%
-% (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(..) -- Exported to ParseIface
+ , mkMachInt, mkMachWord
+ , mkMachInt64, mkMachWord64
+ , isLitLitLit
+ , literalType, literalPrimRep
+ , hashLiteral
+
+ , inIntRange, inWordRange
+
+ , word2IntLit, int2WordLit, char2IntLit, int2CharLit
+ , float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit
+ , addr2IntLit, int2AddrLit, float2DoubleLit, double2FloatLit
+ ) where
+
#include "HsVersions.h"
-module Literal (
- Literal(..),
-
- mkMachInt, mkMachWord,
- literalType, literalPrimRep,
- showLiteral,
- isNoRepLit, isLitLitLit
- ) where
-
-IMP_Ubiq(){-uitous-}
-IMPORT_1_3(Ratio)
-
--- friends:
-import PrimRep ( PrimRep(..) ) -- non-abstract
-import TysPrim ( getPrimRepInfo,
- addrPrimTy, intPrimTy, floatPrimTy,
- doublePrimTy, charPrimTy, wordPrimTy )
-
--- others:
-import CStrings ( stringToC, charToC, charToEasyHaskell )
-import TysWiredIn ( stringTy )
-import Pretty -- pretty-printing stuff
-import PprStyle ( PprStyle(..), codeStyle, ifaceStyle )
-import Util ( thenCmp, panic, pprPanic )
+import TysPrim ( charPrimTy, addrPrimTy, floatPrimTy, doublePrimTy,
+ intPrimTy, wordPrimTy, int64PrimTy, word64PrimTy
+ )
+import Name ( hashName )
+import PrimRep ( PrimRep(..) )
+import TyCon ( isNewTyCon )
+import Type ( Type, typePrimRep )
+import PprType ( pprParendType )
+import Demand ( Demand )
+import CStrings ( charToC, charToEasyHaskell, pprFSInCStyle )
+
+import Outputable
+import Util ( thenCmp )
+
+import Ratio ( numerator, denominator )
+import FastString ( uniqueOfFS )
+import Char ( ord, chr )
+
+#if __GLASGOW_HASKELL__ >= 404
+import GlaExts ( fromInt )
+#endif
+\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
\end{code}
+
+
+%************************************************************************
+%* *
+\subsection{Literals}
+%* *
+%************************************************************************
So-called @Literals@ are {\em either}:
\begin{itemize}
\begin{code}
data Literal
- = MachChar Char
+ = ------------------
+ -- First the primitive guys
+ MachChar Char
| MachStr FAST_STRING
- | MachAddr Integer -- whatever this machine thinks is a "pointer"
+ | MachAddr Integer -- Whatever this machine thinks is a "pointer"
- | MachInt Integer -- for the numeric types, these are
- Bool -- True <=> signed (Int#); False <=> unsigned (Word#)
+ | 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
- | 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-}
-
-instance Ord3 Literal where
- cmp (MachChar a) (MachChar b) = a `tcmp` b
- cmp (MachStr a) (MachStr b) = a `tcmp` b
- cmp (MachAddr a) (MachAddr b) = a `tcmp` b
- cmp (MachInt a b) (MachInt c d) = (a `tcmp` c) `thenCmp` (b `tcmp` d)
- cmp (MachFloat a) (MachFloat b) = a `tcmp` b
- cmp (MachDouble a) (MachDouble b) = a `tcmp` b
- cmp (MachLitLit a b) (MachLitLit c d) = (a `tcmp` c) `thenCmp` (b `tcmp` d)
- cmp (NoRepStr a) (NoRepStr b) = a `tcmp` b
- cmp (NoRepInteger a _) (NoRepInteger b _) = a `tcmp` b
- cmp (NoRepRational a _) (NoRepRational b _) = a `tcmp` b
-
- -- now we *know* the tags are different, so...
- cmp 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)
-
-tcmp x y = case _tagCmp x y of { _LT -> LT_; _EQ -> EQ_; GT__ -> GT_ }
+ | MachLitLit FAST_STRING Type -- Type might be Add# 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 `cmp` b) of { EQ_ -> True; _ -> False }
- a /= b = case (a `cmp` b) of { EQ_ -> False; _ -> True }
+ 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 `cmp` b) of { LT_ -> True; EQ_ -> True; GT__ -> False }
- a < b = case (a `cmp` b) of { LT_ -> True; EQ_ -> False; GT__ -> False }
- a >= b = case (a `cmp` b) of { LT_ -> False; EQ_ -> True; GT__ -> True }
- a > b = case (a `cmp` b) of { LT_ -> False; EQ_ -> False; GT__ -> True }
- _tagCmp a b = case (a `cmp` b) of { LT_ -> _LT; EQ_ -> _EQ; GT__ -> _GT }
+ 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
+
+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?
