[PATCH] Reassociate x + -0.1234 * y into x - 0.1234 * y

[Chad Rosier]

LGTM.  Let me know if you need me to push the commit.

 Chad

================
Comment at: lib/Transforms/Scalar/Reassociate.cpp:1986
@@ +1985,3 @@
+    // Transform:
+    //   FAdd(x, FMul(-0.1234, y)) -> FSub(x, FMul(0.1234, y))
+    // where the FMul can also be an FDiv, and FAdd can be a FSub.
----------------
We can be more generic here:
FAdd(x, FMul(-ConstantFP, y)) -> FSub(x, FMul(ConstantFP, y))

I actually prefer

// Reassociate: x + -ConstantFP * y -> x - ConstantFP * y

but I'll let you decide.

http://reviews.llvm.org/D4904