Fix a bug in optimising division to shift right

Division by an integral log2 can't be directly optimised to a shift
right, because shift right behaves like a division that rounds to
negative infinity, whereas we want one that rounds to zero.  Fix this
by adding (divisor-1) to the dividend when it is negative before
shifting.  We do this without jumps, generating very slightly worse
code than gcc, which uses conditional moves on CPUs that support it.

diff --git a/compiler/cmm/CmmOpt.hs b/compiler/cmm/CmmOpt.hs
index c454ff4..a23a46f 100644 (file)
--- a/compiler/cmm/CmmOpt.hs
+++ b/compiler/cmm/CmmOpt.hs
@@ -385,14 +385,37 @@ cmmMachOpFold mop args@[x, y@(CmmLit (CmmInt 1 rep))]
 cmmMachOpFold mop args@[x, y@(CmmLit (CmmInt n _))]
   = case mop of
        MO_Mul rep
-           -> case exactLog2 n of
-                 Nothing -> unchanged
-                 Just p  -> CmmMachOp (MO_Shl rep) [x, CmmLit (CmmInt p rep)]
+          | Just p <- exactLog2 n ->
+                 CmmMachOp (MO_Shl rep) [x, CmmLit (CmmInt p rep)]
        MO_S_Quot rep
-           -> case exactLog2 n of
-                 Nothing -> unchanged
-                 Just p  -> CmmMachOp (MO_S_Shr rep) [x, CmmLit (CmmInt p rep)]
-       other 
+          | Just p <- exactLog2 n, 
+            CmmReg _ <- x ->   -- We duplicate x below, hence require
+                               -- it is a reg.  FIXME: remove this restriction.
+               -- shift right is not the same as quot, because it rounds
+               -- to minus infinity, whereasq uot rounds toward zero.
+               -- To fix this up, we add one less than the divisor to the
+               -- dividend if it is a negative number.
+               --
+               -- to avoid a test/jump, we use the following sequence:
+               --      x1 = x >> word_size-1  (all 1s if -ve, all 0s if +ve)
+               --      x2 = y & (divisor-1)
+               --      result = (x+x2) >>= log2(divisor)
+               -- this could be done a bit more simply using conditional moves,
+               -- but we're processor independent here.
+               --
+               -- we optimise the divide by 2 case slightly, generating
+               --      x1 = x >> word_size-1  (unsigned)
+               --      return = (x + x1) >>= log2(divisor)
+               let 
+                   bits = fromIntegral (machRepBitWidth rep) - 1
+                   shr = if p == 1 then MO_U_Shr rep else MO_S_Shr rep
+                   x1 = CmmMachOp shr [x, CmmLit (CmmInt bits rep)]
+                   x2 = if p == 1 then x1 else
+                        CmmMachOp (MO_And rep) [x1, CmmLit (CmmInt (n-1) rep)]
+                   x3 = CmmMachOp (MO_Add rep) [x, x2]
+               in
+                CmmMachOp (MO_S_Shr rep) [x3, CmmLit (CmmInt p rep)]
+       other
            -> unchanged
     where
        unchanged = CmmMachOp mop args