+
+inIntRange, inWordRange :: Integer -> Bool
+inIntRange x = x >= tARGET_MIN_INT && x <= tARGET_MAX_INT
+inWordRange x = x >= 0 && x <= tARGET_MAX_WORD
\end{code}
+ Coercions
+ ~~~~~~~~~
\begin{code}
-isNoRepLit (NoRepStr _) = True -- these are not primitive typed!
-isNoRepLit (NoRepInteger _ _) = True
-isNoRepLit (NoRepRational _ _) = True
-isNoRepLit _ = False
+word2IntLit, int2WordLit, char2IntLit, int2CharLit,
+ float2IntLit, int2FloatLit, double2IntLit, int2DoubleLit,
+ addr2IntLit, int2AddrLit, float2DoubleLit, double2FloatLit :: Literal -> Literal
+
+word2IntLit (MachWord w)
+ | w > tARGET_MAX_INT = MachInt ((-1) + tARGET_MAX_WORD - w)
+ | otherwise = MachInt w
+
+int2WordLit (MachInt i)
+ | i < 0 = MachWord (1 + tARGET_MAX_WORD + i) -- (-1) ---> tARGET_MAX_WORD
+ | otherwise = MachWord i
+
+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 (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
\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 (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 (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 (MachLitLit a b) (MachLitLit c d) = (a `compare` c) `thenCmp` (b `compare` d)
+cmpLit lit1 lit2 | litTag lit1 _LT_ 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 (MachLitLit _ _) = ILIT(10)
+\end{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)
+ Printing
+ ~~~~~~~~
+* MachX (i.e. unboxed) things are printed unadornded (e.g. 3, 'a', "foo")
+ exceptions: MachFloat and MachAddr get an initial keyword prefix
-instance Outputable Literal where
- ppr sty (MachChar ch)
- = let
- char_encoding
- = case sty of
- PprForC -> charToC ch
- PprForAsm _ _ -> charToC ch
- PprInterface -> charToEasyHaskell ch
- _ -> [ch]
- in
- ppBesides [ppCast sty SLIT("(C_)"), ppChar '\'', ppStr char_encoding, ppChar '\'']
-
- ppr sty (MachStr s)
- | codeStyle sty = ppBesides [ppChar '"', ppStr (stringToC (_UNPK_ s)), ppChar '"']
- | otherwise = ppStr (show (_UNPK_ s))
-
- ppr sty lit@(NoRepStr s)
- | codeStyle sty = pprPanic "NoRep in code style" (ppr PprDebug lit)
- | otherwise = ppBesides [ppStr "_string_", ppStr (show (_UNPK_ s))]
-
- ppr sty (MachInt i signed)
- | codeStyle sty && out_of_range
- = panic ("ERROR: Int " ++ show i ++ " out of range [" ++
- show range_min ++ " .. " ++ show range_max ++ "]\n")
-
- | otherwise = ppInteger i
-
- where
- range_min = if signed then minInt else 0
- range_max = maxInt
- out_of_range = not (i >= toInteger range_min && i <= toInteger range_max)
-
- ppr sty (MachFloat f)
- | codeStyle sty = ppBesides [ppCast sty SLIT("(StgFloat)"), ppRational f]
- | otherwise = ppBesides [ppStr "_float_", ppRational f]
-
- ppr sty (MachDouble d) = ppRational d
-
- ppr sty (MachAddr p)
- | codeStyle sty = ppBesides [ppCast sty SLIT("(void*)"), ppInteger p]
- | otherwise = ppBesides [ppStr "_addr_", ppInteger p]
-
- ppr sty lit@(NoRepInteger i _)
- | codeStyle sty = pprPanic "NoRep in code style" (ppr PprDebug lit)
- | otherwise = ppCat [ppStr "_integer_", ppInteger i]
-
- ppr sty lit@(NoRepRational r _)
- | codeStyle sty = pprPanic "NoRep in code style" (ppr PprDebug lit)
- | otherwise = ppCat [ppStr "_rational_", ppInteger (numerator r), ppInteger (denominator r)]
-
- ppr sty (MachLitLit s k)
- | codeStyle sty = ppPStr s
- | otherwise = ppBesides [ppStr "_litlit_", ppStr (show (_UNPK_ s))]
-
-showLiteral :: PprStyle -> Literal -> String
-showLiteral sty lit = ppShow 80 (ppr sty lit)
+\begin{code}
+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 -> pprFSInCStyle s
+ | otherwise -> pprFSAsString 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
+
+ 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 | ifaceStyle sty && d < 0 -> parens (rational d)
+ | otherwise -> rational d
+
+ MachAddr p | code_style -> ptext SLIT("(void*)") <> integer p
+ | otherwise -> ptext SLIT("__addr") <+> integer p
+
+ MachLitLit s ty | code_style -> ptext s
+ | otherwise -> parens (hsep [ptext SLIT("__litlit"),
+ pprFSAsString 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) = ord 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 (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}