[PATCH] InstCombine: fsub nsz 0, X ==> fsub nsz -0.0, X

[Sanjay Patel]

Hi hfinkel, majnemer, resistor,

Some day the backend may handle instruction-level fast math flags and make this transform unnecessary, but it's still better practice to use the canonical representation of fneg when possible (use a -0.0).

This is a partial fix for PR20870 ( http://llvm.org/bugs/show_bug.cgi?id=20870 ). See also http://reviews.llvm.org/D6723.

http://reviews.llvm.org/D6731

Files:
  lib/Transforms/InstCombine/InstCombineAddSub.cpp
  test/Transforms/InstCombine/fast-math.ll

Index: lib/Transforms/InstCombine/InstCombineAddSub.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1762,6 +1762,14 @@
                                   TLI, DT, AT))
     return ReplaceInstUsesWith(I, V);
 
+  // fsub nsz 0, X ==> fsub nsz -0.0, X
+  if (I.getFastMathFlags().noSignedZeros() && match(Op0, m_Zero())) {
+    // Subtraction from -0.0 is the canonical form of fneg.
+    Instruction *NewI = BinaryOperator::CreateFNeg(Op1);
+    NewI->copyFastMathFlags(&I);
+    return NewI;
+  }
+
   if (isa<Constant>(Op0))
     if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
       if (Instruction *NV = FoldOpIntoSelect(I, SI))
Index: test/Transforms/InstCombine/fast-math.ll
===================================================================
--- test/Transforms/InstCombine/fast-math.ll
+++ test/Transforms/InstCombine/fast-math.ll
@@ -93,7 +93,7 @@
   ret float %t3
 
 ; CHECK-LABEL: @fold9(
-; CHECK: fsub fast float 0.000000e+00, %f2
+; CHECK: fsub fast float -0.000000e+00, %f2
 }
 
 ; Let C3 = C1 + C2. (f1 + C1) + (f2 + C2) => (f1 + f2) + C3 instead of
@@ -322,6 +322,14 @@
 ; CHECK: fmul float %f1, %f2
 }
 
+define float @fneg2(float %x) {
+  %sub = fsub nsz float 0.0, %x
+  ret float %sub
+; CHECK-LABEL: @fneg2(
+; CHECK-NEXT: fsub nsz float -0.000000e+00, %x
+; CHECK-NEXT: ret float 
+}
+
 ; =========================================================================
 ;
 ;   Testing-cases about div

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D6731.17473.patch
Type: text/x-patch
Size: 1552 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20141218/7b16a6a1/attachment.bin>