module SpecConstr( specConstrProgram #ifdef GHCI , SpecConstrAnnotation(..) #endif ) where #include "HsVersions.h" import CoreSyn import CoreSubst import CoreUtils import CoreUnfold ( couldBeSmallEnoughToInline ) import CoreFVs ( exprsFreeVars ) import CoreMonad import Literal ( litIsLifted ) import HscTypes ( ModGuts(..) ) import WwLib ( mkWorkerArgs ) import DataCon import Coercion hiding( substTy, substCo ) import Rules import Type hiding ( substTy ) import TyCon ( isRecursiveTyCon, tyConName ) import Id import PprCore ( pprParendExpr ) import MkCore ( mkImpossibleExpr ) import Var import VarEnv import VarSet import Name import BasicTypes import DynFlags ( DynFlags(..) ) import StaticFlags ( opt_PprStyle_Debug ) import Maybes ( orElse, catMaybes, isJust, isNothing ) import Demand import Serialized ( deserializeWithData ) import Util import Pair import UniqSupply import Outputable import FastString import UniqFM import MonadUtils import Control.Monad ( zipWithM ) import Data.List import PrelNames ( specTyConName ) -- See Note [Forcing specialisation] #ifndef GHCI type SpecConstrAnnotation = () #else import TyCon ( TyCon ) import GHC.Exts( SpecConstrAnnotation(..) ) #endif\end{code} ----------------------------------------------------- Game plan ----------------------------------------------------- Consider drop n [] = [] drop 0 xs = [] drop n (x:xs) = drop (n-1) xs After the first time round, we could pass n unboxed. This happens in numerical code too. Here's what it looks like in Core: drop n xs = case xs of [] -> [] (y:ys) -> case n of I# n# -> case n# of 0 -> [] _ -> drop (I# (n# -# 1#)) xs Notice that the recursive call has an explicit constructor as argument. Noticing this, we can make a specialised version of drop RULE: drop (I# n#) xs ==> drop' n# xs drop' n# xs = let n = I# n# in ...orig RHS... Now the simplifier will apply the specialisation in the rhs of drop', giving drop' n# xs = case xs of [] -> [] (y:ys) -> case n# of 0 -> [] _ -> drop' (n# -# 1#) xs Much better! We'd also like to catch cases where a parameter is carried along unchanged, but evaluated each time round the loop: f i n = if i>0 || i>n then i else f (i*2) n Here f isn't strict in n, but we'd like to avoid evaluating it each iteration. In Core, by the time we've w/wd (f is strict in i) we get f i# n = case i# ># 0 of False -> I# i# True -> case n of { I# n# -> case i# ># n# of False -> I# i# True -> f (i# *# 2#) n At the call to f, we see that the argument, n is known to be (I# n#), and n is evaluated elsewhere in the body of f, so we can play the same trick as above. Note [Reboxing] ~~~~~~~~~~~~~~~ We must be careful not to allocate the same constructor twice. Consider f p = (...(case p of (a,b) -> e)...p..., ...let t = (r,s) in ...t...(f t)...) At the recursive call to f, we can see that t is a pair. But we do NOT want to make a specialised copy: f' a b = let p = (a,b) in (..., ...) because now t is allocated by the caller, then r and s are passed to the recursive call, which allocates the (r,s) pair again. This happens if (a) the argument p is used in other than a case-scrutinisation way. (b) the argument to the call is not a 'fresh' tuple; you have to look into its unfolding to see that it's a tuple Hence the "OR" part of Note [Good arguments] below. ALTERNATIVE 2: pass both boxed and unboxed versions. This no longer saves allocation, but does perhaps save evals. In the RULE we'd have something like f (I# x#) = f' (I# x#) x# If at the call site the (I# x) was an unfolding, then we'd have to rely on CSE to eliminate the duplicate allocation.... This alternative doesn't look attractive enough to pursue. ALTERNATIVE 3: ignore the reboxing problem. The trouble is that the conservative reboxing story prevents many useful functions from being specialised. Example: foo :: Maybe Int -> Int -> Int foo (Just m) 0 = 0 foo x@(Just m) n = foo x (n-m) Here the use of 'x' will clearly not require boxing in the specialised function. The strictness analyser has the same problem, in fact. Example: f p@(a,b) = ... If we pass just 'a' and 'b' to the worker, it might need to rebox the pair to create (a,b). A more sophisticated analysis might figure out precisely the cases in which this could happen, but the strictness analyser does no such analysis; it just passes 'a' and 'b', and hopes for the best. So my current choice is to make SpecConstr similarly aggressive, and ignore the bad potential of reboxing. Note [Good arguments] ~~~~~~~~~~~~~~~~~~~~~ So we look for * A self-recursive function. Ignore mutual recursion for now, because it's less common, and the code is simpler for self-recursion. * EITHER a) At a recursive call, one or more parameters is an explicit constructor application AND That same parameter is scrutinised by a case somewhere in the RHS of the function OR b) At a recursive call, one or more parameters has an unfolding that is an explicit constructor application AND That same parameter is scrutinised by a case somewhere in the RHS of the function AND Those are the only uses of the parameter (see Note [Reboxing]) What to abstract over ~~~~~~~~~~~~~~~~~~~~~ There's a bit of a complication with type arguments. If the call site looks like f p = ...f ((:) [a] x xs)... then our specialised function look like f_spec x xs = let p = (:) [a] x xs in ....as before.... This only makes sense if either a) the type variable 'a' is in scope at the top of f, or b) the type variable 'a' is an argument to f (and hence fs) Actually, (a) may hold for value arguments too, in which case we may not want to pass them. Supose 'x' is in scope at f's defn, but xs is not. Then we'd like f_spec xs = let p = (:) [a] x xs in ....as before.... Similarly (b) may hold too. If x is already an argument at the call, no need to pass it again. Finally, if 'a' is not in scope at the call site, we could abstract it as we do the term variables: f_spec a x xs = let p = (:) [a] x xs in ...as before... So the grand plan is: * abstract the call site to a constructor-only pattern e.g. C x (D (f p) (g q)) ==> C s1 (D s2 s3) * Find the free variables of the abstracted pattern * Pass these variables, less any that are in scope at the fn defn. But see Note [Shadowing] below. NOTICE that we only abstract over variables that are not in scope, so we're in no danger of shadowing variables used in "higher up" in f_spec's RHS. Note [Shadowing] ~~~~~~~~~~~~~~~~ In this pass we gather up usage information that may mention variables that are bound between the usage site and the definition site; or (more seriously) may be bound to something different at the definition site. For example: f x = letrec g y v = let x = ... in ...(g (a,b) x)... Since 'x' is in scope at the call site, we may make a rewrite rule that looks like RULE forall a,b. g (a,b) x = ... But this rule will never match, because it's really a different 'x' at the call site -- and that difference will be manifest by the time the simplifier gets to it. [A worry: the simplifier doesn't *guarantee* no-shadowing, so perhaps it may not be distinct?] Anyway, the rule isn't actually wrong, it's just not useful. One possibility is to run deShadowBinds before running SpecConstr, but instead we run the simplifier. That gives the simplest possible program for SpecConstr to chew on; and it virtually guarantees no shadowing. Note [Specialising for constant parameters] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This one is about specialising on a *constant* (but not necessarily constructor) argument foo :: Int -> (Int -> Int) -> Int foo 0 f = 0 foo m f = foo (f m) (+1) It produces lvl_rmV :: GHC.Base.Int -> GHC.Base.Int lvl_rmV = \ (ds_dlk :: GHC.Base.Int) -> case ds_dlk of wild_alH { GHC.Base.I# x_alG -> GHC.Base.I# (GHC.Prim.+# x_alG 1) T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) -> GHC.Prim.Int# T.$wfoo = \ (ww_sme :: GHC.Prim.Int#) (w_smg :: GHC.Base.Int -> GHC.Base.Int) -> case ww_sme of ds_Xlw { __DEFAULT -> case w_smg (GHC.Base.I# ds_Xlw) of w1_Xmo { GHC.Base.I# ww1_Xmz -> T.$wfoo ww1_Xmz lvl_rmV }; 0 -> 0 } The recursive call has lvl_rmV as its argument, so we could create a specialised copy with that argument baked in; that is, not passed at all. Now it can perhaps be inlined. When is this worth it? Call the constant 'lvl' - If 'lvl' has an unfolding that is a constructor, see if the corresponding parameter is scrutinised anywhere in the body. - If 'lvl' has an unfolding that is a inlinable function, see if the corresponding parameter is applied (...to enough arguments...?) Also do this is if the function has RULES? Also Note [Specialising for lambda parameters] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ foo :: Int -> (Int -> Int) -> Int foo 0 f = 0 foo m f = foo (f m) (\n -> n-m) This is subtly different from the previous one in that we get an explicit lambda as the argument: T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) -> GHC.Prim.Int# T.$wfoo = \ (ww_sm8 :: GHC.Prim.Int#) (w_sma :: GHC.Base.Int -> GHC.Base.Int) -> case ww_sm8 of ds_Xlr { __DEFAULT -> case w_sma (GHC.Base.I# ds_Xlr) of w1_Xmf { GHC.Base.I# ww1_Xmq -> T.$wfoo ww1_Xmq (\ (n_ad3 :: GHC.Base.Int) -> case n_ad3 of wild_alB { GHC.Base.I# x_alA -> GHC.Base.I# (GHC.Prim.-# x_alA ds_Xlr) }) }; 0 -> 0 } I wonder if SpecConstr couldn't be extended to handle this? After all, lambda is a sort of constructor for functions and perhaps it already has most of the necessary machinery? Furthermore, there's an immediate win, because you don't need to allocate the lamda at the call site; and if perchance it's called in the recursive call, then you may avoid allocating it altogether. Just like for constructors. Looks cool, but probably rare...but it might be easy to implement. Note [SpecConstr for casts] ~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider data family T a :: * data instance T Int = T Int foo n = ... where go (T 0) = 0 go (T n) = go (T (n-1)) The recursive call ends up looking like go (T (I# ...) `cast` g) So we want to spot the constructor application inside the cast. That's why we have the Cast case in argToPat Note [Local recursive groups] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ For a *local* recursive group, we can see all the calls to the function, so we seed the specialisation loop from the calls in the body, not from the calls in the RHS. Consider: bar m n = foo n (n,n) (n,n) (n,n) (n,n) where foo n p q r s | n == 0 = m | n > 3000 = case p of { (p1,p2) -> foo (n-1) (p2,p1) q r s } | n > 2000 = case q of { (q1,q2) -> foo (n-1) p (q2,q1) r s } | n > 1000 = case r of { (r1,r2) -> foo (n-1) p q (r2,r1) s } | otherwise = case s of { (s1,s2) -> foo (n-1) p q r (s2,s1) } If we start with the RHSs of 'foo', we get lots and lots of specialisations, most of which are not needed. But if we start with the (single) call in the rhs of 'bar' we get exactly one fully-specialised copy, and all the recursive calls go to this fully-specialised copy. Indeed, the original function is later collected as dead code. This is very important in specialising the loops arising from stream fusion, for example in NDP where we were getting literally hundreds of (mostly unused) specialisations of a local function. In a case like the above we end up never calling the original un-specialised function. (Although we still leave its code around just in case.) However, if we find any boring calls in the body, including *unsaturated* ones, such as letrec foo x y = ....foo... in map foo xs then we will end up calling the un-specialised function, so then we *should* use the calls in the un-specialised RHS as seeds. We call these "boring call patterns", and callsToPats reports if it finds any of these. Note [Top-level recursive groups] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If all the bindings in a top-level recursive group are local (not exported), then all the calls are in the rest of the top-level bindings. This means we can specialise with those call patterns instead of with the RHSs of the recursive group. (Question: maybe we should *also* use calls in the rest of the top-level bindings as seeds? To get the call usage information, we work backwards through the top-level bindings so we see the usage before we get to the binding of the function. Before we can collect the usage though, we go through all the bindings and add them to the environment. This is necessary because usage is only tracked for functions in the environment. The actual seeding of the specialisation is very similar to Note [Local recursive group]. Note [Do not specialise diverging functions] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Specialising a function that just diverges is a waste of code. Furthermore, it broke GHC (simpl014) thus: {-# STR Sb #-} f = \x. case x of (a,b) -> f x If we specialise f we get f = \x. case x of (a,b) -> fspec a b But fspec doesn't have decent strictness info. As it happened, (f x) :: IO t, so the state hack applied and we eta expanded fspec, and hence f. But now f's strictness is less than its arity, which breaks an invariant. Note [Forcing specialisation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ With stream fusion and in other similar cases, we want to fully specialise some (but not necessarily all!) loops regardless of their size and the number of specialisations. We allow a library to do this, in one of two ways (one which is deprecated): 1) Add a parameter of type GHC.Types.SPEC (from ghc-prim) to the loop body. 2) (Deprecated) Annotate a type with ForceSpecConstr from GHC.Exts, and then add *that* type as a parameter to the loop body The reason #2 is deprecated is because it requires GHCi, which isn't available for things like a cross compiler using stage1. Here's a (simplified) example from the `vector` package. You may bring the special 'force specialization' type into scope by saying: import GHC.Types (SPEC(..)) or by defining your own type (again, deprecated): data SPEC = SPEC | SPEC2 {-# ANN type SPEC ForceSpecConstr #-} (Note this is the exact same definition of GHC.Types.SPEC, just without the annotation.) After that, you say: foldl :: (a -> b -> a) -> a -> Stream b -> a {-# INLINE foldl #-} foldl f z (Stream step s _) = foldl_loop SPEC z s where foldl_loop !sPEC z s = case step s of Yield x s' -> foldl_loop sPEC (f z x) s' Skip -> foldl_loop sPEC z s' Done -> z SpecConstr will spot the SPEC parameter and always fully specialise foldl_loop. Note that * We have to prevent the SPEC argument from being removed by w/w which is why (a) SPEC is a sum type, and (b) we have to seq on the SPEC argument. * And lastly, the SPEC argument is ultimately eliminated by SpecConstr itself so there is no runtime overhead. This is all quite ugly; we ought to come up with a better design. ForceSpecConstr arguments are spotted in scExpr' and scTopBinds which then set sc_force to True when calling specLoop. This flag does four things: * Ignore specConstrThreshold, to specialise functions of arbitrary size (see scTopBind) * Ignore specConstrCount, to make arbitrary numbers of specialisations (see specialise) * Specialise even for arguments that are not scrutinised in the loop (see argToPat; Trac #4488) * Only specialise on recursive types a finite number of times (see is_too_recursive; Trac #5550; Note [Limit recursive specialisation]) This flag is inherited for nested non-recursive bindings (which are likely to be join points and hence should be fully specialised) but reset for nested recursive bindings. What alternatives did I consider? Annotating the loop itself doesn't work because (a) it is local and (b) it will be w/w'ed and having w/w propagating annotations somehow doesn't seem like a good idea. The types of the loop arguments really seem to be the most persistent thing. Annotating the types that make up the loop state doesn't work, either, because (a) it would prevent us from using types like Either or tuples here, (b) we don't want to restrict the set of types that can be used in Stream states and (c) some types are fixed by the user (e.g., the accumulator here) but we still want to specialise as much as possible. Alternatives to ForceSpecConstr ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Instead of giving the loop an extra argument of type SPEC, we also considered *wrapping* arguments in SPEC, thus data SPEC a = SPEC a | SPEC2 loop = \arg -> case arg of SPEC state -> case state of (x,y) -> ... loop (SPEC (x',y')) ... S2 -> error ... The idea is that a SPEC argument says "specialise this argument regardless of whether the function case-analyses it". But this doesn't work well: * SPEC must still be a sum type, else the strictness analyser eliminates it * But that means that 'loop' won't be strict in its real payload This loss of strictness in turn screws up specialisation, because we may end up with calls like loop (SPEC (case z of (p,q) -> (q,p))) Without the SPEC, if 'loop' were strict, the case would move out and we'd see loop applied to a pair. But if 'loop' isn't strict this doesn't look like a specialisable call. Note [Limit recursive specialisation] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ It is possible for ForceSpecConstr to cause an infinite loop of specialisation. Because there is no limit on the number of specialisations, a recursive call with a recursive constructor as an argument (for example, list cons) will generate a specialisation for that constructor. If the resulting specialisation also contains a recursive call with the constructor, this could proceed indefinitely. For example, if ForceSpecConstr is on: loop :: [Int] -> [Int] -> [Int] loop z [] = z loop z (x:xs) = loop (x:z) xs this example will create a specialisation for the pattern loop (a:b) c = loop' a b c loop' a b [] = (a:b) loop' a b (x:xs) = loop (x:(a:b)) xs and a new pattern is found: loop (a:(b:c)) d = loop'' a b c d which can continue indefinitely. Roman's suggestion to fix this was to stop after a couple of times on recursive types, but still specialising on non-recursive types as much as possible. To implement this, we count the number of recursive constructors in each function argument. If the maximum is greater than the specConstrRecursive limit, do not specialise on that pattern. This is only necessary when ForceSpecConstr is on: otherwise the specConstrCount will force termination anyway. See Trac #5550. Note [NoSpecConstr] ~~~~~~~~~~~~~~~~~~~ The ignoreDataCon stuff allows you to say {-# ANN type T NoSpecConstr #-} to mean "don't specialise on arguments of this type". It was added before we had ForceSpecConstr. Lacking ForceSpecConstr we specialised regardless of size; and then we needed a way to turn that *off*. Now that we have ForceSpecConstr, this NoSpecConstr is probably redundant. (Used only for PArray.) ----------------------------------------------------- Stuff not yet handled ----------------------------------------------------- Here are notes arising from Roman's work that I don't want to lose. Example 1 ~~~~~~~~~ data T a = T !a foo :: Int -> T Int -> Int foo 0 t = 0 foo x t | even x = case t of { T n -> foo (x-n) t } | otherwise = foo (x-1) t SpecConstr does no specialisation, because the second recursive call looks like a boxed use of the argument. A pity. $wfoo_sFw :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int# $wfoo_sFw = \ (ww_sFo [Just L] :: GHC.Prim.Int#) (w_sFq [Just L] :: T.T GHC.Base.Int) -> case ww_sFo of ds_Xw6 [Just L] { __DEFAULT -> case GHC.Prim.remInt# ds_Xw6 2 of wild1_aEF [Dead Just A] { __DEFAULT -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 1) w_sFq; 0 -> case w_sFq of wild_Xy [Just L] { T.T n_ad5 [Just U(L)] -> case n_ad5 of wild1_aET [Just A] { GHC.Base.I# y_aES [Just L] -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 y_aES) wild_Xy } } }; 0 -> 0 Example 2 ~~~~~~~~~ data a :*: b = !a :*: !b data T a = T !a foo :: (Int :*: T Int) -> Int foo (0 :*: t) = 0 foo (x :*: t) | even x = case t of { T n -> foo ((x-n) :*: t) } | otherwise = foo ((x-1) :*: t) Very similar to the previous one, except that the parameters are now in a strict tuple. Before SpecConstr, we have $wfoo_sG3 :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int# $wfoo_sG3 = \ (ww_sFU [Just L] :: GHC.Prim.Int#) (ww_sFW [Just L] :: T.T GHC.Base.Int) -> case ww_sFU of ds_Xws [Just L] { __DEFAULT -> case GHC.Prim.remInt# ds_Xws 2 of wild1_aEZ [Dead Just A] { __DEFAULT -> case ww_sFW of tpl_B2 [Just L] { T.T a_sFo [Just A] -> $wfoo_sG3 (GHC.Prim.-# ds_Xws 1) tpl_B2 -- $wfoo1 }; 0 -> case ww_sFW of wild_XB [Just A] { T.T n_ad7 [Just S(L)] -> case n_ad7 of wild1_aFd [Just L] { GHC.Base.I# y_aFc [Just L] -> $wfoo_sG3 (GHC.Prim.-# ds_Xws y_aFc) wild_XB -- $wfoo2 } } }; 0 -> 0 } We get two specialisations: "SC:$wfoo1" [0] __forall {a_sFB :: GHC.Base.Int sc_sGC :: GHC.Prim.Int#} Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int a_sFB) = Foo.$s$wfoo1 a_sFB sc_sGC ; "SC:$wfoo2" [0] __forall {y_aFp :: GHC.Prim.Int# sc_sGC :: GHC.Prim.Int#} Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int (GHC.Base.I# y_aFp)) = Foo.$s$wfoo y_aFp sc_sGC ; But perhaps the first one isn't good. After all, we know that tpl_B2 is a T (I# x) really, because T is strict and Int has one constructor. (We can't unbox the strict fields, because T is polymorphic!) %************************************************************************ %* * \subsection{Top level wrapper stuff} %* * %************************************************************************ \begin{code}
specConstrProgram :: ModGuts -> CoreM ModGuts specConstrProgram guts = do dflags <- getDynFlags us <- getUniqueSupplyM annos <- getFirstAnnotations deserializeWithData guts let binds' = reverse $ fst $ initUs us $ do -- Note [Top-level recursive groups] (env, binds) <- goEnv (initScEnv dflags annos) (mg_binds guts) go env nullUsage (reverse binds) return (guts { mg_binds = binds' }) where goEnv env [] = return (env, []) goEnv env (bind:binds) = do (env', bind') <- scTopBindEnv env bind (env'', binds') <- goEnv env' binds return (env'', bind' : binds') go _ _ [] = return [] go env usg (bind:binds) = do (usg', bind') <- scTopBind env usg bind binds' <- go env usg' binds return (bind' : binds')\end{code} %************************************************************************ %* * \subsection{Environment: goes downwards} %* * %************************************************************************ Note [Work-free values only in environment] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The sc_vals field keeps track of in-scope value bindings, so that if we come across (case x of Just y ->...) we can reduce the case from knowing that x is bound to a pair. But only *work-free* values are ok here. For example if the envt had x -> Just (expensive v) then we do NOT want to expand to let y = expensive v in ... because the x-binding still exists and we've now duplicated (expensive v). This seldom happens because let-bound constructor applications are ANF-ised, but it can happen as a result of on-the-fly transformations in SpecConstr itself. Here is Trac #7865: let { a'_shr = case xs_af8 of _ { [] -> acc_af6; : ds_dgt [Dmd=
