Refactor unboxed fields in preparation of reusable SIMD boxes

runtime/vm/intermediate_language_x64.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_X64.
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/intermediate_language.h"
 
 #include "vm/dart_entry.h"
@@ skipping 1474 matching lines @@
         UNREACHABLE();
       }
     }
   }
   __ Bind(&ok);
 }
 
 
 class StoreInstanceFieldSlowPath : public SlowPathCode {
  public:
-  explicit StoreInstanceFieldSlowPath(StoreInstanceFieldInstr* instruction)
-      : instruction_(instruction) { }
+  explicit StoreInstanceFieldSlowPath(const Class& cls,
+                                      StoreInstanceFieldInstr* instruction)
+      : instruction_(instruction), cls_(cls) { }
 
   virtual void EmitNativeCode(FlowGraphCompiler* compiler) {
     __ Comment("StoreInstanceFieldSlowPath");
     __ Bind(entry_label());
-    const Class& double_class = compiler->double_class();
     const Code& stub =
-        Code::Handle(StubCode::GetAllocationStubForClass(double_class));
-    const ExternalLabel label(double_class.ToCString(), stub.EntryPoint());
+        Code::Handle(StubCode::GetAllocationStubForClass(cls_));
+    const ExternalLabel label(cls_.ToCString(), stub.EntryPoint());
 
     LocationSummary* locs = instruction_->locs();
     locs->live_registers()->Remove(locs->out());
 
     compiler->SaveLiveRegisters(locs);
     compiler->GenerateCall(Scanner::kNoSourcePos,  // No token position.
                            &label,
                            PcDescriptors::kOther,
                            locs);
     __ MoveRegister(locs->temp(0).reg(), RAX);
     compiler->RestoreLiveRegisters(locs);
 
     __ jmp(exit_label());
   }
 
  private:
   StoreInstanceFieldInstr* instruction_;
+  const Class& cls_;
 };
 
 
 LocationSummary* StoreInstanceFieldInstr::MakeLocationSummary(bool opt) const {
   const intptr_t kNumInputs = 2;
   const intptr_t kNumTemps = 0;
   LocationSummary* summary =
       new LocationSummary(kNumInputs, kNumTemps,
           (field().guarded_cid() == kIllegalCid) || (is_initialization_)
           ? LocationSummary::kCallOnSlowPath
@@ skipping 23 matching lines @@
 
 void StoreInstanceFieldInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   Label skip_store;
 
   Register instance_reg = locs()->in(0).reg();
 
   if (IsUnboxedStore() && compiler->is_optimizing()) {
     XmmRegister value = locs()->in(1).fpu_reg();
     Register temp = locs()->temp(0).reg();
     Register temp2 = locs()->temp(1).reg();
+    const intptr_t cid = field().UnboxedFieldCid();
 
     if (is_initialization_) {
+      const Class* cls = NULL;
+      switch (cid) {
+        case kDoubleCid:
+          cls = &compiler->double_class();
+          break;
+        // TODO(johnmccutchan): Add kFloat32x4Cid here.
+        default:
+          UNREACHABLE();
+      }
       StoreInstanceFieldSlowPath* slow_path =
-          new StoreInstanceFieldSlowPath(this);
+          new StoreInstanceFieldSlowPath(*cls, this);
       compiler->AddSlowPathCode(slow_path);
 
-      __ TryAllocate(compiler->double_class(),
+      __ TryAllocate(*cls,
                      slow_path->entry_label(),
                      Assembler::kFarJump,
                      temp,
                      PP);
       __ Bind(slow_path->exit_label());
       __ movq(temp2, temp);
       __ StoreIntoObject(instance_reg,
                          FieldAddress(instance_reg, field().Offset()),
                          temp2);
     } else {
       __ movq(temp, FieldAddress(instance_reg, field().Offset()));
     }
-    __ movsd(FieldAddress(temp, Double::value_offset()), value);
+    switch (cid) {
+      case kDoubleCid:
+      __ movsd(FieldAddress(temp, Double::value_offset()), value);
+      // TODO(johnmccutchan): Add kFloat32x4Cid here.
+      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();
     FpuRegister fpu_temp = locs()->temp(2).fpu_reg();
 
