Unboxed Float64x2 fields and some misc fixes

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 1669 matching lines @@
 
     if (is_initialization_) {
       const Class* cls = NULL;
       switch (cid) {
         case kDoubleCid:
           cls = &compiler->double_class();
           break;
         case kFloat32x4Cid:
           cls = &compiler->float32x4_class();
           break;
+        case kFloat64x2Cid:
+          cls = &compiler->float64x2_class();
+          break;
         default:
           UNREACHABLE();
       }
 
       StoreInstanceFieldSlowPath* slow_path =
           new StoreInstanceFieldSlowPath(this, *cls);
       compiler->AddSlowPathCode(slow_path);
 
       __ TryAllocate(*cls,
                      slow_path->entry_label(),
@@ skipping 12 matching lines @@
         __ Comment("UnboxedDoubleStoreInstanceFieldInstr");
         __ StoreDToOffset(value, temp, Double::value_offset() - kHeapObjectTag);
         break;
       case kFloat32x4Cid:
         __ Comment("UnboxedFloat32x4StoreInstanceFieldInstr");
         __ StoreDToOffset(value, temp,
             Float32x4::value_offset() - kHeapObjectTag);
         __ StoreDToOffset(value_odd, temp,
             Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
         break;
+      case kFloat64x2Cid:
+        __ Comment("UnboxedFloat64x2StoreInstanceFieldInstr");
+        __ StoreDToOffset(value, temp,
+            Float64x2::value_offset() - kHeapObjectTag);
+        __ StoreDToOffset(value_odd, temp,
+            Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+        break;
       default:
         UNREACHABLE();
     }
 
     return;
   }
 
   if (IsPotentialUnboxedStore()) {
     Register value_reg = locs()->in(1).reg();
     Register temp = locs()->temp(0).reg();
     Register temp2 = locs()->temp(1).reg();
     DRegister fpu_temp = EvenDRegisterOf(locs()->temp(2).fpu_reg());
     DRegister fpu_temp_odd = OddDRegisterOf(locs()->temp(2).fpu_reg());
 
     Label store_pointer;
     Label store_double;
     Label store_float32x4;
+    Label store_float64x2;
 
     __ LoadObject(temp, Field::ZoneHandle(field().raw()));
 
     __ ldr(temp2, FieldAddress(temp, Field::is_nullable_offset()));
     __ CompareImmediate(temp2, kNullCid);
     __ b(&store_pointer, EQ);
 
     __ ldrb(temp2, FieldAddress(temp, Field::kind_bits_offset()));
     __ tst(temp2, ShifterOperand(1 << Field::kUnboxingCandidateBit));
     __ b(&store_pointer, EQ);
 
     __ ldr(temp2, FieldAddress(temp, Field::guarded_cid_offset()));
     __ CompareImmediate(temp2, kDoubleCid);
     __ b(&store_double, EQ);
 
     __ ldr(temp2, FieldAddress(temp, Field::guarded_cid_offset()));
     __ CompareImmediate(temp2, kFloat32x4Cid);
     __ b(&store_float32x4, EQ);
 
+    __ ldr(temp2, FieldAddress(temp, Field::guarded_cid_offset()));
+    __ CompareImmediate(temp2, kFloat64x2Cid);
+    __ b(&store_float64x2, EQ);
+
     // Fall through.
     __ b(&store_pointer);
 
     if (!compiler->is_optimizing()) {
         locs()->live_registers()->Add(locs()->in(0));
         locs()->live_registers()->Add(locs()->in(1));
     }
 
     {
       __ Bind(&store_double);
@@ skipping 53 matching lines @@
           Float32x4::value_offset() - kHeapObjectTag);
       __ LoadDFromOffset(fpu_temp_odd, value_reg,
           Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
       __ StoreDToOffset(fpu_temp, temp,
           Float32x4::value_offset() - kHeapObjectTag);
       __ StoreDToOffset(fpu_temp_odd, temp,
           Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
       __ b(&skip_store);
     }
 
+    {
+      __ Bind(&store_float64x2);
+      Label copy_float64x2;
+      StoreInstanceFieldSlowPath* slow_path =
+          new StoreInstanceFieldSlowPath(this, compiler->float64x2_class());
+      compiler->AddSlowPathCode(slow_path);
+
+      __ ldr(temp, FieldAddress(instance_reg, offset_in_bytes_));
+      __ CompareImmediate(temp,
+                          reinterpret_cast<intptr_t>(Object::null()));
+      __ b(&copy_float64x2, NE);
+
+      __ TryAllocate(compiler->float64x2_class(),
+                     slow_path->entry_label(),
+                     temp,
+                     temp2);
+      __ Bind(slow_path->exit_label());
+      __ MoveRegister(temp2, temp);
+      __ StoreIntoObject(instance_reg,
+                         FieldAddress(instance_reg, offset_in_bytes_),
+                         temp2);
+      __ Bind(&copy_float64x2);
+      // TODO(zra): Maybe use vldmd here.
+      __ LoadDFromOffset(fpu_temp, value_reg,
+          Float64x2::value_offset() - kHeapObjectTag);
+      __ LoadDFromOffset(fpu_temp_odd, value_reg,
+          Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+      __ StoreDToOffset(fpu_temp, temp,
+          Float64x2::value_offset() - kHeapObjectTag);
+      __ StoreDToOffset(fpu_temp_odd, temp,
+          Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+      __ b(&skip_store);
+    }
+
     __ Bind(&store_pointer);
   }
 
