From f807f6a05fbaa06a2a37cf01dc1df7dc357648e9 Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Mon, 5 Dec 2016 15:45:27 +0000
Subject: [PATCH] [x86] fold fand (fxor X, -1) Y --> fandn X, Y

I noticed this gap in the scalar FP-logic matching with:
D26712
and:
rL287171

Differential Revision: https://reviews.llvm.org/D27385

llvm-svn: 288675
---
 llvm/lib/Target/X86/X86ISelLowering.cpp   | 31 +++++++++++++++++++++++++++++++
 llvm/test/CodeGen/X86/fp-logic-replace.ll |  9 +++------
 2 files changed, 34 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6b23b61..128dbff 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -31764,6 +31764,34 @@ static SDValue getNullFPConstForNullVal(SDValue V, SelectionDAG &DAG,
   return V;
 }
 
+static SDValue combineFAndFNotToFAndn(SDNode *N, SelectionDAG &DAG,
+                                      const X86Subtarget &Subtarget) {
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+  EVT VT = N->getValueType(0);
+  SDLoc DL(N);
+
+  // Vector types are handled in combineANDXORWithAllOnesIntoANDNP().
+  if (!((VT == MVT::f32 && Subtarget.hasSSE1()) ||
+        (VT == MVT::f64 && Subtarget.hasSSE2())))
+    return SDValue();
+
+  auto isAllOnesConstantFP = [](SDValue V) {
+    auto *C = dyn_cast<ConstantFPSDNode>(V);
+    return C && C->getConstantFPValue()->isAllOnesValue();
+  };
+
+  // fand (fxor X, -1), Y --> fandn X, Y
+  if (N0.getOpcode() == X86ISD::FXOR && isAllOnesConstantFP(N0.getOperand(1)))
+    return DAG.getNode(X86ISD::FANDN, DL, VT, N0.getOperand(0), N1);
+
+  // fand X, (fxor Y, -1) --> fandn Y, X
+  if (N1.getOpcode() == X86ISD::FXOR && isAllOnesConstantFP(N1.getOperand(1)))
+    return DAG.getNode(X86ISD::FANDN, DL, VT, N1.getOperand(0), N0);
+
+  return SDValue();
+}
+
 /// Do target-specific dag combines on X86ISD::FAND nodes.
 static SDValue combineFAnd(SDNode *N, SelectionDAG &DAG,
                            const X86Subtarget &Subtarget) {
@@ -31775,6 +31803,9 @@ static SDValue combineFAnd(SDNode *N, SelectionDAG &DAG,
   if (SDValue V = getNullFPConstForNullVal(N->getOperand(1), DAG, Subtarget))
     return V;
 
+  if (SDValue V = combineFAndFNotToFAndn(N, DAG, Subtarget))
+    return V;
+
   return lowerX86FPLogicOp(N, DAG, Subtarget);
 }
 
diff --git a/llvm/test/CodeGen/X86/fp-logic-replace.ll b/llvm/test/CodeGen/X86/fp-logic-replace.ll
index 50e2c1b..0a233fd 100644
--- a/llvm/test/CodeGen/X86/fp-logic-replace.ll
+++ b/llvm/test/CodeGen/X86/fp-logic-replace.ll
@@ -29,16 +29,13 @@ define double @FsANDPSrr(double %x, double %y) {
 define double @FsANDNPSrr(double %x, double %y) {
 ; SSE-LABEL: FsANDNPSrr:
 ; SSE:       # BB#0:
-; SSE-NEXT:    movsd {{.*#+}} xmm2 = mem[0],zero
-; SSE-NEXT:    xorpd %xmm1, %xmm2
-; SSE-NEXT:    andpd %xmm2, %xmm0
+; SSE-NEXT:    andnps %xmm0, %xmm1
+; SSE-NEXT:    movaps %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: FsANDNPSrr:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; AVX-NEXT:    vxorpd %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vandpd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vandnps %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
   %bc1 = bitcast double %x to i64
-- 
2.7.4