%
-% (c) The GRASP/AQUA Project, Glasgow University, 1992-1995
+% (c) The GRASP/AQUA Project, Glasgow University, 1998
%
\section[Literal]{@Literal@: Machine literals (unboxed, of course)}
\begin{code}
+module Literal
+ ( Literal(..) -- Exported to ParseIface
+ , mkMachInt, mkMachWord
+ , mkMachInt64, mkMachWord64
+ , isLitLitLit, maybeLitLit, litSize, litIsDupable,
+ , literalType, literalPrimRep
+ , hashLiteral
+
+ , inIntRange, inWordRange, tARGET_MAX_INT, inCharRange
+
+ , word2IntLit, int2WordLit
+ , intToInt8Lit, intToInt16Lit, intToInt32Lit
+ , wordToWord8Lit, wordToWord16Lit, wordToWord32Lit
+ , char2IntLit, int2CharLit
+ , float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit
+ , addr2IntLit, int2AddrLit, float2DoubleLit, double2FloatLit
+ ) where
+
#include "HsVersions.h"
-module Literal (
- Literal(..),
+import TysPrim ( charPrimTy, addrPrimTy, floatPrimTy, doublePrimTy,
+ intPrimTy, wordPrimTy, int64PrimTy, word64PrimTy
+ )
+import PrimRep ( PrimRep(..) )
+import Type ( Type, typePrimRep )
+import PprType ( pprParendType )
+import CStrings ( pprFSInCStyle )
+
+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}
+
- mkMachInt, mkMachWord,
- literalType, literalPrimRep,
- showLiteral,
- isNoRepLit, isLitLitLit
- -- and to make the interface self-sufficient....
- ) where
+%************************************************************************
+%* *
+\subsection{Sizes}
+%* *
+%************************************************************************
-import Ubiq{-uitous-}
+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).
--- friends:
-import PrimRep ( PrimRep(..) ) -- non-abstract
-import TysPrim ( getPrimRepInfo,
- addrPrimTy, intPrimTy, floatPrimTy,
- doublePrimTy, charPrimTy, wordPrimTy )
+\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
--- others:
-import CStrings ( stringToC, charToC, charToEasyHaskell )
-import TysWiredIn ( integerTy, rationalTy, stringTy )
-import Pretty -- pretty-printing stuff
-import PprStyle ( PprStyle(..), codeStyle )
-import Util ( panic )
+tARGET_MAX_CHAR :: Int
+tARGET_MAX_CHAR = 0x10ffff
\end{code}
+
+
+%************************************************************************
+%* *
+\subsection{Literals}
+%* *
+%************************************************************************
So-called @Literals@ are {\em either}:
\begin{itemize}
\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"
+
+ | MachInt Integer -- Int# At least 32 bits
+ | MachInt64 Integer -- Int64# At least 64 bits
+ | MachWord Integer -- Word# At least 32 bits
+ | MachWord64 Integer -- Word64# At least 64 bits
+
| MachFloat Rational
| MachDouble Rational
- | MachLitLit FAST_STRING
- PrimRep
- | NoRepStr FAST_STRING -- the uncommitted ones
- | NoRepInteger Integer
- | NoRepRational Rational
+ -- 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)
+
+instance Eq Literal where
+ a == b = case (a `compare` b) of { EQ -> True; _ -> False }
+ a /= b = case (a `compare` b) of { EQ -> False; _ -> True }
+
+instance Ord Literal where
+ a <= b = case (a `compare` b) of { LT -> True; EQ -> True; GT -> False }
+ a < b = case (a `compare` b) of { LT -> True; EQ -> False; GT -> False }
+ a >= b = case (a `compare` b) of { LT -> False; EQ -> True; GT -> True }
+ a > b = case (a `compare` b) of { LT -> False; EQ -> False; GT -> True }
+ compare a b = cmpLit a b
+\end{code}
+
+
+ Construction
+ ~~~~~~~~~~~~
+\begin{code}
+mkMachInt, mkMachWord, mkMachInt64, mkMachWord64 :: Integer -> Literal
- deriving (Eq, Ord)
- -- 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 x = ASSERT2( inIntRange x, integer x ) MachInt x
+mkMachWord x = ASSERT2( inWordRange x, integer x ) MachWord x
+mkMachInt64 x = MachInt64 x -- Assertions?
+mkMachWord64 x = MachWord64 x -- Ditto?
-mkMachInt, mkMachWord :: Integer -> Literal
+inIntRange, inWordRange :: Integer -> Bool
+inIntRange x = x >= tARGET_MIN_INT && x <= tARGET_MAX_INT
+inWordRange x = x >= 0 && x <= tARGET_MAX_WORD
-mkMachInt x = MachInt x True{-signed-}
-mkMachWord x = MachInt x False{-unsigned-}
+inCharRange :: Int -> Bool
+inCharRange c = c >= 0 && c <= tARGET_MAX_CHAR
\end{code}
+ Coercions
+ ~~~~~~~~~
\begin{code}
-isNoRepLit (NoRepStr _) = True -- these are not primitive typed!
-isNoRepLit (NoRepInteger _) = True
-isNoRepLit (NoRepRational _) = True
-isNoRepLit _ = False
+word2IntLit, int2WordLit,
+ intToInt8Lit, intToInt16Lit, intToInt32Lit,
+ wordToWord8Lit, wordToWord16Lit, wordToWord32Lit,
+ char2IntLit, int2CharLit,
+ float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit,
+ addr2IntLit, int2AddrLit, 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
+
+intToInt8Lit (MachInt i) = MachInt (toInteger (fromInteger i :: Int8))
+intToInt16Lit (MachInt i) = MachInt (toInteger (fromInteger i :: Int16))
+intToInt32Lit (MachInt i) = MachInt (toInteger (fromInteger i :: Int32))
+wordToWord8Lit (MachWord w) = MachWord (toInteger (fromInteger w :: Word8))
+wordToWord16Lit (MachWord w) = MachWord (toInteger (fromInteger w :: Word16))
+wordToWord32Lit (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 (MachFloat f) = MachInt (truncate f)
+int2DoubleLit (MachInt i) = MachDouble (fromInteger i)
+
+addr2IntLit (MachAddr a) = MachInt a
+int2AddrLit (MachInt i) = MachAddr i
+
+float2DoubleLit (MachFloat f) = MachDouble f
+double2FloatLit (MachDouble d) = MachFloat d
+\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 _) = integerTy
-literalType (NoRepRational _)= rationalTy
-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 (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}
-ppCast :: PprStyle -> FAST_STRING -> Pretty
-ppCast PprForC cast = ppPStr cast
-ppCast _ _ = ppNil
+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 `compare` 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 sty (MachChar ch)
- = let
- char_encoding
- = case sty of
- PprForC -> charToC ch
- PprForAsm _ _ -> charToC ch
- PprUnfolding -> charToEasyHaskell ch
- _ -> [ch]
- in
- ppBeside (ppBesides [ppCast sty SLIT("(C_)"), ppChar '\'', ppStr char_encoding, ppChar '\''])
- (if_ubxd sty)
-
- ppr sty (MachStr s)
- = ppBeside (if codeStyle sty
- then ppBesides [ppChar '"', ppStr (stringToC (_UNPK_ s)), ppChar '"']
- else ppStr (show (_UNPK_ s)))
- (if_ubxd sty)
-
- ppr sty (MachAddr p) = ppBesides [ppCast sty SLIT("(void*)"), ppInteger p, if_ubxd sty]
- ppr sty (MachInt i signed)
- | codeStyle sty
- && ((signed && (i >= toInteger minInt && i <= toInteger maxInt))
- || (not signed && (i >= toInteger 0 && i <= toInteger maxInt)))
- -- ToDo: Think about these ranges!
- = ppBesides [ppInteger i, if_ubxd sty]
-
- | not (codeStyle sty) -- we'd prefer the code to the error message
- = ppBesides [ppInteger i, if_ubxd sty]
-
- | otherwise
- = error ("ERROR: Int " ++ show i ++ " out of range [" ++
- show range_min ++ " .. " ++ show maxInt ++ "]\n")
- where
- range_min = if signed then minInt else 0
-
- ppr sty (MachFloat f) = ppBesides [ppCast sty SLIT("(StgFloat)"), ppRational f, if_ubxd sty]
- ppr sty (MachDouble d) = ppBesides [ppRational d, if_ubxd sty, if_ubxd sty]
-
- ppr sty (NoRepInteger i)
- | codeStyle sty = ppInteger i
- | ufStyle sty = ppCat [ppStr "_NOREP_I_", ppInteger i]
- | otherwise = ppBesides [ppInteger i, ppChar 'I']
-
- ppr sty (NoRepRational r)
- | ufStyle sty = ppCat [ppStr "_NOREP_R_", ppInteger (numerator r), ppInteger (denominator r)]
- | codeStyle sty = panic "ppr.ForC.NoRepRational"
- | otherwise = ppBesides [ppRational r, ppChar 'R']
-
- ppr sty (NoRepStr s)
- | codeStyle sty = ppBesides [ppStr (show (_UNPK_ s))]
- | ufStyle sty = ppCat [ppStr "_NOREP_S_", ppStr (show (_UNPK_ s))]
- | otherwise = ppBesides [ppStr (show (_UNPK_ s)), ppChar 'S']
-
- ppr sty (MachLitLit s k)
- | codeStyle sty = ppPStr s
- | ufStyle sty = ppBesides [ppStr "``", ppPStr s, ppStr "'' _K_ ", ppr sty k]
- | otherwise = ppBesides [ppStr "``", ppPStr s, ppStr "''"]
-
-ufStyle PprUnfolding = True
-ufStyle _ = False
-
-if_ubxd sty = if codeStyle sty then ppNil else ppChar '#'
-
-showLiteral :: PprStyle -> Literal -> String
-
-showLiteral sty lit = ppShow 80 (ppr sty 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
+ iface_style = ifaceStyle sty
+ in
+ case lit of
+ 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'!
+
+ 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
+
+ MachInt64 i | code_style -> pprIntVal i -- Same problem with gcc???
+ | otherwise -> ptext SLIT("__int64") <+> integer i
+
+ MachWord w | code_style -> pprHexVal w
+ | otherwise -> ptext SLIT("__word") <+> integer w
+
+ 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
+
+ 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
+
+ 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}
+
+
+%************************************************************************
+%* *
+\subsection{Hashing}
+%* *
+%************************************************************************
+
+Hash values should be zero or a positive integer. No negatives please.
+(They mess up the UniqFM for some reason.)
+
+\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}
projects / ghc-hetmet.git / blobdiff