A64: Update DoAllocate() to use a W register for the size

src/a64/lithium-codegen-a64.cc

Index: src/a64/lithium-codegen-a64.cc
diff --git a/src/a64/lithium-codegen-a64.cc b/src/a64/lithium-codegen-a64.cc
index 40ef5c6f124501e42e5c4c05735718a84eaa2627..63f540636a28fd6ae1a18821837c05eac6fa1a3b 100644
--- a/src/a64/lithium-codegen-a64.cc
+++ b/src/a64/lithium-codegen-a64.cc
@@ -1505,7 +1505,7 @@ void LCodeGen::DoAllocate(LAllocate* instr) {
     int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
     __ Allocate(size, result, temp1, temp2, deferred->entry(), flags);
   } else {
-    Register size = ToRegister(instr->size());
+    Register size = ToRegister32(instr->size());

[jochen (gone - plz use gerrit), 2014/02/17 16:17:27]

I have to extend this to X() as well

     __ Allocate(size, result, temp1, temp2, deferred->entry(), flags);
   }
 
@@ -1516,7 +1516,7 @@ void LCodeGen::DoAllocate(LAllocate* instr) {
       int32_t size = ToInteger32(LConstantOperand::cast(instr->size()));
       __ Mov(temp1, size - kPointerSize);
     } else {
-      __ Sub(temp1, ToRegister(instr->size()), kPointerSize);
+      __ Sub(temp1, ToRegister32(instr->size()), kPointerSize);

[m.m.capewell, 2014/02/17 15:48:44]

I'm surprised that this builds, as temp1 is X and the result of ToRegister32, W.
Use temp1.W() if you're certain that size > kPointerSize (and thus can't be
negative.)

[jochen (gone - plz use gerrit), 2014/02/17 16:17:27]

size has to be at leat 2*kPointerSize, otherwise the loop below will run havoc
anyway

     }
     __ Sub(result, result, kHeapObjectTag);
 
@@ -1545,7 +1545,8 @@ void LCodeGen::DoDeferredAllocate(LAllocate* instr) {
   if (instr->size()->IsConstantOperand()) {
     __ Mov(size, Operand(ToSmi(LConstantOperand::cast(instr->size()))));
   } else {
-    __ SmiTag(size, ToRegister(instr->size()));
+    __ Mov(x10, ToRegister32(instr->size()));
+    __ SmiTag(size, x10);

[m.m.capewell, 2014/02/17 15:48:44]

I don't think you need a mov here. You can call the X() method on a register to
get its X-sized equivalent, so you could:

Register untagged_size = ToRegister32(instr->size());
__ SmiTag(size, untagged_size.X());

[jochen (gone - plz use gerrit), 2014/02/17 16:17:27]

Because the shift will kill the upper 32bits anyway, makes sense.

   }
   int flags = AllocateDoubleAlignFlag::encode(
       instr->hydrogen()->MustAllocateDoubleAligned());