data ScEnv = SCE { sc_dflags :: DynFlags, sc_size :: Maybe Int, -- Size threshold sc_count :: Maybe Int, -- Max # of specialisations for any one fn -- See Note [Avoiding exponential blowup] sc_recursive :: Int, -- Max # of specialisations over recursive type. -- Stops ForceSpecConstr from diverging. sc_force :: Bool, -- Force specialisation? -- See Note [Forcing specialisation] sc_subst :: Subst, -- Current substitution -- Maps InIds to OutExprs sc_how_bound :: HowBoundEnv, -- Binds interesting non-top-level variables -- Domain is OutVars (*after* applying the substitution) sc_vals :: ValueEnv, -- Domain is OutIds (*after* applying the substitution) -- Used even for top-level bindings (but not imported ones) -- The range of the ValueEnv is *work-free* values -- such as (\x. blah), or (Just v) -- but NOT (Just (expensive v)) -- See Note [Work-free values only in environment] sc_annotations :: UniqFM SpecConstrAnnotation } --------------------- -- As we go, we apply a substitution (sc_subst) to the current term type InExpr = CoreExpr -- _Before_ applying the subst type InVar = Var type OutExpr = CoreExpr -- _After_ applying the subst type OutId = Id type OutVar = Var --------------------- type HowBoundEnv = VarEnv HowBound -- Domain is OutVars --------------------- type ValueEnv = IdEnv Value -- Domain is OutIds data Value = ConVal AltCon [CoreArg] -- _Saturated_ constructors -- The AltCon is never DEFAULT | LambdaVal -- Inlinable lambdas or PAPs instance Outputable Value where ppr (ConVal con args) = ppr con <+> interpp'SP args ppr LambdaVal = ptext (sLit "<Lambda>") --------------------- initScEnv :: DynFlags -> UniqFM SpecConstrAnnotation -> ScEnv initScEnv dflags anns = SCE { sc_dflags = dflags, sc_size = specConstrThreshold dflags, sc_count = specConstrCount dflags, sc_recursive = specConstrRecursive dflags, sc_force = False, sc_subst = emptySubst, sc_how_bound = emptyVarEnv, sc_vals = emptyVarEnv, sc_annotations = anns } data HowBound = RecFun -- These are the recursive functions for which -- we seek interesting call patterns | RecArg -- These are those functions' arguments, or their sub-components; -- we gather occurrence information for these instance Outputable HowBound where ppr RecFun = text "RecFun" ppr RecArg = text "RecArg" scForce :: ScEnv -> Bool -> ScEnv scForce env b = env { sc_force = b } lookupHowBound :: ScEnv -> Id -> Maybe HowBound lookupHowBound env id = lookupVarEnv (sc_how_bound env) id scSubstId :: ScEnv -> Id -> CoreExpr scSubstId env v = lookupIdSubst (text "scSubstId") (sc_subst env) v scSubstTy :: ScEnv -> Type -> Type scSubstTy env ty = substTy (sc_subst env) ty scSubstCo :: ScEnv -> Coercion -> Coercion scSubstCo env co = substCo (sc_subst env) co zapScSubst :: ScEnv -> ScEnv zapScSubst env = env { sc_subst = zapSubstEnv (sc_subst env) } extendScInScope :: ScEnv -> [Var] -> ScEnv -- Bring the quantified variables into scope extendScInScope env qvars = env { sc_subst = extendInScopeList (sc_subst env) qvars } -- Extend the substitution extendScSubst :: ScEnv -> Var -> OutExpr -> ScEnv extendScSubst env var expr = env { sc_subst = extendSubst (sc_subst env) var expr } extendScSubstList :: ScEnv -> [(Var,OutExpr)] -> ScEnv extendScSubstList env prs = env { sc_subst = extendSubstList (sc_subst env) prs } extendHowBound :: ScEnv -> [Var] -> HowBound -> ScEnv extendHowBound env bndrs how_bound = env { sc_how_bound = extendVarEnvList (sc_how_bound env) [(bndr,how_bound) | bndr <- bndrs] } extendBndrsWith :: HowBound -> ScEnv -> [Var] -> (ScEnv, [Var]) extendBndrsWith how_bound env bndrs = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndrs') where (subst', bndrs') = substBndrs (sc_subst env) bndrs hb_env' = sc_how_bound env `extendVarEnvList` [(bndr,how_bound) | bndr <- bndrs'] extendBndrWith :: HowBound -> ScEnv -> Var -> (ScEnv, Var) extendBndrWith how_bound env bndr = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndr') where (subst', bndr') = substBndr (sc_subst env) bndr hb_env' = extendVarEnv (sc_how_bound env) bndr' how_bound extendRecBndrs :: ScEnv -> [Var] -> (ScEnv, [Var]) extendRecBndrs env bndrs = (env { sc_subst = subst' }, bndrs') where (subst', bndrs') = substRecBndrs (sc_subst env) bndrs extendBndr :: ScEnv -> Var -> (ScEnv, Var) extendBndr env bndr = (env { sc_subst = subst' }, bndr') where (subst', bndr') = substBndr (sc_subst env) bndr extendValEnv :: ScEnv -> Id -> Maybe Value -> ScEnv extendValEnv env _ Nothing = env extendValEnv env id (Just cv) | valueIsWorkFree cv -- Don't duplicate work!! Trac #7865 = env { sc_vals = extendVarEnv (sc_vals env) id cv } extendValEnv env _ _ = env extendCaseBndrs :: ScEnv -> OutExpr -> OutId -> AltCon -> [Var] -> (ScEnv, [Var]) -- When we encounter -- case scrut of b -- C x y -> ... -- we want to bind b, to (C x y) -- NB1: Extends only the sc_vals part of the envt -- NB2: Kill the dead-ness info on the pattern binders x,y, since -- they are potentially made alive by the [b -> C x y] binding extendCaseBndrs env scrut case_bndr con alt_bndrs = (env2, alt_bndrs') where live_case_bndr = not (isDeadBinder case_bndr) env1 | Var v <- scrut = extendValEnv env v cval | otherwise = env -- See Note [Add scrutinee to ValueEnv too] env2 | live_case_bndr = extendValEnv env1 case_bndr cval | otherwise = env1 alt_bndrs' | case scrut of { Var {} -> True; _ -> live_case_bndr } = map zap alt_bndrs | otherwise = alt_bndrs cval = case con of DEFAULT -> Nothing LitAlt {} -> Just (ConVal con []) DataAlt {} -> Just (ConVal con vanilla_args) where vanilla_args = map Type (tyConAppArgs (idType case_bndr)) ++ varsToCoreExprs alt_bndrs zap v | isTyVar v = v -- See NB2 above | otherwise = zapIdOccInfo v decreaseSpecCount :: ScEnv -> Int -> ScEnv -- See Note [Avoiding exponential blowup] decreaseSpecCount env n_specs = env { sc_count = case sc_count env of Nothing -> Nothing Just n -> Just (n `div` (n_specs + 1)) } -- The "+1" takes account of the original function; -- See Note [Avoiding exponential blowup] --------------------------------------------------- -- See Note [Forcing specialisation] ignoreType :: ScEnv -> Type -> Bool ignoreDataCon :: ScEnv -> DataCon -> Bool forceSpecBndr :: ScEnv -> Var -> Bool #ifndef GHCI ignoreType _ _ = False ignoreDataCon _ _ = False #else /* GHCI */ ignoreDataCon env dc = ignoreTyCon env (dataConTyCon dc) ignoreType env ty = case tyConAppTyCon_maybe ty of Just tycon -> ignoreTyCon env tycon _ -> False ignoreTyCon :: ScEnv -> TyCon -> Bool ignoreTyCon env tycon = lookupUFM (sc_annotations env) tycon == Just NoSpecConstr #endif /* GHCI */ forceSpecBndr env var = forceSpecFunTy env . snd . splitForAllTys . varType $ var forceSpecFunTy :: ScEnv -> Type -> Bool forceSpecFunTy env = any (forceSpecArgTy env) . fst . splitFunTys forceSpecArgTy :: ScEnv -> Type -> Bool forceSpecArgTy env ty | Just ty' <- coreView ty = forceSpecArgTy env ty' forceSpecArgTy env ty | Just (tycon, tys) <- splitTyConApp_maybe ty , tycon /= funTyCon = tyConName tycon == specTyConName #ifdef GHCI || lookupUFM (sc_annotations env) tycon == Just ForceSpecConstr #endif || any (forceSpecArgTy env) tys forceSpecArgTy _ _ = False\end{code} Note [Add scrutinee to ValueEnv too] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider this: case x of y (a,b) -> case b of c I# v -> ...(f y)... By the time we get to the call (f y), the ValueEnv will have a binding for y, and for c y -> (a,b) c -> I# v BUT that's not enough! Looking at the call (f y) we see that y is pair (a,b), but we also need to know what 'b' is. So in extendCaseBndrs we must *also* add the binding b -> I# v else we lose a useful specialisation for f. This is necessary even though the simplifier has systematically replaced uses of 'x' with 'y' and 'b' with 'c' in the code. The use of 'b' in the ValueEnv came from outside the case. See Trac #4908 for the live example. Note [Avoiding exponential blowup] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The sc_count field of the ScEnv says how many times we are prepared to duplicate a single function. But we must take care with recursive specialisations. Consider let $j1 = let $j2 = let $j3 = ... in ...$j3... in ...$j2... in ...$j1... If we specialise $j1 then in each specialisation (as well as the original) we can specialise $j2, and similarly $j3. Even if we make just *one* specialisation of each, because we also have the original we'll get 2^n copies of $j3, which is not good. So when recursively specialising we divide the sc_count by the number of copies we are making at this level, including the original. %************************************************************************ %* * \subsection{Usage information: flows upwards} %* * %************************************************************************ \begin{code}
