Use (more) xor on FP register to load it with 0.0.

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 1591 matching lines @@
       __ LoadMultipleDFromOffset(dtmp0, 2, FP, source_offset);
       __ StoreMultipleDToOffset(dtmp0, 2, FP, dest_offset);
     }
   } else {
     ASSERT(source.IsConstant());
     const Object& constant = source.constant();
     if (destination.IsRegister()) {
       __ LoadObject(destination.reg(), constant);
     } else if (destination.IsFpuRegister()) {
       const DRegister dst = EvenDRegisterOf(destination.fpu_reg());
-      __ LoadObject(TMP, constant);
-      __ AddImmediate(TMP, TMP, Double::value_offset() - kHeapObjectTag);
-      __ vldrd(dst, Address(TMP, 0));
+      if (Utils::DoublesBitEqual(Double::Cast(constant).value(), 0.0) &&
+          TargetCPUFeatures::neon_supported()) {
+        QRegister qdst = destination.fpu_reg();
+        __ veorq(qdst, qdst, qdst);

[regis, 2014/07/29 21:43:17]

Why not veord?
Mmm, it looks like Zach did not implement it. I guess we only allocate Q
registers and veord is probably not faster than veorq.

+      } else {
+        __ LoadObject(TMP, constant);
+        __ AddImmediate(TMP, TMP, Double::value_offset() - kHeapObjectTag);
+        __ vldrd(dst, Address(TMP, 0));
+      }
     } else if (destination.IsDoubleStackSlot()) {
+      if (Utils::DoublesBitEqual(Double::Cast(constant).value(), 0.0) &&
+          TargetCPUFeatures::neon_supported()) {
+        __ veorq(QTMP, QTMP, QTMP);
+      } else {
+        __ LoadObject(TMP, constant);
+        __ AddImmediate(TMP, TMP, Double::value_offset() - kHeapObjectTag);
+        __ vldrd(DTMP, Address(TMP, 0));
+      }
       const intptr_t dest_offset = destination.ToStackSlotOffset();
-      __ LoadObject(TMP, constant);
-      __ AddImmediate(TMP, TMP, Double::value_offset() - kHeapObjectTag);
-      __ vldrd(DTMP, Address(TMP, 0));
       __ StoreDToOffset(DTMP, FP, dest_offset);
     } else {
       ASSERT(destination.IsStackSlot());
       const intptr_t dest_offset = destination.ToStackSlotOffset();
       __ LoadObject(TMP, constant);
       __ StoreToOffset(kWord, TMP, FP, dest_offset);
     }
   }
 
   move->Eliminate();
@@ skipping 142 matching lines @@
   DRegister dreg = EvenDRegisterOf(reg);
   __ vldrd(dreg, Address(SP, kDoubleSize, Address::PostIndex));
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM