[turbofan] Add backend support for float32 operations.

src/compiler/x64/instruction-selector-x64.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include <algorithm>
 
 #include "src/compiler/instruction-selector-impl.h"
 #include "src/compiler/node-matchers.h"
 #include "src/compiler/node-properties.h"
 
@@ skipping 717 matching lines @@
 }
 
 
 void InstructionSelector::VisitUint32MulHigh(Node* node) {
   VisitMulHigh(this, node, kX64UmulHigh32);
 }
 
 
 void InstructionSelector::VisitChangeFloat32ToFloat64(Node* node) {
   X64OperandGenerator g(this);
-  Emit(kSSECvtss2sd, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
+  Emit(kSSEFloat32ToFloat64, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
 }
 
 
 void InstructionSelector::VisitChangeInt32ToFloat64(Node* node) {
   X64OperandGenerator g(this);
   Emit(kSSEInt32ToFloat64, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
 }
 
 
 void InstructionSelector::VisitChangeUint32ToFloat64(Node* node) {
@@ skipping 52 matching lines @@
     }
     default:
       break;
   }
   Emit(kX64Movl, g.DefineAsRegister(node), g.Use(value));
 }
 
 
 void InstructionSelector::VisitTruncateFloat64ToFloat32(Node* node) {
   X64OperandGenerator g(this);
-  Emit(kSSECvtsd2ss, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
+  Emit(kSSEFloat64ToFloat32, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
 }
 
 
 void InstructionSelector::VisitTruncateInt64ToInt32(Node* node) {
   X64OperandGenerator g(this);
   Node* value = node->InputAt(0);
   if (CanCover(node, value)) {
     switch (value->opcode()) {
       case IrOpcode::kWord64Sar:
       case IrOpcode::kWord64Shr: {
         Int64BinopMatcher m(value);
         if (m.right().Is(32)) {
           Emit(kX64Shr, g.DefineSameAsFirst(node),
                g.UseRegister(m.left().node()), g.TempImmediate(32));
           return;
         }
         break;
       }
       default:
         break;
     }
   }
   Emit(kX64Movl, g.DefineAsRegister(node), g.Use(value));
 }
 
 
+namespace {
+
+void VisitFloatBinop(InstructionSelector* selector, Node* node,
+                     ArchOpcode avx_opcode, ArchOpcode sse_opcode) {
+  X64OperandGenerator g(selector);
+  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
+  InstructionOperand operand1 = g.Use(node->InputAt(1));
+  if (selector->IsSupported(AVX)) {
+    selector->Emit(avx_opcode, g.DefineAsRegister(node), operand0, operand1);
+  } else {
+    selector->Emit(sse_opcode, g.DefineSameAsFirst(node), operand0, operand1);
+  }
+}
+
+}  // namespace
+
+
+void InstructionSelector::VisitFloat32Add(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Add, kSSEFloat32Add);
+}
+
+
+void InstructionSelector::VisitFloat32Sub(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Sub, kSSEFloat32Sub);
+}
+
+
+void InstructionSelector::VisitFloat32Mul(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Mul, kSSEFloat32Mul);
+}
+
+
+void InstructionSelector::VisitFloat32Div(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Div, kSSEFloat32Div);
+}
+
+
+void InstructionSelector::VisitFloat32Max(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Max, kSSEFloat32Max);
+}
+
+
+void InstructionSelector::VisitFloat32Min(Node* node) {
+  VisitFloatBinop(this, node, kAVXFloat32Min, kSSEFloat32Min);
+}
+
+
+void InstructionSelector::VisitFloat32Sqrt(Node* node) {
+  X64OperandGenerator g(this);
+  Emit(kSSEFloat32Sqrt, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
+}
+
+
 void InstructionSelector::VisitFloat64Add(Node* node) {
-  X64OperandGenerator g(this);
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Add, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Add, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Add, kSSEFloat64Add);
 }
 
 
 void InstructionSelector::VisitFloat64Sub(Node* node) {
   X64OperandGenerator g(this);
   Float64BinopMatcher m(node);
   if (m.left().IsMinusZero() && m.right().IsFloat64RoundDown() &&
       CanCover(m.node(), m.right().node())) {
     if (m.right().InputAt(0)->opcode() == IrOpcode::kFloat64Sub &&
         CanCover(m.right().node(), m.right().InputAt(0))) {
       Float64BinopMatcher mright0(m.right().InputAt(0));
       if (mright0.left().IsMinusZero()) {
         Emit(kSSEFloat64Round | MiscField::encode(kRoundUp),
              g.DefineAsRegister(node), g.UseRegister(mright0.right().node()));
         return;
       }
     }
   }
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Sub, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Sub, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Sub, kSSEFloat64Sub);
 }
 
 
 void InstructionSelector::VisitFloat64Mul(Node* node) {
-  X64OperandGenerator g(this);
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Mul, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Mul, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Mul, kSSEFloat64Mul);
 }
 
 
 void InstructionSelector::VisitFloat64Div(Node* node) {
-  X64OperandGenerator g(this);
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Div, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Div, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Div, kSSEFloat64Div);
 }
 
 
 void InstructionSelector::VisitFloat64Mod(Node* node) {
   X64OperandGenerator g(this);
   InstructionOperand temps[] = {g.TempRegister(rax)};
   Emit(kSSEFloat64Mod, g.DefineSameAsFirst(node),
        g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)), 1,
        temps);
 }
 
 
 void InstructionSelector::VisitFloat64Max(Node* node) {
-  X64OperandGenerator g(this);
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Max, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Max, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Max, kSSEFloat64Max);
 }
 
 
 void InstructionSelector::VisitFloat64Min(Node* node) {
-  X64OperandGenerator g(this);
-  if (IsSupported(AVX)) {
-    Emit(kAVXFloat64Min, g.DefineAsRegister(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  } else {
-    Emit(kSSEFloat64Min, g.DefineSameAsFirst(node),
-         g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)));
-  }
+  VisitFloatBinop(this, node, kAVXFloat64Min, kSSEFloat64Min);
 }
 
 
 void InstructionSelector::VisitFloat64Sqrt(Node* node) {
   X64OperandGenerator g(this);
   Emit(kSSEFloat64Sqrt, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
 }
 
 
 namespace {
@@ skipping 161 matching lines @@
 
 
 // Shared routine for comparison with zero.
 void VisitCompareZero(InstructionSelector* selector, Node* node,
                       InstructionCode opcode, FlagsContinuation* cont) {
   X64OperandGenerator g(selector);
   VisitCompare(selector, opcode, g.Use(node), g.TempImmediate(0), cont);
 }
 
 
+// Shared routine for multiple float32 compare operations (inputs commuted).
+void VisitFloat32Compare(InstructionSelector* selector, Node* node,
+                         FlagsContinuation* cont) {
+  Node* const left = node->InputAt(0);
+  Node* const right = node->InputAt(1);
+  InstructionCode const opcode =
+      selector->IsSupported(AVX) ? kAVXFloat32Cmp : kSSEFloat32Cmp;
+  VisitCompare(selector, opcode, right, left, cont, false);
+}
+
+
 // Shared routine for multiple float64 compare operations (inputs commuted).
 void VisitFloat64Compare(InstructionSelector* selector, Node* node,
                          FlagsContinuation* cont) {
   Node* const left = node->InputAt(0);
   Node* const right = node->InputAt(1);
-  VisitCompare(selector, kSSEFloat64Cmp, right, left, cont, false);
+  InstructionCode const opcode =
+      selector->IsSupported(AVX) ? kAVXFloat64Cmp : kSSEFloat64Cmp;
+  VisitCompare(selector, opcode, right, left, cont, false);
 }
 
 }  // namespace
 
 
 void InstructionSelector::VisitBranch(Node* branch, BasicBlock* tbranch,
                                       BasicBlock* fbranch) {
   X64OperandGenerator g(this);
   Node* user = branch;
   Node* value = branch->InputAt(0);
@@ skipping 35 matching lines @@
         return VisitWord64Compare(this, value, &cont);
       case IrOpcode::kInt64LessThan:
         cont.OverwriteAndNegateIfEqual(kSignedLessThan);
         return VisitWord64Compare(this, value, &cont);
       case IrOpcode::kInt64LessThanOrEqual:
         cont.OverwriteAndNegateIfEqual(kSignedLessThanOrEqual);
         return VisitWord64Compare(this, value, &cont);
       case IrOpcode::kUint64LessThan:
         cont.OverwriteAndNegateIfEqual(kUnsignedLessThan);
         return VisitWord64Compare(this, value, &cont);
+      case IrOpcode::kFloat32Equal:
+        cont.OverwriteAndNegateIfEqual(kUnorderedEqual);
+        return VisitFloat32Compare(this, value, &cont);
+      case IrOpcode::kFloat32LessThan:
+        cont.OverwriteAndNegateIfEqual(kUnsignedGreaterThan);
+        return VisitFloat32Compare(this, value, &cont);
+      case IrOpcode::kFloat32LessThanOrEqual:
+        cont.OverwriteAndNegateIfEqual(kUnsignedGreaterThanOrEqual);
+        return VisitFloat32Compare(this, value, &cont);
       case IrOpcode::kFloat64Equal:
         cont.OverwriteAndNegateIfEqual(kUnorderedEqual);
         return VisitFloat64Compare(this, value, &cont);
       case IrOpcode::kFloat64LessThan:
         cont.OverwriteAndNegateIfEqual(kUnsignedGreaterThan);
         return VisitFloat64Compare(this, value, &cont);
       case IrOpcode::kFloat64LessThanOrEqual:
         cont.OverwriteAndNegateIfEqual(kUnsignedGreaterThanOrEqual);
         return VisitFloat64Compare(this, value, &cont);
       case IrOpcode::kProjection:
@@ skipping 197 matching lines @@
   VisitWord64Compare(this, node, &cont);
 }
 
 
 void InstructionSelector::VisitUint64LessThan(Node* node) {
   FlagsContinuation cont(kUnsignedLessThan, node);
   VisitWord64Compare(this, node, &cont);
 }
 
 
+void InstructionSelector::VisitFloat32Equal(Node* node) {
+  FlagsContinuation cont(kUnorderedEqual, node);
+  VisitFloat32Compare(this, node, &cont);
+}
+
+
+void InstructionSelector::VisitFloat32LessThan(Node* node) {
+  FlagsContinuation cont(kUnsignedGreaterThan, node);
+  VisitFloat32Compare(this, node, &cont);
+}
+
+
+void InstructionSelector::VisitFloat32LessThanOrEqual(Node* node) {
+  FlagsContinuation cont(kUnsignedGreaterThanOrEqual, node);
+  VisitFloat32Compare(this, node, &cont);
+}
+
+
 void InstructionSelector::VisitFloat64Equal(Node* node) {
   FlagsContinuation cont(kUnorderedEqual, node);
   VisitFloat64Compare(this, node, &cont);
 }
 
 
 void InstructionSelector::VisitFloat64LessThan(Node* node) {
   FlagsContinuation cont(kUnsignedGreaterThan, node);
   VisitFloat64Compare(this, node, &cont);
 }
@@ skipping 39 matching lines @@
   Node* right = node->InputAt(1);
   Emit(kSSEFloat64InsertHighWord32, g.DefineSameAsFirst(node),
        g.UseRegister(left), g.Use(right));
 }
 
 
 // static
 MachineOperatorBuilder::Flags
 InstructionSelector::SupportedMachineOperatorFlags() {
   MachineOperatorBuilder::Flags flags =
+      MachineOperatorBuilder::kFloat32Max |
+      MachineOperatorBuilder::kFloat32Min |
       MachineOperatorBuilder::kFloat64Max |
       MachineOperatorBuilder::kFloat64Min |
       MachineOperatorBuilder::kWord32ShiftIsSafe;
   if (CpuFeatures::IsSupported(SSE4_1)) {
     flags |= MachineOperatorBuilder::kFloat64RoundDown |
              MachineOperatorBuilder::kFloat64RoundTruncate;
   }
   return flags;
 }
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8