X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Fcompiler%2FbasicTypes%2FLiteral.lhs;h=2167ba025753d81722ca7957bfe6c21973156489;hb=fcfe16433bd582d0e00404ea652806d13d14103c;hp=ad7111848a0dbb1076827ee904146916a7049ee8;hpb=6ecaa998306b102c6e8824ad4f83ab4a93edfba6;p=ghc-hetmet.git diff --git a/ghc/compiler/basicTypes/Literal.lhs b/ghc/compiler/basicTypes/Literal.lhs index ad71118..2167ba0 100644 --- a/ghc/compiler/basicTypes/Literal.lhs +++ b/ghc/compiler/basicTypes/Literal.lhs @@ -1,41 +1,85 @@ % -% (c) The GRASP/AQUA Project, Glasgow University, 1992-1996 +% (c) The GRASP/AQUA Project, Glasgow University, 1998 % \section[Literal]{@Literal@: Machine literals (unboxed, of course)} \begin{code} module Literal - ( - Literal(..) - - , mkMachInt - , mkMachInt_safe - , mkMachWord - , literalType - , literalPrimRep - , showLiteral - , isNoRepLit - , isLitLitLit - ) where + ( Literal(..) -- Exported to ParseIface + , mkMachInt, mkMachWord + , mkMachInt64, mkMachWord64 + , isLitLitLit, maybeLitLit, litSize, litIsDupable, + , literalType, literalPrimRep + , hashLiteral + + , inIntRange, inWordRange, tARGET_MAX_INT, inCharRange + , isZeroLit, + + , word2IntLit, int2WordLit + , narrow8IntLit, narrow16IntLit, narrow32IntLit + , narrow8WordLit, narrow16WordLit, narrow32WordLit + , char2IntLit, int2CharLit + , float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit + , nullAddrLit, float2DoubleLit, double2FloatLit + ) where #include "HsVersions.h" --- friends: -import PrimRep ( PrimRep(..), ppPrimRep ) -- non-abstract -import TysPrim ( getPrimRepInfo, - addrPrimTy, intPrimTy, floatPrimTy, - doublePrimTy, charPrimTy, wordPrimTy +import TysPrim ( charPrimTy, addrPrimTy, floatPrimTy, doublePrimTy, + intPrimTy, wordPrimTy, int64PrimTy, word64PrimTy ) +import PrimRep ( PrimRep(..) ) +import TcType ( Type, tcCmpType ) +import Type ( typePrimRep ) +import PprType ( pprParendType ) +import CStrings ( pprFSInCStyle ) --- others: -import Type ( Type ) -import CStrings ( stringToC, charToC, charToEasyHaskell ) -import TysWiredIn ( stringTy ) import Outputable +import FastTypes import Util ( thenCmp ) +import Ratio ( numerator ) +import FastString ( uniqueOfFS, lengthFS ) +import Int ( Int8, Int16, Int32 ) +import Word ( Word8, Word16, Word32 ) +import Char ( ord, chr ) \end{code} + + +%************************************************************************ +%* * +\subsection{Sizes} +%* * +%************************************************************************ + +If we're compiling with GHC (and we're not cross-compiling), then we +know that minBound and maxBound :: Int are the right values for the +target architecture. Otherwise, we assume -2^31 and 2^31-1 +respectively (which will be wrong on a 64-bit machine). + +\begin{code} +tARGET_MIN_INT, tARGET_MAX_INT, tARGET_MAX_WORD :: Integer +#if __GLASGOW_HASKELL__ +tARGET_MIN_INT = toInteger (minBound :: Int) +tARGET_MAX_INT = toInteger (maxBound :: Int) +#else +tARGET_MIN_INT = -2147483648 +tARGET_MAX_INT = 2147483647 +#endif +tARGET_MAX_WORD = (tARGET_MAX_INT * 2) + 1 + +tARGET_MAX_CHAR :: Int +tARGET_MAX_CHAR = 0x10ffff +\end{code} + + +%************************************************************************ +%* * +\subsection{Literals} +%* * +%************************************************************************ + So-called @Literals@ are {\em either}: \begin{itemize} \item @@ -50,87 +94,41 @@ function applications, etc., etc., has not yet been done. \begin{code} data Literal - = MachChar Char + = ------------------ + -- First the primitive guys + MachChar Int -- Char# At least 31 bits | MachStr FAST_STRING - | MachAddr Integer -- whatever this machine thinks is a "pointer" - - | MachInt Integer -- for the numeric types, these are - Bool -- True <=> signed (Int#); False <=> unsigned (Word#) + | MachAddr Integer -- Whatever this machine thinks is a "pointer" - | MachInt64 Integer -- guaranteed 64-bit versions of the above. - Bool -- True <=> signed (Int#); False <=> unsigned (Word#) + | MachInt Integer -- Int# At least WORD_SIZE_IN_BITS bits + | MachInt64 Integer -- Int64# At least 64 bits + | MachWord Integer -- Word# At least WORD_SIZE_IN_BITS bits + | MachWord64 Integer -- Word64# At least 64 bits | MachFloat Rational | MachDouble Rational - | MachLitLit FAST_STRING - PrimRep - - | NoRepStr FAST_STRING - | NoRepInteger Integer Type -- This Type is always Integer - | NoRepRational Rational Type -- This Type is always Rational - -- We keep these Types in the literal because Rational isn't - -- (currently) wired in, so we can't conjure up its type out of - -- thin air. Integer is, so the type here is really redundant. - - -- deriving (Eq, Ord): no, don't want to compare Types - -- The Ord is needed for the FiniteMap used in the lookForConstructor - -- in SimplEnv. If you declared that lookForConstructor *ignores* - -- constructor-applications with LitArg args, then you could get - -- rid of this Ord. - -mkMachInt, mkMachWord :: Integer -> Literal - -mkMachInt x = MachInt x True{-signed-} -mkMachWord x = MachInt x False{-unsigned-} - --- check if the int is within range -mkMachInt_safe :: Integer -> Literal -mkMachInt_safe i - | out_of_range = - pprPanic "mkMachInt_safe" - (hsep [text "ERROR: Int ", text (show i), text "out of range", - brackets (int minInt <+> text ".." <+> int maxInt)]) - | otherwise = MachInt i True{-signed-} - where - out_of_range = --- i < fromInt minBound || - i > fromInt maxInt - -mkMachInt64 x = MachInt64 x True{-signed-} -mkMachWord64 x = MachInt64 x False{-unsigned-} + -- MachLabel is used (only) for the literal derived from a + -- "foreign label" declaration. + -- string argument is the name of a symbol. This literal + -- refers to the *address* of the label. + | MachLabel FAST_STRING -- always an Addr# + + -- lit-lits only work for via-C compilation, hence they + -- are deprecated. The string is emitted verbatim into + -- the C file, and can therefore be any C expression, + -- macro call, #defined constant etc. + | MachLitLit FAST_STRING Type -- Type might be Addr# or Int# etc +\end{code} + +\begin{code} +instance Outputable Literal where + ppr lit = pprLit lit + +instance Show Literal where + showsPrec p lit = showsPrecSDoc p (ppr lit) -cmpLit (MachChar a) (MachChar b) = a `compare` b -cmpLit (MachStr a) (MachStr b) = a `compare` b -cmpLit (MachAddr a) (MachAddr b) = a `compare` b -cmpLit (MachInt a b) (MachInt c