Fix parallel move code for swapping SIMD registers on ARM and ARM64.

runtime/vm/flow_graph_compiler_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/flow_graph_compiler.h"
 
 #include "vm/ast_printer.h"
@@ skipping 1672 matching lines @@
   } else if (source.IsRegister() && destination.IsStackSlot()) {
     Exchange(source.reg(),
              destination.base_reg(), destination.ToStackSlotOffset());
   } else if (source.IsStackSlot() && destination.IsRegister()) {
     Exchange(destination.reg(),
              source.base_reg(), source.ToStackSlotOffset());
   } else if (source.IsStackSlot() && destination.IsStackSlot()) {
     Exchange(source.base_reg(), source.ToStackSlotOffset(),
              destination.base_reg(), destination.ToStackSlotOffset());
   } else if (source.IsFpuRegister() && destination.IsFpuRegister()) {
-    const DRegister dst = EvenDRegisterOf(destination.fpu_reg());
-    DRegister src = EvenDRegisterOf(source.fpu_reg());
-    __ vmovd(DTMP, src);
-    __ vmovd(src, dst);
-    __ vmovd(dst, DTMP);
+    if (TargetCPUFeatures::neon_supported()) {
+      QRegister dst = destination.fpu_reg();

[zerny-google, 2014/12/01 15:23:02]

Nit: should we keep const here and below?

[Florian Schneider, 2014/12/01 15:24:40]

Done.

+      QRegister src = source.fpu_reg();
+      __ vmovq(QTMP, src);
+      __ vmovq(src, dst);
+      __ vmovq(dst, QTMP);
+    } else {
+      DRegister dst = EvenDRegisterOf(destination.fpu_reg());
+      DRegister src = EvenDRegisterOf(source.fpu_reg());
+      __ vmovd(DTMP, src);
+      __ vmovd(src, dst);
+      __ vmovd(dst, DTMP);
+    }
   } else if (source.IsFpuRegister() || destination.IsFpuRegister()) {
     ASSERT(destination.IsDoubleStackSlot() ||
            destination.IsQuadStackSlot() ||
            source.IsDoubleStackSlot() ||
            source.IsQuadStackSlot());
     bool double_width = destination.IsDoubleStackSlot() ||
                         source.IsDoubleStackSlot();
     QRegister qreg = source.IsFpuRegister() ? source.fpu_reg()
                                             : destination.fpu_reg();
     DRegister reg = EvenDRegisterOf(qreg);
@@ skipping 124 matching lines @@
   DRegister dreg = EvenDRegisterOf(reg);
   __ vldrd(dreg, Address(SP, kDoubleSize, Address::PostIndex));
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM