Inline Float64x2 methods that take 1 argument

runtime/vm/intermediate_language_arm.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_ARM.
 #if defined(TARGET_ARCH_ARM)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/dart_entry.h"
@@ skipping 4075 matching lines @@
       break;
     case MethodRecognizer::kFloat64x2Sqrt:
       __ vsqrtd(dresult0, dvalue0);
       __ vsqrtd(dresult1, dvalue1);
       break;
     default: UNREACHABLE();
   }
 }
 
 
+LocationSummary* Float64x2OneArgInstr::MakeLocationSummary(bool opt) const {
+  const intptr_t kNumInputs = 2;
+  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_out(Location::SameAsFirstInput());
+  return summary;
+}
+
+
+void Float64x2OneArgInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  QRegister left = locs()->in(0).fpu_reg();
+  DRegister left0 = EvenDRegisterOf(left);
+  DRegister left1 = OddDRegisterOf(left);
+  QRegister right = locs()->in(1).fpu_reg();
+  DRegister right0 = EvenDRegisterOf(right);
+  DRegister right1 = OddDRegisterOf(right);
+  QRegister out = locs()->out().fpu_reg();
+  ASSERT(left == out);
+
+  switch (op_kind()) {
+    case MethodRecognizer::kFloat64x2Scale:
+      __ vmuld(left0, left0, right0);
+      __ vmuld(left1, left1, right0);
+      break;
+    case MethodRecognizer::kFloat64x2WithX:
+      __ vmovd(left0, right0);
+      break;
+    case MethodRecognizer::kFloat64x2WithY:
+      __ vmovd(left1, right0);
+      break;
+    case MethodRecognizer::kFloat64x2Min:
+      __ vcmpd(left0, right0);
+      __ vmstat();
+      __ vmovd(left0, right0, GT);

[zra, 2014/03/10 22:14:08]

I double checked the performance guide, and it looks like it recommends
preferring branches instead of conditional FPU instructions. If it's not too
much trouble, why don't we go with that first? Something like:

Label le0;
__ vcmpd(left0, right0);
__ vmstat();
__ b(&le0, LE);
__ vmovd(left0, right0);
__ Bind(&le0)

[Cutch, 2014/03/10 22:26:45]

Done.

+      __ vcmpd(left1, right1);
+      __ vmstat();
+      __ vmovd(left1, right1, GT);
+      break;
+    case MethodRecognizer::kFloat64x2Max:
+      __ vcmpd(left0, right0);
+      __ vmstat();
+      __ vmovd(left0, right0, LT);

[zra, 2014/03/10 22:14:08]

Same here.

[Cutch, 2014/03/10 22:26:45]

Done.

+      __ vcmpd(left1, right1);
+      __ vmstat();
+      __ vmovd(left1, right1, LT);
+      break;
+    default: UNREACHABLE();
+  }
+}
+
+
 LocationSummary* Int32x4BoolConstructorInstr::MakeLocationSummary(
     bool opt) const {
   const intptr_t kNumInputs = 4;
   const intptr_t kNumTemps = 1;
   LocationSummary* summary =
       new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
   summary->set_in(0, Location::RequiresRegister());
   summary->set_in(1, Location::RequiresRegister());
   summary->set_in(2, Location::RequiresRegister());
   summary->set_in(3, Location::RequiresRegister());
@@ skipping 1174 matching lines @@
   compiler->GenerateCall(token_pos(),
                          &label,
                          PcDescriptors::kOther,
                          locs());
   __ Drop(ArgumentCount());  // Discard arguments.
 }
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM