{- (c) The AQUA Project, Glasgow University, 1993-1998 \section[SimplUtils]{The simplifier utilities} -} {-# LANGUAGE CPP #-} module SimplUtils ( -- Rebuilding mkLam, mkCase, prepareAlts, tryEtaExpandRhs, -- Inlining, preInlineUnconditionally, postInlineUnconditionally, activeUnfolding, activeRule, getUnfoldingInRuleMatch, simplEnvForGHCi, updModeForStableUnfoldings, -- The continuation type SimplCont(..), DupFlag(..), isSimplified, contIsDupable, contResultType, contHoleType, contIsTrivial, contArgs, countValArgs, countArgs, mkBoringStop, mkRhsStop, mkLazyArgStop, contIsRhsOrArg, interestingCallContext, -- ArgInfo ArgInfo(..), ArgSpec(..), mkArgInfo, addValArgTo, addCastTo, addTyArgTo, argInfoExpr, argInfoAppArgs, pushSimplifiedArgs, abstractFloats ) where #include "HsVersions.h" import SimplEnv import CoreMonad ( SimplifierMode(..), Tick(..) ) import MkCore ( sortQuantVars ) import DynFlags import CoreSyn import qualified CoreSubst import PprCore import CoreFVs import CoreUtils import CoreArity import CoreUnfold import Name import Id import Var import Demand import SimplMonad import Type hiding( substTy ) import Coercion hiding( substCo, substTy ) import DataCon ( dataConWorkId ) import VarEnv import VarSet import BasicTypes import Util import MonadUtils import Outputable import FastString import Pair import ListSetOps ( minusList ) import Control.Monad ( when ) import Data.List ( partition ) {- ************************************************************************ * * The SimplCont and DupFlag types * * ************************************************************************ A SimplCont allows the simplifier to traverse the expression in a zipper-like fashion. The SimplCont represents the rest of the expression, "above" the point of interest. You can also think of a SimplCont as an "evaluation context", using that term in the way it is used for operational semantics. This is the way I usually think of it, For example you'll often see a syntax for evaluation context looking like C ::= [] | C e | case C of alts | C `cast` co That's the kind of thing we are doing here, and I use that syntax in the comments. Key points: * A SimplCont describes a *strict* context (just like evaluation contexts do). E.g. Just [] is not a SimplCont * A SimplCont describes a context that *does not* bind any variables. E.g. \x. [] is not a SimplCont -} data SimplCont = Stop -- An empty context, or <hole> OutType -- Type of the <hole> CallCtxt -- Tells if there is something interesting about -- the context, and hence the inliner -- should be a bit keener (see interestingCallContext) -- Specifically: -- This is an argument of a function that has RULES -- Inlining the call might allow the rule to fire -- Never ValAppCxt (use ApplyToVal instead) -- or CaseCtxt (use Select instead) | CastIt -- <hole> `cast` co OutCoercion -- The coercion simplified -- Invariant: never an identity coercion SimplCont | ApplyToVal { -- <hole> arg sc_dup :: DupFlag, -- See Note [DupFlag invariants] sc_arg :: InExpr, -- The argument, sc_env :: StaticEnv, -- and its static env sc_cont :: SimplCont } | ApplyToTy { -- <hole> ty sc_arg_ty :: OutType, -- Argument type sc_hole_ty :: OutType, -- Type of the function, presumably (forall a. blah) -- See Note [The hole type in ApplyToTy] sc_cont :: SimplCont } | Select -- case <hole> of alts DupFlag -- See Note [DupFlag invariants] InId [InAlt] StaticEnv -- The case binder, alts type, alts, and subst-env SimplCont -- The two strict forms have no DupFlag, because we never duplicate them | StrictBind -- (\x* \xs. e) <hole> InId [InBndr] -- let x* = <hole> in e InExpr StaticEnv -- is a special case SimplCont | StrictArg -- f e1 ..en <hole> ArgInfo -- Specifies f, e1..en, Whether f has rules, etc -- plus strictness flags for *further* args CallCtxt -- Whether *this* argument position is interesting SimplCont | TickIt (Tickish Id) -- Tick tickish <hole> SimplCont data DupFlag = NoDup -- Unsimplified, might be big | Simplified -- Simplified | OkToDup -- Simplified and small isSimplified :: DupFlag -> Bool isSimplified NoDup = False isSimplified _ = True -- Invariant: the subst-env is empty perhapsSubstTy :: DupFlag -> StaticEnv -> Type -> Type perhapsSubstTy dup env ty | isSimplified dup = ty | otherwise = substTy env ty {- Note [DupFlag invariants] ~~~~~~~~~~~~~~~~~~~~~~~~~ In both (ApplyToVal dup _ env k) and (Select dup _ _ env k) the following invariants hold (a) if dup = OkToDup, then continuation k is also ok-to-dup (b) if dup = OkToDup or Simplified, the subst-env is empty (and and hence no need to re-simplify) -} instance Outputable DupFlag where ppr OkToDup = ptext (sLit "ok") ppr NoDup = ptext (sLit "nodup") ppr Simplified = ptext (sLit "simpl") instance Outputable SimplCont where ppr (Stop ty interesting) = ptext (sLit "Stop") <> brackets (ppr interesting) <+> ppr ty ppr (ApplyToTy { sc_arg_ty = ty , sc_cont = cont }) = (ptext (sLit "ApplyToTy") <+> pprParendType ty) $$ ppr cont ppr (ApplyToVal { sc_arg = arg , sc_dup = dup , sc_cont = cont }) = (ptext (sLit "ApplyToVal") <+> ppr dup <+> pprParendExpr arg) $$ ppr cont ppr (StrictBind b _ _ _ cont) = (ptext (sLit "StrictBind") <+> ppr b) $$ ppr cont ppr (StrictArg ai _ cont) = (ptext (sLit "StrictArg") <+> ppr (ai_fun ai)) $$ ppr cont ppr (Select dup bndr alts se cont) = (ptext (sLit "Select") <+> ppr dup <+> ppr bndr) $$ ifPprDebug (nest 2 $ vcat [ppr (seTvSubst se), ppr alts]) $$ ppr cont ppr (CastIt co cont ) = (ptext (sLit "CastIt") <+> ppr co) $$ ppr cont ppr (TickIt t cont) = (ptext (sLit "TickIt") <+> ppr t) $$ ppr cont {- Note [The hole type in ApplyToTy] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The sc_hole_ty field of ApplyToTy records the type of the "hole" in the continuation. It is absolutely necessary to compute contHoleType, but it is not used for anything else (and hence may not be evaluated). Why is it necessary for contHoleType? Consider the continuation ApplyToType Int (Stop Int) corresponding to (<hole> @Int) :: Int What is the type of <hole>? It could be (forall a. Int) or (forall a. a), and there is no way to know which, so we must record it. In a chain of applications (f @t1 @t2 @t3) we'll lazily compute exprType for (f @t1) and (f @t1 @t2), which is potentially non-linear; but it probably doesn't matter because we'll never compute them all. ************************************************************************ * * ArgInfo and ArgSpec * * ************************************************************************ -} data ArgInfo = ArgInfo { ai_fun :: OutId, -- The function ai_args :: [ArgSpec], -- ...applied to these args (which are in *reverse* order) ai_type :: OutType, -- Type of (f a1 ... an) ai_rules :: [CoreRule], -- Rules for this function ai_encl :: Bool, -- Flag saying whether this function -- or an enclosing one has rules (recursively) -- True => be keener to inline in all args ai_strs :: [Bool], -- Strictness of remaining arguments -- Usually infinite, but if it is finite it guarantees -- that the function diverges after being given -- that number of args ai_discs :: [Int] -- Discounts for remaining arguments; non-zero => be keener to inline -- Always infinite } data ArgSpec = ValArg OutExpr -- Apply to this (coercion or value); c.f. ApplyToVal | TyArg { as_arg_ty :: OutType -- Apply to this type; c.f. ApplyToTy , as_hole_ty :: OutType } -- Type of the function (presumably forall a. blah) | CastBy OutCoercion -- Cast by this; c.f. CastIt instance Outputable ArgSpec where ppr (ValArg e) = ptext (sLit "ValArg") <+> ppr e ppr (TyArg { as_arg_ty = ty }) = ptext (sLit "TyArg") <+> ppr ty ppr (CastBy c) = ptext (sLit "CastBy") <+> ppr c addValArgTo :: ArgInfo -> OutExpr -> ArgInfo addValArgTo ai arg = ai { ai_args = ValArg arg : ai_args ai , ai_type = funResultTy (ai_type ai) } addTyArgTo :: ArgInfo -> OutType -> ArgInfo addTyArgTo ai arg_ty = ai { ai_args = arg_spec : ai_args ai , ai_type = applyTy poly_fun_ty arg_ty } where poly_fun_ty = ai_type ai arg_spec = TyArg { as_arg_ty = arg_ty, as_hole_ty = poly_fun_ty } addCastTo :: ArgInfo -> OutCoercion -> ArgInfo addCastTo ai co = ai { ai_args = CastBy co : ai_args ai , ai_type = pSnd (coercionKind co) } argInfoAppArgs :: [ArgSpec] -> [OutExpr] argInfoAppArgs [] = [] argInfoAppArgs (CastBy {} : _) = [] -- Stop at a cast argInfoAppArgs (ValArg e : as) = e : argInfoAppArgs as argInfoAppArgs (TyArg { as_arg_ty = ty } : as) = Type ty : argInfoAppArgs as pushSimplifiedArgs :: SimplEnv -> [ArgSpec] -> SimplCont -> SimplCont pushSimplifiedArgs _env [] k = k pushSimplifiedArgs env (arg : args) k = case arg of TyArg { as_arg_ty = arg_ty, as_hole_ty = hole_ty } -> ApplyToTy { sc_arg_ty = arg_ty, sc_hole_ty = hole_ty, sc_cont = rest } ValArg e -> ApplyToVal { sc_arg = e, sc_env = env, sc_dup = Simplified, sc_cont = rest } CastBy c -> CastIt c rest where rest = pushSimplifiedArgs env args k -- The env has an empty SubstEnv argInfoExpr :: OutId -> [ArgSpec] -> OutExpr -- NB: the [ArgSpec] is reversed so that the first arg -- in the list is the last one in the application argInfoExpr fun rev_args = go rev_args where go [] = Var fun go (ValArg a : as) = go as `App` a go (TyArg { as_arg_ty = ty } : as) = go as `App` Type ty go (CastBy co : as) = mkCast (go as) co {- ************************************************************************ * * Functions on SimplCont * * ************************************************************************ -} mkBoringStop :: OutType -> SimplCont mkBoringStop ty = Stop ty BoringCtxt mkRhsStop :: OutType -> SimplCont -- See Note [RHS of lets] in CoreUnfold mkRhsStop ty = Stop ty RhsCtxt mkLazyArgStop :: OutType -> CallCtxt -> SimplCont mkLazyArgStop ty cci = Stop ty cci ------------------- contIsRhsOrArg :: SimplCont -> Bool contIsRhsOrArg (Stop {}) = True contIsRhsOrArg (StrictBind {}) = True contIsRhsOrArg (StrictArg {}) = True contIsRhsOrArg _ = False contIsRhs :: SimplCont -> Bool contIsRhs (Stop _ RhsCtxt) = True contIsRhs _ = False ------------------- contIsDupable :: SimplCont -> Bool contIsDupable (Stop {}) = True contIsDupable (ApplyToTy { sc_cont = k }) = contIsDupable k contIsDupable (ApplyToVal { sc_dup = OkToDup }) = True -- See Note [DupFlag invariants] contIsDupable (Select OkToDup _ _ _ _) = True -- ...ditto... contIsDupable (CastIt _ k) = contIsDupable k contIsDupable _ = False ------------------- contIsTrivial :: SimplCont -> Bool contIsTrivial (Stop {}) = True contIsTrivial (ApplyToTy { sc_cont = k }) = contIsTrivial k contIsTrivial (ApplyToVal { sc_arg = Coercion _, sc_cont = k }) = contIsTrivial k contIsTrivial (CastIt _ k) = contIsTrivial k contIsTrivial _ = False ------------------- contResultType :: SimplCont -> OutType contResultType (Stop ty _) = ty contResultType (CastIt _ k) = contResultType k contResultType (StrictBind _ _ _ _ k) = contResultType k contResultType (StrictArg _ _ k) = contResultType k contResultType (Select _ _ _ _ k) = contResultType k contResultType (ApplyToTy { sc_cont = k }) = contResultType k contResultType (ApplyToVal { sc_cont = k }) = contResultType k contResultType (TickIt _ k) = contResultType k contHoleType :: SimplCont -> OutType contHoleType (Stop ty _) = ty contHoleType (TickIt _ k) = contHoleType k contHoleType (CastIt co _) = pFst (coercionKind co) contHoleType (Select d b _ se _) = perhapsSubstTy d se (idType b) contHoleType (StrictBind b _ _ se _) = substTy se (idType b) contHoleType (StrictArg ai _ _) = funArgTy (ai_type ai) contHoleType (ApplyToTy { sc_hole_ty = ty }) = ty -- See Note [The hole type in ApplyToTy] contHoleType (ApplyToVal { sc_arg = e, sc_env = se, sc_dup = dup, sc_cont = k }) = mkFunTy (perhapsSubstTy dup se (exprType e)) (contHoleType k) ------------------- countValArgs :: SimplCont -> Int -- Count value arguments excluding coercions countValArgs (ApplyToVal { sc_arg = arg, sc_cont = cont }) | Coercion {} <- arg = countValArgs cont | otherwise = 1 + countValArgs cont countValArgs _ = 0 countArgs :: SimplCont -> Int -- Count all arguments, including types, coercions, and other values countArgs (ApplyToTy { sc_cont = cont }) = 1 + countArgs cont countArgs (ApplyToVal { sc_cont = cont }) = 1 + countArgs cont countArgs _ = 0 contArgs :: SimplCont -> (Bool, [ArgSummary], SimplCont) -- Summarises value args, discards type args and coercions -- The returned continuation of the call is only used to -- answer questions like "are you interesting?" contArgs cont | lone cont = (True, [], cont) | otherwise = go [] cont where lone (ApplyToTy {}) = False -- See Note [Lone variables] in CoreUnfold lone (ApplyToVal {}) = False lone (CastIt {}) = False lone _ = True go args (ApplyToVal { sc_arg = arg, sc_env = se, sc_cont = k }) = go (is_interesting arg se : args) k go args (ApplyToTy { sc_cont = k }) = go args k go args (CastIt _ k) = go args k go args k = (False, reverse args, k) is_interesting arg se = interestingArg se arg -- Do *not* use short-cutting substitution here -- because we want to get as much IdInfo as possible ------------------- mkArgInfo :: Id -> [CoreRule] -- Rules for function -> Int -- Number of value args -> SimplCont -- Context of the call -> ArgInfo mkArgInfo fun rules n_val_args call_cont | n_val_args < idArity fun -- Note [Unsaturated functions] = ArgInfo { ai_fun = fun, ai_args = [], ai_type = fun_ty , ai_rules = rules, ai_encl = False , ai_strs = vanilla_stricts , ai_discs = vanilla_discounts } | otherwise = ArgInfo { ai_fun = fun, ai_args = [], ai_type = fun_ty , ai_rules = rules , ai_encl = interestingArgContext rules call_cont , ai_strs = add_type_str fun_ty arg_stricts , ai_discs = arg_discounts } where fun_ty = idType fun vanilla_discounts, arg_discounts :: [Int] vanilla_discounts = repeat 0 arg_discounts = case idUnfolding fun of CoreUnfolding {uf_guidance = UnfIfGoodArgs {ug_args = discounts}} -> discounts ++ vanilla_discounts _ -> vanilla_discounts vanilla_stricts, arg_stricts :: [Bool] vanilla_stricts = repeat False arg_stricts = case splitStrictSig (idStrictness fun) of (demands, result_info) | not (demands `lengthExceeds` n_val_args) -> -- Enough args, use the strictness given. -- For bottoming functions we used to pretend that the arg -- is lazy, so that we don't treat the arg as an -- interesting context. This avoids substituting -- top-level bindings for (say) strings into -- calls to error. But now we are more careful about -- inlining lone variables, so its ok (see SimplUtils.analyseCont) if isBotRes result_info then map isStrictDmd demands -- Finite => result is bottom else map isStrictDmd demands ++ vanilla_stricts | otherwise -> WARN( True, text "More demands than arity" <+> ppr fun <+> ppr (idArity fun) <+> ppr n_val_args <+> ppr demands ) vanilla_stricts -- Not enough args, or no strictness add_type_str :: Type -> [Bool] -> [Bool] -- If the function arg types are strict, record that in the 'strictness bits' -- No need to instantiate because unboxed types (which dominate the strict -- types) can't instantiate type variables. -- add_type_str is done repeatedly (for each call); might be better -- once-for-all in the function -- But beware primops/datacons with no strictness add_type_str _ [] = [] add_type_str fun_ty strs -- Look through foralls | Just (_, fun_ty') <- splitForAllTy_maybe fun_ty -- Includes coercions = add_type_str fun_ty' strs add_type_str fun_ty (str:strs) -- Add strict-type info | Just (arg_ty, fun_ty') <- splitFunTy_maybe fun_ty = (str || isStrictType arg_ty) : add_type_str fun_ty' strs add_type_str _ strs = strs {- Note [Unsaturated functions] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider (test eyeball/inline4) x = a:as y = f x where f has arity 2. Then we do not want to inline 'x', because it'll just be floated out again. Even if f has lots of discounts on its first argument -- it must be saturated for these to kick in -} {- ************************************************************************ * * Interesting arguments * * ************************************************************************ Note [Interesting call context] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We want to avoid inlining an expression where there can't possibly be any gain, such as in an argument position. Hence, if the continuation is interesting (eg. a case scrutinee, application etc.) then we inline, otherwise we don't. Previously some_benefit used to return True only if the variable was applied to some value arguments. This didn't work: let x = _coerce_ (T Int) Int (I# 3) in case _coerce_ Int (T Int) x of I# y -> .... we want to inline x, but can't see that it's a constructor in a case scrutinee position, and some_benefit is False. Another example: dMonadST = _/\_ t -> :Monad (g1 _@_ t, g2 _@_ t, g3 _@_ t) .... case dMonadST _@_ x0 of (a,b,c) -> .... we'd really like to inline dMonadST here, but we *don't* want to inline if the case expression is just case x of y { DEFAULT -> ... } since we can just eliminate this case instead (x is in WHNF). Similar applies when x is bound to a lambda expression. Hence contIsInteresting looks for case expressions with just a single default