d) = (a `compare` c) `thenCmp` (b `compare` d) -cmpLit (MachFloat a) (MachFloat b) = a `compare` b -cmpLit (MachDouble a) (MachDouble b) = a `compare` b -cmpLit (MachLitLit a b) (MachLitLit c d) = (a `compare` c) `thenCmp` (b `compare` d) -cmpLit (NoRepStr a) (NoRepStr b) = a `compare` b -cmpLit (NoRepInteger a _) (NoRepInteger b _) = a `compare` b -cmpLit (NoRepRational a _) (NoRepRational b _) = a `compare` b - - -- now we *know* the tags are different, so... -cmpLit other_1 other_2 - | tag1 _LT_ tag2 = LT - | otherwise = GT - where - tag1 = tagof other_1 - tag2 = tagof other_2 - - tagof (MachChar _) = ILIT(1) - tagof (MachStr _) = ILIT(2) - tagof (MachAddr _) = ILIT(3) - tagof (MachInt _ _) = ILIT(4) - tagof (MachFloat _) = ILIT(5) - tagof (MachDouble _) = ILIT(6) - tagof (MachLitLit _ _) = ILIT(7) - tagof (NoRepStr _) = ILIT(8) - tagof (NoRepInteger _ _) = ILIT(9) - tagof (NoRepRational _ _) = ILIT(10) - instance Eq Literal where a == b = case (a `compare` b) of { EQ -> True; _ -> False } a /= b = case (a `compare` b) of { EQ -> False; _ -> True } @@ -143,107 +141,269 @@ instance Ord Literal where compare a b = cmpLit a b \end{code} + + Construction + ~~~~~~~~~~~~ \begin{code} -isNoRepLit (NoRepStr _) = True -- these are not primitive typed! -isNoRepLit (NoRepInteger _ _) = True -isNoRepLit (NoRepRational _ _) = True -isNoRepLit _ = False +mkMachInt, mkMachWord, mkMachInt64, mkMachWord64 :: Integer -> Literal + +mkMachInt x = -- ASSERT2( inIntRange x, integer x ) + -- Not true: you can write out of range Int# literals + -- For example, one can write (intToWord# 0xffff0000) to + -- get a particular Word bit-pattern, and there's no other + -- convenient way to write such literals, which is why we allow it. + MachInt x +mkMachWord x = -- ASSERT2( inWordRange x, integer x ) + MachWord x +mkMachInt64 x = MachInt64 x +mkMachWord64 x = MachWord64 x + +inIntRange, inWordRange :: Integer -> Bool +inIntRange x = x >= tARGET_MIN_INT && x <= tARGET_MAX_INT +inWordRange x = x >= 0 && x <= tARGET_MAX_WORD + +inCharRange :: Int -> Bool +inCharRange c = c >= 0 && c <= tARGET_MAX_CHAR + +isZeroLit :: Literal -> Bool +isZeroLit (MachInt 0) = True +isZeroLit (MachInt64 0) = True +isZeroLit (MachWord 0) = True +isZeroLit (MachWord64 0) = True +isZeroLit (MachFloat 0) = True +isZeroLit (MachDouble 0) = True +isZeroLit other = False +\end{code} + Coercions + ~~~~~~~~~ +\begin{code} +word2IntLit, int2WordLit, + narrow8IntLit, narrow16IntLit, narrow32IntLit, + narrow8WordLit, narrow16WordLit, narrow32WordLit, + char2IntLit, int2CharLit, + float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit, + float2DoubleLit, double2FloatLit + :: Literal -> Literal + +word2IntLit (MachWord w) + | w > tARGET_MAX_INT = MachInt (w - tARGET_MAX_WORD - 1) + | otherwise = MachInt w + +int2WordLit (MachInt i) + | i < 0 = MachWord (1 + tARGET_MAX_WORD + i) -- (-1) ---> tARGET_MAX_WORD + | otherwise = MachWord i + +narrow8IntLit (MachInt i) = MachInt (toInteger (fromInteger i :: Int8)) +narrow16IntLit (MachInt i) = MachInt (toInteger (fromInteger i :: Int16)) +narrow32IntLit (MachInt i) = MachInt (toInteger (fromInteger i :: Int32)) +narrow8WordLit (MachWord w) = MachWord (toInteger (fromInteger w :: Word8)) +narrow16WordLit (MachWord w) = MachWord (toInteger (fromInteger w :: Word16)) +narrow32WordLit (MachWord w) = MachWord (toInteger (fromInteger w :: Word32)) + +char2IntLit (MachChar c) = MachInt (toInteger c) +int2CharLit (MachInt i) = MachChar (fromInteger i) + +float2IntLit (MachFloat f) = MachInt (truncate f) +int2FloatLit (MachInt i) = MachFloat (fromInteger i) + +double2IntLit (MachDouble f) = MachInt (truncate f) +int2DoubleLit (MachInt i) = MachDouble (fromInteger i) + +float2DoubleLit (MachFloat f) = MachDouble f +double2FloatLit (MachDouble d) = MachFloat d + +nullAddrLit :: Literal +nullAddrLit = MachAddr 0 +\end{code} + + Predicates + ~~~~~~~~~~ +\begin{code} isLitLitLit (MachLitLit _ _) = True isLitLitLit _ = False + +maybeLitLit (MachLitLit s t) = Just (s,t) +maybeLitLit _ = Nothing + +litIsDupable :: Literal -> Bool + -- True if code space does not go bad if we duplicate this literal + -- False principally of strings +litIsDupable (MachStr _) = False +litIsDupable other = True + +litSize :: Literal -> Int + -- used by CoreUnfold.sizeExpr +litSize (MachStr str) = lengthFS str `div` 4 +litSize _other = 1 \end{code} + Types + ~~~~~ \begin{code} literalType :: Literal -> Type - -literalType (MachChar _) = charPrimTy -literalType (MachStr _) = addrPrimTy -literalType (MachAddr _) = addrPrimTy -literalType (MachInt _ signed) = if signed then intPrimTy else wordPrimTy -literalType (MachFloat _) = floatPrimTy -literalType (MachDouble _) = doublePrimTy -literalType (MachLitLit _ k) = case (getPrimRepInfo k) of { (_,t,_) -> t } -literalType (NoRepInteger _ t) = t -literalType (NoRepRational _ t) = t -literalType (NoRepStr _) = stringTy +literalType (MachChar _) = charPrimTy +literalType (MachStr _) = addrPrimTy +literalType (MachAddr _) = addrPrimTy +literalType (MachInt _) = intPrimTy +literalType (MachWord _) = wordPrimTy +literalType (MachInt64 _) = int64PrimTy +literalType (MachWord64 _) = word64PrimTy +literalType (MachFloat _) = floatPrimTy +literalType (MachDouble _) = doublePrimTy +literalType (MachLabel _) = addrPrimTy +literalType (MachLitLit _ ty) = ty \end{code} \begin{code} literalPrimRep :: Literal -> PrimRep -literalPrimRep (MachChar _) = CharRep -literalPrimRep (MachStr _) = AddrRep -- specifically: "char *" -literalPrimRep (MachAddr _) = AddrRep -literalPrimRep (MachInt _ signed) = if signed then IntRep else WordRep -literalPrimRep (MachInt64 _ signed) = if signed then Int64Rep else Word64Rep -literalPrimRep (MachFloat _) = FloatRep -literalPrimRep (MachDouble _) = DoubleRep -literalPrimRep (MachLitLit _ k) = k -#ifdef DEBUG -literalPrimRep (NoRepInteger _ _) = panic "literalPrimRep:NoRepInteger" -literalPrimRep (NoRepRational _ _) = panic "literalPrimRep:NoRepRational" -literalPrimRep (NoRepStr _) = panic "literalPrimRep:NoRepString" -#endif +literalPrimRep (MachChar _) = CharRep +literalPrimRep (MachStr _) = AddrRep -- specifically: "char *" +literalPrimRep (MachAddr _) = AddrRep +literalPrimRep (MachInt _) = IntRep +literalPrimRep (MachWord _) = WordRep +literalPrimRep (MachInt64 _) = Int64Rep +literalPrimRep (MachWord64 _) = Word64Rep +literalPrimRep (MachFloat _) = FloatRep +literalPrimRep (MachDouble _) = DoubleRep +literalPrimRep (MachLabel _) = AddrRep +literalPrimRep (MachLitLit _ ty) = typePrimRep ty \end{code} -The boring old output stuff: + + Comparison + ~~~~~~~~~~ \begin{code} --- MachX (i.e. unboxed) things are printed unadornded (e.g. 3, 'a', "foo") --- exceptions: MachFloat and MachAddr get an initial keyword prefix --- --- NoRep things get an initial keyword prefix (e.g. _integer_ 3) +cmpLit (MachChar a) (MachChar b) = a `compare` b +cmpLit (MachStr a) (MachStr b) = a `compare` b +cmpLit (MachAddr a) (MachAddr b) = a `compare` b +cmpLit (MachInt a) (MachInt b) = a `compare` b +cmpLit (MachWord a) (MachWord b) = a `compare` b +cmpLit (MachInt64 a) (MachInt64 b) = a `compare` b +cmpLit (MachWord64 a) (MachWord64 b) = a `compare` b +cmpLit (MachFloat a) (MachFloat b) = a `compare` b +cmpLit (MachDouble a) (MachDouble b) = a `compare` b +cmpLit (MachLabel a) (MachLabel b) = a `compare` b +cmpLit (MachLitLit a b) (MachLitLit c d) = (a `compare` c) `thenCmp` (b `tcCmpType` d) +cmpLit lit1 lit2 | litTag lit1 <# litTag lit2 = LT + | otherwise = GT + +litTag (MachChar _) = _ILIT(1) +litTag (MachStr _) = _ILIT(2) +litTag (MachAddr _) = _ILIT(3) +litTag (MachInt _) = _ILIT(4) +litTag (MachWord _) = _ILIT(5) +litTag (MachInt64 _) = _ILIT(6) +litTag (MachWord64 _) = _ILIT(7) +litTag (MachFloat _) = _ILIT(8) +litTag (MachDouble _) = _ILIT(9) +litTag (MachLabel _) = _ILIT(10) +litTag (MachLitLit _ _) = _ILIT(11) +\end{code} -instance Outputable Literal where - ppr lit = pprLit lit + Printing + ~~~~~~~~ +* MachX (i.e. unboxed) things are printed unadornded (e.g. 3, 'a', "foo") + exceptions: MachFloat and MachAddr get an initial keyword prefix +\begin{code} pprLit lit = getPprStyle $ \ sty -> let - code_style = codeStyle sty + code_style = codeStyle sty + iface_style = ifaceStyle sty in case lit of - MachChar ch | code_style -> hcat [ptext SLIT("(C_)"), char '\'', text (charToC ch), char '\''] - | ifaceStyle sty -> char '\'' <> text (charToEasyHaskell ch) <> char '\'' - | otherwise -> text ['\'', ch, '\''] + MachChar ch | code_style -> hcat [ptext SLIT("(C_)"), text (show ch)] + | otherwise -> pprHsChar ch + + MachStr s | code_style -> pprFSInCStyle s + | otherwise -> pprHsString s + -- Warning: printing MachStr in code_style assumes it contains + -- only characters '\0'..'\xFF'! - MachStr s | code_style -> doubleQuotes (text (stringToC (_UNPK_ s))) - | otherwise -> text (show (_UNPK_ s)) + MachInt i | code_style && i == tARGET_MIN_INT -> parens (integer (i+1) <> text "-1") + -- Avoid a problem whereby gcc interprets + -- the constant minInt as unsigned. + | otherwise -> pprIntVal i - NoRepStr s | code_style -> pprPanic "NoRep in code style" (ppr lit) - | otherwise -> ptext SLIT("_string_") <+> text (show (_UNPK_ s)) + MachInt64 i | code_style -> pprIntVal i -- Same problem with gcc??? + | otherwise -> ptext SLIT("__int64") <+> integer i - MachInt i _ -> integer i -{- - | code_style && out_of_range - -> pprPanic "" (hsep [text "ERROR: Int ", text (show i), text "out of range", - brackets (ppr range_min <+> text ".." <+> ppr range_max)]) - | otherwise -> integer i + MachWord w | code_style -> pprHexVal w + | otherwise -> ptext SLIT("__word") <+> integer w - where - range_min = if signed then minInt else 0 - range_max = maxInt - out_of_range = not (i >= toInteger range_min && i <= toInteger range_max) --} + MachWord64 w | code_style -> pprHexVal w + | otherwise -> ptext SLIT("__word64") <+> integer w MachFloat f | code_style -> ptext SLIT("(StgFloat)") <> rational f - | otherwise -> ptext SLIT("_float_") <+> rational f + | otherwise -> ptext SLIT("__float") <+> rational f - MachDouble d -> rational d + MachDouble d | iface_style && d < 0 -> parens (rational d) + | otherwise -> rational d MachAddr p | code_style -> ptext SLIT("(void*)") <> integer p - | otherwise -> ptext SLIT("_addr_") <+> integer p + | otherwise -> ptext SLIT("__addr") <+> integer p + + MachLabel l | code_style -> ptext SLIT("(&") <> ptext l <> char ')' + | otherwise -> ptext SLIT("__label") <+> pprHsString l + + MachLitLit s ty | code_style -> ptext s + | otherwise -> parens (hsep [ptext SLIT("__litlit"), + pprHsString s, + pprParendType ty]) + +pprIntVal :: Integer -> SDoc +-- Print negative integers with parens to be sure it's unambiguous +pprIntVal i | i < 0 = parens (integer i) + | otherwise = integer i + +pprHexVal :: Integer -> SDoc +-- Print in C hex format: 0x13fa +pprHexVal 0 = ptext SLIT("0x0") +pprHexVal w = ptext SLIT("0x") <> go w + where + go 0 = empty + go w = go quot <> dig + where + (quot,rem) = w `quotRem` 16 + dig | rem < 10 = char (chr (fromInteger rem + ord '0')) + | otherwise = char (chr (fromInteger rem - 10 + ord 'a')) +\end{code} - NoRepInteger i _ | code_style -> pprPanic "NoRep in code style" (ppr lit) - | otherwise -> ptext SLIT("_integer_") <+> integer i - NoRepRational r _ | code_style -> pprPanic "NoRep in code style" (ppr lit) - | otherwise -> hsep [ptext SLIT("_rational_"), integer (numerator r), - integer (denominator r)] +%************************************************************************ +%* * +\subsection{Hashing} +%* * +%************************************************************************ - MachLitLit s k | code_style -> ptext s - | otherwise -> hsep [ptext SLIT("_litlit_"), ppPrimRep k, text (show (_UNPK_ s))] +Hash values should be zero or a positive integer. No negatives please. +(They mess up the UniqFM for some reason.) -showLiteral :: Literal -> String -showLiteral lit = showSDoc (ppr lit) +\begin{code} +hashLiteral :: Literal -> Int +hashLiteral (MachChar c) = c + 1000 -- Keep it out of range of common ints +hashLiteral (MachStr s) = hashFS s +hashLiteral (MachAddr i) = hashInteger i +hashLiteral (MachInt i) = hashInteger i +hashLiteral (MachInt64 i) = hashInteger i +hashLiteral (MachWord i) = hashInteger i +hashLiteral (MachWord64 i) = hashInteger i +hashLiteral (MachFloat r) = hashRational r +hashLiteral (MachDouble r) = hashRational r +hashLiteral (MachLabel s) = hashFS s +hashLiteral (MachLitLit s _) = hashFS s + +hashRational :: Rational -> Int +hashRational r = hashInteger (numerator r) + +hashInteger :: Integer -> Int +hashInteger i = 1 + abs (fromInteger (i `rem` 10000)) + -- The 1+ is to avoid zero, which is a Bad Number + -- since we use * to combine hash values + +hashFS :: FAST_STRING -> Int +hashFS s = iBox (uniqueOfFS s) \end{code} -