   if (ShouldEmitStoreBarrier()) {
     Register value_reg = locs()->in(1).reg();
     __ StoreIntoObject(instance_reg,
                        FieldAddress(instance_reg, offset_in_bytes_),
                        value_reg,
                        CanValueBeSmi());
   } else {
@@ skipping 246 matching lines @@
                            Double::value_offset() - kHeapObjectTag);
         break;
       case kFloat32x4Cid:
         __ Comment("UnboxedFloat32x4LoadFieldInstr");
         // TODO(zra): Maybe use vldmd here.
         __ LoadDFromOffset(result, temp,
             Float32x4::value_offset() - kHeapObjectTag);
         __ LoadDFromOffset(result_odd, temp,
             Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
         break;
+      case kFloat64x2Cid:
+        __ Comment("UnboxedFloat64x2LoadFieldInstr");
+        // TODO(zra): Maybe use vldmd here.
+        __ LoadDFromOffset(result, temp,
+            Float64x2::value_offset() - kHeapObjectTag);
+        __ LoadDFromOffset(result_odd, temp,
+            Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+        break;
       default:
         UNREACHABLE();
     }
     return;
   }
 
   Label done;
   Register result_reg = locs()->out().reg();
   if (IsPotentialUnboxedLoad()) {
     Register temp = locs()->temp(1).reg();
     DRegister value = EvenDRegisterOf(locs()->temp(0).fpu_reg());
     DRegister value_odd = OddDRegisterOf(locs()->temp(0).fpu_reg());
 
     Label load_pointer;
     Label load_double;
     Label load_float32x4;
+    Label load_float64x2;
 
     __ LoadObject(result_reg, Field::ZoneHandle(field()->raw()));
 
     FieldAddress field_cid_operand(result_reg, Field::guarded_cid_offset());
     FieldAddress field_nullability_operand(result_reg,
                                            Field::is_nullable_offset());
 
     __ ldr(temp, field_nullability_operand);
     __ CompareImmediate(temp, kNullCid);
     __ b(&load_pointer, EQ);
 
     __ ldr(temp, field_cid_operand);
     __ CompareImmediate(temp, kDoubleCid);
     __ b(&load_double, EQ);
 
     __ ldr(temp, field_cid_operand);
     __ CompareImmediate(temp, kFloat32x4Cid);
     __ b(&load_float32x4, EQ);
 
+    __ ldr(temp, field_cid_operand);
+    __ CompareImmediate(temp, kFloat64x2Cid);
+    __ b(&load_float64x2, EQ);
+
     // Fall through.
     __ b(&load_pointer);
 
     if (!compiler->is_optimizing()) {
       locs()->live_registers()->Add(locs()->in(0));
     }
 
     {
       __ Bind(&load_double);
       BoxDoubleSlowPath* slow_path = new BoxDoubleSlowPath(this);
@@ skipping 28 matching lines @@
           Float32x4::value_offset() - kHeapObjectTag);
       __ LoadDFromOffset(value_odd, temp,
           Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
       __ StoreDToOffset(value, result_reg,
           Float32x4::value_offset() - kHeapObjectTag);
       __ StoreDToOffset(value_odd, result_reg,
           Float32x4::value_offset() + 2 * kWordSize - kHeapObjectTag);
       __ b(&done);
     }
 
+    {
+      __ Bind(&load_float64x2);
+      BoxFloat64x2SlowPath* slow_path = new BoxFloat64x2SlowPath(this);
+      compiler->AddSlowPathCode(slow_path);
+
+      __ TryAllocate(compiler->float64x2_class(),
+                     slow_path->entry_label(),
+                     result_reg,
+                     temp);
+      __ Bind(slow_path->exit_label());
+      __ ldr(temp, FieldAddress(instance_reg, offset_in_bytes()));
+      // TODO(zra): Maybe use vldmd here.
+      __ LoadDFromOffset(value, temp,
+          Float64x2::value_offset() - kHeapObjectTag);
+      __ LoadDFromOffset(value_odd, temp,
+          Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+      __ StoreDToOffset(value, result_reg,
+          Float64x2::value_offset() - kHeapObjectTag);
+      __ StoreDToOffset(value_odd, result_reg,
+          Float64x2::value_offset() + 2 * kWordSize - kHeapObjectTag);
+      __ b(&done);
+    }
+
     __ Bind(&load_pointer);
   }
   __ LoadFromOffset(kWord, result_reg,
                     instance_reg, offset_in_bytes() - kHeapObjectTag);
   __ Bind(&done);
 }
 
 
 LocationSummary* InstantiateTypeInstr::MakeLocationSummary(bool opt) const {
   const intptr_t kNumInputs = 1;
@@ skipping 3162 matching lines @@
   compiler->GenerateCall(token_pos(),
                          &label,
                          PcDescriptors::kOther,
                          locs());
   __ Drop(ArgumentCount());  // Discard arguments.
 }
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_ARM