[project @ 2000-07-16 21:10:48 by panne]

This commit tries to fix the discrepancies between the results of
floating point calculations during runtime and compile time, see
e.g. fptools/ghc/tests/numeric/should_run/arith008.hs. Part of the
story was the fact that floating point values are represented as
Rationals in GHC and therefore never lost precision, at least for the
operations for which constant folding is done. To compensate for this,
before and after floating point operations the operands are
temporarily converted to/from Float/Double. This is wrong, because
host architecture and target architecture are confused this way, but
in a non-cross-compiling scenario it works.

diff --git a/ghc/compiler/prelude/PrelRules.lhs b/ghc/compiler/prelude/PrelRules.lhs
index 801095e..0b54318 100644 (file)
--- a/ghc/compiler/prelude/PrelRules.lhs
+++ b/ghc/compiler/prelude/PrelRules.lhs
@@ -3,6 +3,10 @@
 %
 \section[ConFold]{Constant Folder}
 
+Conceptually, constant folding should be parameterized with the kind
+of target machine to get identical behaviour during compilation time
+and runtime. We cheat a little bit here...
+
 ToDo:
    check boundaries before folding, e.g. we can fold the Float addition
    (i1 + i2) only if it results        in a valid Float.
@@ -33,6 +37,7 @@ import Unique         ( unpackCStringFoldrIdKey, hasKey )
 import Bits            ( Bits(..) )
 import Word            ( Word64 )
 import Outputable
+import CmdLineOpts      ( opt_SimplStrictFP )
 \end{code}
 
 
@@ -286,21 +291,31 @@ or_rule r1 r2 args = case r1 args of
                   Nothing    -> r2 args
 
 twoLits :: (Literal -> Literal -> Maybe (RuleName, CoreExpr)) -> RuleFun
-twoLits rule [Lit l1, Lit l2] = rule l1 l2
+twoLits rule [Lit l1, Lit l2] = rule (convFloating l1) (convFloating l2)
 twoLits rule other           = Nothing
 
 oneLit :: (Literal -> Maybe (RuleName, CoreExpr)) -> RuleFun
-oneLit rule [Lit l1] = rule l1
+oneLit rule [Lit l1] = rule (convFloating l1)
 oneLit rule other    = Nothing
 
+-- When we strictfp is requested, cut down the precision of the Rational value
+-- to that of Float/Double. We confuse host architecture and target architecture
+-- here, but it's convenient (and wrong :-).
+convFloating :: Literal -> Literal
+convFloating (MachFloat  f) | opt_SimplStrictFP =
+   MachFloat  (toRational ((fromRational f) :: Float ))
+convFloating (MachDouble d) | opt_SimplStrictFP =
+   MachDouble (toRational ((fromRational d) :: Double))
+convFloating l = l
+
 
 trueVal       = Var trueDataConId
 falseVal      = Var falseDataConId
 mkIntVal    i = Lit (mkMachInt  i)
 mkWordVal   w = Lit (mkMachWord w)
 mkCharVal   c = Lit (MachChar   c)
-mkFloatVal  f = Lit (MachFloat  f)
-mkDoubleVal d = Lit (MachDouble d)
+mkFloatVal  f = Lit (convFloating (MachFloat  f))
+mkDoubleVal d = Lit (convFloating (MachDouble d))
 \end{code}