case. -} interestingCallContext :: SimplCont -> CallCtxt -- See Note [Interesting call context] interestingCallContext cont = interesting cont where interesting (Select _ _bndr _ _ _) = CaseCtxt interesting (ApplyToVal {}) = ValAppCtxt -- Can happen if we have (f Int |> co) y -- If f has an INLINE prag we need to give it some -- motivation to inline. See Note [Cast then apply] -- in CoreUnfold interesting (StrictArg _ cci _) = cci interesting (StrictBind {}) = BoringCtxt interesting (Stop _ cci) = cci interesting (TickIt _ k) = interesting k interesting (ApplyToTy { sc_cont = k }) = interesting k interesting (CastIt _ k) = interesting k -- If this call is the arg of a strict function, the context -- is a bit interesting. If we inline here, we may get useful -- evaluation information to avoid repeated evals: e.g. -- x + (y * z) -- Here the contIsInteresting makes the '*' keener to inline, -- which in turn exposes a constructor which makes the '+' inline. -- Assuming that +,* aren't small enough to inline regardless. -- -- It's also very important to inline in a strict context for things -- like -- foldr k z (f x) -- Here, the context of (f x) is strict, and if f's unfolding is -- a build it's *great* to inline it here. So we must ensure that -- the context for (f x) is not totally uninteresting. interestingArgContext :: [CoreRule] -> SimplCont -> Bool -- If the argument has form (f x y), where x,y are boring, -- and f is marked INLINE, then we don't want to inline f. -- But if the context of the argument is -- g (f x y) -- where g has rules, then we *do* want to inline f, in case it -- exposes a rule that might fire. Similarly, if the context is -- h (g (f x x)) -- where h has rules, then we do want to inline f; hence the -- call_cont argument to interestingArgContext -- -- The ai-rules flag makes this happen; if it's -- set, the inliner gets just enough keener to inline f -- regardless of how boring f's arguments are, if it's marked INLINE -- -- The alternative would be to *always* inline an INLINE function, -- regardless of how boring its context is; but that seems overkill -- For example, it'd mean that wrapper functions were always inlined -- -- The call_cont passed to interestingArgContext is the context of -- the call itself, e.g. g <hole> in the example above interestingArgContext rules call_cont = notNull rules || enclosing_fn_has_rules where enclosing_fn_has_rules = go call_cont go (Select {}) = False go (ApplyToVal {}) = False -- Shouldn't really happen go (ApplyToTy {}) = False -- Ditto go (StrictArg _ cci _) = interesting cci go (StrictBind {}) = False -- ?? go (CastIt _ c) = go c go (Stop _ cci) = interesting cci go (TickIt _ c) = go c interesting RuleArgCtxt = True interesting _ = False {- Note [Interesting arguments] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ An argument is interesting if it deserves a discount for unfoldings with a discount in that argument position. The idea is to avoid unfolding a function that is applied only to variables that have no unfolding (i.e. they are probably lambda bound): f x y z There is little point in inlining f here. Generally, *values* (like (C a b) and (\x.e)) deserve discounts. But we must look through lets, eg (let x = e in C a b), because the let will float, exposing the value, if we inline. That makes it different to exprIsHNF. Before 2009 we said it was interesting if the argument had *any* structure at all; i.e. (hasSomeUnfolding v). But does too much inlining; see Trac #3016. But we don't regard (f x y) as interesting, unless f is unsaturated. If it's saturated and f hasn't inlined, then it's probably not going to now! Note [Conlike is interesting] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider f d = ...((*) d x y)... ... f (df d')... where df is con-like. Then we'd really like to inline 'f' so that the rule for (*) (df d) can fire. To do this a) we give a discount for being an argument of a class-op (eg (*) d) b) we say that a con-like argument (eg (df d)) is interesting -} interestingArg :: SimplEnv -> CoreExpr -> ArgSummary -- See Note [Interesting arguments] interestingArg env e = go env 0 e where -- n is # value args to which the expression is applied go env n (Var v) | SimplEnv { seIdSubst = ids, seInScope = in_scope } <- env = case lookupVarEnv ids v of Nothing -> go_var n (refineFromInScope in_scope v) Just (DoneId v') -> go_var n (refineFromInScope in_scope v') Just (DoneEx e) -> go (zapSubstEnv env) n e Just (ContEx tvs cvs ids e) -> go (setSubstEnv env tvs cvs ids) n e go _ _ (Lit {}) = ValueArg go _ _ (Type _) = TrivArg go _ _ (Coercion _) = TrivArg go env n (App fn (Type _)) = go env n fn go env n (App fn (Coercion _)) = go env n fn go env n (App fn _) = go env (n+1) fn go env n (Tick _ a) = go env n a go env n (Cast e _) = go env n e go env n (Lam v e) | isTyVar v = go env n e | n>0 = go env (n-1) e | otherwise = ValueArg go env n (Let _ e) = case go env n e of { ValueArg -> ValueArg; _ -> NonTrivArg } go _ _ (Case {}) = NonTrivArg go_var n v | isConLikeId v = ValueArg -- Experimenting with 'conlike' rather that -- data constructors here | idArity v > n = ValueArg -- Catches (eg) primops with arity but no unfolding | n > 0 = NonTrivArg -- Saturated or unknown call | conlike_unfolding = ValueArg -- n==0; look for an interesting unfolding -- See Note [Conlike is interesting] | otherwise = TrivArg -- n==0, no useful unfolding where conlike_unfolding = isConLikeUnfolding (idUnfolding v) {- ************************************************************************ * * SimplifierMode * * ************************************************************************ The SimplifierMode controls several switches; see its definition in CoreMonad sm_rules :: Bool -- Whether RULES are enabled sm_inline :: Bool -- Whether inlining is enabled sm_case_case :: Bool -- Whether case-of-case is enabled sm_eta_expand :: Bool -- Whether eta-expansion is enabled -} simplEnvForGHCi :: DynFlags -> SimplEnv simplEnvForGHCi dflags = mkSimplEnv $ SimplMode { sm_names = ["GHCi"] , sm_phase = InitialPhase , sm_rules = rules_on , sm_inline = False , sm_eta_expand = eta_expand_on , sm_case_case = True } where rules_on = gopt Opt_EnableRewriteRules dflags eta_expand_on = gopt Opt_DoLambdaEtaExpansion dflags -- Do not do any inlining, in case we expose some unboxed -- tuple stuff that confuses the bytecode interpreter updModeForStableUnfoldings :: Activation -> SimplifierMode -> SimplifierMode -- See Note [Simplifying inside stable unfoldings] updModeForStableUnfoldings inline_rule_act current_mode = current_mode { sm_phase = phaseFromActivation inline_rule_act , sm_inline = True , sm_eta_expand = False } -- For sm_rules, just inherit; sm_rules might be "off" -- because of -fno-enable-rewrite-rules where phaseFromActivation (ActiveAfter n) = Phase n phaseFromActivation _ = InitialPhase {- Note [Inlining in gentle mode] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Something is inlined if (i) the sm_inline flag is on, AND (ii) the thing has an INLINE pragma, AND (iii) the thing is inlinable in the earliest phase. Example of why (iii) is important: {-# INLINE [~1] g #-} g = ... {-# INLINE f #-} f x = g (g x) If we were to inline g into f's inlining, then an importing module would never be able to do f e --> g (g e) ---> RULE fires because the stable unfolding for f has had g inlined into it. On the other hand, it is bad not to do ANY inlining into an stable unfolding, because then recursive knots in instance declarations don't get unravelled. However, *sometimes* SimplGently must do no call-site inlining at all (hence sm_inline = False). Before full laziness we must be careful not to inline wrappers, because doing so inhibits floating e.g. ...(case f x of ...)... ==> ...(case (case x of I# x# -> fw x#) of ...)... ==> ...(case x of I# x# -> case fw x# of ...)