Use UnboxedInt32 and UnboxedUint32 representation for LoadIndexedInstr

runtime/vm/intermediate_language_arm64.cc

Index: runtime/vm/intermediate_language_arm64.cc
diff --git a/runtime/vm/intermediate_language_arm64.cc b/runtime/vm/intermediate_language_arm64.cc
index 651371dfa933af4185a4d75942a4857217684b28..3c27c59fa14df7a196e18232a3deabbfacb6f04b 100644
--- a/runtime/vm/intermediate_language_arm64.cc
+++ b/runtime/vm/intermediate_language_arm64.cc
@@ -993,9 +993,11 @@ Representation LoadIndexedInstr::representation() const {
     case kTypedDataUint16ArrayCid:
     case kOneByteStringCid:
     case kTwoByteStringCid:
+      return kTagged;
     case kTypedDataInt32ArrayCid:
+      return kUnboxedInt32;
     case kTypedDataUint32ArrayCid:
-      return kTagged;
+      return kUnboxedUint32;
     case kTypedDataFloat32ArrayCid:
     case kTypedDataFloat64ArrayCid:
       return kUnboxedDouble;
@@ -1093,6 +1095,25 @@ void LoadIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
     return;
   }
 
+  if ((representation() == kUnboxedInt32) ||
+      (representation() == kUnboxedUint32)) {
+    const Register result = locs()->out(0).reg();
+    switch (class_id()) {
+      case kTypedDataInt32ArrayCid:
+        ASSERT(representation() == kUnboxedInt32);
+        __ ldr(result, element_address, kWord);
+        break;
+      case kTypedDataUint32ArrayCid:
+        ASSERT(representation() == kUnboxedUint32);
+        __ ldr(result, element_address, kUnsignedWord);
+        break;
+      default:
+        UNREACHABLE();
+      }
+    return;
+  }
+
+  ASSERT(representation() == kTagged);
   const Register result = locs()->out(0).reg();
   switch (class_id()) {
     case kTypedDataInt8ArrayCid:
@@ -1118,14 +1139,6 @@ void LoadIndexedInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
       __ ldr(result, element_address, kUnsignedHalfword);
       __ SmiTag(result);
       break;
-    case kTypedDataInt32ArrayCid:
-      __ ldr(result, element_address, kWord);
-      __ SmiTag(result);
-      break;
-    case kTypedDataUint32ArrayCid:
-      __ ldr(result, element_address, kUnsignedWord);
-      __ SmiTag(result);
-      break;
     default:
       ASSERT((class_id() == kArrayCid) || (class_id() == kImmutableArrayCid));
       __ ldr(result, element_address);
@@ -4387,7 +4400,23 @@ void UnaryDoubleOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
 }
 
 
-DEFINE_UNIMPLEMENTED_INSTRUCTION(Int32ToDoubleInstr)
+LocationSummary* Int32ToDoubleInstr::MakeLocationSummary(Isolate* isolate,
+                                                       bool opt) const {
+  const intptr_t kNumInputs = 1;
+  const intptr_t kNumTemps = 0;
+  LocationSummary* result = new(isolate) LocationSummary(
+      isolate, kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  result->set_in(0, Location::RequiresRegister());
+  result->set_out(0, Location::RequiresFpuRegister());
+  return result;
+}
+
+
+void Int32ToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  const Register value = locs()->in(0).reg();
+  const VRegister result = locs()->out(0).fpu_reg();
+  __ scvtfd(result, value);

[Vyacheslav Egorov (Google), 2014/10/03 13:07:41]

Consider using 32-bit version of scvtfd (instruction is the same except 31-bit -
0 for 32-bit version and 1 for 64-bit version). See C3.6.30 in ARMv8 ISA

The reason for using 32-bit version is because truncating conversions right now
leave garbage in the upper bits under certain conditions.

+}
 
 
 LocationSummary* SmiToDoubleInstr::MakeLocationSummary(Isolate* isolate,