module Literal
( Literal(..) -- Exported to ParseIface
, mkMachInt, mkMachWord
- , mkMachInt64, mkMachWord64
- , isLitLitLit, maybeLitLit, litSize, litIsDupable,
- , literalType, literalPrimRep
+ , mkMachInt64, mkMachWord64, mkStringLit
+ , litSize
+ , litIsDupable, litIsTrivial
+ , literalType
, hashLiteral
, inIntRange, inWordRange, tARGET_MAX_INT, inCharRange
- , isZeroLit,
+ , isZeroLit
, word2IntLit, int2WordLit
, narrow8IntLit, narrow16IntLit, narrow32IntLit
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 )
-
+import Type ( Type )
import Outputable
import FastTypes
import FastString
import Binary
-import Util ( thenCmp )
+import UnicodeUtil ( stringToUtf8 )
import Ratio ( numerator )
import FastString ( uniqueOfFS, lengthFS )
-import Int ( Int8, Int16, Int32 )
-import Word ( Word8, Word16, Word32 )
+import DATA_INT ( Int8, Int16, Int32 )
+import DATA_WORD ( Word8, Word16, Word32 )
import Char ( ord, chr )
\end{code}
data Literal
= ------------------
-- First the primitive guys
- MachChar Int -- Char# At least 31 bits
+ MachChar Char -- Char# At least 31 bits
| MachStr FastString
- | 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 -- Int# At least WORD_SIZE_IN_BITS bits
| MachInt64 Integer -- Int64# At least 64 bits
-- string argument is the name of a symbol. This literal
-- refers to the *address* of the label.
| MachLabel FastString -- 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 FastString Type -- Type might be Addr# or Int# etc
+ (Maybe Int) -- the size (in bytes) of the arguments
+ -- the label expects. Only applicable with
+ -- 'stdcall' labels.
+ -- Just x => "@<x>" will be appended to label
+ -- name when emitting asm.
\end{code}
-Binary instance: must do this manually, because we don't want the type
-arg of MachLitLit involved.
+Binary instance
\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 (MachAddr ac) = do putByte bh 2; put_ bh ac
+ 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) = do putByte bh 9; put_ bh aj
- put_ bh (MachLitLit ak _) = do putByte bh 10; put_ bh ak
+ put_ bh (MachLabel aj mb) = do putByte bh 9; put_ bh aj ; put_ bh mb
get bh = do
h <- getByte bh
case h of
ab <- get bh
return (MachStr ab)
2 -> do
- ac <- get bh
- return (MachAddr ac)
+ return (MachNullAddr)
3 -> do
ad <- get bh
return (MachInt ad)
return (MachDouble ai)
9 -> do
aj <- get bh
- return (MachLabel aj)
- 10 -> do
- ak <- get bh
- return (MachLitLit ak (error "MachLitLit: no type"))
+ mb <- get bh
+ return (MachLabel aj mb)
\end{code}
\begin{code}
mkMachInt64 x = MachInt64 x
mkMachWord64 x = MachWord64 x
+mkStringLit :: String -> Literal
+mkStringLit s = MachStr (mkFastString (stringToUtf8 s))
+
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
+inCharRange :: Char -> Bool
+inCharRange c = c >= '\0' && c <= chr tARGET_MAX_CHAR
isZeroLit :: Literal -> Bool
isZeroLit (MachInt 0) = True
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)
+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)
double2FloatLit (MachDouble d) = MachFloat d
nullAddrLit :: Literal
-nullAddrLit = MachAddr 0
+nullAddrLit = MachNullAddr
\end{code}
Predicates
~~~~~~~~~~
\begin{code}
-isLitLitLit (MachLitLit _ _) = True
-isLitLitLit _ = False
-
-maybeLitLit (MachLitLit s t) = Just (s,t)
-maybeLitLit _ = Nothing
+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
- -- False principally of strings
+-- 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) = lengthFS str `div` 4
+-- 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}
~~~~~
\begin{code}
literalType :: Literal -> Type
-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 _) = IntRep
-literalPrimRep (MachWord _) = WordRep
-literalPrimRep (MachInt64 _) = Int64Rep
-literalPrimRep (MachWord64 _) = Word64Rep
-literalPrimRep (MachFloat _) = FloatRep
-literalPrimRep (MachDouble _) = DoubleRep
-literalPrimRep (MachLabel _) = AddrRep
-literalPrimRep (MachLitLit _ ty) = typePrimRep ty
+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}
\begin{code}
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 (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 (MachLitLit a b) (MachLitLit c d) = (a `compare` c) `thenCmp` (b `tcCmpType` d)
+cmpLit (MachLabel a _) (MachLabel b _) = a `compare` b
cmpLit lit1 lit2 | litTag lit1 <# litTag lit2 = LT
- | otherwise = GT
+ | otherwise = GT
litTag (MachChar _) = _ILIT(1)
litTag (MachStr _) = _ILIT(2)
-litTag (MachAddr _) = _ILIT(3)
+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)
-litTag (MachLitLit _ _) = _ILIT(11)
+litTag (MachLabel _ _) = _ILIT(10)
\end{code}
Printing
~~~~~~~~
* MachX (i.e. unboxed) things are printed unadornded (e.g. 3, 'a', "foo")
- exceptions: MachFloat and MachAddr get an initial keyword prefix
+ exceptions: MachFloat gets an initial keyword prefix.
\begin{code}
-pprLit lit
- = getPprStyle $ \ sty ->
- let
- code_style = codeStyle 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)") <> code_rational f
- | otherwise -> ptext SLIT("__float") <+> rational f
-
- MachDouble d | code_style -> code_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("(&") <> ftext l <> char ')'
- | otherwise -> ptext SLIT("__label") <+> pprHsString l
-
- MachLitLit s ty | code_style -> ftext s
- | otherwise -> parens (hsep [ptext SLIT("__litlit"),
- pprHsString s,
- pprParendType ty])
-
--- negative floating literals in code style need parentheses to avoid
--- interacting with surrounding syntax.
-code_rational d | d < 0 = parens (rational d)
- | otherwise = rational d
+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
-
-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}
\begin{code}
hashLiteral :: Literal -> Int
-hashLiteral (MachChar c) = c + 1000 -- Keep it out of range of common ints
+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 (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
-hashLiteral (MachLitLit s _) = hashFS s
+hashLiteral (MachLabel s _) = hashFS s
hashRational :: Rational -> Int
hashRational r = hashInteger (numerator r)