... and now the redex (f x) isn't floatable any more. The no-inlining thing is also important for Template Haskell. You might be compiling in one-shot mode with -O2; but when TH compiles a splice before running it, we don't want to use -O2. Indeed, we don't want to inline anything, because the byte-code interpreter might get confused about unboxed tuples and suchlike. Note [Simplifying inside stable unfoldings] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We must take care with simplification inside stable unfoldings (which come from INLINE pragmas). First, consider the following example let f = \pq -> BIG in let g = \y -> f y y {-# INLINE g #-} in ...g...g...g...g...g... Now, if that's the ONLY occurrence of f, it might be inlined inside g, and thence copied multiple times when g is inlined. HENCE we treat any occurrence in a stable unfolding as a multiple occurrence, not a single one; see OccurAnal.addRuleUsage. Second, we do want *do* to some modest rules/inlining stuff in stable unfoldings, partly to eliminate senseless crap, and partly to break the recursive knots generated by instance declarations. However, suppose we have {-# INLINE <act> f #-} f = <rhs> meaning "inline f in phases p where activation <act>(p) holds". Then what inlinings/rules can we apply to the copy of <rhs> captured in f's stable unfolding? Our model is that literally <rhs> is substituted for f when it is inlined. So our conservative plan (implemented by updModeForStableUnfoldings) is this: ------------------------------------------------------------- When simplifying the RHS of an stable unfolding, set the phase to the phase in which the stable unfolding first becomes active ------------------------------------------------------------- That ensures that a) Rules/inlinings that *cease* being active before p will not apply to the stable unfolding, consistent with it being inlined in its *original* form in phase p. b) Rules/inlinings that only become active *after* p will not apply to the stable unfolding, again to be consistent with inlining the *original* rhs in phase p. For example, {-# INLINE f #-} f x = ...g... {-# NOINLINE [1] g #-} g y = ... {-# RULE h g = ... #-} Here we must not inline g into f's RHS, even when we get to phase 0, because when f is later inlined into some other module we want the rule for h to fire. Similarly, consider {-# INLINE f #-} f x = ...g... g y = ... and suppose that there are auto-generated specialisations and a strictness wrapper for g. The specialisations get activation AlwaysActive, and the strictness wrapper get activation (ActiveAfter 0). So the strictness wrepper fails the test and won't be inlined into f's stable unfolding. That means f can inline, expose the specialised call to g, so the specialisation rules can fire. A note about wrappers ~~~~~~~~~~~~~~~~~~~~~ It's also important not to inline a worker back into a wrapper. A wrapper looks like wraper = inline_me (\x -> ...worker... ) Normally, the inline_me prevents the worker getting inlined into the wrapper (initially, the worker's only call site!). But, if the wrapper is sure to be called, the strictness analyser will mark it 'demanded', so when the RHS is simplified, it'll get an ArgOf continuation. -} activeUnfolding :: SimplEnv -> Id -> Bool activeUnfolding env | not (sm_inline mode) = active_unfolding_minimal | otherwise = case sm_phase mode of InitialPhase -> active_unfolding_gentle Phase n -> active_unfolding n where mode = getMode env getUnfoldingInRuleMatch :: SimplEnv -> InScopeEnv -- When matching in RULE, we want to "look through" an unfolding -- (to see a constructor) if *rules* are on, even if *inlinings* -- are not. A notable example is DFuns, which really we want to -- match in rules like (op dfun) in gentle mode. Another example -- is 'otherwise' which we want exprIsConApp_maybe to be able to -- see very early on getUnfoldingInRuleMatch env = (in_scope, id_unf) where in_scope = seInScope env mode = getMode env id_unf id | unf_is_active id = idUnfolding id | otherwise = NoUnfolding unf_is_active id | not (sm_rules mode) = active_unfolding_minimal id | otherwise = isActive (sm_phase mode) (idInlineActivation id) active_unfolding_minimal :: Id -> Bool -- Compuslory unfoldings only -- Ignore SimplGently, because we want to inline regardless; -- the Id has no top-level binding at all -- -- NB: we used to have a second exception, for data con wrappers. -- On the grounds that we use gentle mode for rule LHSs, and -- they match better when data con wrappers are inlined. -- But that only really applies to the trivial wrappers (like (:)), -- and they are now constructed as Compulsory unfoldings (in MkId) -- so they'll happen anyway. active_unfolding_minimal id = isCompulsoryUnfolding (realIdUnfolding id) active_unfolding :: PhaseNum -> Id -> Bool active_unfolding n id = isActiveIn n (idInlineActivation id) active_unfolding_gentle :: Id -> Bool -- Anything that is early-active -- See Note [Gentle mode] active_unfolding_gentle id = isInlinePragma prag && isEarlyActive (inlinePragmaActivation prag) -- NB: wrappers are not early-active where prag = idInlinePragma id ---------------------- activeRule :: SimplEnv -> Activation -> Bool -- Nothing => No rules at all activeRule env | not (sm_rules mode) = \_ -> False -- Rewriting is off | otherwise = isActive (sm_phase mode) where mode = getMode env {- ************************************************************************ * * preInlineUnconditionally * * ************************************************************************ preInlineUnconditionally ~~~~~~~~~~~~~~~~~~~~~~~~ @preInlineUnconditionally@ examines a bndr to see if it is used just once in a completely safe way, so that it is safe to discard the binding inline its RHS at the (unique) usage site, REGARDLESS of how big the RHS might be. If this is the case we don't simplify the RHS first, but just inline it un-simplified. This is much better than first simplifying a perhaps-huge RHS and then inlining and re-simplifying it. Indeed, it can be at least quadratically better. Consider x1 = e1 x2 = e2[x1] x3 = e3[x2] ...etc... xN = eN[xN-1] We may end up simplifying e1 N times, e2 N-1 times, e3 N-3 times etc. This can happen with cascades of functions too: f1 = \x1.e1 f2 = \xs.e2[f1] f3 = \xs.e3[f3] ...etc... THE MAIN INVARIANT is this: ---- preInlineUnconditionally invariant ----- IF preInlineUnconditionally chooses to inline x = <rhs> THEN doing the inlining should not change the occurrence info for the free vars of <rhs> ---------------------------------------------- For example, it's tempting to look at trivial binding like x = y and inline it unconditionally. But suppose x is used many times, but this is the unique occurrence of y. Then inlining x would change y's occurrence info, which breaks the invariant. It matters: y might have a BIG rhs, which will now be dup'd at every occurrenc of x. Even RHSs labelled InlineMe aren't caught here, because there might be no benefit from inlining at the call site. [Sept 01] Don't unconditionally inline a top-level thing, because that can simply make a static thing into something built dynamically. E.g. x = (a,b) main = \s -> h x [Remember that we treat \s as a one-shot lambda.] No point in inlining x unless there is something interesting about the call site. But watch out: if you aren't careful, some useful foldr/build fusion can be lost (most notably in spectral/hartel/parstof) because the foldr didn't see the build. Doing the dynamic allocation isn't a big deal, in fact, but losing the fusion can be. But the right thing here seems to be to do a callSiteInline based on the fact that there is something interesting about the call site (it's strict). Hmm. That seems a bit fragile. Conclusion: inline top level things gaily until Phase 0 (the last phase), at which point don't. Note [pre/postInlineUnconditionally in gentle mode] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Even in gentle mode we want to do preInlineUnconditionally. The reason is that too little clean-up happens if you don't inline use-once things. Also a bit of inlining is *good* for full laziness; it can expose constant sub-expressions. Example in spectral/mandel/Mandel.hs, where the mandelset function gets a useful let-float if you inline windowToViewport However, as usual for Gentle mode, do not inline things that are inactive in the intial stages. See Note [Gentle mode]. Note [Stable unfoldings and preInlineUnconditionally] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Surprisingly, do not pre-inline-unconditionally Ids with INLINE pragmas! Example {-# INLINE f #-} f :: Eq a => a -> a f x = ... fInt :: Int -> Int fInt = f Int dEqInt ...fInt...fInt...fInt... Here f occurs just once, in the RHS of f1. But if we inline it there we'll lose the opportunity to inline at each of fInt's call sites. The INLINE pragma will only inline when the application is saturated for exactly this reason; and we don't want PreInlineUnconditionally to second-guess it. A live example is Trac #3736. c.f. Note [Stable unfoldings and postInlineUnconditionally] Note [Top-level botomming Ids] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Don't inline top-level