#13502: Static argument transformation should also run after specialisation -------------------------------------+------------------------------------- Reporter: mpickering | Owner: (none) Type: feature request | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Description changed by mpickering: Old description:
Consider the following program where we eventually want `thepayload` to simplify to the same code as `direct`.
{{{#!hs {-# LANGUAGE ExistentialQuantification, RankNTypes, DeriveFunctor #-} module Foo where
newtype Q a b t = Q { getQ :: forall f . Applicative f => (a -> f b) -> f t } deriving Functor
instance Applicative (Q a b) where pure a = Q (\_ -> pure a) (Q ab) <*> (Q a) = (Q (\v -> ab v <*> a v))
singleQ :: a -> Q a b b singleQ a = Q (\f -> f a)
data L a = Nil | L a (L a) deriving Show
traverseList :: Applicative f => (a -> f b) -> L a -> f (L b) traverseList f Nil = pure Nil traverseList f (L a la) = L <$> f a <*> traverseList f la
newtype Identity a = Identity { runIdentity :: a } deriving (Functor, Show)
instance Applicative Identity where pure = Identity (Identity f) <*> (Identity x) = Identity (f x)
thepayload :: L String -> L String thepayload l = runIdentity $ (getQ $ (traverseList singleQ l)) Identity
direct :: L String -> L String direct Nil = Nil direct (L a b) = L a b }}}
With `ghc-8.0.2` and `-fstatic-argument-transformation`, the specialiser will specialise the call to `traverseList` and leave us with a definition like,
{{{#!hs rec (20) $ssat_worker= λ sg sc l → case l of Nil→ pure sc Nil L a la → <*> sc (fmap ($p1Applicative sc) (L a)) ($ssat_worker sg sc la)
(4) thepayload = λl→ $ssat_worker $fApplicativeIdentity l }}}
`$ssat_worked` is recursive in `l` but not in the other two arguments so we can also apply SAT here.
Notice that `$ssat_worker` is called with a statically known dictionary in `thepayload` and so if we can inline `$ssat_worker` we would get the same code as the naive `direct`, as desired.
I verified that inserting another SAT pass later in the compilation pipeline does have the desired effect but am not sure where exactly the right place would be or whether it is in general desirable.
New description: Consider the following program where we eventually want `thepayload` to simplify to the same code as `direct`. {{{#!hs {-# LANGUAGE ExistentialQuantification, RankNTypes, DeriveFunctor #-} module Foo where newtype Q a b t = Q { getQ :: forall f . Applicative f => (a -> f b) -> f t } deriving Functor instance Applicative (Q a b) where pure a = Q (\_ -> pure a) (Q ab) <*> (Q a) = (Q (\v -> ab v <*> a v)) singleQ :: a -> Q a b b singleQ a = Q (\f -> f a) data L a = Nil | L a (L a) deriving Show traverseList :: Applicative f => (a -> f b) -> L a -> f (L b) traverseList f Nil = pure Nil traverseList f (L a la) = L <$> f a <*> traverseList f la newtype Identity a = Identity { runIdentity :: a } deriving (Functor, Show) instance Applicative Identity where pure = Identity (Identity f) <*> (Identity x) = Identity (f x) thepayload :: L String -> L String thepayload l = runIdentity $ (getQ $ (traverseList singleQ l)) Identity direct :: L String -> L String direct Nil = Nil direct (L a b) = L a (direct b) }}} With `ghc-8.0.2` and `-fstatic-argument-transformation`, the specialiser will specialise the call to `traverseList` and leave us with a definition like, {{{#!hs rec (20) $ssat_worker= λ sg sc l → case l of Nil→ pure sc Nil L a la → <*> sc (fmap ($p1Applicative sc) (L a)) ($ssat_worker sg sc la) (4) thepayload = λl→ $ssat_worker $fApplicativeIdentity l }}} `$ssat_worked` is recursive in `l` but not in the other two arguments so we can also apply SAT here. Notice that `$ssat_worker` is called with a statically known dictionary in `thepayload` and so if we can inline `$ssat_worker` we would get the same code as the naive `direct`, as desired. I verified that inserting another SAT pass later in the compilation pipeline does have the desired effect but am not sure where exactly the right place would be or whether it is in general desirable. -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13502#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler