X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Fcompiler%2FbasicTypes%2FLiteral.lhs;h=e83ea9db74e006ea9b6e9305f8678d7c648074e9;hb=9d7da331989abcd1844e9d03b8d1e4163796fa85;hp=eeddb56823648ab3217f5f2ce6017c70dbc861cd;hpb=9dd6e1c216993624a2cd74b62ca0f0569c02c26b;p=ghc-hetmet.git diff --git a/ghc/compiler/basicTypes/Literal.lhs b/ghc/compiler/basicTypes/Literal.lhs index eeddb56..e83ea9d 100644 --- a/ghc/compiler/basicTypes/Literal.lhs +++ b/ghc/compiler/basicTypes/Literal.lhs @@ -1,36 +1,81 @@ % -% (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, mkMachWord, - literalType, literalPrimRep, - showLiteral, - isNoRepLit, isLitLitLit - ) where +module Literal + ( Literal(..) -- Exported to ParseIface + , mkMachInt, mkMachWord + , mkMachInt64, mkMachWord64, mkStringLit + , litSize + , litIsDupable, litIsTrivial + , literalType + , 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 ) - --- others: import Type ( Type ) -import CStrings ( stringToC, charToC, charToEasyHaskell ) -import TysWiredIn ( stringTy ) import Outputable -import Util ( thenCmp ) +import FastTypes +import FastString +import Binary + +import Ratio ( numerator ) +import FastString ( uniqueOfFS, lengthFS ) +import DATA_INT ( Int8, Int16, Int32 ) +import DATA_WORD ( Word8, Word16, Word32 ) +import Char ( ord, chr ) +\end{code} + + + +%************************************************************************ +%* * +\subsection{Sizes} +%* * +%************************************************************************ -import GlaExts ( (<#) ) +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} @@ -46,68 +91,94 @@ function applications, etc., etc., has not yet been done. \begin{code} data Literal - = MachChar Char - | MachStr FAST_STRING + = ------------------ + -- First the primitive guys + MachChar Char -- Char# At least 31 bits + + | MachStr FastString -- A string-literal: stored and emitted + -- UTF-8 encoded, we'll arrange to decode it + -- at runtime. Also emitted with a '\0' + -- terminator. - | MachAddr Integer -- whatever this machine thinks is a "pointer" + | MachNullAddr -- the NULL pointer, the only pointer value + -- that can be represented as a Literal. - | MachInt Integer -- for the numeric types, these are - 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 + -- 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 FastString -- always an Addr# + (Maybe Int) -- the size (in bytes) of the arguments + -- the label expects. Only applicable with + -- 'stdcall' labels. + -- Just x => "@" will be appended to label + -- name when emitting asm. +\end{code} - | 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. +Binary instance - -- 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. +\begin{code} +instance Binary Literal where + put_ bh (MachChar aa) = do putByte bh 0; put_ bh aa + put_ bh (MachStr ab) = do putByte bh 1; put_ bh ab + put_ bh (MachNullAddr) = do putByte bh 2 + put_ bh (MachInt ad) = do putByte bh 3; put_ bh ad + put_ bh (MachInt64 ae) = do putByte bh 4; put_ bh ae + put_ bh (MachWord af) = do putByte bh 5; put_ bh af + put_ bh (MachWord64 ag) = do putByte bh 6; put_ bh ag + put_ bh (MachFloat ah) = do putByte bh 7; put_ bh ah + put_ bh (MachDouble ai) = do putByte bh 8; put_ bh ai + put_ bh (MachLabel aj mb) = do putByte bh 9; put_ bh aj ; put_ bh mb + get bh = do + h <- getByte bh + case h of + 0 -> do + aa <- get bh + return (MachChar aa) + 1 -> do + ab <- get bh + return (MachStr ab) + 2 -> do + return (MachNullAddr) + 3 -> do + ad <- get bh + return (MachInt ad) + 4 -> do + ae <- get bh + return (MachInt64 ae) + 5 -> do + af <- get bh + return (MachWord af) + 6 -> do + ag <- get bh + return (MachWord64 ag) + 7 -> do + ah <- get bh + return (MachFloat ah) + 8 -> do + ai <- get bh + return (MachDouble ai) + 9 -> do + aj <- get bh + mb <- get bh + return (MachLabel aj mb) +\end{code} -mkMachInt, mkMachWord :: Integer -> Literal +\begin{code} +instance Outputable Literal where + ppr lit = pprLit lit -mkMachInt x = MachInt x True{-signed-} -mkMachWord x = MachInt x False{-unsigned-} +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 } @@ -120,103 +191,215 @@ instance Ord Literal where compare a b = cmpLit a b \end{code} -\begin{code} -isNoRepLit (NoRepStr _) = True -- these are not primitive typed! -isNoRepLit (NoRepInteger _ _) = True -isNoRepLit (NoRepRational _ _) = True -isNoRepLit _ = False -isLitLitLit (MachLitLit _ _) = True -isLitLitLit _ = False + Construction + ~~~~~~~~~~~~ +\begin{code} +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 + +mkStringLit :: String -> Literal +mkStringLit s = MachStr (mkFastString s) -- stored UTF-8 encoded + +inIntRange, inWordRange :: Integer -> Bool +inIntRange x = x >= tARGET_MIN_INT && x <= tARGET_MAX_INT +inWordRange x = x >= 0 && x <= tARGET_MAX_WORD + +inCharRange :: Char -> Bool +inCharRange c = c >= '\0' && c <= chr 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} -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 +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 (ord c)) +int2CharLit (MachInt i) = MachChar (chr (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 = MachNullAddr \end{code} + Predicates + ~~~~~~~~~~ \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 (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 +litIsTrivial :: Literal -> Bool +-- True if there is absolutely no penalty to duplicating the literal +-- c.f. CoreUtils.exprIsTrivial +-- False principally of strings +litIsTrivial (MachStr _) = False +litIsTrivial other = True + +litIsDupable :: Literal -> Bool +-- True if code space does not go bad if we duplicate this literal +-- c.f. CoreUtils.exprIsDupable +-- Currently we treat it just like litIsTrivial +litIsDupable (MachStr _) = False +litIsDupable other = True + +litSize :: Literal -> Int +-- Used by CoreUnfold.sizeExpr +litSize (MachStr str) = 1 + ((lengthFS str + 3) `div` 4) + -- Every literal has size at least 1, otherwise + -- f "x" + -- might be too small + -- [Sept03: make literal strings a bit bigger to avoid fruitless + -- duplication of little strings] +litSize _other = 1 \end{code} -The boring old output stuff: + Types + ~~~~~ \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) - -instance Outputable Literal where - ppr lit = pprLit lit - -pprLit lit - = getPprStyle $ \ sty -> - let - code_style = codeStyle 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, '\''] - - MachStr s | code_style -> doubleQuotes (text (stringToC (_UNPK_ s))) - | otherwise -> text (show (_UNPK_ s)) - - NoRepStr s | code_style -> pprPanic "NoRep in code style" (ppr lit) - | otherwise -> ptext SLIT("_string_") <+> text (show (_UNPK_ s)) - - MachInt i signed | 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 +literalType :: Literal -> Type +literalType MachNullAddr = addrPrimTy +literalType (MachChar _) = charPrimTy +literalType (MachStr _) = addrPrimTy +literalType (MachInt _) = intPrimTy +literalType (MachWord _) = wordPrimTy +literalType (MachInt64 _) = int64PrimTy +literalType (MachWord64 _) = word64PrimTy +literalType (MachFloat _) = floatPrimTy +literalType (MachDouble _) = doublePrimTy +literalType (MachLabel _ _) = addrPrimTy +\end{code} - where - range_min = if signed then minInt else 0 - range_max = maxInt - out_of_range = not (i >= toInteger range_min && i <= toInteger range_max) - MachFloat f | code_style -> ptext SLIT("(StgFloat)") <> rational f - | otherwise -> ptext SLIT("_float_") <+> rational f + Comparison + ~~~~~~~~~~ +\begin{code} +cmpLit (MachChar a) (MachChar b) = a `compare` b +cmpLit (MachStr a) (MachStr b) = a `compare` b +cmpLit (MachNullAddr) (MachNullAddr) = EQ +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 lit1 lit2 | litTag lit1 <# litTag lit2 = LT + | otherwise = GT + +litTag (MachChar _) = _ILIT(1) +litTag (MachStr _) = _ILIT(2) +litTag (MachNullAddr) = _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) +\end{code} - MachDouble d -> rational d + Printing + ~~~~~~~~ +* MachX (i.e. unboxed) things are printed unadornded (e.g. 3, 'a', "foo") + exceptions: MachFloat gets an initial keyword prefix. - MachAddr p | code_style -> ptext SLIT("(void*)") <> integer p - | otherwise -> ptext SLIT("_addr_") <+> integer p +\begin{code} +pprLit (MachChar ch) = pprHsChar ch +pprLit (MachStr s) = pprHsString s +pprLit (MachInt i) = pprIntVal i +pprLit (MachInt64 i) = ptext SLIT("__int64") <+> integer i +pprLit (MachWord w) = ptext SLIT("__word") <+> integer w +pprLit (MachWord64 w) = ptext SLIT("__word64") <+> integer w +pprLit (MachFloat f) = ptext SLIT("__float") <+> rational f +pprLit (MachDouble d) = rational d +pprLit (MachNullAddr) = ptext SLIT("__NULL") +pprLit (MachLabel l mb) = ptext SLIT("__label") <+> + case mb of + Nothing -> pprHsString l + Just x -> doubleQuotes (text (unpackFS l ++ '@':show x)) + +pprIntVal :: Integer -> SDoc +-- Print negative integers with parens to be sure it's unambiguous +pprIntVal i | i < 0 = parens (integer i) + | otherwise = integer i +\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) = ord c + 1000 -- Keep it out of range of common ints +hashLiteral (MachStr s) = hashFS s +hashLiteral (MachNullAddr) = 0 +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 + +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 :: FastString -> Int +hashFS s = iBox (uniqueOfFS s) \end{code} -