[llvm] a253a2a - [SDAG] fold fsub -0.0, undef to undef rather than NaN

[Sanjay Patel via llvm-commits]

Author: Sanjay Patel
Date: 2020-02-23T11:36:53-05:00
New Revision: a253a2a793cda34d1f6421ee9b7ca76a03fdfc59

URL: https://github.com/llvm/llvm-project/commit/a253a2a793cda34d1f6421ee9b7ca76a03fdfc59
DIFF: https://github.com/llvm/llvm-project/commit/a253a2a793cda34d1f6421ee9b7ca76a03fdfc59.diff

LOG: [SDAG] fold fsub -0.0, undef to undef rather than NaN

A question about this behavior came up on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2020-February/139003.html
...and as part of backend improvements in D73978.

We decided not to implement a more general change that would have
folded any FP binop with nearly arbitrary constant + undef operand
to undef because that is not theoretically correct (even if it is
practically correct).

This is the SDAG-equivalent to the IR change in D74713.

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
    llvm/test/CodeGen/X86/vec_fneg.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index e809816d68be..d2db4bccc3ac 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5112,8 +5112,13 @@ SDValue SelectionDAG::foldConstantFPMath(unsigned Opcode, const SDLoc &DL,
   }
 
   switch (Opcode) {
-  case ISD::FADD:
   case ISD::FSUB:
+    // -0.0 - undef --> undef (consistent with "fneg undef")
+    if (N1CFP && N1CFP->getValueAPF().isNegZero() && N2.isUndef())
+      return getUNDEF(VT);
+    LLVM_FALLTHROUGH;
+
+  case ISD::FADD:
   case ISD::FMUL:
   case ISD::FDIV:
   case ISD::FREM:

diff  --git a/llvm/test/CodeGen/X86/vec_fneg.ll b/llvm/test/CodeGen/X86/vec_fneg.ll
index 4d5539feef3c..c3c1932c2311 100644
--- a/llvm/test/CodeGen/X86/vec_fneg.ll
+++ b/llvm/test/CodeGen/X86/vec_fneg.ll
@@ -76,12 +76,10 @@ define <4 x float> @fneg_undef(<4 x float> %Q) nounwind {
 define <4 x float> @fsub_neg0_undef_elts_undef(<4 x float> %x) {
 ; X32-SSE-LABEL: fsub_neg0_undef_elts_undef:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movaps {{.*#+}} xmm0 = <NaN,u,u,NaN>
 ; X32-SSE-NEXT:    retl
 ;
 ; X64-SSE-LABEL: fsub_neg0_undef_elts_undef:
 ; X64-SSE:       # %bb.0:
-; X64-SSE-NEXT:    movaps {{.*#+}} xmm0 = <NaN,u,u,NaN>
 ; X64-SSE-NEXT:    retq
   %r = fsub <4 x float> <float -0.0, float undef, float undef, float -0.0>, undef
   ret <4 x float> %r