data ScUsage = SCU { scu_calls :: CallEnv, -- Calls -- The functions are a subset of the -- RecFuns in the ScEnv scu_occs :: !(IdEnv ArgOcc) -- Information on argument occurrences } -- The domain is OutIds type CallEnv = IdEnv [Call] data Call = Call Id [CoreArg] ValueEnv -- The arguments of the call, together with the -- env giving the constructor bindings at the call site -- We keep the function mainly for debug output instance Outputable Call where ppr (Call fn args _) = ppr fn <+> fsep (map pprParendExpr args) nullUsage :: ScUsage nullUsage = SCU { scu_calls = emptyVarEnv, scu_occs = emptyVarEnv } combineCalls :: CallEnv -> CallEnv -> CallEnv combineCalls = plusVarEnv_C (++) where -- plus cs ds | length res > 1 -- = pprTrace "combineCalls" (vcat [ ptext (sLit "cs:") <+> ppr cs -- , ptext (sLit "ds:") <+> ppr ds]) -- res -- | otherwise = res -- where -- res = cs ++ ds combineUsage :: ScUsage -> ScUsage -> ScUsage combineUsage u1 u2 = SCU { scu_calls = combineCalls (scu_calls u1) (scu_calls u2), scu_occs = plusVarEnv_C combineOcc (scu_occs u1) (scu_occs u2) } combineUsages :: [ScUsage] -> ScUsage combineUsages [] = nullUsage combineUsages us = foldr1 combineUsage us lookupOccs :: ScUsage -> [OutVar] -> (ScUsage, [ArgOcc]) lookupOccs (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndrs = (SCU {scu_calls = sc_calls, scu_occs = delVarEnvList sc_occs bndrs}, [lookupVarEnv sc_occs b `orElse` NoOcc | b <- bndrs]) data ArgOcc = NoOcc -- Doesn't occur at all; or a type argument | UnkOcc -- Used in some unknown way | ScrutOcc -- See Note [ScrutOcc] (DataConEnv [ArgOcc]) -- How the sub-components are used type DataConEnv a = UniqFM a -- Keyed by DataCon {- Note [ScrutOcc] ~~~~~~~~~~~~~~~~~~~ An occurrence of ScrutOcc indicates that the thing, or a `cast` version of the thing, is *only* taken apart or applied. Functions, literal: ScrutOcc emptyUFM Data constructors: ScrutOcc subs, where (subs :: UniqFM [ArgOcc]) gives usage of the *pattern-bound* components, The domain of the UniqFM is the Unique of the data constructor The [ArgOcc] is the occurrences of the *pattern-bound* components of the data structure. E.g. data T a = forall b. MkT a b (b->a) A pattern binds b, x::a, y::b, z::b->a, but not 'a'! -} instance Outputable ArgOcc where ppr (ScrutOcc xs) = ptext (sLit "scrut-occ") <> ppr xs ppr UnkOcc = ptext (sLit "unk-occ") ppr NoOcc = ptext (sLit "no-occ") evalScrutOcc :: ArgOcc evalScrutOcc = ScrutOcc emptyUFM -- Experimentally, this vesion of combineOcc makes ScrutOcc "win", so -- that if the thing is scrutinised anywhere then we get to see that -- in the overall result, even if it's also used in a boxed way -- This might be too agressive; see Note [Reboxing] Alternative 3 combineOcc :: ArgOcc -> ArgOcc -> ArgOcc combineOcc NoOcc occ = occ combineOcc occ NoOcc = occ combineOcc (ScrutOcc xs) (ScrutOcc ys) = ScrutOcc (plusUFM_C combineOccs xs ys) combineOcc UnkOcc (ScrutOcc ys) = ScrutOcc ys combineOcc (ScrutOcc xs) UnkOcc = ScrutOcc xs combineOcc UnkOcc UnkOcc = UnkOcc combineOccs :: [ArgOcc] -> [ArgOcc] -> [ArgOcc] combineOccs xs ys = zipWithEqual "combineOccs" combineOcc xs ys setScrutOcc :: ScEnv -> ScUsage -> OutExpr -> ArgOcc -> ScUsage -- _Overwrite_ the occurrence info for the scrutinee, if the scrutinee -- is a variable, and an interesting variable setScrutOcc env usg (Cast e _) occ = setScrutOcc env usg e occ setScrutOcc env usg (Tick _ e) occ = setScrutOcc env usg e occ setScrutOcc env usg (Var v) occ | Just RecArg <- lookupHowBound env v = usg { scu_occs = extendVarEnv (scu_occs usg) v occ } | otherwise = usg setScrutOcc _env usg _other _occ -- Catch-all = usg\end{code} %************************************************************************ %* * \subsection{The main recursive function} %* * %************************************************************************ The main recursive function gathers up usage information, and creates specialised versions of functions. \begin{code}
scExpr, scExpr' :: ScEnv -> CoreExpr -> UniqSM (ScUsage, CoreExpr) -- The unique supply is needed when we invent -- a new name for the specialised function and its args scExpr env e = scExpr' env e scExpr' env (Var v) = case scSubstId env v of Var v' -> return (mkVarUsage env v' [], Var v') e' -> scExpr (zapScSubst env) e' scExpr' env (Type t) = return (nullUsage, Type (scSubstTy env t)) scExpr' env (Coercion c) = return (nullUsage, Coercion (scSubstCo env c)) scExpr' _ e@(Lit {}) = return (nullUsage, e) scExpr' env (Tick t e) = do (usg, e') <- scExpr env e return (usg, Tick t e') scExpr' env (Cast e co) = do (usg, e') <- scExpr env e return (usg, Cast e' (scSubstCo env co)) scExpr' env e@(App _ _) = scApp env (collectArgs e) scExpr' env (Lam b e) = do let (env', b') = extendBndr env b (usg, e') <- scExpr env' e return (usg, Lam b' e') scExpr' env (Case scrut b ty alts) = do { (scrut_usg, scrut') <- scExpr env scrut ; case isValue (sc_vals env) scrut' of Just (ConVal con args) -> sc_con_app con args scrut' _other -> sc_vanilla scrut_usg scrut' } where sc_con_app con args scrut' -- Known constructor; simplify = do { let (_, bs, rhs) = findAlt con alts `orElse` (DEFAULT, [], mkImpossibleExpr ty) alt_env' = extendScSubstList env ((b,scrut') : bs `zip` trimConArgs con args) ; scExpr alt_env' rhs } sc_vanilla scrut_usg scrut' -- Normal case = do { let (alt_env,b') = extendBndrWith RecArg env b -- Record RecArg for the components ; (alt_usgs, alt_occs, alts') <- mapAndUnzip3M (sc_alt alt_env scrut' b') alts ; let scrut_occ = foldr combineOcc NoOcc alt_occs scrut_usg' = setScrutOcc env scrut_usg scrut' scrut_occ -- The combined usage of the scrutinee is given -- by scrut_occ, which is passed to scScrut, which -- in turn treats a bare-variable scrutinee specially ; return (foldr combineUsage scrut_usg' alt_usgs, Case scrut' b' (scSubstTy env ty) alts') } sc_alt env scrut' b' (con,bs,rhs) = do { let (env1, bs1) = extendBndrsWith RecArg env bs (env2, bs2) = extendCaseBndrs env1 scrut' b' con bs1 ; (usg, rhs') <- scExpr env2 rhs ; let (usg', b_occ:arg_occs) = lookupOccs usg (b':bs2) scrut_occ = case con of DataAlt dc -> ScrutOcc (unitUFM dc arg_occs) _ -> ScrutOcc emptyUFM ; return (usg', b_occ `combineOcc` scrut_occ, (con, bs2, rhs')) } scExpr' env (Let (NonRec bndr rhs) body) | isTyVar bndr -- Type-lets may be created by doBeta = scExpr' (extendScSubst env bndr rhs) body | otherwise = do { let (body_env, bndr') = extendBndr env bndr ; (rhs_usg, rhs_info) <- scRecRhs env (bndr',rhs) ; let body_env2 = extendHowBound body_env [bndr'] RecFun -- Note [Local let bindings] RI _ rhs' _ _ _ = rhs_info body_env3 = extendValEnv body_env2 bndr' (isValue (sc_vals env) rhs') ; (body_usg, body') <- scExpr body_env3 body -- NB: For non-recursive bindings we inherit sc_force flag from -- the parent function (see Note [Forcing specialisation]) ; (spec_usg, specs) <- specialise env (scu_calls body_usg) rhs_info (SI [] 0 (Just rhs_usg)) ; return (body_usg { scu_calls = scu_calls body_usg `delVarEnv` bndr' } `combineUsage` spec_usg, -- Note [spec_usg includes rhs_usg] mkLets [NonRec b r | (b,r) <- specInfoBinds rhs_info specs] body') } -- A *local* recursive group: see Note [Local recursive groups] scExpr' env (Let (Rec prs) body) = do { let (bndrs,rhss) = unzip prs (rhs_env1,bndrs') = extendRecBndrs env bndrs rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun force_spec = any (forceSpecBndr env) bndrs' -- Note [Forcing specialisation] ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss) ; (body_usg, body') <- scExpr rhs_env2 body -- NB: start specLoop from body_usg ; (spec_usg, specs) <- specLoop (scForce rhs_env2 force_spec) (scu_calls body_usg) rhs_infos nullUsage [SI [] 0 (Just usg) | usg <- rhs_usgs] -- Do not unconditionally generate specialisations from rhs_usgs -- Instead use them only if we find an unspecialised call -- See Note [Local recursive groups] ; let all_usg = spec_usg `combineUsage` body_usg -- Note [spec_usg includes rhs_usg] bind' = Rec (concat (zipWith specInfoBinds rhs_infos specs)) ; return (all_usg { scu_calls = scu_calls all_usg `delVarEnvList` bndrs' }, Let bind' body') }\end{code} Note [Local let bindings] ~~~~~~~~~~~~~~~~~~~~~~~~~ It is not uncommon to find this let $j = \x.
