Inline Float32x4 constructors

runtime/vm/intermediate_language_x64.cc

 // Copyright (c) 2013, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_X64.
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/intermediate_language.h"
 
 #include "lib/error.h"
@@ skipping 2892 matching lines @@
       break;
     case MethodRecognizer::kFloat32x4ShuffleW:
       __ shufps(value, value, Immediate(0xFF));
       __ cvtss2sd(value, value);
       break;
     default: UNREACHABLE();
   }
 }
 
 
+LocationSummary* Float32x4ConstructorInstr::MakeLocationSummary() const {
+  const intptr_t kNumInputs = 4;
+  const intptr_t kNumTemps = 0;
+  LocationSummary* summary =
+      new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  summary->set_in(0, Location::RequiresFpuRegister());
+  summary->set_in(1, Location::RequiresFpuRegister());
+  summary->set_in(2, Location::RequiresFpuRegister());
+  summary->set_in(3, Location::RequiresFpuRegister());
+  summary->set_out(Location::SameAsFirstInput());
+  return summary;
+}
+
+
+void Float32x4ConstructorInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  XmmRegister v0 = locs()->in(0).fpu_reg();
+  XmmRegister v1 = locs()->in(1).fpu_reg();
+  XmmRegister v2 = locs()->in(2).fpu_reg();
+  XmmRegister v3 = locs()->in(3).fpu_reg();
+  ASSERT(v0 == locs()->out().fpu_reg());
+  switch (op_kind()) {
+    case MethodRecognizer::kFloat32x4Constructor:
+      __ subq(RSP, Immediate(16));
+      __ cvtsd2ss(v0, v0);
+      __ movss(Address(RSP, -16), v0);
+      __ movsd(v0, v1);
+      __ cvtsd2ss(v0, v0);
+      __ movss(Address(RSP, -12), v0);
+      __ movsd(v0, v2);
+      __ cvtsd2ss(v0, v0);
+      __ movss(Address(RSP, -8), v0);
+      __ movsd(v0, v3);
+      __ cvtsd2ss(v0, v0);
+      __ movss(Address(RSP, -4), v0);
+      __ movups(v0, Address(RSP, -16));
+      __ addq(RSP, Immediate(16));
+    break;
+    default: UNREACHABLE();
+  }
+}
+
+
+LocationSummary* Float32x4ZeroInstr::MakeLocationSummary() const {
+  const intptr_t kNumInputs = 0;
+  const intptr_t kNumTemps = 0;
+  LocationSummary* summary =
+      new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  summary->set_out(Location::RequiresFpuRegister());
+  return summary;
+}
+
+
+void Float32x4ZeroInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  XmmRegister value = locs()->out().fpu_reg();
+  switch (op_kind()) {
+    case MethodRecognizer::kFloat32x4Zero:
+      __ xorps(value, value);
+    break;
+    default: UNREACHABLE();
+  }
+}
+
+
+LocationSummary* Float32x4SplatInstr::MakeLocationSummary() const {
+  const intptr_t kNumInputs = 1;
+  const intptr_t kNumTemps = 0;
+  LocationSummary* summary =
+      new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  summary->set_in(0, Location::RequiresFpuRegister());
+  summary->set_out(Location::SameAsFirstInput());
+  return summary;
+}
+
+
+void Float32x4SplatInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  XmmRegister value = locs()->out().fpu_reg();
+  switch (op_kind()) {
+    case MethodRecognizer::kFloat32x4Splat:
+      // Convert to Float32.
+      __ cvtsd2ss(value, value);
+      // Splat across all lanes.
+      __ shufps(value, value, Immediate(0x00));
+    break;
+    default: UNREACHABLE();
+  }
+}
+
+
 LocationSummary* MathSqrtInstr::MakeLocationSummary() const {
   const intptr_t kNumInputs = 1;
   const intptr_t kNumTemps = 0;
   LocationSummary* summary =
       new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
   summary->set_in(0, Location::RequiresFpuRegister());
   summary->set_out(Location::RequiresFpuRegister());
   return summary;
 }
 
@@ skipping 736 matching lines @@
                          PcDescriptors::kOther,
                          locs());
   __ Drop(2);  // Discard type arguments and receiver.
 }
 
 }  // namespace dart
 
 #undef __
 
 #endif  // defined TARGET_ARCH_X64