X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Fcompiler%2FbasicTypes%2FIdInfo.lhs;h=76cb1f9f5573cf6f40fccee4ffd89f044bec9cf3;hb=ddbf5c13453b6a740167ca90b007c45b6030d0aa;hp=40b3c1ff7d8451830cd57847718a896a51596a4f;hpb=7a3bd641457666e10d0a47be9f22762e03defbf0;p=ghc-hetmet.git diff --git a/ghc/compiler/basicTypes/IdInfo.lhs b/ghc/compiler/basicTypes/IdInfo.lhs index 40b3c1f..76cb1f9 100644 --- a/ghc/compiler/basicTypes/IdInfo.lhs +++ b/ghc/compiler/basicTypes/IdInfo.lhs @@ -1,5 +1,5 @@ % -% (c) The GRASP/AQUA Project, Glasgow University, 1993-1996 +% (c) The GRASP/AQUA Project, Glasgow University, 1993-1998 % \section[IdInfo]{@IdInfos@: Non-essential information about @Ids@} @@ -7,547 +7,783 @@ Haskell. [WDP 94/11]) \begin{code} -#include "HsVersions.h" - module IdInfo ( IdInfo, -- Abstract - noIdInfo, - ppIdInfo, - applySubstToIdInfo, apply_to_IdInfo, -- not for general use, please + vanillaIdInfo, constantIdInfo, mkIdInfo, seqIdInfo, megaSeqIdInfo, - ArityInfo(..), - exactArity, atLeastArity, unknownArity, - arityInfo, addArityInfo, ppArityInfo, + -- Zapping + zapFragileInfo, zapLamInfo, zapSpecPragInfo, shortableIdInfo, copyIdInfo, - DemandInfo, - noDemandInfo, mkDemandInfo, demandInfo, ppDemandInfo, addDemandInfo, willBeDemanded, + -- Flavour + IdFlavour(..), flavourInfo, + setNoDiscardInfo, setFlavourInfo, + ppFlavourInfo, - StrictnessInfo(..), -- Non-abstract - Demand(..), -- Non-abstract - wwLazy, wwStrict, wwUnpack, wwPrim, wwEnum, + -- Arity + ArityInfo(..), + exactArity, atLeastArity, unknownArity, hasArity, + arityInfo, setArityInfo, ppArityInfo, arityLowerBound, - getWorkerId_maybe, - workerExists, - mkStrictnessInfo, mkBottomStrictnessInfo, noStrictnessInfo, bottomIsGuaranteed, - strictnessInfo, ppStrictnessInfo, addStrictnessInfo, + -- Strictness; imported from Demand + StrictnessInfo(..), + mkStrictnessInfo, noStrictnessInfo, + ppStrictnessInfo,isBottomingStrictness, + strictnessInfo, setStrictnessInfo, - unfoldInfo, addUnfoldInfo, + -- Usage generalisation + TyGenInfo(..), + tyGenInfo, setTyGenInfo, + noTyGenInfo, isNoTyGenInfo, ppTyGenInfo, tyGenInfoString, - specInfo, addSpecInfo, + -- Worker + WorkerInfo(..), workerExists, wrapperArity, workerId, + workerInfo, setWorkerInfo, ppWorkerInfo, - UpdateInfo, SYN_IE(UpdateSpec), - mkUpdateInfo, updateInfo, updateInfoMaybe, ppUpdateInfo, addUpdateInfo, + -- Unfolding + unfoldingInfo, setUnfoldingInfo, - DeforestInfo(..), - deforestInfo, ppDeforestInfo, addDeforestInfo, + -- DemandInfo + demandInfo, setDemandInfo, - ArgUsageInfo, ArgUsage(..), SYN_IE(ArgUsageType), - mkArgUsageInfo, argUsageInfo, addArgUsageInfo, getArgUsage, + -- Inline prags + InlinePragInfo(..), + inlinePragInfo, setInlinePragInfo, pprInlinePragInfo, + isNeverInlinePrag, neverInlinePrag, - FBTypeInfo, FBType(..), FBConsum(..), FBProd(..), - fbTypeInfo, ppFBTypeInfo, addFBTypeInfo, mkFBTypeInfo, getFBType - ) where + -- Occurrence info + OccInfo(..), isFragileOcc, isDeadOcc, isLoopBreaker, + InsideLam, OneBranch, insideLam, notInsideLam, oneBranch, notOneBranch, + occInfo, setOccInfo, + + -- Specialisation + specInfo, setSpecInfo, -IMP_Ubiq() -IMPORT_1_3(Char(toLower)) + -- CAF info + CafInfo(..), cafInfo, setCafInfo, mayHaveCafRefs, ppCafInfo, -IMPORT_DELOOPER(IdLoop) -- IdInfo is a dependency-loop ranch, and - -- we break those loops by using IdLoop and - -- *not* importing much of anything else, - -- except from the very general "utils". + -- Constructed Product Result Info + CprInfo(..), cprInfo, setCprInfo, ppCprInfo, noCprInfo, -import Type ( eqSimpleTy, splitFunTyExpandingDicts ) -import CmdLineOpts ( opt_OmitInterfacePragmas ) + -- Lambda-bound variable info + LBVarInfo(..), lbvarInfo, setLBVarInfo, noLBVarInfo + ) where -import Demand -import Maybes ( firstJust ) -import Outputable ( ifPprInterface, Outputable(..){-instances-} ) -import PprStyle ( PprStyle(..) ) -import Pretty -import Unique ( pprUnique ) -import Util ( mapAccumL, panic, assertPanic, pprPanic ) +#include "HsVersions.h" -#ifdef REALLY_HASKELL_1_3 -ord = fromEnum :: Char -> Int -#endif -applySubstToTy = panic "IdInfo.applySubstToTy" -showTypeCategory = panic "IdInfo.showTypeCategory" +import CoreSyn +import Type ( Type, usOnce ) +import PrimOp ( PrimOp ) +import Var ( Id ) +import BasicTypes ( OccInfo(..), isFragileOcc, isDeadOcc, seqOccInfo, isLoopBreaker, + InsideLam, insideLam, notInsideLam, + OneBranch, oneBranch, notOneBranch, + Arity + ) +import DataCon ( DataCon ) +import FieldLabel ( FieldLabel ) +import Type ( usOnce, usMany ) +import Demand -- Lots of stuff +import Outputable +import Util ( seqList ) + +infixl 1 `setDemandInfo`, + `setTyGenInfo`, + `setStrictnessInfo`, + `setSpecInfo`, + `setArityInfo`, + `setInlinePragInfo`, + `setUnfoldingInfo`, + `setCprInfo`, + `setWorkerInfo`, + `setLBVarInfo`, + `setCafInfo`, + `setOccInfo` + -- infixl so you can say (id `set` a `set` b) \end{code} An @IdInfo@ gives {\em optional} information about an @Id@. If present it never lies, but it may not be present, in which case there is always a conservative assumption which can be made. + There is one exception: the 'flavour' is *not* optional. + You must not discard it. + It used to be in Var.lhs, but that seems unclean. + Two @Id@s may have different info even though they have the same @Unique@ (and are hence the same @Id@); for example, one might lack the properties attached to the other. The @IdInfo@ gives information about the value, or definition, of the @Id@. It does {\em not} contain information about the @Id@'s usage -(except for @DemandInfo@? ToDo). +(except for @DemandInfo@? ToDo). (@lbvarInfo@ is also a marginal +case. KSW 1999-04). \begin{code} data IdInfo - = IdInfo - ArityInfo -- Its arity - - DemandInfo -- Whether or not it is definitely - -- demanded - - SpecEnv - -- Specialisations of this function which exist - - (StrictnessInfo Id) - -- Strictness properties, notably - -- how to conjure up "worker" functions - - Unfolding - -- Its unfolding; for locally-defined - -- things, this can *only* be NoUnfolding - - UpdateInfo -- Which args should be updated - - DeforestInfo -- Whether its definition should be - -- unfolded during deforestation - - ArgUsageInfo -- how this Id uses its arguments - - FBTypeInfo -- the Foldr/Build W/W property of this function. + = IdInfo { + flavourInfo :: IdFlavour, -- NOT OPTIONAL + arityInfo :: ArityInfo, -- Its arity + demandInfo :: Demand, -- Whether or not it is definitely demanded + specInfo :: CoreRules, -- Specialisations of this function which exist + tyGenInfo :: TyGenInfo, -- Restrictions on usage-generalisation of this Id + strictnessInfo :: StrictnessInfo, -- Strictness properties + workerInfo :: WorkerInfo, -- Pointer to Worker Function + unfoldingInfo :: Unfolding, -- Its unfolding + cafInfo :: CafInfo, -- whether it refers (indirectly) to any CAFs + cprInfo :: CprInfo, -- Function always constructs a product result + lbvarInfo :: LBVarInfo, -- Info about a lambda-bound variable + inlinePragInfo :: InlinePragInfo, -- Inline pragma + occInfo :: OccInfo -- How it occurs + } + +seqIdInfo :: IdInfo -> () +seqIdInfo (IdInfo {}) = () + +megaSeqIdInfo :: IdInfo -> () +megaSeqIdInfo info + = seqFlavour (flavourInfo info) `seq` + seqArity (arityInfo info) `seq` + seqDemand (demandInfo info) `seq` + seqRules (specInfo info) `seq` + seqTyGenInfo (tyGenInfo info) `seq` + seqStrictnessInfo (strictnessInfo info) `seq` + seqWorker (workerInfo info) `seq` + +-- seqUnfolding (unfoldingInfo info) `seq` +-- Omitting this improves runtimes a little, presumably because +-- some unfoldings are not calculated at all + + seqCaf (cafInfo info) `seq` + seqCpr (cprInfo info) `seq` + seqLBVar (lbvarInfo info) `seq` + seqOccInfo (occInfo info) \end{code} -\begin{code} -noIdInfo = IdInfo UnknownArity UnknownDemand nullSpecEnv NoStrictnessInfo noUnfolding - NoUpdateInfo Don'tDeforest NoArgUsageInfo NoFBTypeInfo -\end{code} +Setters -Simply turgid. But BE CAREFUL: don't @apply_to_Id@ if that @Id@ -will in turn @apply_to_IdInfo@ of the self-same @IdInfo@. (A very -nasty loop, friends...) \begin{code} -apply_to_IdInfo ty_fn idinfo@(IdInfo arity demand spec strictness unfold - update deforest arg_usage fb_ww) - | isNullSpecEnv spec - = idinfo +setFlavourInfo info fl = fl `seq` info { flavourInfo = fl } +setWorkerInfo info wk = wk `seq` info { workerInfo = wk } +setSpecInfo info sp = PSEQ sp (info { specInfo = sp }) +setTyGenInfo info tg = tg `seq` info { tyGenInfo = tg } +setInlinePragInfo info pr = pr `seq` info { inlinePragInfo = pr } +setOccInfo info oc = oc `seq` info { occInfo = oc } +setStrictnessInfo info st = st `seq` info { strictnessInfo = st } + -- Try to avoid spack leaks by seq'ing + +setUnfoldingInfo info uf + | isEvaldUnfolding uf && isStrict (demandInfo info) + -- If the unfolding is a value, the demand info may + -- go pear-shaped, so we nuke it. Example: + -- let x = (a,b) in + -- case x of (p,q) -> h p q x + -- Here x is certainly demanded. But after we've nuked + -- the case, we'll get just + -- let x = (a,b) in h a b x + -- and now x is not demanded (I'm assuming h is lazy) + -- This really happens. The solution here is a bit ad hoc... + = info { unfoldingInfo = uf, demandInfo = wwLazy } + | otherwise - = panic "IdInfo:apply_to_IdInfo" -{- LATER: - let - new_spec = apply_spec spec - - -- NOT a good idea: - -- apply_strict strictness `thenLft` \ new_strict -> - -- apply_wrap wrap `thenLft` \ new_wrap -> - in - IdInfo arity demand new_spec strictness unfold - update deforest arg_usage fb_ww - where - apply_spec (SpecEnv is) - = SpecEnv (map do_one is) - where - do_one (SpecInfo ty_maybes ds spec_id) - = --apply_to_Id ty_fn spec_id `thenLft` \ new_spec_id -> - SpecInfo (map apply_to_maybe ty_maybes) ds spec_id - where - apply_to_maybe Nothing = Nothing - apply_to_maybe (Just ty) = Just (ty_fn ty) --} - -{- NOT a good idea; - apply_strict info@NoStrictnessInfo = returnLft info - apply_strict BottomGuaranteed = ??? - apply_strict (StrictnessInfo wrap_arg_info id_maybe) - = (case id_maybe of - Nothing -> returnLft Nothing - Just xx -> applySubstToId subst xx `thenLft` \ new_xx -> - returnLft (Just new_xx) - ) `thenLft` \ new_id_maybe -> - returnLft (StrictnessInfo wrap_arg_info new_id_maybe) --} + -- We do *not* seq on the unfolding info, For some reason, doing so + -- actually increases residency significantly. + = info { unfoldingInfo = uf } + +setDemandInfo info dd = info { demandInfo = dd } +setArityInfo info ar = info { arityInfo = ar } +setCafInfo info cf = info { cafInfo = cf } +setCprInfo info cp = info { cprInfo = cp } +setLBVarInfo info lb = info { lbvarInfo = lb } + +setNoDiscardInfo info = case flavourInfo info of + VanillaId -> info { flavourInfo = ExportedId } + other -> info +zapSpecPragInfo info = case flavourInfo info of + SpecPragmaId -> info { flavourInfo = VanillaId } + other -> info \end{code} -Variant of the same thing for the typechecker. + \begin{code} -applySubstToIdInfo s0 (IdInfo arity demand spec strictness unfold - update deforest arg_usage fb_ww) - = panic "IdInfo:applySubstToIdInfo" -{- LATER: - case (apply_spec s0 spec) of { (s1, new_spec) -> - (s1, IdInfo arity demand new_spec strictness unfold update deforest arg_usage fb_ww) } - where - apply_spec s0 (SpecEnv is) - = case (mapAccumL do_one s0 is) of { (s1, new_is) -> - (s1, SpecEnv new_is) } - where - do_one s0 (SpecInfo ty_maybes ds spec_id) - = case (mapAccumL apply_to_maybe s0 ty_maybes) of { (s1, new_maybes) -> - (s1, SpecInfo new_maybes ds spec_id) } - where - apply_to_maybe s0 Nothing = (s0, Nothing) - apply_to_maybe s0 (Just ty) - = case (applySubstToTy s0 ty) of { (s1, new_ty) -> - (s1, Just new_ty) } --} +vanillaIdInfo :: IdInfo + -- Used for locally-defined Ids + -- We are going to calculate correct CAF information at the end +vanillaIdInfo = mkIdInfo VanillaId NoCafRefs + +constantIdInfo :: IdInfo + -- Used for imported Ids + -- The default is that they *do* have CAFs; an interface-file pragma + -- may say "oh no it doesn't", but in the absence of such a pragma + -- we'd better assume it does +constantIdInfo = mkIdInfo ConstantId MayHaveCafRefs + +mkIdInfo :: IdFlavour -> CafInfo -> IdInfo +mkIdInfo flv caf + = IdInfo { + flavourInfo = flv, + cafInfo = caf, + arityInfo = UnknownArity, + demandInfo = wwLazy, + specInfo = emptyCoreRules, + tyGenInfo = noTyGenInfo, + workerInfo = NoWorker, + strictnessInfo = NoStrictnessInfo, + unfoldingInfo = noUnfolding, + cprInfo = NoCPRInfo, + lbvarInfo = NoLBVarInfo, + inlinePragInfo = NoInlinePragInfo, + occInfo = NoOccInfo + } \end{code} + +%************************************************************************ +%* * +\subsection{Flavour} +%* * +%************************************************************************ + \begin{code} -ppIdInfo :: PprStyle - -> Bool -- True <=> print specialisations, please - -> IdInfo - -> Pretty - -ppIdInfo sty specs_please - (IdInfo arity demand specenv strictness unfold update deforest arg_usage fbtype) - = ppCat [ - -- order is important!: - ppArityInfo sty arity, - ppUpdateInfo sty update, - ppDeforestInfo sty deforest, - - ppStrictnessInfo sty strictness, - - if specs_please - then ppNil -- ToDo -- sty (not (isDataCon for_this_id)) - -- better_id_fn inline_env (mEnvToList specenv) - else ppNil, - - -- DemandInfo needn't be printed since it has no effect on interfaces - ppDemandInfo sty demand, - ppFBTypeInfo sty fbtype - ] +data IdFlavour + = VanillaId -- Locally defined, not exported + | ExportedId -- Locally defined, exported + | SpecPragmaId -- Locally defined, RHS holds specialised call + + | ConstantId -- Imported from elsewhere, or a default method Id. + + | DictFunId -- We flag dictionary functions so that we can + -- conveniently extract the DictFuns from a set of + -- bindings when building a module's interface + + | DataConId DataCon -- The Id for a data constructor *worker* + | DataConWrapId DataCon -- The Id for a data constructor *wrapper* + -- [the only reasons we need to know is so that + -- a) we can suppress printing a definition in the interface file + -- b) when typechecking a pattern we can get from the + -- Id back to the data con] + | PrimOpId PrimOp -- The Id for a primitive operator + | RecordSelId FieldLabel -- The Id for a record selector + + +ppFlavourInfo :: IdFlavour -> SDoc +ppFlavourInfo VanillaId = empty +ppFlavourInfo ExportedId = ptext SLIT("[Exported]") +ppFlavourInfo SpecPragmaId = ptext SLIT("[SpecPrag]") +ppFlavourInfo ConstantId = ptext SLIT("[Constant]") +ppFlavourInfo DictFunId = ptext SLIT("[DictFun]") +ppFlavourInfo (DataConId _) = ptext SLIT("[DataCon]") +ppFlavourInfo (DataConWrapId _) = ptext SLIT("[DataConWrapper]") +ppFlavourInfo (PrimOpId _) = ptext SLIT("[PrimOp]") +ppFlavourInfo (RecordSelId _) = ptext SLIT("[RecSel]") + +seqFlavour :: IdFlavour -> () +seqFlavour f = f `seq` () \end{code} +The @SpecPragmaId@ exists only to make Ids that are +on the *LHS* of bindings created by SPECIALISE pragmas; +eg: s = f Int d +The SpecPragmaId is never itself mentioned; it +exists solely so that the specialiser will find +the call to f, and make specialised version of it. +The SpecPragmaId binding is discarded by the specialiser +when it gathers up overloaded calls. +Meanwhile, it is not discarded as dead code. + + %************************************************************************ %* * \subsection[arity-IdInfo]{Arity info about an @Id@} %* * %************************************************************************ +For locally-defined Ids, the code generator maintains its own notion +of their arities; so it should not be asking... (but other things +besides the code-generator need arity info!) + \begin{code} data ArityInfo = UnknownArity -- No idea - | ArityExactly Int -- Arity is exactly this - | ArityAtLeast Int -- Arity is this or greater -\end{code} -\begin{code} + | ArityExactly Arity -- Arity is exactly this. We use this when importing a + -- function; it's already been compiled and we know its + -- arity for sure. + + | ArityAtLeast Arity -- A partial application of this Id to up to n-1 value arguments + -- does essentially no work. That is not necessarily the + -- same as saying that it has n leading lambdas, because coerces + -- may get in the way. + + -- functions in the module being compiled. Their arity + -- might increase later in the compilation process, if + -- an extra lambda floats up to the binding site. + deriving( Eq ) + +seqArity :: ArityInfo -> () +seqArity a = arityLowerBound a `seq` () + exactArity = ArityExactly atLeastArity = ArityAtLeast unknownArity = UnknownArity -arityInfo (IdInfo arity _ _ _ _ _ _ _ _) = arity +arityLowerBound :: ArityInfo -> Arity +arityLowerBound UnknownArity = 0 +arityLowerBound (ArityAtLeast n) = n +arityLowerBound (ArityExactly n) = n -addArityInfo id_info UnknownArity = id_info -addArityInfo (IdInfo _ a c d e f g h i) arity = IdInfo arity a c d e f g h i +hasArity :: ArityInfo -> Bool +hasArity UnknownArity = False +hasArity other = True -ppArityInfo sty UnknownArity = ppNil -ppArityInfo sty (ArityExactly arity) = ppCat [ppPStr SLIT("_A_"), ppInt arity] -ppArityInfo sty (ArityAtLeast arity) = ppCat [ppPStr SLIT("_A>_"), ppInt arity] +ppArityInfo UnknownArity = empty +ppArityInfo (ArityExactly arity) = hsep [ptext SLIT("__A"), int arity] +ppArityInfo (ArityAtLeast arity) = hsep [ptext SLIT("__AL"), int arity] \end{code} %************************************************************************ %* * -\subsection[demand-IdInfo]{Demand info about an @Id@} +\subsection{Inline-pragma information} %* * %************************************************************************ -Whether a value is certain to be demanded or not. (This is the -information that is computed by the ``front-end'' of the strictness -analyser.) - -This information is only used within a module, it is not exported -(obviously). - \begin{code} -data DemandInfo - = UnknownDemand - | DemandedAsPer Demand +data InlinePragInfo + = NoInlinePragInfo + | IMustNotBeINLINEd Bool -- True <=> came from an INLINE prag, False <=> came from a NOINLINE prag + (Maybe Int) -- Phase number from pragma, if any + deriving( Eq ) + -- The True, Nothing case doesn't need to be recorded + + -- SEE COMMENTS WITH CoreUnfold.blackListed on the + -- exact significance of the IMustNotBeINLINEd pragma + +isNeverInlinePrag :: InlinePragInfo -> Bool +isNeverInlinePrag (IMustNotBeINLINEd _ Nothing) = True +isNeverInlinePrag other = False + +neverInlinePrag :: InlinePragInfo +neverInlinePrag = IMustNotBeINLINEd True{-should be False? --SDM -} Nothing + +instance Outputable InlinePragInfo where + -- This is now parsed in interface files + ppr NoInlinePragInfo = empty + ppr other_prag = ptext SLIT("__U") <> pprInlinePragInfo other_prag + +pprInlinePragInfo NoInlinePragInfo = empty +pprInlinePragInfo (IMustNotBeINLINEd True Nothing) = empty +pprInlinePragInfo (IMustNotBeINLINEd True (Just n)) = brackets (int n) +pprInlinePragInfo (IMustNotBeINLINEd False Nothing) = brackets (char '!') +pprInlinePragInfo (IMustNotBeINLINEd False (Just n)) = brackets (char '!' <> int n) + +instance Show InlinePragInfo where + showsPrec p prag = showsPrecSDoc p (ppr prag) \end{code} -\begin{code} -noDemandInfo = UnknownDemand -mkDemandInfo :: Demand -> DemandInfo -mkDemandInfo demand = DemandedAsPer demand +%************************************************************************ +%* * +\subsection[TyGen-IdInfo]{Type generalisation info about an @Id@} +%* * +%************************************************************************ -willBeDemanded :: DemandInfo -> Bool -willBeDemanded (DemandedAsPer demand) = isStrict demand -willBeDemanded _ = False -\end{code} +Certain passes (notably usage inference) may change the type of an +identifier, modifying all in-scope uses of that