X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=compiler%2FsimplCore%2FSimplUtils.lhs;h=693644f258096743b55c8022ad9da8508c4f0fae;hb=088962101d8ad4ba5455e295258df009c72315f2;hp=cfd6830d33809ebdcc60a4a969f52195d948907c;hpb=2763f56de2097a34176aa883dd4f0b3de1cb896c;p=ghc-hetmet.git diff --git a/compiler/simplCore/SimplUtils.lhs b/compiler/simplCore/SimplUtils.lhs index cfd6830..693644f 100644 --- a/compiler/simplCore/SimplUtils.lhs +++ b/compiler/simplCore/SimplUtils.lhs @@ -15,8 +15,9 @@ module SimplUtils ( SimplCont(..), DupFlag(..), LetRhsFlag(..), contIsDupable, contResultType, countValArgs, countArgs, pushContArgs, - mkBoringStop, mkRhsStop, contIsRhs, contIsRhsOrArg, - getContArgs, interestingCallContext, interestingArg, isStrictType + mkBoringStop, mkLazyArgStop, mkRhsStop, contIsRhs, contIsRhsOrArg, + getContArgs, interestingCallContext, interestingArgContext, + interestingArg, isStrictType ) where @@ -24,21 +25,21 @@ module SimplUtils ( import SimplEnv import DynFlags ( SimplifierSwitch(..), SimplifierMode(..), - DynFlag(..), dopt ) + DynFlags, DynFlag(..), dopt ) import StaticFlags ( opt_UF_UpdateInPlace, opt_SimplNoPreInlining, opt_RulesOff ) import CoreSyn import CoreFVs ( exprFreeVars ) -import CoreUtils ( cheapEqExpr, exprType, exprIsTrivial, exprIsCheap, +import CoreUtils ( cheapEqExpr, exprType, exprIsTrivial, etaExpand, exprEtaExpandArity, bindNonRec, mkCoerce2, findDefault, exprOkForSpeculation, exprIsHNF, mergeAlts ) import Literal ( mkStringLit ) import CoreUnfold ( smallEnoughToInline ) import MkId ( eRROR_ID ) -import Id ( idType, isDataConWorkId, idOccInfo, isDictId, +import Id ( Id, idType, isDataConWorkId, idOccInfo, isDictId, isDeadBinder, idNewDemandInfo, isExportedId, - idUnfolding, idNewStrictness, idInlinePragma, + idUnfolding, idNewStrictness, idInlinePragma, idHasRules ) import NewDemand ( isStrictDmd, isBotRes, splitStrictSig ) import SimplMonad @@ -63,21 +64,25 @@ import Outputable \begin{code} data SimplCont -- Strict contexts - = Stop OutType -- Type of the result + = Stop OutType -- Type of the result LetRhsFlag - Bool -- True <=> This is the RHS of a thunk whose type suggests - -- that update-in-place would be possible - -- (This makes the inliner a little keener.) + Bool -- True <=> There is something interesting about + -- the context, and hence the inliner + -- should be a bit keener (see interestingCallContext) + -- Two cases: + -- (a) This is the RHS of a thunk whose type suggests + -- that update-in-place would be possible + -- (b) This is an argument of a function that has RULES + -- Inlining the call might allow the rule to fire | CoerceIt OutType -- The To-type, simplified SimplCont - | InlinePlease -- This continuation makes a function very - SimplCont -- keen to inline itelf - | ApplyTo DupFlag - InExpr SimplEnv -- The argument, as yet unsimplified, - SimplCont -- and its environment + CoreExpr -- The argument + (Maybe SimplEnv) -- (Just se) => the arg is un-simplified and this is its subst-env + -- Nothing => the arg is already simplified; don't repeatedly simplify it! + SimplCont -- and its environment | Select DupFlag InId [InAlt] SimplEnv -- The case binder, alts, and subst-env @@ -86,7 +91,7 @@ data SimplCont -- Strict contexts | ArgOf LetRhsFlag -- An arbitrary strict context: the argument -- of a strict function, or a primitive-arg fn -- or a PrimOp - -- No DupFlag because we never duplicate it + -- No DupFlag, because we never duplicate it OutType -- arg_ty: type of the argument itself OutType -- cont_ty: the type of the expression being sought by the context -- f (error "foo") ==> coerce t (error "foo") @@ -110,7 +115,6 @@ instance Outputable SimplCont where ppr (Select dup bndr alts se cont) = (ptext SLIT("Select") <+> ppr dup <+> ppr bndr) $$ (nest 4 (ppr alts)) $$ ppr cont ppr (CoerceIt ty cont) = (ptext SLIT("CoerceIt") <+> ppr ty) $$ ppr cont - ppr (InlinePlease cont) = ptext SLIT("InlinePlease") $$ ppr cont data DupFlag = OkToDup | NoDup @@ -120,9 +124,14 @@ instance Outputable DupFlag where ------------------- -mkBoringStop, mkRhsStop :: OutType -> SimplCont -mkBoringStop ty = Stop ty AnArg (canUpdateInPlace ty) -mkRhsStop ty = Stop ty AnRhs (canUpdateInPlace ty) +mkBoringStop :: OutType -> SimplCont +mkBoringStop ty = Stop ty AnArg False + +mkLazyArgStop :: OutType -> Bool -> SimplCont +mkLazyArgStop ty has_rules = Stop ty AnArg (canUpdateInPlace ty || has_rules) + +mkRhsStop :: OutType -> SimplCont +mkRhsStop ty = Stop ty AnRhs (canUpdateInPlace ty) contIsRhs :: SimplCont -> Bool contIsRhs (Stop _ AnRhs _) = True @@ -139,14 +148,12 @@ contIsDupable (Stop _ _ _) = True contIsDupable (ApplyTo OkToDup _ _ _) = True contIsDupable (Select OkToDup _ _ _ _) = True contIsDupable (CoerceIt _ cont) = contIsDupable cont -contIsDupable (InlinePlease cont) = contIsDupable cont contIsDupable other = False ------------------- discardableCont :: SimplCont -> Bool discardableCont (Stop _ _ _) = False discardableCont (CoerceIt _ cont) = discardableCont cont -discardableCont (InlinePlease cont) = discardableCont cont discardableCont other = True discardCont :: SimplCont -- A continuation, expecting @@ -163,7 +170,6 @@ contResultType (Stop to_ty _ _) = to_ty contResultType (ArgOf _ _ to_ty _) = to_ty contResultType (ApplyTo _ _ _ cont) = contResultType cont contResultType (CoerceIt _ cont) = contResultType cont -contResultType (InlinePlease cont) = contResultType cont contResultType (Select _ _ _ _ cont) = contResultType cont ------------------- @@ -177,19 +183,18 @@ countArgs (ApplyTo _ arg se cont) = 1 + countArgs cont countArgs other = 0 ------------------- -pushContArgs :: SimplEnv -> [OutArg] -> SimplCont -> SimplCont +pushContArgs ::[OutArg] -> SimplCont -> SimplCont -- Pushes args with the specified environment -pushContArgs env [] cont = cont -pushContArgs env (arg : args) cont = ApplyTo NoDup arg env (pushContArgs env args cont) +pushContArgs [] cont = cont +pushContArgs (arg : args) cont = ApplyTo NoDup arg Nothing (pushContArgs args cont) \end{code} \begin{code} getContArgs :: SwitchChecker -> OutId -> SimplCont - -> ([(InExpr, SimplEnv, Bool)], -- Arguments; the Bool is true for strict args - SimplCont, -- Remaining continuation - Bool) -- Whether we came across an InlineCall + -> ([(InExpr, Maybe SimplEnv, Bool)], -- Arguments; the Bool is true for strict args + SimplCont) -- Remaining continuation -- getContArgs id k = (args, k', inl) -- args are the leading ApplyTo items in k -- (i.e. outermost comes first) @@ -202,22 +207,18 @@ getContArgs chkr fun orig_cont stricts | switchIsOn chkr NoCaseOfCase = vanilla_stricts | otherwise = computed_stricts in - go [] stricts False orig_cont + go [] stricts orig_cont where ---------------------------- -- Type argument - go acc ss inl (ApplyTo _ arg@(Type _) se cont) - = go ((arg,se,False) : acc) ss inl cont + go acc ss (ApplyTo _ arg@(Type _) se cont) + = go ((arg,se,False) : acc) ss cont -- NB: don't bother to instantiate the function type -- Value argument - go acc (s:ss) inl (ApplyTo _ arg se cont) - = go ((arg,se,s) : acc) ss inl cont - - -- An Inline continuation - go acc ss inl (InlinePlease cont) - = go acc ss True cont + go acc (s:ss) (ApplyTo _ arg se cont) + = go ((arg,se,s) : acc) ss cont -- We're run out of arguments, or else we've run out of demands -- The latter only happens if the result is guaranteed bottom @@ -229,9 +230,9 @@ getContArgs chkr fun orig_cont -- Then, especially in the first of these cases, we'd like to discard -- the continuation, leaving just the bottoming expression. But the -- type might not be right, so we may have to add a coerce. - go acc ss inl cont - | null ss && discardableCont cont = (reverse acc, discardCont cont, inl) - | otherwise = (reverse acc, cont, inl) + go acc ss cont + | null ss && discardableCont cont = (reverse acc, discardCont cont) + | otherwise = (reverse acc, cont) ---------------------------- vanilla_stricts, computed_stricts :: [Bool] @@ -375,14 +376,13 @@ interestingCallContext :: Bool -- False <=> no args at all interestingCallContext some_args some_val_args cont = interesting cont where - interesting (InlinePlease _) = True - interesting (Select _ _ _ _ _) = some_args - interesting (ApplyTo _ _ _ _) = True -- Can happen if we have (coerce t (f x)) y + interesting (Select {}) = some_args + interesting (ApplyTo {}) = True -- Can happen if we have (coerce t (f x)) y -- Perhaps True is a bit over-keen, but I've -- seen (coerce f) x, where f has an INLINE prag, -- So we have to give some motivaiton for inlining it - interesting (ArgOf _ _ _ _) = some_val_args - interesting (Stop ty _ upd_in_place) = some_val_args && upd_in_place + interesting (ArgOf {}) = some_val_args + interesting (Stop ty _ interesting) = some_val_args && interesting interesting (CoerceIt _ cont) = interesting cont -- If this call is the arg of a strict function, the context -- is a bit interesting. If we inline here, we may get useful @@ -401,6 +401,32 @@ interestingCallContext some_args some_val_args cont ------------------- +interestingArgContext :: Id -> 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. +-- The interesting_arg_ctxt 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 +interestingArgContext fn cont + = idHasRules fn || go cont + where + go (Select {}) = False + go (ApplyTo {}) = False + go (ArgOf {}) = True + go (CoerceIt _ c) = go c + go (Stop _ _ interesting) = interesting + +------------------- canUpdateInPlace :: Type -> Bool -- Consider let x = in ... -- If returns an explicit constructor, we might be able @@ -449,7 +475,7 @@ because doing so inhibits floating ==> ...(case x of I# x# -> case fw x# of ...)... and now the redex (f x) isn't floatable any more. -The no-inling thing is also important for Template Haskell. You might be +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 @@ -671,7 +697,13 @@ our new view that inlining is like a RULE, so I'm sticking to the 'active' story for now. \begin{code} -postInlineUnconditionally :: SimplEnv -> TopLevelFlag -> OutId -> OccInfo -> OutExpr -> Unfolding -> Bool +postInlineUnconditionally + :: SimplEnv -> TopLevelFlag + -> InId -- The binder (an OutId would be fine too) + -> OccInfo -- From the InId + -> OutExpr + -> Unfolding + -> Bool postInlineUnconditionally env top_lvl bndr occ_info rhs unfolding | not active = False | isLoopBreaker occ_info = False @@ -679,19 +711,28 @@ postInlineUnconditionally env top_lvl bndr occ_info rhs unfolding | exprIsTrivial rhs = True | otherwise = case occ_info of - OneOcc in_lam one_br int_cxt - -> (one_br || smallEnoughToInline unfolding) -- Small enough to dup + -- 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 (becuase 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 ... + -- I'm not sure how important this is in practice + OneOcc in_lam one_br int_cxt -- OneOcc => no work-duplication issue + -> smallEnoughToInline unfolding -- Small enough to dup -- ToDo: consider discount on smallEnoughToInline if int_cxt is true -- - -- NB: Do we want to inline arbitrarily big things becuase - -- one_br is True? that can lead to inline cascades. But - -- preInlineUnconditionlly has dealt with all the common cases - -- so perhaps it's worth the risk. Here's an example - -- let f = if b then Left (\x.BIG) else Right (\y.BIG) - -- in \y. ....f.... - -- We can't preInlineUnconditionally because that woud invalidate - -- the occ info for b. Yet f is used just once, and duplicating - -- the case work is fine (exprIsCheap). + -- 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. && ((isNotTopLevel top_lvl && not in_lam) || -- But outside a lambda, we want to be reasonably aggressive @@ -707,18 +748,25 @@ postInlineUnconditionally env top_lvl bndr occ_info rhs unfolding -- 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 ... + other -> False - -- The point here is that for *non-values* that occur - -- outside a lambda, the call-site inliner won't have - -- a chance (becuase 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 ... - -- I'm not sure how important this is in practice + +-- 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 = case getMode env of SimplGently -> isAlwaysActive prag @@ -794,7 +842,7 @@ mkLam env bndrs body cont | dopt Opt_DoLambdaEtaExpansion dflags, any isRuntimeVar bndrs - = tryEtaExpansion body `thenSmpl` \ body' -> + = tryEtaExpansion dflags body `thenSmpl` \ body' -> returnSmpl (emptyFloats env, mkLams bndrs body') {- Sept 01: I'm experimenting with getting the @@ -877,13 +925,13 @@ when computing arity; and etaExpand adds the coerces as necessary when actually computing the expansion. \begin{code} -tryEtaExpansion :: OutExpr -> SimplM OutExpr +tryEtaExpansion :: DynFlags -> OutExpr -> SimplM OutExpr -- There is at least one runtime binder in the binders -tryEtaExpansion body +tryEtaExpansion dflags body = getUniquesSmpl `thenSmpl` \ us -> returnSmpl (etaExpand fun_arity us body (exprType body)) where - fun_arity = exprEtaExpandArity body + fun_arity = exprEtaExpandArity dflags body \end{code}