Ids that are bottoming, even if they are used just once, because FloatOut has gone to some trouble to extract them out. Inlining them won't make the program run faster! Note [Do not inline CoVars unconditionally] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Coercion variables appear inside coercions, and the RHS of a let-binding is a term (not a coercion) so we can't necessarily inline the latter in the former. -} preInlineUnconditionally :: DynFlags -> SimplEnv -> TopLevelFlag -> InId -> InExpr -> Bool -- Precondition: rhs satisfies the let/app invariant -- See Note [CoreSyn let/app invariant] in CoreSyn -- Reason: we don't want to inline single uses, or discard dead bindings, -- for unlifted, side-effect-full bindings preInlineUnconditionally dflags env top_lvl bndr rhs | not active = False | isStableUnfolding (idUnfolding bndr) = False -- Note [Stable unfoldings and preInlineUnconditionally] | isTopLevel top_lvl && isBottomingId bndr = False -- Note [Top-level bottoming Ids] | not (gopt Opt_SimplPreInlining dflags) = False | isCoVar bndr = False -- Note [Do not inline CoVars unconditionally] | otherwise = case idOccInfo bndr of IAmDead -> True -- Happens in ((\x.1) v) OneOcc in_lam True int_cxt -> try_once in_lam int_cxt _ -> False where mode = getMode env active = isActive (sm_phase mode) act -- See Note [pre/postInlineUnconditionally in gentle mode] act = idInlineActivation bndr try_once in_lam int_cxt -- There's one textual occurrence | not in_lam = isNotTopLevel top_lvl || early_phase | otherwise = int_cxt && canInlineInLam rhs -- Be very careful before inlining inside a lambda, because (a) we must not -- invalidate occurrence information, and (b) we want to avoid pushing a -- single allocation (here) into multiple allocations (inside lambda). -- Inlining a *function* with a single *saturated* call would be ok, mind you. -- || (if is_cheap && not (canInlineInLam rhs) then pprTrace "preinline" (ppr bndr <+> ppr rhs) ok else ok) -- where -- is_cheap = exprIsCheap rhs -- ok = is_cheap && int_cxt -- int_cxt The context isn't totally boring -- E.g. let f = \ab.BIG in \y. map f xs -- Don't want to substitute for f, because then we allocate -- its closure every time the \y is called -- But: let f = \ab.BIG in \y. map (f y) xs -- Now we do want to substitute for f, even though it's not -- saturated, because we're going to allocate a closure for -- (f y) every time round the loop anyhow. -- canInlineInLam => free vars of rhs are (Once in_lam) or Many, -- so substituting rhs inside a lambda doesn't change the occ info. -- Sadly, not quite the same as exprIsHNF. canInlineInLam (Lit _) = True canInlineInLam (Lam b e) = isRuntimeVar b || canInlineInLam e canInlineInLam (Tick t e) = not (tickishIsCode t) && canInlineInLam e canInlineInLam _ = False -- not ticks. Counting ticks cannot be duplicated, and non-counting -- ticks around a Lam will disappear anyway. early_phase = case sm_phase mode of Phase 0 -> False _ -> True -- If we don't have this early_phase test, consider -- x = length [1,2,3] -- The full laziness pass carefully floats all the cons cells to -- top level, and preInlineUnconditionally floats them all back in. -- Result is (a) static allocation replaced by dynamic allocation -- (b) many simplifier iterations because this tickles -- a related problem; only one inlining per pass -- -- On the other hand, I have seen cases where top-level fusion is -- lost if we don't inline top level thing (e.g. string constants) -- Hence the test for phase zero (which is the phase for all the final -- simplifications). Until phase zero we take no special notice of -- top level things, but then we become more leery about inlining -- them. {- ************************************************************************ * * postInlineUnconditionally * * ************************************************************************ postInlineUnconditionally ~~~~~~~~~~~~~~~~~~~~~~~~~ @postInlineUnconditionally@ decides whether to unconditionally inline a thing based on the form of its RHS; in particular if it has a trivial RHS. If so, we can inline and discard the binding altogether. NB: a loop breaker has must_keep_binding = True and non-loop-breakers only have *forward* references. Hence, it's safe to discard the binding NOTE: This isn't our last opportunity to inline. We're at the binding site right now, and we'll get another opportunity when we get to the ocurrence(s) Note that we do this unconditional inlining only for trival RHSs. Don't inline even WHNFs inside lambdas; doing so may simply increase allocation when the function is called. This isn't the last chance; see NOTE above. NB: Even inline pragmas (e.g. IMustBeINLINEd) are ignored here Why? Because we don't even want to inline them into the RHS of constructor arguments. See NOTE above NB: At one time even NOINLINE was ignored here: if the rhs is trivial it's best to inline it anyway. We often get a=E; b=a from desugaring, with both a and b marked NOINLINE. But that seems incompatible with our new view that inlining is like a RULE, so I'm sticking to the 'active' story for now. -} postInlineUnconditionally :: DynFlags -> SimplEnv -> TopLevelFlag -> OutId -- The binder (an InId would be fine too) -- (*not* a CoVar) -> OccInfo -- From the InId -> OutExpr -> Unfolding -> Bool -- Precondition: rhs satisfies the let/app invariant -- See Note [CoreSyn let/app invariant] in CoreSyn -- Reason: we don't want to inline single uses, or discard dead bindings, -- for unlifted, side-effect-full bindings postInlineUnconditionally dflags env top_lvl bndr occ_info rhs unfolding | not active = False | isWeakLoopBreaker occ_info = False -- If it's a loop-breaker of any kind, don't inline -- because it might be referred to "earlier" | isExportedId bndr = False | isStableUnfolding unfolding = False -- Note [Stable unfoldings and postInlineUnconditionally] | isTopLevel top_lvl = False -- Note [Top level and postInlineUnconditionally] | exprIsTrivial rhs = True | otherwise = case occ_info of -- The point of examining occ_info here is that for *non-values* -- that occur outside a lambda, the call-site inliner won't have -- a chance (because it doesn't know that the thing -- only occurs once). The pre-inliner won't have gotten -- it either, if the thing occurs in more than one branch -- So the main target is things like -- let x = f y in -- case v of -- True -> case x of ... -- False -> case x of ... -- This is very important in practice; e.g. wheel-seive1 doubles -- in allocation if you miss this out OneOcc in_lam _one_br int_cxt -- OneOcc => no code-duplication issue -> smallEnoughToInline dflags unfolding -- Small enough to dup -- ToDo: consider discount on smallEnoughToInline if int_cxt is true -- -- NB: Do NOT inline arbitrarily big things, even if one_br is True -- Reason: doing so risks exponential behaviour. We simplify a big -- expression, inline it, and simplify it again. But if the -- very same thing happens in the big expression, we get -- exponential cost! -- PRINCIPLE: when we've already simplified an expression once, -- make sure that we only inline it if it's reasonably small. && (not in_lam || -- Outside a lambda, we want to be reasonably aggressive -- about inlining into multiple branches of case -- e.g. let x = <non-value> -- in case y of { C1 -> ..x..; C2 -> ..x..; C3 -> ... } -- Inlining can be a big win if C3 is the hot-spot, even if -- the uses in C1, C2 are not 'interesting' -- An example that gets worse if you add int_cxt here is 'clausify' (isCheapUnfolding unfolding && int_cxt)) -- isCheap => acceptable work duplication; in_lam may be true -- int_cxt to prevent us inlining inside a lambda without some -- good reason. See the notes on int_cxt in preInlineUnconditionally IAmDead -> True -- This happens; for example, the case_bndr during case of -- known constructor: case (a,b) of x { (p,q) -> ... } -- Here x isn't mentioned in the RHS, so we don't want to -- create the (dead) let-binding let x = (a,b) in ... _ -> False -- Here's an example that we don't handle well: -- let f = if b then Left (\x.BIG) else Right (\y.BIG) -- in \y. ....case f of {...