From bb5cba3cca070e0a7eb85ef25c513cd6b86ddd01 Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Tue, 26 Mar 2019 20:54:15 +0000
Subject: [PATCH] [SDAG] add simplifications for FP at node creation time

We have the folds for fadd/fsub/fmul already in DAGCombiner,
so it may be possible to remove that code if we can guarantee that
these ops are zapped before they can exist.

llvm-svn: 357029
---
 llvm/include/llvm/CodeGen/SelectionDAG.h       |  4 ++++
 llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 27 ++++++++++++++++++++++++++
 llvm/test/CodeGen/X86/extract-fp.ll            |  3 +--
 3 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/llvm/include/llvm/CodeGen/SelectionDAG.h b/llvm/include/llvm/CodeGen/SelectionDAG.h
index 548c28f..aac3e0f 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAG.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAG.h
@@ -976,6 +976,10 @@ public:
   /// Try to simplify a shift into 1 of its operands or a constant.
   SDValue simplifyShift(SDValue X, SDValue Y);
 
+  /// Try to simplify a floating-point binary operation into 1 of its operands
+  /// or a constant.
+  SDValue simplifyFPBinop(unsigned Opcode, SDValue X, SDValue Y);
+
   /// VAArg produces a result and token chain, and takes a pointer
   /// and a source value as input.
   SDValue getVAArg(EVT VT, const SDLoc &dl, SDValue Chain, SDValue Ptr,
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index e9ef415..9534704 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -4931,6 +4931,8 @@ SDValue SelectionDAG::getNode(unsigned Opcode, const SDLoc &DL, EVT VT,
     assert(VT.isFloatingPoint() && "This operator only applies to FP types!");
     assert(N1.getValueType() == N2.getValueType() &&
            N1.getValueType() == VT && "Binary operator types must match!");
+    if (SDValue V = simplifyFPBinop(Opcode, N1, N2))
+      return V;
     break;
   case ISD::FCOPYSIGN:   // N1 and result must match.  N1/N2 need not match.
     assert(N1.getValueType() == VT &&
@@ -7053,6 +7055,31 @@ SDValue SelectionDAG::simplifyShift(SDValue X, SDValue Y) {
   return SDValue();
 }
 
+// TODO: Use fast-math-flags to enable more simplifications.
+SDValue SelectionDAG::simplifyFPBinop(unsigned Opcode, SDValue X, SDValue Y) {
+  ConstantFPSDNode *YC = isConstOrConstSplatFP(Y, /* AllowUndefs */ true);
+  if (!YC)
+    return SDValue();
+
+  // X + -0.0 --> X
+  if (Opcode == ISD::FADD)
+    if (YC->getValueAPF().isNegZero())
+      return X;
+
+  // X - +0.0 --> X
+  if (Opcode == ISD::FSUB)
+    if (YC->getValueAPF().isPosZero())
+      return X;
+
+  // X * 1.0 --> X
+  // X / 1.0 --> X
+  if (Opcode == ISD::FMUL || Opcode == ISD::FDIV)
+    if (YC->getValueAPF().isExactlyValue(1.0))
+      return X;
+
+  return SDValue();
+}
+
 SDValue SelectionDAG::getVAArg(EVT VT, const SDLoc &dl, SDValue Chain,
                                SDValue Ptr, SDValue SV, unsigned Align) {
   SDValue Ops[] = { Chain, Ptr, SV, getTargetConstant(Align, dl, MVT::i32) };
diff --git a/llvm/test/CodeGen/X86/extract-fp.ll b/llvm/test/CodeGen/X86/extract-fp.ll
index 06ba30b..5632450 100644
--- a/llvm/test/CodeGen/X86/extract-fp.ll
+++ b/llvm/test/CodeGen/X86/extract-fp.ll
@@ -36,12 +36,11 @@ define float @ext_fmul_v4f32(<4 x float> %x) {
   ret float %ext
 }
 
-; TODO: X / 1.0 --> X
+; X / 1.0 --> X
 
 define float @ext_fdiv_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: ext_fdiv_v4f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    divss {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %bo = fdiv <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 42.0>
   %ext = extractelement <4 x float> %bo, i32 0
-- 
2.7.4