identifier +appropriately to maintain type safety. + +However, some identifiers must not have their types changed in this +way, because their types are conjured up in the front end of the +compiler rather than being read from the interface file. Default +methods, dictionary functions, record selectors, and others are in +this category. (see comment at TcClassDcl.tcClassSig). + +To indicate this property, such identifiers are marked TyGenNever. + +Furthermore, if the usage inference generates a usage-specialised +variant of a function, we must NOT re-infer a fully-generalised type +at the next inference. This finer property is indicated by a +TyGenUInfo on the identifier. \begin{code} -demandInfo (IdInfo _ demand _ _ _ _ _ _ _) = demand +data TyGenInfo + = NoTyGenInfo -- no restriction on type generalisation -addDemandInfo (IdInfo a _ c d e f g h i) demand = IdInfo a demand c d e f g h i + | TyGenUInfo [Maybe Type] -- restrict generalisation of this Id to + -- preserve specified usage annotations -ppDemandInfo PprInterface _ = ppNil -ppDemandInfo sty UnknownDemand = ppStr "{-# L #-}" -ppDemandInfo sty (DemandedAsPer info) = ppCat [ppStr "{-#", ppStr (showList [info] ""), ppStr "#-}"] -\end{code} + | TyGenNever -- never generalise the type of this Id -%************************************************************************ -%* * -\subsection[specialisation-IdInfo]{Specialisation info about an @Id@} -%* * -%************************************************************************ + deriving ( Eq ) +\end{code} -See SpecEnv.lhs +For TyGenUInfo, the list has one entry for each usage annotation on +the type of the Id, in left-to-right pre-order (annotations come +before the type they annotate). Nothing means no restriction; Just +usOnce or Just usMany forces that annotation to that value. Other +usage annotations are illegal. \begin{code} -specInfo (IdInfo _ _ spec _ _ _ _ _ _) = spec - -addSpecInfo id_info spec | isNullSpecEnv spec = id_info -addSpecInfo (IdInfo a b _ d e f g h i) spec = IdInfo a b spec d e f g h i +seqTyGenInfo :: TyGenInfo -> () +seqTyGenInfo NoTyGenInfo = () +seqTyGenInfo (TyGenUInfo us) = seqList us () +seqTyGenInfo TyGenNever = () + +noTyGenInfo :: TyGenInfo +noTyGenInfo = NoTyGenInfo + +isNoTyGenInfo :: TyGenInfo -> Bool +isNoTyGenInfo NoTyGenInfo = True +isNoTyGenInfo _ = False + +-- NB: There's probably no need to write this information out to the interface file. +-- Why? Simply because imported identifiers never get their types re-inferred. +-- But it's definitely nice to see in dumps, it for debugging purposes. + +ppTyGenInfo :: TyGenInfo -> SDoc +ppTyGenInfo NoTyGenInfo = empty +ppTyGenInfo (TyGenUInfo us) = ptext SLIT("__G") <+> text (tyGenInfoString us) +ppTyGenInfo TyGenNever = ptext SLIT("__G N") + +tyGenInfoString us = map go us + where go Nothing = 'x' -- for legibility, choose + go (Just u) | u == usOnce = '1' -- chars with identity + | u == usMany = 'M' -- Z-encoding. + go other = pprPanic "IdInfo.tyGenInfoString: unexpected annotation" (ppr other) + +instance Outputable TyGenInfo where + ppr = ppTyGenInfo + +instance Show TyGenInfo where + showsPrec p c = showsPrecSDoc p (ppr c) \end{code} + %************************************************************************ %* * -\subsection[strictness-IdInfo]{Strictness info about an @Id@} +\subsection[worker-IdInfo]{Worker info about an @Id@} %* * %************************************************************************ -We specify the strictness of a function by giving information about -each of the ``wrapper's'' arguments (see the description about -worker/wrapper-style transformations in the PJ/Launchbury paper on -unboxed types). +If this Id has a worker then we store a reference to it. Worker +functions are generated by the worker/wrapper pass. This uses +information from the strictness and CPR analyses. -The list of @Demands@ specifies: (a)~the strictness properties -of a function's arguments; (b)~the {\em existence} of a ``worker'' -version of the function; and (c)~the type signature of that worker (if -it exists); i.e. its calling convention. +There might not be a worker, even for a strict function, because: +(a) the function might be small enough to inline, so no need + for w/w split +(b) the strictness info might be "SSS" or something, so no w/w split. \begin{code} -data StrictnessInfo bdee - = NoStrictnessInfo - - | BottomGuaranteed -- This Id guarantees never to return; - -- it is bottom regardless of its arguments. - -- Useful for "error" and other disguised - -- variants thereof. - - | StrictnessInfo [Demand] -- The main stuff; see below. - (Maybe bdee) -- Worker's Id, if applicable. - -- (It may not be applicable because the strictness info - -- might say just "SSS" or something; so there's no w/w split.) -\end{code} - -\begin{code} -mkStrictnessInfo :: [Demand] -> Maybe bdee -> StrictnessInfo bdee - -mkStrictnessInfo xs wrkr - | all is_lazy xs = NoStrictnessInfo -- Uninteresting - | otherwise = StrictnessInfo xs wrkr - where - is_lazy (WwLazy False) = True -- NB "Absent" args do *not* count! - is_lazy _ = False -- (as they imply a worker) -noStrictnessInfo = NoStrictnessInfo -mkBottomStrictnessInfo = BottomGuaranteed +data WorkerInfo = NoWorker + | HasWorker Id Arity + -- The Arity is the arity of the *wrapper* at the moment of the + -- w/w split. See comments in MkIface.ifaceId, with the 'Worker' code. -bottomIsGuaranteed BottomGuaranteed = True -bottomIsGuaranteed other = False +seqWorker :: WorkerInfo -> () +seqWorker (HasWorker id _) = id `seq` () +seqWorker NoWorker = () -strictnessInfo (IdInfo _ _ _ strict _ _ _ _ _) = strict +ppWorkerInfo NoWorker = empty +ppWorkerInfo (HasWorker wk_id _) = ptext SLIT("__P") <+> ppr wk_id -addStrictnessInfo id_info NoStrictnessInfo = id_info -addStrictnessInfo (IdInfo a b d _ e f g h i) strict = IdInfo a b d strict e f g h i +noWorkerInfo = NoWorker -ppStrictnessInfo sty NoStrictnessInfo = ppNil -ppStrictnessInfo sty BottomGuaranteed = ppPStr SLIT("_S_ _!_") +workerExists :: WorkerInfo -> Bool +workerExists NoWorker = False +workerExists (HasWorker _ _) = True -ppStrictnessInfo sty (StrictnessInfo wrapper_args wrkr_maybe) - = ppCat [ppPStr SLIT("_S_"), ppStr (showList wrapper_args ""), pp_wrkr] - where - pp_wrkr = case wrkr_maybe of - Nothing -> ppNil - Just wrkr -> ppr sty wrkr -\end{code} +workerId :: WorkerInfo -> Id +workerId (HasWorker id _) = id - -\begin{code} -workerExists :: StrictnessInfo bdee -> Bool -workerExists (StrictnessInfo _ (Just worker_id)) = True -workerExists other = False - -getWorkerId_maybe :: StrictnessInfo bdee -> Maybe bdee -getWorkerId_maybe (StrictnessInfo _ maybe_worker_id) = maybe_worker_id -getWorkerId_maybe other = Nothing +wrapperArity :: WorkerInfo -> Arity +wrapperArity (HasWorker _ a) = a \end{code} %************************************************************************ %* * -\subsection[unfolding-IdInfo]{Unfolding info about an @Id@} +\subsection[CAF-IdInfo]{CAF-related information} %* * %************************************************************************ -\begin{code} -unfoldInfo (IdInfo _ _ _ _ unfolding _ _ _ _) = unfolding +This information is used to build Static Reference Tables (see +simplStg/ComputeSRT.lhs). -addUnfoldInfo (IdInfo a b d e _ f g h i) uf = IdInfo a b d e uf f g h i -\end{code} +\begin{code} +data CafInfo + = MayHaveCafRefs -- either: + -- (1) A function or static constructor + -- that refers to one or more CAFs, + -- (2) A real live CAF -%************************************************************************ -%* * -\subsection[update-IdInfo]{Update-analysis info about an @Id@} -%* * -%************************************************************************ + | NoCafRefs -- A function or static constructor + -- that refers to no CAFs. -\begin{code} -data UpdateInfo - = NoUpdateInfo - | SomeUpdateInfo UpdateSpec - deriving (Eq, Ord) - -- we need Eq/Ord to cross-chk update infos in interfaces - --- the form in which we pass update-analysis info between modules: -type UpdateSpec = [Int] -\end{code} +-- LATER: not sure how easy this is... +-- | OneCafRef Id -\begin{code} -mkUpdateInfo = SomeUpdateInfo -updateInfoMaybe NoUpdateInfo = Nothing -updateInfoMaybe (SomeUpdateInfo []) = Nothing -updateInfoMaybe (SomeUpdateInfo u) = Just u -\end{code} +mayHaveCafRefs MayHaveCafRefs = True +mayHaveCafRefs _ = False -Text instance so that the update annotations can be read in. +seqCaf c = c `seq` () -\begin{code} -#ifdef REALLY_HASKELL_1_3 -instance Read UpdateInfo where -#else -instance Text UpdateInfo where -#endif - readsPrec p s | null s = panic "IdInfo: empty update pragma?!" - | otherwise = [(SomeUpdateInfo (map ok_digit s),"")] - where - ok_digit c | c >= '0' && c <= '2' = ord c - ord '0' - | otherwise = panic "IdInfo: not a digit while reading update pragma" - -updateInfo (IdInfo _ _ _ _ _ update _ _ _) = update - -addUpdateInfo id_info NoUpdateInfo = id_info -addUpdateInfo (IdInfo a b d e f _ g h i) upd_info = IdInfo a b d e f upd_info g h i - -ppUpdateInfo sty NoUpdateInfo = ppNil -ppUpdateInfo sty (SomeUpdateInfo []) = ppNil -ppUpdateInfo sty (SomeUpdateInfo spec) = ppBeside (ppPStr SLIT("_U_ ")) (ppBesides (map ppInt spec)) +ppCafInfo NoCafRefs = ptext SLIT("__C") +ppCafInfo MayHaveCafRefs = empty \end{code} + %************************************************************************ -%* * -\subsection[deforest-IdInfo]{Deforestation info about an @Id@} -%* * +%* * +\subsection[cpr-IdInfo]{Constructed Product Result info about an @Id@} +%* * %************************************************************************ -The deforest info says whether this Id is to be unfolded during -deforestation. Therefore, when the deforest pragma is true, we must -also have the unfolding information available for this Id. +If the @Id@ is a function then it may have CPR info. A CPR analysis +phase detects whether: + +\begin{enumerate} +\item +The function's return value has a product type, i.e. an algebraic type +with a single constructor. Examples of such types are tuples and boxed +primitive values. +\item +The function always 'constructs' the value that it is returning. It +must do this on every path through, and it's OK if it calls another +function which constructs the result. +\end{enumerate} + +If this is the case then we store a template which tells us the +function has the CPR property and which components of the result are +also CPRs. \begin{code} -data DeforestInfo - = Don'tDeforest -- just a bool, might extend this - | DoDeforest -- later. - -- deriving (Eq, Ord) +data CprInfo + = NoCPRInfo + | ReturnsCPR -- Yes, this function returns a constructed product + -- Implicitly, this means "after the function has been applied + -- to all its arguments", so the worker/wrapper builder in + -- WwLib.mkWWcpr checks that that it is indeed saturated before + -- making use of the CPR info + + -- We used to keep nested info about sub-components, but + -- we never used it so I threw it away \end{code} \begin{code} -deforestInfo (IdInfo _ _ _ _ _ _ deforest _ _) = deforest +seqCpr :: CprInfo -> () +seqCpr ReturnsCPR = () +seqCpr NoCPRInfo = () -addDeforestInfo id_info Don'tDeforest = id_info -addDeforestInfo (IdInfo a b d e f g _ h i) deforest = IdInfo a b d e f g deforest h i +noCprInfo = NoCPRInfo -ppDeforestInfo sty Don'tDeforest = ppNil -ppDeforestInfo sty DoDeforest = ppPStr SLIT("_DEFOREST_") +ppCprInfo NoCPRInfo = empty +ppCprInfo ReturnsCPR = ptext SLIT("__M") + +instance Outputable CprInfo where + ppr = ppCprInfo + +instance Show CprInfo where + showsPrec p c = showsPrecSDoc p (ppr c) \end{code} + %************************************************************************ %* * -\subsection[argUsage-IdInfo]{Argument Usage info about an @Id@} +\subsection[lbvar-IdInfo]{Lambda-bound var info about an @Id@} %* * %************************************************************************ -\begin{code} -data ArgUsageInfo - = NoArgUsageInfo - | SomeArgUsageInfo ArgUsageType - -- ??? deriving (Eq, Ord) - -data ArgUsage = ArgUsage Int -- number of arguments (is linear!) - | UnknownArgUsage -type ArgUsageType = [ArgUsage] -- c_1 -> ... -> BLOB -\end{code} +If the @Id@ is a lambda-bound variable then it may have lambda-bound +var info. The usage analysis (UsageSP) detects whether the lambda +binding this var is a ``one-shot'' lambda; that is, whether it is +applied at most once. -\begin{code} -mkArgUsageInfo [] = NoArgUsageInfo -mkArgUsageInfo au = SomeArgUsageInfo au +This information may be useful in optimisation, as computations may +safely be floated inside such a lambda without risk of duplicating +work. -getArgUsage :: ArgUsageInfo -> ArgUsageType -getArgUsage NoArgUsageInfo = [] -getArgUsage (SomeArgUsageInfo u) = u +\begin{code} +data LBVarInfo + = NoLBVarInfo + + | LBVarInfo Type -- The lambda that binds this Id has this usage + -- annotation (i.e., if ==usOnce, then the + -- lambda is applied at most once). + -- The annotation's kind must be `$' + -- HACK ALERT! placing this info here is a short-term hack, + -- but it minimises changes to the rest of the compiler. + -- Hack agreed by SLPJ/KSW 1999-04. + +seqLBVar l = l `seq` () \end{code} \begin{code} -argUsageInfo (IdInfo _ _ _ _ _ _ _ au _) = au - -addArgUsageInfo id_info NoArgUsageInfo = id_info -addArgUsageInfo (IdInfo a b d e f g h _ i) au_info = IdInfo a b d e f g h au_info i - -ppArgUsageInfo sty NoArgUsageInfo = ppNil -ppArgUsageInfo sty (SomeArgUsageInfo aut) = ppBeside (ppPStr SLIT("_L_ ")) (ppArgUsageType aut) - -ppArgUsage (ArgUsage n) = ppInt n -ppArgUsage (UnknownArgUsage) = ppChar '-' - -ppArgUsageType aut = ppBesides - [ ppChar '"' , - ppIntersperse ppComma (map ppArgUsage aut), - ppChar '"' ] +noLBVarInfo = NoLBVarInfo + +-- not safe to print or parse LBVarInfo because it is not really a +-- property of the definition, but a property of the context. +pprLBVarInfo NoLBVarInfo = empty +pprLBVarInfo (LBVarInfo u) | u == usOnce + = getPprStyle $ \ sty -> + if ifaceStyle sty + then empty + else ptext SLIT("OneShot") + | otherwise + = empty + +instance Outputable LBVarInfo where + ppr = pprLBVarInfo + +instance Show LBVarInfo where + showsPrec p c = showsPrecSDoc p (ppr c) \end{code} + %************************************************************************ %* * -\subsection[FBType-IdInfo]{Type of an expression through Foldr/build's eyes} +\subsection{Bulk operations on IdInfo} %* * %************************************************************************ -\begin{code} -data FBTypeInfo - = NoFBTypeInfo - | SomeFBTypeInfo FBType +zapFragileInfo is used when cloning binders, mainly in the +simplifier. We must forget about used-once information because that +isn't necessarily correct in the transformed program. +Also forget specialisations and unfoldings because they would need +substitution to be correct. (They