From e4a0fc7d75e896577364e839e19a73bfa028f3fb Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Sat, 30 Mar 2019 18:59:17 +0000
Subject: [PATCH] [X86] Teach isel for RMW binops to handle negate

Negate updates flags like a subtract. We should be able to use the flags from the RMW form of negate when we have (store (X86ISD::SUB 0, load A), A)

Differential Revision: https://reviews.llvm.org/D60007

llvm-svn: 357353
---
 llvm/lib/Target/X86/X86ISelDAGToDAG.cpp | 17 +++++++++++++++--
 llvm/test/CodeGen/X86/fold-rmw-ops.ll   | 20 ++++----------------
 2 files changed, 19 insertions(+), 18 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 444df1a..8d18a1f 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2634,10 +2634,13 @@ bool X86DAGToDAGISel::foldLoadStoreIntoMemOperand(SDNode *Node) {
     return false;
 
   bool IsCommutable = false;
+  bool IsNegate = false;
   switch (Opc) {
   default:
     return false;
   case X86ISD::SUB:
+    IsNegate = isNullConstant(StoredVal.getOperand(0));
+    break;
   case X86ISD::SBB:
     break;
   case X86ISD::ADD:
@@ -2649,7 +2652,7 @@ bool X86DAGToDAGISel::foldLoadStoreIntoMemOperand(SDNode *Node) {
     break;
   }
 
-  unsigned LoadOpNo = 0;
+  unsigned LoadOpNo = IsNegate ? 1 : 0;
   LoadSDNode *LoadNode = nullptr;
   SDValue InputChain;
   if (!isFusableLoadOpStorePattern(StoreNode, StoredVal, CurDAG, LoadOpNo,
@@ -2687,8 +2690,18 @@ bool X86DAGToDAGISel::foldLoadStoreIntoMemOperand(SDNode *Node) {
 
   MachineSDNode *Result;
   switch (Opc) {
-  case X86ISD::ADD:
   case X86ISD::SUB:
+    // Handle negate.
+    if (IsNegate) {
+      unsigned NewOpc = SelectOpcode(X86::NEG64m, X86::NEG32m, X86::NEG16m,
+                                     X86::NEG8m);
+      const SDValue Ops[] = {Base, Scale, Index, Disp, Segment, InputChain};
+      Result = CurDAG->getMachineNode(NewOpc, SDLoc(Node), MVT::i32,
+                                      MVT::Other, Ops);
+      break;
+    }
+   LLVM_FALLTHROUGH;
+  case X86ISD::ADD:
     // Try to match inc/dec.
     if (!Subtarget->slowIncDec() || OptForSize) {
       bool IsOne = isOneConstant(StoredVal.getOperand(1));
diff --git a/llvm/test/CodeGen/X86/fold-rmw-ops.ll b/llvm/test/CodeGen/X86/fold-rmw-ops.ll
index ad8b4e4..87008b1 100644
--- a/llvm/test/CodeGen/X86/fold-rmw-ops.ll
+++ b/llvm/test/CodeGen/X86/fold-rmw-ops.ll
@@ -2434,10 +2434,7 @@ b:
 define void @neg64_br() nounwind {
 ; CHECK-LABEL: neg64_br:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
-; CHECK-NEXT:    subq {{.*}}(%rip), %rax # encoding: [0x48,0x2b,0x05,A,A,A,A]
-; CHECK-NEXT:    # fixup A - offset: 3, value: g64-4, kind: reloc_riprel_4byte_relax_rex
-; CHECK-NEXT:    movq %rax, {{.*}}(%rip) # encoding: [0x48,0x89,0x05,A,A,A,A]
+; CHECK-NEXT:    negq {{.*}}(%rip) # encoding: [0x48,0xf7,0x1d,A,A,A,A]
 ; CHECK-NEXT:    # fixup A - offset: 3, value: g64-4, kind: reloc_riprel_4byte
 ; CHECK-NEXT:    js .LBB76_1 # encoding: [0x78,A]
 ; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB76_1-1, kind: FK_PCRel_1
@@ -2468,10 +2465,7 @@ b:
 define void @neg32_br() nounwind {
 ; CHECK-LABEL: neg32_br:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
-; CHECK-NEXT:    subl {{.*}}(%rip), %eax # encoding: [0x2b,0x05,A,A,A,A]
-; CHECK-NEXT:    # fixup A - offset: 2, value: g32-4, kind: reloc_riprel_4byte
-; CHECK-NEXT:    movl %eax, {{.*}}(%rip) # encoding: [0x89,0x05,A,A,A,A]
+; CHECK-NEXT:    negl {{.*}}(%rip) # encoding: [0xf7,0x1d,A,A,A,A]
 ; CHECK-NEXT:    # fixup A - offset: 2, value: g32-4, kind: reloc_riprel_4byte
 ; CHECK-NEXT:    js .LBB77_1 # encoding: [0x78,A]
 ; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB77_1-1, kind: FK_PCRel_1
@@ -2502,10 +2496,7 @@ b:
 define void @neg16_br() nounwind {
 ; CHECK-LABEL: neg16_br:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
-; CHECK-NEXT:    subw {{.*}}(%rip), %ax # encoding: [0x66,0x2b,0x05,A,A,A,A]
-; CHECK-NEXT:    # fixup A - offset: 3, value: g16-4, kind: reloc_riprel_4byte
-; CHECK-NEXT:    movw %ax, {{.*}}(%rip) # encoding: [0x66,0x89,0x05,A,A,A,A]
+; CHECK-NEXT:    negw {{.*}}(%rip) # encoding: [0x66,0xf7,0x1d,A,A,A,A]
 ; CHECK-NEXT:    # fixup A - offset: 3, value: g16-4, kind: reloc_riprel_4byte
 ; CHECK-NEXT:    js .LBB78_1 # encoding: [0x78,A]
 ; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB78_1-1, kind: FK_PCRel_1
@@ -2536,10 +2527,7 @@ b:
 define void @neg8_br() nounwind {
 ; CHECK-LABEL: neg8_br:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
-; CHECK-NEXT:    subb {{.*}}(%rip), %al # encoding: [0x2a,0x05,A,A,A,A]
-; CHECK-NEXT:    # fixup A - offset: 2, value: g8-4, kind: reloc_riprel_4byte
-; CHECK-NEXT:    movb %al, {{.*}}(%rip) # encoding: [0x88,0x05,A,A,A,A]
+; CHECK-NEXT:    negb {{.*}}(%rip) # encoding: [0xf6,0x1d,A,A,A,A]
 ; CHECK-NEXT:    # fixup A - offset: 2, value: g8-4, kind: reloc_riprel_4byte
 ; CHECK-NEXT:    js .LBB79_1 # encoding: [0x78,A]
 ; CHECK-NEXT:    # fixup A - offset: 1, value: .LBB79_1-1, kind: FK_PCRel_1
-- 
2.7.4