} .... -- Here f is used just once, and duplicating the case work is fine (exprIsCheap). -- But -- - We can't preInlineUnconditionally because that woud invalidate -- the occ info for b. -- - We can't postInlineUnconditionally because the RHS is big, and -- that risks exponential behaviour -- - We can't call-site inline, because the rhs is big -- Alas! where active = isActive (sm_phase (getMode env)) (idInlineActivation bndr) -- See Note [pre/postInlineUnconditionally in gentle mode] {- Note [Top level and postInlineUnconditionally] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We don't do postInlineUnconditionally for top-level things (even for ones that are trivial): * Doing so will inline top-level error expressions that have been carefully floated out by FloatOut. More generally, it might replace static allocation with dynamic. * Even for trivial expressions there's a problem. Consider {-# RULE "foo" forall (xs::[T]). reverse xs = ruggle xs #-} blah xs = reverse xs ruggle = sort In one simplifier pass we might fire the rule, getting blah xs = ruggle xs but in *that* simplifier pass we must not do postInlineUnconditionally on 'ruggle' because then we'll have an unbound occurrence of 'ruggle' If the rhs is trivial it'll be inlined by callSiteInline, and then the binding will be dead and discarded by the next use of OccurAnal * There is less point, because the main goal is to get rid of local bindings used in multiple case branches. * The inliner should inline trivial things at call sites anyway. Note [Stable unfoldings and postInlineUnconditionally] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Do not do postInlineUnconditionally if the Id has an stable unfolding, otherwise we lose the unfolding. Example -- f has stable unfolding with rhs (e |> co) -- where 'e' is big f = e |> co Then there's a danger we'll optimise to f' = e f = f' |> co and now postInlineUnconditionally, losing the stable unfolding on f. Now f' won't inline because 'e' is too big. c.f. Note [Stable unfoldings and preInlineUnconditionally] ************************************************************************ * * Rebuilding a lambda * * ************************************************************************ -} mkLam :: [OutBndr] -> OutExpr -> SimplCont -> SimplM OutExpr -- mkLam tries three things -- a) eta reduction, if that gives a trivial expression -- b) eta expansion [only if there are some value lambdas] mkLam [] body _cont = return body mkLam bndrs body cont = do { dflags <- getDynFlags ; mkLam' dflags bndrs body } where mkLam' :: DynFlags -> [OutBndr] -> OutExpr -> SimplM OutExpr mkLam' dflags bndrs (Cast body co) | not (any bad bndrs) -- Note [Casts and lambdas] = do { lam <- mkLam' dflags bndrs body ; return (mkCast lam (mkPiCos Representational bndrs co)) } where co_vars = tyCoVarsOfCo co bad bndr = isCoVar bndr && bndr `elemVarSet` co_vars mkLam' dflags bndrs body@(Lam {}) = mkLam' dflags (bndrs ++ bndrs1) body1 where (bndrs1, body1) = collectBinders body mkLam' dflags bndrs (Tick t expr) | tickishFloatable t = mkTick t <$> mkLam' dflags bndrs expr mkLam' dflags bndrs body | gopt Opt_DoEtaReduction dflags , Just etad_lam <- tryEtaReduce bndrs body = do { tick (EtaReduction (head bndrs)) ; return etad_lam } | not (contIsRhs cont) -- See Note [Eta-expanding lambdas] , gopt Opt_DoLambdaEtaExpansion dflags , any isRuntimeVar bndrs , let body_arity = exprEtaExpandArity dflags body , body_arity > 0 = do { tick (EtaExpansion (head bndrs)) ; return (mkLams bndrs (etaExpand body_arity body)) } | otherwise = return (mkLams bndrs body) {- Note [Eta expanding lambdas] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ In general we *do* want to eta-expand lambdas. Consider f (\x -> case x of (a,b) -> \s -> blah) where 's' is a state token, and hence can be eta expanded. This showed up in the code for GHc.IO.Handle.Text.hPutChar, a rather important function! The eta-expansion will never happen unless we do it now. (Well, it's possible that CorePrep will do it, but CorePrep only has a half-baked eta-expander that can't deal with casts. So it's much better to do it here.) However, when the lambda is let-bound, as the RHS of a let, we have a better eta-expander (in the form of tryEtaExpandRhs), so we don't bother to try expansion in mkLam in that case; hence the contIsRhs guard. Note [Casts and lambdas] ~~~~~~~~~~~~~~~~~~~~~~~~ Consider (\x. (\y. e) `cast` g1) `cast` g2 There is a danger here that the two lambdas look separated, and the full laziness pass might float an expression to between the two. So this equation in mkLam' floats the g1 out, thus: (\x. e `cast` g1) --> (\x.e) `cast` (tx -> g1) where x:tx. In general, this floats casts outside lambdas, where (I hope) they might meet and cancel with some other cast: \x. e `cast` co ===> (\x. e) `cast` (tx -> co) /\a. e `cast` co ===> (/\a. e) `cast` (/\a. co) /\g. e `cast` co ===> (/\g. e) `cast` (/\g. co) (if not (g `in` co)) Notice that it works regardless of 'e'. Originally it worked only if 'e' was itself a lambda, but in some cases that resulted in fruitless iteration in the simplifier. A good example was when compiling Text.ParserCombinators.ReadPrec, where we had a definition like (\x. Get `cast` g) where Get is a constructor with nonzero arity. Then mkLam eta-expanded the Get, and the next iteration eta-reduced it, and then eta-expanded it again. Note also the side condition for the case of coercion binders. It does not make sense to transform /\g. e `cast` g ==> (/\g.e) `cast` (/\g.g) because the latter is not well-kinded. ************************************************************************ * * Eta expansion * * ************************************************************************ -} tryEtaExpandRhs :: SimplEnv -> OutId -> OutExpr -> SimplM (Arity, OutExpr) -- See Note [Eta-expanding at let bindings] tryEtaExpandRhs env bndr rhs = do { dflags <- getDynFlags ; (new_arity, new_rhs) <- try_expand dflags ; WARN( new_arity < old_id_arity, (ptext (sLit "Arity decrease:") <+> (ppr bndr <+> ppr old_id_arity <+> ppr old_arity <+> ppr new_arity) $$ ppr new_rhs) ) -- Note [Arity decrease] in Simplify return (new_arity, new_rhs) } where try_expand dflags | exprIsTrivial rhs = return (exprArity rhs, rhs) | sm_eta_expand (getMode env) -- Provided eta-expansion is on , let new_arity1 = findRhsArity dflags bndr rhs old_arity new_arity2 = idCallArity bndr new_arity = max new_arity1 new_arity2 , new_arity > old_arity -- And the current manifest arity isn't enough = do { tick (EtaExpansion bndr) ; return (new_arity, etaExpand new_arity rhs) } | otherwise = return (old_arity, rhs) old_arity = exprArity rhs -- See Note [Do not expand eta-expand PAPs] old_id_arity = idArity bndr {- Note [Eta-expanding at let bindings] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We now eta expand at let-bindings, which is where the payoff comes. The most significant thing is that we can do a simple arity analysis (in CoreArity.findRhsArity), which we can't do for free-floating lambdas One useful consequence of not eta-expanding lambdas is this example: genMap :: C a => ... {-# INLINE genMap #-} genMap f xs = ... myMap :: D a => ... {-# INLINE myMap #-} myMap = genMap Notice that 'genMap' should only inline if applied to two arguments. In the stable unfolding for myMap we'll have the unfolding (\d -> genMap Int (..d..)) We do not want to eta-expand to (\d f xs -> genMap Int (..d..) f xs) because then 'genMap' will inline, and it really shouldn't: at least as far as the programmer is concerned, it's not applied to two arguments! Note [Do not eta-expand PAPs] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ We used to have old_arity = manifestArity rhs, which meant that we would eta-expand even PAPs. But this gives no particular advantage, and can lead to a massive blow-up in code size, exhibited by Trac #9020. Suppose we have a PAP foo :: IO () foo = returnIO () Then we can eta-expand do foo = (\eta. (returnIO () |> sym g) eta) |> g where g :: IO () ~ State# RealWorld -> (# State# RealWorld, () #) But there is really no point in doing this, and it generates masses of coercions and whatnot that eventually disappear again. For T9020, GHC allocated 6.6G beore, and 0.8G afterwards; and residency dropped from 1.8G to 45M. But note that this won't eta-expand, say f = \g -> map g Does it matter not eta-expanding such functions? I'm not sure. Perhaps strictness analysis will have less to bite on? ************************************************************************ * * \subsection{Floating lets out of big lambdas} * * ************************************************************************ Note [Floating and type abstraction] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Consider this: x = /\a. C e1 e2 We'd like to float this to y1 = /\a. e1 y2 = /\a. e2 x = /\a. C (y1 a) (y2 a) for the usual reasons: we want to inline x rather vigorously. You