get pinned back on separately.) -data FBType = FBType [FBConsum] FBProd deriving (Eq) - -data FBConsum = FBGoodConsum | FBBadConsum deriving(Eq) -data FBProd = FBGoodProd | FBBadProd deriving(Eq) -\end{code} +Hoever, we REMEMBER loop-breaker and dead-variable information. The loop-breaker +information is used (for example) in MkIface to avoid exposing the unfolding of +a loop breaker. \begin{code} -mkFBTypeInfo = SomeFBTypeInfo +zapFragileInfo :: IdInfo -> Maybe IdInfo +zapFragileInfo info@(IdInfo {occInfo = occ, + workerInfo = wrkr, + specInfo = rules, + unfoldingInfo = unfolding}) + | not (isFragileOcc occ) + -- We must forget about whether it was marked safe-to-inline, + -- because that isn't necessarily true in the simplified expression. + -- This is important because expressions may be re-simplified + -- We don't zap deadness or loop-breaker-ness. + -- The latter is important because it tells MkIface not to + -- spit out an inlining for the thing. The former doesn't + -- seem so important, but there's no harm. + + && isEmptyCoreRules rules + -- Specialisations would need substituting. They get pinned + -- back on separately. + + && not (workerExists wrkr) + + && not (hasUnfolding unfolding) + -- This is very important; occasionally a let-bound binder is used + -- as a binder in some lambda, in which case its unfolding is utterly + -- bogus. Also the unfolding uses old binders so if we left it we'd + -- have to substitute it. Much better simply to give the Id a new + -- unfolding each time, which is what the simplifier does. + = Nothing -getFBType :: FBTypeInfo -> Maybe FBType -getFBType NoFBTypeInfo = Nothing -getFBType (SomeFBTypeInfo u) = Just u + | otherwise + = Just (info {occInfo = robust_occ_info, + workerInfo = noWorkerInfo, + specInfo = emptyCoreRules, + unfoldingInfo = noUnfolding}) + where + -- It's important to keep the loop-breaker info, + -- because the substitution doesn't remember it. + robust_occ_info = case occ of + OneOcc _ _ -> NoOccInfo + other -> occ \end{code} +@zapLamInfo@ is used for lambda binders that turn out to to be +part of an unsaturated lambda + \begin{code} -fbTypeInfo (IdInfo _ _ _ _ _ _ _ _ fb) = fb +zapLamInfo :: IdInfo -> Maybe IdInfo +zapLamInfo info@(IdInfo {occInfo = occ, demandInfo = demand}) + | is_safe_occ && not (isStrict demand) + = Nothing + | otherwise + = Just (info {occInfo = safe_occ, + demandInfo = wwLazy}) + where + -- The "unsafe" occ info is the ones that say I'm not in a lambda + -- because that might not be true for an unsaturated lambda + is_safe_occ = case occ of + OneOcc in_lam once -> in_lam + other -> True + + safe_occ = case occ of + OneOcc _ once -> OneOcc insideLam once + other -> occ +\end{code} -addFBTypeInfo id_info NoFBTypeInfo = id_info -addFBTypeInfo (IdInfo a b d e f g h i _) fb_info = IdInfo a b d e f g h i fb_info -ppFBTypeInfo sty NoFBTypeInfo = ppNil -ppFBTypeInfo sty (SomeFBTypeInfo (FBType cons prod)) - = ppBeside (ppPStr SLIT("_F_ ")) (ppFBType cons prod) +copyIdInfo is used when shorting out a top-level binding + f_local = BIG + f = f_local +where f is exported. We are going to swizzle it around to + f = BIG + f_local = f + +BUT (a) we must be careful about messing up rules + (b) we must ensure f's IdInfo ends up right + +(a) Messing up the rules +~~~~~~~~~~~~~~~~~~~~ +The example that went bad on me was this one: + + iterate :: (a -> a) -> a -> [a] + iterate = iterateList + + iterateFB c f x = x `c` iterateFB c f (f x) + iterateList f x = x : iterateList f (f x) + + {-# RULES + "iterate" forall f x. iterate f x = build (\c _n -> iterateFB c f x) + "iterateFB" iterateFB (:) = iterateList + #-} + +This got shorted out to: + + iterateList :: (a -> a) -> a -> [a] + iterateList = iterate + + iterateFB c f x = x `c` iterateFB c f (f x) + iterate f x = x : iterate f (f x) + + {-# RULES + "iterate" forall f x. iterate f x = build (\c _n -> iterateFB c f x) + "iterateFB" iterateFB (:) = iterate + #-} + +And now we get an infinite loop in the rule system + iterate f x -> build (\cn -> iterateFB c f x + -> iterateFB (:) f x + -> iterate f x + +Tiresome solution: don't do shorting out if f has rewrite rules. +Hence shortableIdInfo. + +(b) Keeping the IdInfo right +~~~~~~~~~~~~~~~~~~~~~~~~ +We want to move strictness/worker info from f_local to f, but keep the rest. +Hence copyIdInfo. -ppFBType cons prod = ppBesides - ([ ppChar '"' ] ++ map ppCons cons ++ [ ppChar '-', ppProd prod, ppChar '"' ]) - where - ppCons FBGoodConsum = ppChar 'G' - ppCons FBBadConsum = ppChar 'B' - ppProd FBGoodProd = ppChar 'G' - ppProd FBBadProd = ppChar 'B' +\begin{code} +shortableIdInfo :: IdInfo -> Bool +shortableIdInfo info = isEmptyCoreRules (specInfo info) + +copyIdInfo :: IdInfo -- f_local + -> IdInfo -- f (the exported one) + -> IdInfo -- New info for f +copyIdInfo f_local f = f { strictnessInfo = strictnessInfo f_local, + workerInfo = workerInfo f_local, + cprInfo = cprInfo f_local + } \end{code}