scApp :: ScEnv -> (InExpr, [InExpr]) -> UniqSM (ScUsage, CoreExpr) scApp env (Var fn, args) -- Function is a variable = ASSERT( not (null args) ) do { args_w_usgs <- mapM (scExpr env) args ; let (arg_usgs, args') = unzip args_w_usgs arg_usg = combineUsages arg_usgs ; case scSubstId env fn of fn'@(Lam {}) -> scExpr (zapScSubst env) (doBeta fn' args') -- Do beta-reduction and try again Var fn' -> return (arg_usg `combineUsage` mkVarUsage env fn' args', mkApps (Var fn') args') other_fn' -> return (arg_usg, mkApps other_fn' args') } -- NB: doing this ignores any usage info from the substituted -- function, but I don't think that matters. If it does -- we can fix it. where doBeta :: OutExpr -> [OutExpr] -> OutExpr -- ToDo: adjust for System IF doBeta (Lam bndr body) (arg : args) = Let (NonRec bndr arg) (doBeta body args) doBeta fn args = mkApps fn args -- The function is almost always a variable, but not always. -- In particular, if this pass follows float-in, -- which it may, we can get -- (let f = ...f... in f) arg1 arg2 scApp env (other_fn, args) = do { (fn_usg, fn') <- scExpr env other_fn ; (arg_usgs, args') <- mapAndUnzipM (scExpr env) args ; return (combineUsages arg_usgs `combineUsage` fn_usg, mkApps fn' args') } ---------------------- mkVarUsage :: ScEnv -> Id -> [CoreExpr] -> ScUsage mkVarUsage env fn args = case lookupHowBound env fn of Just RecFun -> SCU { scu_calls = unitVarEnv fn [Call fn args (sc_vals env)] , scu_occs = emptyVarEnv } Just RecArg -> SCU { scu_calls = emptyVarEnv , scu_occs = unitVarEnv fn arg_occ } Nothing -> nullUsage where -- I rather think we could use UnkOcc all the time arg_occ | null args = UnkOcc | otherwise = evalScrutOcc ---------------------- scTopBindEnv :: ScEnv -> CoreBind -> UniqSM (ScEnv, CoreBind) scTopBindEnv env (Rec prs) = do { let (rhs_env1,bndrs') = extendRecBndrs env bndrs rhs_env2 = extendHowBound rhs_env1 bndrs RecFun prs' = zip bndrs' rhss ; return (rhs_env2, Rec prs') } where (bndrs,rhss) = unzip prs scTopBindEnv env (NonRec bndr rhs) = do { let (env1, bndr') = extendBndr env bndr env2 = extendValEnv env1 bndr' (isValue (sc_vals env) rhs) ; return (env2, NonRec bndr' rhs) } ---------------------- scTopBind :: ScEnv -> ScUsage -> CoreBind -> UniqSM (ScUsage, CoreBind) {- scTopBind _ usage _ | pprTrace "scTopBind_usage" (ppr (scu_calls usage)) False = error "false" -} scTopBind env body_usage (Rec prs) | Just threshold <- sc_size env , not force_spec , not (all (couldBeSmallEnoughToInline (sc_dflags env) threshold) rhss) -- No specialisation = do { (rhs_usgs, rhss') <- mapAndUnzipM (scExpr env) rhss ; return (body_usage `combineUsage` combineUsages rhs_usgs, Rec (bndrs `zip` rhss')) } | otherwise -- Do specialisation = do { (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs env) prs -- ; pprTrace "scTopBind" (ppr bndrs $$ ppr (map (lookupVarEnv (scu_calls body_usage)) bndrs)) (return ()) -- Note [Top-level recursive groups] ; let (usg,rest) | any isExportedId bndrs -- Seed from RHSs = ( combineUsages rhs_usgs, [SI [] 0 Nothing | _ <- rhs_usgs] ) | otherwise -- Seed from body only = ( body_usage, [SI [] 0 (Just us) | us <- rhs_usgs] ) ; (spec_usage, specs) <- specLoop (scForce env force_spec) (scu_calls usg) rhs_infos nullUsage rest ; return (body_usage `combineUsage` spec_usage, Rec (concat (zipWith specInfoBinds rhs_infos specs))) } where (bndrs,rhss) = unzip prs force_spec = any (forceSpecBndr env) bndrs -- Note [Forcing specialisation] scTopBind env usage (NonRec bndr rhs) -- Oddly, we don't seem to specialise top-level non-rec functions = do { (rhs_usg', rhs') <- scExpr env rhs ; return (usage `combineUsage` rhs_usg', NonRec bndr rhs') } ---------------------- scRecRhs :: ScEnv -> (OutId, InExpr) -> UniqSM (ScUsage, RhsInfo) scRecRhs env (bndr,rhs) = do { let (arg_bndrs,body) = collectBinders rhs (body_env, arg_bndrs') = extendBndrsWith RecArg env arg_bndrs ; (body_usg, body') <- scExpr body_env body ; let (rhs_usg, arg_occs) = lookupOccs body_usg arg_bndrs' ; return (rhs_usg, RI bndr (mkLams arg_bndrs' body') arg_bndrs body arg_occs) } -- The arg_occs says how the visible, -- lambda-bound binders of the RHS are used -- (including the TyVar binders) -- Two pats are the same if they match both ways ---------------------- specInfoBinds :: RhsInfo -> SpecInfo -> [(Id,CoreExpr)] specInfoBinds (RI fn new_rhs _ _ _) (SI specs _ _) = [(id,rhs) | OS _ _ id rhs <- specs] ++ -- First the specialised bindings [(fn `addIdSpecialisations` rules, new_rhs)] -- And now the original binding where rules = [r | OS _ r _ _ <- specs]\end{code} %************************************************************************ %* * The specialiser itself %* * %************************************************************************ \begin{code}
data RhsInfo = RI OutId -- The binder OutExpr -- The new RHS [InVar] InExpr -- The *original* RHS (\xs.body) -- Note [Specialise original body] [ArgOcc] -- Info on how the xs occur in body data SpecInfo = SI [OneSpec] -- The specialisations we have generated Int -- Length of specs; used for numbering them (Maybe ScUsage) -- Just cs => we have not yet used calls in the -- from calls in the *original* RHS as -- seeds for new specialisations; -- if you decide to do so, here is the -- RHS usage (which has not yet been -- unleashed) -- Nothing => we have -- See Note [Local recursive groups] -- See Note [spec_usg includes rhs_usg] -- One specialisation: Rule plus definition data OneSpec = OS CallPat -- Call pattern that generated this specialisation CoreRule -- Rule connecting original id with the specialisation OutId OutExpr -- Spec id + its rhs specLoop :: ScEnv -> CallEnv -> [RhsInfo] -> ScUsage -> [SpecInfo] -- One per binder; acccumulating parameter -> UniqSM (ScUsage, [SpecInfo]) -- ...ditto... specLoop env all_calls rhs_infos usg_so_far specs_so_far = do { specs_w_usg <- zipWithM (specialise env all_calls) rhs_infos specs_so_far ; let (new_usg_s, all_specs) = unzip specs_w_usg new_usg = combineUsages new_usg_s new_calls = scu_calls new_usg all_usg = usg_so_far `combineUsage` new_usg ; if isEmptyVarEnv new_calls then return (all_usg, all_specs) else specLoop env new_calls rhs_infos all_usg all_specs } specialise :: ScEnv -> CallEnv -- Info on newly-discovered calls to this function -> RhsInfo -> SpecInfo -- Original RHS plus patterns dealt with -> UniqSM (ScUsage, SpecInfo) -- New specialised versions and their usage -- See Note [spec_usg includes rhs_usg] -- Note: this only generates *specialised* bindings -- The original binding is added by specInfoBinds -- -- Note: the rhs here is the optimised version of the original rhs -- So when we make a specialised copy of the RHS, we're starting -- from an RHS whose nested functions have been optimised already. specialise env bind_calls (RI fn _ arg_bndrs body arg_occs) spec_info@(SI specs spec_count mb_unspec) | isBottomingId fn -- Note [Do not specialise diverging functions] -- and do not generate specialisation seeds from its RHS = return (nullUsage, spec_info) | isNeverActive (idInlineActivation fn) -- See Note [Transfer activation] || null arg_bndrs -- Only specialise functions = case mb_unspec of -- Behave as if there was a single, boring call Just rhs_usg -> return (rhs_usg, SI specs spec_count Nothing) -- See Note [spec_usg includes rhs_usg] Nothing -> return (nullUsage, spec_info) | Just all_calls <- lookupVarEnv bind_calls fn = -- pprTrace "specialise entry {" (ppr fn <+> ppr (length all_calls)) $ do { (boring_call, pats) <- callsToPats env specs arg_occs all_calls -- Bale out if too many specialisations ; let n_pats = length pats spec_count' = n_pats + spec_count ; case sc_count env of Just max | not (sc_force env) && spec_count' > max -> if (debugIsOn || opt_PprStyle_Debug) -- Suppress this scary message for then pprTrace "SpecConstr" msg $ -- ordinary users! Trac #5125 return (nullUsage, spec_info) else return (nullUsage, spec_info) where msg = vcat [ sep [ ptext (sLit "Function") <+> quotes (ppr fn) , nest 2 (ptext (sLit "has") <+> speakNOf spec_count' (ptext (sLit "call pattern")) <> comma <+> ptext (sLit "but the limit is") <+> int max) ] , ptext (sLit "Use -fspec-constr-count=n to set the bound") , extra ] extra | not opt_PprStyle_Debug = ptext (sLit "Use -dppr-debug to see specialisations") | otherwise = ptext (sLit "Specialisations:") <+> ppr (pats ++ [p | OS p _ _ _ <- specs]) _normal_case -> do { -- ; if (not (null pats) || isJust mb_unspec) then -- pprTrace "specialise" (vcat [ ppr fn <+> text "with" <+> int (length pats) <+> text "good patterns" -- , text "mb_unspec" <+> ppr (isJust mb_unspec) -- , text "arg_occs" <+> ppr arg_occs -- , text "good pats" <+> ppr pats]) $ -- return () -- else return () ; let spec_env = decreaseSpecCount env n_pats ; (spec_usgs, new_specs) <- mapAndUnzipM (spec_one spec_env fn arg_bndrs body) (pats `zip` [spec_count..]) -- See Note [Specialise original body] ; let spec_usg = combineUsages spec_usgs -- If there were any boring calls among the seeds (= all_calls), then those -- calls will call the un-specialised function. So we should use the seeds -- from the _unspecialised_ function's RHS, which are in mb_unspec, by returning -- then in new_usg. (new_usg, mb_unspec') = case mb_unspec of Just rhs_usg | boring_call -> (spec_usg `combineUsage` rhs_usg, Nothing) _ -> (spec_usg, mb_unspec) -- ; pprTrace "specialise return }" (ppr fn -- <+> ppr (scu_calls new_usg)) ; return (new_usg, SI (new_specs ++ specs) spec_count' mb_unspec') } } | otherwise -- No new seeds, so return nullUsage = return (nullUsage, spec_info) --------------------- spec_one :: ScEnv -> OutId -- Function -> [InVar] -- Lambda-binders of RHS; should match patterns -> InExpr -- Body of the original function -> (CallPat, Int) -> UniqSM (ScUsage, OneSpec) -- Rule and binding -- spec_one creates a specialised copy of the function, together -- with a rule for using it. I'm very proud of how short this -- function is, considering what it does :-). {- Example In-scope: a, x::a f = /\b \y::[(a,b)] -> ....f (b,c) ((:) (a,(b,c)) (x,v) (h w))... [c::*, v::(b,c) are presumably bound by the (...) part] ==> f_spec = /\ b c \ v::(b,c) hw::[(a,(b,c))] -> (...entire body of f...) [b -> (b,c), y -> ((:) (a,(b,c)) (x,v) hw)] RULE: forall b::* c::*, -- Note, *not* forall a, x v::(b,c), hw::[(a,(b,c))] . f (b,c) ((:) (a,(b,c)) (x,v) hw) = f_spec b c v hw -} spec_one env fn arg_bndrs body (call_pat@(qvars, pats), rule_number) = do { spec_uniq <- getUniqueUs ; let spec_env = extendScSubstList (extendScInScope env qvars) (arg_bndrs `zip` pats) fn_name = idName fn fn_loc = nameSrcSpan fn_name fn_occ = nameOccName fn_name spec_occ = mkSpecOcc fn_occ -- We use fn_occ rather than fn in the rule_name string -- as we don't want the uniq to end up in the rule, and -- hence in the ABI, as that can cause spurious ABI -- changes (#4012). rule_name = mkFastString ("SC:" ++ occNameString fn_occ ++ show rule_number) spec_name = mkInternalName spec_uniq spec_occ fn_loc -- ; pprTrace "{spec_one" (ppr (sc_count env) <+> ppr fn <+> ppr pats <+> text "-->" <+> ppr spec_name) $ -- return () -- Specialise the body ; (spec_usg, spec_body) <- scExpr spec_env body -- ; pprTrace "done spec_one}" (ppr fn) $ -- return () -- And build the results ; let spec_id = mkLocalId spec_name (mkPiTypes spec_lam_args body_ty) -- See Note [Transfer strictness] `setIdStrictness` spec_str `setIdArity` count isId spec_lam_args spec_str = calcSpecStrictness fn spec_lam_args pats -- Conditionally use result of new worker-wrapper transform (spec_lam_args, spec_call_args) = mkWorkerArgs (sc_dflags env) qvars NoOneShotInfo body_ty -- Usual w/w hack to avoid generating -- a spec_rhs of unlifted type and no args spec_rhs = mkLams spec_lam_args spec_body body_ty = exprType spec_body rule_rhs = mkVarApps (Var spec_id) spec_call_args inline_act = idInlineActivation fn rule = mkRule True {- Auto -} True {- Local -} rule_name inline_act fn_name qvars pats rule_rhs -- See Note [Transfer activation] ; return (spec_usg, OS call_pat rule spec_id spec_rhs) } calcSpecStrictness :: Id -- The original function -> [Var] -> [CoreExpr] -- Call pattern -> StrictSig -- Strictness of specialised thing -- See Note [Transfer strictness] calcSpecStrictness fn qvars pats = mkClosedStrictSig spec_dmds topRes where spec_dmds = [ lookupVarEnv dmd_env qv `orElse` topDmd | qv <- qvars, isId qv ] StrictSig (DmdType _ dmds _) = idStrictness fn dmd_env = go emptyVarEnv dmds pats go :: DmdEnv -> [Demand] -> [CoreExpr] -> DmdEnv go env ds (Type {} : pats) = go env ds pats go env ds (Coercion {} : pats) = go env ds pats go env (d:ds) (pat : pats) = go (go_one env d pat) ds pats go env _ _ = env go_one :: DmdEnv -> Demand -> CoreExpr -> DmdEnv go_one env d (Var v) = extendVarEnv_C bothDmd env v d go_one env d e | Just ds <- splitProdDmd_maybe d -- NB: d does not have to be strict , (Var _, args) <- collectArgs e = go env ds args go_one env _ _ = env\end{code} Note [spec_usg includes rhs_usg] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ In calls to 'specialise', the returned ScUsage must include the rhs_usg in the passed-in SpecInfo, unless there are no calls at all to the function. The caller can, indeed must, assume this. He should not combine in rhs_usg himself, or he'll get rhs_usg twice -- and that can lead to an exponential blowup of duplicates in the CallEnv. This is what gave rise to the massive performace loss in Trac #8852. Note [Specialise original body] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The RhsInfo for a binding keeps the *original* body of the binding. We must specialise that, *not* the result of applying specExpr to the RHS (which is also kept in RhsInfo). Otherwise we end up specialising a specialised RHS, and that can lead directly to exponential behaviour. Note [Transfer activation] ~~~~~~~~~~~~~~~~~~~~~~~~~~ This note is for SpecConstr, but exactly the same thing happens in the overloading specialiser; see Note [Auto-specialisation and RULES] in Specialise. In which phase should the specialise-constructor rules be active? Originally I made them always-active, but Manuel found that this defeated some clever user-written rules. Then I made them active only in Phase 0; after all, currently, the specConstr transformation is only run after the simplifier has reached Phase 0, but that meant that specialisations didn't fire inside wrappers; see test simplCore/should_compile/spec-inline. So now I just use the inline-activation of the parent Id, as the activation for the specialiation RULE, just like the main specialiser; This in turn means there is no point in specialising NOINLINE things, so we test for that. Note [Transfer strictness] ~~~~~~~~~~~~~~~~~~~~~~~~~~ We must transfer strictness information from the original function to the specialised one. Suppose, for example f has strictness SS and a RULE f (a:as) b = f_spec a as b Now we want f_spec to have strictness LLS, otherwise we'll use call-by-need when calling f_spec instead of call-by-value. And that can result in unbounded worsening in space (cf the classic foldl vs foldl') See Trac #3437 for a good example. The function calcSpecStrictness performs the calculation. %************************************************************************ %* * \subsection{Argument analysis} %* * %************************************************************************ This code deals with analysing call-site arguments to see whether they are constructor applications. Note [Free type variables of the qvar types] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ In a call (f @a x True), that we want to specialise, what variables should we quantify over. Clearly over 'a' and 'x', but what about any type variables free in x's type? In fact we don't need to worry about them because (f @a) can only be a well-typed application if its type is compatible with x, so any variables free in x's type must be free in (f @a), and hence either be gathered via 'a' itself, or be in scope at f's defn. Hence we just take (exprsFreeVars pats). BUT phantom type synonyms can mess this reasoning up, eg x::T b with type T b = Int So we apply expandTypeSynonyms to the bound Ids. See Trac # 5458. Yuk. \begin{code}
type CallPat = ([Var], [CoreExpr]) -- Quantified variables and arguments callsToPats :: ScEnv -> [OneSpec] -> [ArgOcc] -> [Call] -> UniqSM (Bool, [CallPat]) -- Result has no duplicate patterns, -- nor ones mentioned in done_pats -- Bool indicates that there was at least one boring pattern callsToPats env done_specs bndr_occs calls = do { mb_pats <- mapM (callToPats env bndr_occs) calls ; let good_pats :: [(CallPat, ValueEnv)] good_pats = catMaybes mb_pats done_pats = [p | OS p _ _ _ <- done_specs] is_done p = any (samePat p) done_pats no_recursive = map fst (filterOut (is_too_recursive env) good_pats) ; return (any isNothing mb_pats, filterOut is_done (nubBy samePat no_recursive)) } is_too_recursive :: ScEnv -> (CallPat, ValueEnv) -> Bool -- Count the number of recursive constructors in a call pattern, -- filter out if there are more than the maximum. -- This is only necessary if ForceSpecConstr is in effect: -- otherwise specConstrCount will cause specialisation to terminate. -- See Note [Limit recursive specialisation] is_too_recursive env ((_,exprs), val_env) = sc_force env && maximum (map go exprs) > sc_recursive env where go e | Just (ConVal (DataAlt dc) args) <- isValue val_env e , isRecursiveTyCon (dataConTyCon dc) = 1 + sum (map go args) |App f a <- e = go f + go a | otherwise = 0 callToPats :: ScEnv -> [ArgOcc] -> Call -> UniqSM (Maybe (CallPat, ValueEnv)) -- The [Var] is the variables to quantify over in the rule -- Type variables come first, since they may scope -- over the following term variables -- The [CoreExpr] are the argument patterns for the rule callToPats env bndr_occs (Call _ args con_env) | length args < length bndr_occs -- Check saturated = return Nothing | otherwise = do { let in_scope = substInScope (sc_subst env) ; (interesting, pats) <- argsToPats env in_scope con_env args bndr_occs ; let pat_fvs = varSetElems (exprsFreeVars pats) in_scope_vars = getInScopeVars in_scope qvars = filterOut (`elemVarSet` in_scope_vars) pat_fvs -- Quantify over variables that are not in scope -- at the call site -- See Note [Free type variables of the qvar types] -- See Note [Shadowing] at the top (tvs, ids) = partition isTyVar qvars qvars' = tvs ++ map sanitise ids -- Put the type variables first; the type of a term -- variable may mention a type variable sanitise id = id `setIdType` expandTypeSynonyms (idType id) -- See Note [Free type variables of the qvar types] ; -- pprTrace "callToPats" (ppr args $$ ppr bndr_occs) $ if interesting then return (Just ((qvars', pats), con_env)) else return Nothing } -- argToPat takes an actual argument, and returns an abstracted -- version, consisting of just the "constructor skeleton" of the -- argument, with non-constructor sub-expression replaced by new -- placeholder variables. For example: -- C a (D (f x) (g y)) ==> C p1 (D p2 p3) argToPat :: ScEnv -> InScopeSet -- What's in scope at the fn defn site -> ValueEnv -- ValueEnv at the call site -> CoreArg -- A call arg (or component thereof) -> ArgOcc -> UniqSM (Bool, CoreArg) -- Returns (interesting, pat), -- where pat is the pattern derived from the argument -- interesting=True if the pattern is non-trivial (not a variable or type) -- E.g. x:xs --> (True, x:xs) -- f xs --> (False, w) where w is a fresh wildcard -- (f xs, 'c') --> (True, (w, 'c')) where w is a fresh wildcard -- \x. x+y --> (True, \x. x+y) -- lvl7 --> (True, lvl7) if lvl7 is bound -- somewhere further out argToPat _env _in_scope _val_env arg@(Type {}) _arg_occ = return (False, arg) argToPat env in_scope val_env (Tick _ arg) arg_occ = argToPat env in_scope val_env arg arg_occ -- Note [Notes in call patterns] -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -- Ignore Notes. In particular, we want to ignore any InlineMe notes -- Perhaps we should not ignore profiling notes, but I'm going to -- ride roughshod over them all for now. --- See Note [Notes in RULE matching] in Rules argToPat env in_scope val_env (Let _ arg) arg_occ = argToPat env in_scope val_env arg arg_occ -- See Note [Matching lets] in Rule.lhs -- Look through let expressions -- e.g. f (let v = rhs in (v,w)) -- Here we can specialise for f (v,w) -- because the rule-matcher will look through the let. {- Disabled; see Note [Matching cases] in Rule.lhs argToPat env in_scope val_env (Case scrut _ _ [(_, _, rhs)]) arg_occ | exprOkForSpeculation scrut -- See Note [Matching cases] in Rule.hhs = argToPat env in_scope val_env rhs arg_occ -} argToPat env in_scope val_env (Cast arg co) arg_occ | isReflCo co -- Substitution in the SpecConstr itself -- can lead to identity coercions = argToPat env in_scope val_env arg arg_occ | not (ignoreType env ty2) = do { (interesting, arg') <- argToPat env in_scope val_env arg arg_occ ; if not interesting then wildCardPat ty2 else do { -- Make a wild-card pattern for the coercion uniq <- getUniqueUs ; let co_name = mkSysTvName uniq (fsLit "sg") co_var = mkCoVar co_name (mkCoercionType Representational ty1 ty2) ; return (interesting, Cast arg' (mkCoVarCo co_var)) } } where Pair ty1 ty2 = coercionKind co {- Disabling lambda specialisation for now It's fragile, and the spec_loop can be infinite argToPat in_scope val_env arg arg_occ | is_value_lam arg = return (True, arg) where is_value_lam (Lam v e) -- Spot a value lambda, even if | isId v = True -- it is inside a type lambda | otherwise = is_value_lam e is_value_lam other = False -} -- Check for a constructor application -- NB: this *precedes* the Var case, so that we catch nullary constrs argToPat env in_scope val_env arg arg_occ | Just (ConVal (DataAlt dc) args) <- isValue val_env arg , not (ignoreDataCon env dc) -- See Note [NoSpecConstr] , Just arg_occs <- mb_scrut dc = do { let (ty_args, rest_args) = splitAtList (dataConUnivTyVars dc) args ; (_, args') <- argsToPats env in_scope val_env rest_args arg_occs ; return (True, mkConApp dc (ty_args ++ args')) } where mb_scrut dc = case arg_occ of ScrutOcc bs | Just occs <- lookupUFM bs dc -> Just (occs) -- See Note [Reboxing] _other | sc_force env -> Just (repeat UnkOcc) | otherwise -> Nothing -- Check if the argument is a variable that -- (a) is used in an interesting way in the body -- (b) we know what its value is -- In that case it counts as "interesting" argToPat env in_scope val_env (Var v) arg_occ | sc_force env || case arg_occ of { UnkOcc -> False; _other -> True }, -- (a) is_value, -- (b) not (ignoreType env (varType v)) = return (True, Var v) where is_value | isLocalId v = v `elemInScopeSet` in_scope && isJust (lookupVarEnv val_env v) -- Local variables have values in val_env | otherwise = isValueUnfolding (idUnfolding v) -- Imports have unfoldings -- I'm really not sure what this comment means -- And by not wild-carding we tend to get forall'd -- variables that are in scope, which in turn can -- expose the weakness in let-matching -- See Note [Matching lets] in Rules -- Check for a variable bound inside the function. -- Don't make a wild-card, because we may usefully share -- e.g. f a = let x = ... in f (x,x) -- NB: this case follows the lambda and con-app cases!! -- argToPat _in_scope _val_env (Var v) _arg_occ -- = return (False, Var v) -- SLPJ : disabling this to avoid proliferation of versions -- also works badly when thinking about seeding the loop -- from the body of the let -- f x y = letrec g z = ... in g (x,y) -- We don't want to specialise for that *particular* x,y -- The default case: make a wild-card -- We use this for coercions too argToPat _env _in_scope _val_env arg _arg_occ = wildCardPat (exprType arg) wildCardPat :: Type -> UniqSM (Bool, CoreArg) wildCardPat ty = do { uniq <- getUniqueUs ; let id = mkSysLocal (fsLit "sc") uniq ty ; return (False, varToCoreExpr id) } argsToPats :: ScEnv -> InScopeSet -> ValueEnv -> [CoreArg] -> [ArgOcc] -- Should be same length -> UniqSM (Bool, [CoreArg]) argsToPats env in_scope val_env args occs = do { stuff <- zipWithM (argToPat env in_scope val_env) args occs ; let (interesting_s, args') = unzip stuff ; return (or interesting_s, args') }\end{code} \begin{code}
isValue :: ValueEnv -> CoreExpr -> Maybe Value isValue _env (Lit lit) | litIsLifted lit = Nothing | otherwise = Just (ConVal (LitAlt lit) []) isValue env (Var v) | Just cval <- lookupVarEnv env v = Just cval -- You might think we could look in the idUnfolding here -- but that doesn't take account of which branch of a -- case we are in, which is the whole point | not (isLocalId v) && isCheapUnfolding unf = isValue env (unfoldingTemplate unf) where unf = idUnfolding v -- However we do want to consult the unfolding -- as well, for let-bound constructors! isValue env (Lam b e) | isTyVar b = case isValue env e of Just _ -> Just LambdaVal Nothing -> Nothing | otherwise = Just LambdaVal isValue _env expr -- Maybe it's a constructor application | (Var fun, args) <- collectArgs expr = case isDataConWorkId_maybe fun of Just con | args `lengthAtLeast` dataConRepArity con -- Check saturated; might be > because the -- arity excludes type args -> Just (ConVal (DataAlt con) args) _other | valArgCount args < idArity fun -- Under-applied function -> Just LambdaVal -- Partial application _other -> Nothing isValue _env _expr = Nothing valueIsWorkFree :: Value -> Bool valueIsWorkFree LambdaVal = True valueIsWorkFree (ConVal _ args) = all exprIsWorkFree args samePat :: CallPat -> CallPat -> Bool samePat (vs1, as1) (vs2, as2) = all2 same as1 as2 where same (Var v1) (Var v2) | v1 `elem` vs1 = v2 `elem` vs2 | v2 `elem` vs2 = False | otherwise = v1 == v2 same (Lit l1) (Lit l2) = l1==l2 same (App f1 a1) (App f2 a2) = same f1 f2 && same a1 a2 same (Type {}) (Type {}) = True -- Note [Ignore type differences] same (Coercion {}) (Coercion {}) = True same (Tick _ e1) e2 = same e1 e2 -- Ignore casts and notes same (Cast e1 _) e2 = same e1 e2 same e1 (Tick _ e2) = same e1 e2 same e1 (Cast e2 _) = same e1 e2 same e1 e2 = WARN( bad e1 || bad e2, ppr e1 $$ ppr e2) False -- Let, lambda, case should not occur bad (Case {}) = True bad (Let {}) = True bad (Lam {}) = True bad _other = False\end{code} Note [Ignore type differences] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We do not want to generate specialisations where the call patterns differ only in their type arguments! Not only is it utterly useless, but it also means that (with polymorphic recursion) we can generate an infinite number of specialisations. Example is Data.Sequence.adjustTree, I think.