may think that this kind of thing is rare. But in some programs it is common. For example, if you do closure conversion you might get: data a :-> b = forall e. (e -> a -> b) :$ e f_cc :: forall a. a :-> a f_cc = /\a. (\e. id a) :$ () Now we really want to inline that f_cc thing so that the construction of the closure goes away. So I have elaborated simplLazyBind to understand right-hand sides that look like /\ a1..an. body and treat them specially. The real work is done in SimplUtils.abstractFloats, but there is quite a bit of plumbing in simplLazyBind as well. The same transformation is good when there are lets in the body: /\abc -> let(rec) x = e in b ==> let(rec) x' = /\abc -> let x = x' a b c in e in /\abc -> let x = x' a b c in b This is good because it can turn things like: let f = /\a -> letrec g = ... g ... in g into letrec g' = /\a -> ... g' a ... in let f = /\ a -> g' a which is better. In effect, it means that big lambdas don't impede let-floating. This optimisation is CRUCIAL in eliminating the junk introduced by desugaring mutually recursive definitions. Don't eliminate it lightly! [May 1999] If we do this transformation *regardless* then we can end up with some pretty silly stuff. For example, let st = /\ s -> let { x1=r1 ; x2=r2 } in ... in .. becomes let y1 = /\s -> r1 y2 = /\s -> r2 st = /\s -> ...[y1 s/x1, y2 s/x2] in .. Unless the "..." is a WHNF there is really no point in doing this. Indeed it can make things worse. Suppose x1 is used strictly, and is of the form x1* = case f y of { (a,b) -> e } If we abstract this wrt the tyvar we then can't do the case inline as we would normally do. That's why the whole transformation is part of the same process that floats let-bindings and constructor arguments out of RHSs. In particular, it is guarded by the doFloatFromRhs call in simplLazyBind. -} abstractFloats :: [OutTyVar] -> SimplEnv -> OutExpr -> SimplM ([OutBind], OutExpr) abstractFloats main_tvs body_env body = ASSERT( notNull body_floats ) do { (subst, float_binds) <- mapAccumLM abstract empty_subst body_floats ; return (float_binds, CoreSubst.substExpr (text "abstract_floats1") subst body) } where main_tv_set = mkVarSet main_tvs body_floats = getFloatBinds body_env empty_subst = CoreSubst.mkEmptySubst (seInScope body_env) abstract :: CoreSubst.Subst -> OutBind -> SimplM (CoreSubst.Subst, OutBind) abstract subst (NonRec id rhs) = do { (poly_id, poly_app) <- mk_poly tvs_here id ; let poly_rhs = mkLams tvs_here rhs' subst' = CoreSubst.extendIdSubst subst id poly_app ; return (subst', (NonRec poly_id poly_rhs)) } where rhs' = CoreSubst.substExpr (text "abstract_floats2") subst rhs tvs_here = varSetElemsKvsFirst (main_tv_set `intersectVarSet` exprSomeFreeVars isTyVar rhs') -- Abstract only over the type variables free in the rhs -- wrt which the new binding is abstracted. But the naive -- approach of abstract wrt the tyvars free in the Id's type -- fails. Consider: -- /\ a b -> let t :: (a,b) = (e1, e2) -- x :: a = fst t -- in ... -- Here, b isn't free in x's type, but we must nevertheless -- abstract wrt b as well, because t's type mentions b. -- Since t is floated too, we'd end up with the bogus: -- poly_t = /\ a b -> (e1, e2) -- poly_x = /\ a -> fst (poly_t a *b*) -- So for now we adopt the even more naive approach of -- abstracting wrt *all* the tyvars. We'll see if that -- gives rise to problems. SLPJ June 98 abstract subst (Rec prs) = do { (poly_ids, poly_apps) <- mapAndUnzipM (mk_poly tvs_here) ids ; let subst' = CoreSubst.extendSubstList subst (ids `zip` poly_apps) poly_rhss = [mkLams tvs_here (CoreSubst.substExpr (text "abstract_floats3") subst' rhs) | rhs <- rhss] ; return (subst', Rec (poly_ids `zip` poly_rhss)) } where (ids,rhss) = unzip prs -- For a recursive group, it's a bit of a pain to work out the minimal -- set of tyvars over which to abstract: -- /\ a b c. let x = ...a... in -- letrec { p = ...x...q... -- q = .....p...b... } in -- ... -- Since 'x' is abstracted over 'a', the {p,q} group must be abstracted -- over 'a' (because x is replaced by (poly_x a)) as well as 'b'. -- Since it's a pain, we just use the whole set, which is always safe -- -- If you ever want to be more selective, remember this bizarre case too: -- x::a = x -- Here, we must abstract 'x' over 'a'. tvs_here = sortQuantVars main_tvs mk_poly tvs_here var = do { uniq <- getUniqueM ; let poly_name = setNameUnique (idName var) uniq -- Keep same name poly_ty = mkForAllTys tvs_here (idType var) -- But new type of course poly_id = transferPolyIdInfo var tvs_here $ -- Note [transferPolyIdInfo] in Id.lhs mkLocalId poly_name poly_ty ; return (poly_id, mkTyApps (Var poly_id) (mkTyVarTys tvs_here)) } -- In the olden days, it was crucial to copy the occInfo of the original var, -- because we were looking at occurrence-analysed but as yet unsimplified code! -- In particular, we mustn't lose the loop breakers. BUT NOW we are looking -- at already simplified code, so it doesn't matter -- -- It's even right to retain single-occurrence or dead-var info: -- Suppose we started with /\a -> let x = E in B -- where x occurs once in B. Then we transform to: -- let x' = /\a -> E in /\a -> let x* = x' a in B -- where x* has an INLINE prag on it. Now, once x* is inlined, -- the occurrences of x' will be just the occurrences originally -- pinned on x. {- Note [Abstract over coercions] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If a coercion variable (g :: a ~ Int) is free in the RHS, then so is the type variable a. Rather than sort this mess out, we simply bale out and abstract wrt all the type variables if any of them are coercion variables. Historical note: if you use let-bindings instead of a substitution, beware of this: -- Suppose we start with: -- -- x = /\ a -> let g = G in E -- -- Then we'll float to get -- -- x = let poly_g = /\ a -> G -- in /\ a -> let g = poly_g a in E -- -- But now the occurrence analyser will see just one occurrence -- of poly_g, not inside a lambda, so the simplifier will -- PreInlineUnconditionally poly_g back into g! Badk to square 1! -- (I used to think that the "don't inline lone occurrences" stuff -- would stop this happening, but since it's the *only* occurrence, -- PreInlineUnconditionally kicks in first!) -- -- Solution: put an INLINE note on g's RHS, so that poly_g seems -- to appear many times. (NB: mkInlineMe eliminates -- such notes on trivial RHSs, so do it manually.) ************************************************************************ * * prepareAlts * * ************************************************************************ prepareAlts tries these things: 1. Eliminate alternatives that cannot match, including the DEFAULT alternative. 2. If the DEFAULT alternative can match only one possible constructor, then make that constructor explicit. e.g. case e of x { DEFAULT -> rhs } ===> case e of x { (a,b) -> rhs } where the type is a single constructor type. This gives better code when rhs also scrutinises x or e. 3. Returns a list of the constructors that cannot holds in the DEFAULT alternative (if there is one) Here "cannot match" includes knowledge from GADTs It's a good idea to do this stuff before simplifying the alternatives, to avoid simplifying alternatives we know can't happen, and to come up with the list of constructors that are handled, to put into the IdInfo of the case binder, for use when simplifying the alternatives. Eliminating the default alternative in (1) isn't so obvious, but it can happen: data Colour = Red | Green | Blue f x = case x of Red -> .. Green -> .. DEFAULT -> h x h y = case y of Blue -> .. DEFAULT -> [ case y of ... ] If we inline h into f, the default case of the inlined h can't happen. If we don't notice this, we may end up filtering out *all* the cases of the inner case y, which give us nowhere to go! -} prepareAlts :: OutExpr -> OutId -> [InAlt] -> SimplM ([AltCon], [InAlt]) -- The returned alternatives can be empty, none are possible prepareAlts scrut case_bndr' alts -- Case binder is needed just for its type. Note that as an -- OutId, it has maximum information; this is important. -- Test simpl013 is an example = do { us <- getUniquesM ; let (imposs_deflt_cons', refined_deflt, alts') = filterAlts us (varType case_bndr') imposs_cons alts (combining_done, imposs_deflt_cons'', alts'') = combineIdenticalAlts imposs_deflt_cons' alts' ; when refined_deflt $ tick (FillInCaseDefault case_bndr') ; when combining_done $ tick (AltMerge case_bndr') ; return (imposs_deflt_cons'', alts'') } where imposs_cons = case scrut of