sig.ChunkSize
-> EventList.T NonNeg.Int (Sig.T v)
{-^ A list of pairs: (relative start time, signal part),
The start time is relative to the start time
of the previous event. -}
-> Sig.T v
{-^ The mixed signal. -}
arrangeList size evs =
let xs = EventList.getBodies evs
in case EventList.getTimes evs of
t:ts -> Sig.replicate size (NonNeg.toNumber t) zero `Sig.append`
addShiftedMany size ts xs
[] -> Sig.empty
addShiftedMany :: (Storable a, Additive.C a) =>
Sig.ChunkSize -> [NonNeg.Int] -> [Sig.T a] -> Sig.T a
addShiftedMany size ds xss =
foldr (uncurry (addShifted size)) Sig.empty (zip (ds++[0]) xss)
{-
It is crucial that 'mix' uses the chunk size structure of the second operand.
This way we avoid unnecessary and even infinite look-ahead.
-}
addShifted :: (Storable a, Additive.C a) =>
Sig.ChunkSize -> NonNeg.Int -> Sig.T a -> Sig.T a -> Sig.T a
addShifted size delNN px py =
let del = NonNeg.toNumber delNN
in uncurry Sig.append $
mapSnd (flip Sig.mixSndPattern py) $
Sig.splitAtPad size del px
{-
arrangeEquidist (Sig.chunkSize 2) (EventList.fromPairList [(10, SVL.pack SVL.defaultChunkSize [1..8::Double]), (2, SVL.pack (Sig.chunkSize 2) $ [4,3,2,1::Double] ++ undefined)])
-}
{- |
The result is a Lazy StorableVector with chunks of the given size.
-}
{-# INLINE arrangeEquidist #-}
arrangeEquidist :: (Storable v, Additive.C v) =>
Sig.ChunkSize
-> EventList.T NonNeg.Int (Sig.T v)
{-^ A list of pairs: (relative start time, signal part),
The start time is relative to the start time
of the previous event. -}
-> Sig.T v
{-^ The mixed signal. -}
arrangeEquidist (SVL.ChunkSize sz) =
let sznn = NonNeg.fromNumberMsg "arrangeEquidist" sz
go acc evs =
let (now,future) = EventListTM.splitAtTime sznn evs
xs =
AbsEventList.toPairList $
EventList.toAbsoluteEventList 0 $
EventListTM.switchTimeR const now
(chunk,newAcc) =
runST
(do v <- SVST.new sz zero
newAcc0 <- mapM (addToBuffer v 0) acc
-- newAcc1 <- AbsEventList.mapM (addToBuffer v) xs
newAcc1 <-
mapM (\(i,s) -> addToBuffer v (NonNeg.toNumber i) s) xs
vf <- SVST.freeze v
return (vf, newAcc0++newAcc1))
(ends, suffixes) = unzip $ newAcc
prefix =
{- if there are more events to come,
we must pad with zeros -}
if EventList.null future
then SV.take (foldl max 0 ends) chunk
else chunk
in if SV.null prefix
then []
else prefix : go (filter (not . Sig.null) suffixes) future
in Sig.fromChunks . go []
{-
{-# INLINE addToBuffer #-}
addToBuffer :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> Sig.T a -> ST s (Int, Sig.T a)
addToBuffer v start =
let n = SVST.length v
go i [] = return (i, [])
go i (c:cs) =
let end = i + SV.length c
in addChunkToBuffer v i c >>
if end<n
then go end cs
else return (n, SV.drop (end-n) c : cs)
in fmap (mapSnd SigSt.fromChunks) . go start . SigSt.chunks
addChunkToBuffer :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> SV.Vector a -> ST s ()
addChunkToBuffer v start xs =
let n = SVST.length v
in SV.foldr
(\x continue i ->
SVST.modify v i (x Additive.+) >>
continue (succ i))
(\_i -> return ())
(Sig.take (n Additive.- start) xs)
start
-}
{-# INLINE addToBuffer #-}
addToBuffer :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> Sig.T a -> ST s (Int, Sig.T a)
addToBuffer v start xs =
let n = SVST.length v
(now,future) = Sig.splitAt (n Additive.- start) xs
go i [] = return i
go i (c:cs) =
addChunkToBuffer v i c >>
go (i Additive.+ SV.length c) cs
in fmap (flip (,) future) . go start . Sig.chunks $ now
{- | chunk must fit into the buffer -}
{- This implementation will be faster as long as 'SV.foldr' is inefficient. -}
{-# INLINE addChunkToBuffer #-}
addChunkToBuffer :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> SV.Vector a -> ST s ()
addChunkToBuffer v start xs =
let go i j =
if j >= SV.length xs
then return ()
else
SVST.unsafeModify v i (SV.index xs j Additive.+) >>
go (i Additive.+ 1) (j Additive.+ 1)
in go start 0
{- | chunk must fit into the buffer -}
{-# INLINE addChunkToBufferFoldr #-}
addChunkToBufferFoldr :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> SV.Vector a -> ST s ()
addChunkToBufferFoldr v start xs =
SV.foldr
(\x continue i ->
SVST.unsafeModify v i (x Additive.+) >>
continue (succ i))
(\_i -> return ())
xs start
-- most elegant solution, but slow because StorableVector.foldr is slow
{-# INLINE addToBufferFoldr #-}
addToBufferFoldr :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> Sig.T a -> ST s (Int, Sig.T a)
addToBufferFoldr v start xs =
let n = SVST.length v
(now,future) = Sig.splitAt (n Additive.- start) xs
in Sig.foldr
(\x continue i ->
SVST.modify v i (x Additive.+) >>
continue (succ i))
(\i -> return (i, future))
now start
{-
Using @Sig.switchL@ in an inner loop
is slower than using @Sig.foldr@.
Using a StorableVectorPointer would be faster,
but I think still slower than @foldr@.
-}
addToBufferSwitchL :: (Storable a, Additive.C a) =>
SVST.Vector s a -> Int -> Sig.T a -> ST s (Int, Sig.T a)
addToBufferSwitchL v start =
let n = SVST.length v
{-# INLINE go #-}
go i =
if i>=n
then return . (,) i
else
Sig.switchL
(return (i, Sig.empty))
(\x xs ->
SVST.modify v i (x Additive.+) >>
go (succ i) xs)
in go start