Var v -> otherCons (idUnfolding v) _ -> [] {- Note [Combine identical alternatives] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If several alternatives are identical, merge them into a single DEFAULT alternative. I've occasionally seen this making a big difference: case e of =====> case e of C _ -> f x D v -> ....v.... D v -> ....v.... DEFAULT -> f x DEFAULT -> f x The point is that we merge common RHSs, at least for the DEFAULT case. [One could do something more elaborate but I've never seen it needed.] To avoid an expensive test, we just merge branches equal to the *first* alternative; this picks up the common cases a) all branches equal b) some branches equal to the DEFAULT (which occurs first) The case where Combine Identical Alternatives transformation showed up was like this (base/Foreign/C/Err/Error.lhs): x | p `is` 1 -> e1 | p `is` 2 -> e2 ...etc... where @is@ was something like p `is` n = p /= (-1) && p == n This gave rise to a horrible sequence of cases case p of (-1) -> $j p 1 -> e1 DEFAULT -> $j p and similarly in cascade for all the join points! NB: it's important that all this is done in [InAlt], *before* we work on the alternatives themselves, because Simpify.simplAlt may zap the occurrence info on the binders in the alternatives, which in turn defeats combineIdenticalAlts (see Trac #7360). Note [Care with impossible-constructors when combining alternatives] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Suppose we have (Trac #10538) data T = A | B | C ... case x::T of DEFAULT -> e1 A -> e2 B -> e1 When calling combineIdentialAlts, we'll have computed that the "impossible constructors" for the DEFAULT alt is {A,B}, since if x is A or B we'll take the other alternatives. But suppose we combine B into the DEFAULT, to get ... case x::T of DEFAULT -> e1 A -> e2 Then we must be careful to trim the impossible constructors to just {A}, else we risk compiling 'e1' wrong! -} combineIdenticalAlts :: [AltCon] -> [InAlt] -> (Bool, [AltCon], [InAlt]) -- See Note [Combine identical alternatives] -- See Note [Care with impossible-constructors when combining alternatives] -- True <=> we did some combining, result is a single DEFAULT alternative combineIdenticalAlts imposs_cons ((_con1,bndrs1,rhs1) : con_alts) | all isDeadBinder bndrs1 -- Remember the default , not (null eliminated_alts) -- alternative comes first = (True, imposs_cons', deflt_alt : filtered_alts) where (eliminated_alts, filtered_alts) = partition identical_to_alt1 con_alts deflt_alt = (DEFAULT, [], mkTicks (concat tickss) rhs1) imposs_cons' = imposs_cons `minusList` map fstOf3 eliminated_alts cheapEqTicked e1 e2 = cheapEqExpr' tickishFloatable e1 e2 identical_to_alt1 (_con,bndrs,rhs) = all isDeadBinder bndrs && rhs `cheapEqTicked` rhs1 tickss = map (stripTicksT tickishFloatable . thirdOf3) eliminated_alts combineIdenticalAlts imposs_cons alts = (False, imposs_cons, alts) {- ************************************************************************ * * mkCase * * ************************************************************************ mkCase tries these things 1. Merge Nested Cases case e of b { ==> case e of b { p1 -> rhs1 p1 -> rhs1 ... ... pm -> rhsm pm -> rhsm _ -> case b of b' { pn -> let b'=b in rhsn pn -> rhsn ... ... po -> let b'=b in rhso po -> rhso _ -> let b'=b in rhsd _ -> rhsd } which merges two cases in one case when -- the default alternative of the outer case scrutises the same variable as the outer case. This transformation is called Case Merging. It avoids that the same variable is scrutinised multiple times. 2. Eliminate Identity Case case e of ===> e True -> True; False -> False and similar friends. -} mkCase, mkCase1, mkCase2 :: DynFlags -> OutExpr -> OutId -> OutType -> [OutAlt] -- Alternatives in standard (increasing) order -> SimplM OutExpr -------------------------------------------------- -- 1. Merge Nested Cases -------------------------------------------------- mkCase dflags scrut outer_bndr alts_ty ((DEFAULT, _, deflt_rhs) : outer_alts) | gopt Opt_CaseMerge dflags , (ticks, Case (Var inner_scrut_var) inner_bndr _ inner_alts) <- stripTicksTop tickishFloatable deflt_rhs , inner_scrut_var == outer_bndr = do { tick (CaseMerge outer_bndr) ; let wrap_alt (con, args, rhs) = ASSERT( outer_bndr `notElem` args ) (con, args, wrap_rhs rhs) -- Simplifier's no-shadowing invariant should ensure -- that outer_bndr is not shadowed by the inner patterns wrap_rhs rhs = Let (NonRec inner_bndr (Var outer_bndr)) rhs -- The let is OK even for unboxed binders, wrapped_alts | isDeadBinder inner_bndr = inner_alts | otherwise = map wrap_alt inner_alts merged_alts = mergeAlts outer_alts wrapped_alts -- NB: mergeAlts gives priority to the left -- case x of -- A -> e1 -- DEFAULT -> case x of -- A -> e2 -- B -> e3 -- When we merge, we must ensure that e1 takes -- precedence over e2 as the value for A! ; fmap (mkTicks ticks) $ mkCase1 dflags scrut outer_bndr alts_ty merged_alts } -- Warning: don't call mkCase recursively! -- Firstly, there's no point, because inner alts have already had -- mkCase applied to them, so they won't have a case in their default -- Secondly, if you do, you get an infinite loop, because the bindCaseBndr -- in munge_rhs may put a case into the DEFAULT branch! mkCase dflags scrut bndr alts_ty alts = mkCase1 dflags scrut bndr alts_ty alts -------------------------------------------------- -- 2. Eliminate Identity Case -------------------------------------------------- mkCase1 _dflags scrut case_bndr _ alts@((_,_,rhs1) : _) -- Identity case | all identity_alt alts = do { tick (CaseIdentity case_bndr) ; return (mkTicks ticks $ re_cast scrut rhs1) } where ticks = concatMap (stripTicksT tickishFloatable . thirdOf3) (tail alts) identity_alt (con, args, rhs) = check_eq rhs con args check_eq (Cast rhs co) con args = not (any (`elemVarSet` tyCoVarsOfCo co) args) && check_eq rhs con args -- See Note [RHS casts] check_eq (Lit lit) (LitAlt lit') _ = lit == lit' check_eq (Var v) _ _ | v == case_bndr = True check_eq (Var v) (DataAlt con) [] = v == dataConWorkId con -- Optimisation only check_eq (Tick t e) alt args = tickishFloatable t && check_eq e alt args check_eq rhs (DataAlt con) args = cheapEqExpr' tickishFloatable rhs $ mkConApp con (arg_tys ++ varsToCoreExprs args) check_eq _ _ _ = False arg_tys = map Type (tyConAppArgs (idType case_bndr)) -- Note [RHS casts] -- ~~~~~~~~~~~~~~~~ -- We've seen this: -- case e of x { _ -> x `cast` c } -- And we definitely want to eliminate this case, to give -- e `cast` c -- So we throw away the cast from the RHS, and reconstruct -- it at the other end. All the RHS casts must be the same -- if (all identity_alt alts) holds. -- -- Don't worry about nested casts, because the simplifier combines them re_cast scrut (Cast rhs co) = Cast (re_cast scrut rhs) co re_cast scrut _ = scrut mkCase1 dflags scrut bndr alts_ty alts = mkCase2 dflags scrut bndr alts_ty alts -------------------------------------------------- -- Catch-all -------------------------------------------------- mkCase2 _dflags scrut bndr alts_ty alts = return (Case scrut bndr alts_ty alts) {- Note [Dead binders] ~~~~~~~~~~~~~~~~~~~~ Note that dead-ness is maintained by the simplifier, so that it is accurate after simplification as well as before. Note [Cascading case merge] ~~~~~~~~~~~~~~~~~~~~~~~~~~~ Case merging should cascade in one sweep, because it happens bottom-up case e of a { DEFAULT -> case a of b DEFAULT -> case b of c { DEFAULT -> e A -> ea B -> eb C -> ec ==> case e of a { DEFAULT -> case a of b DEFAULT -> let c = b in e A -> let c = b in ea B -> eb C -> ec ==> case e of a { DEFAULT -> let b = a in let c = b in e A -> let b = a in let c = b in ea B -> let b = a in eb C -> ec However here's a tricky case that we still don't catch, and I don't see how to catch it in one pass: case x of c1 { I# a1 -> case a1 of c2 -> 0 -> ... DEFAULT -> case x of c3 { I# a2 -> case a2 of ... After occurrence analysis (and its binder-swap) we get this case x of c1 { I# a1 -> let x = c1 in -- Binder-swap addition case a1 of c2 -> 0 -> ... DEFAULT -> case x of c3 { I# a2 -> case a2 of ... When we simplify the inner case x, we'll see that x=c1=I# a1. So we'll bind a2 to a1, and get case x of c1 { I# a1 -> case a1 of c2 -> 0 -> ... DEFAULT -> case a1 of ... This is corect, but we can't do a case merge in this sweep because c2 /= a1. Reason: the binding c1=I# a1 went inwards without getting changed to c1=I# c2. I don't think this is worth fixing, even if I knew how. It'll all come out in the next pass anyway. -}