-    Label store_pointer, copy_payload;
+    Label store_pointer;
+    Label copy_double;
+    Label store_double;
+
     __ LoadObject(temp, Field::ZoneHandle(field().raw()), PP);
-    __ cmpq(FieldAddress(temp, Field::guarded_cid_offset()),
-            Immediate(kDoubleCid));
-    __ j(NOT_EQUAL, &store_pointer);
+
     __ cmpq(FieldAddress(temp, Field::is_nullable_offset()),
             Immediate(kNullCid));
     __ j(EQUAL, &store_pointer);
+
     __ movzxb(temp2, FieldAddress(temp, Field::kind_bits_offset()));
     __ testq(temp2, Immediate(1 << Field::kUnboxingCandidateBit));
     __ j(ZERO, &store_pointer);
 
+    __ cmpq(FieldAddress(temp, Field::guarded_cid_offset()),
+            Immediate(kDoubleCid));
+    __ j(EQUAL, &store_double);
+
+    // Fall through.
+    __ jmp(&store_pointer);
+
+    __ Bind(&store_double);
+
     __ movq(temp, FieldAddress(instance_reg, field().Offset()));
     __ CompareObject(temp, Object::null_object(), PP);
-    __ j(NOT_EQUAL, &copy_payload);
+    __ j(NOT_EQUAL, &copy_double);
 
     StoreInstanceFieldSlowPath* slow_path =
-        new StoreInstanceFieldSlowPath(this);
+        new StoreInstanceFieldSlowPath(compiler->double_class(), this);
     compiler->AddSlowPathCode(slow_path);
 
     if (!compiler->is_optimizing()) {
       locs()->live_registers()->Add(locs()->in(0));
       locs()->live_registers()->Add(locs()->in(1));
     }
     __ TryAllocate(compiler->double_class(),
                    slow_path->entry_label(),
                    Assembler::kFarJump,
                    temp,
                    PP);
     __ Bind(slow_path->exit_label());
     __ movq(temp2, temp);
     __ StoreIntoObject(instance_reg,
                        FieldAddress(instance_reg, field().Offset()),
                        temp2);
 
-    __ Bind(&copy_payload);
+    __ Bind(&copy_double);
     __ movsd(fpu_temp, FieldAddress(value_reg, Double::value_offset()));
     __ movsd(FieldAddress(temp, Double::value_offset()), fpu_temp);
     __ jmp(&skip_store);
     __ Bind(&store_pointer);
   }
 
   if (ShouldEmitStoreBarrier()) {
     Register value_reg = locs()->in(1).reg();
     __ StoreIntoObject(instance_reg,
                        FieldAddress(instance_reg, field().Offset()),
@@ skipping 184 matching lines @@
   return locs;
 }
 
 
 void LoadFieldInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   Register instance_reg = locs()->in(0).reg();
   if (IsUnboxedLoad() && compiler->is_optimizing()) {
     XmmRegister result = locs()->out().fpu_reg();
     Register temp = locs()->temp(0).reg();
     __ movq(temp, FieldAddress(instance_reg, offset_in_bytes()));
-    __ movsd(result, FieldAddress(temp, Double::value_offset()));
+    intptr_t cid = field()->UnboxedFieldCid();
+    switch (cid) {
+      case kDoubleCid:
+        __ movsd(result, FieldAddress(temp, Double::value_offset()));
+      break;
+      // TODO(johnmccutchan): Add Float32x4 path here.
+      default:
+        UNREACHABLE();
+    }
     return;
   }
 
   Label done;
   Register result = locs()->out().reg();
   if (IsPotentialUnboxedLoad()) {
     Register temp = locs()->temp(1).reg();
     XmmRegister value = locs()->temp(0).fpu_reg();
 
     Label load_pointer;
+    Label load_double;
+
     __ LoadObject(result, Field::ZoneHandle(field()->raw()), PP);
 
-
-    __ cmpq(FieldAddress(result, Field::guarded_cid_offset()),
-            Immediate(kDoubleCid));
-    __ j(NOT_EQUAL, &load_pointer);
     __ cmpq(FieldAddress(result, Field::is_nullable_offset()),
             Immediate(kNullCid));
     __ j(EQUAL, &load_pointer);
 
+    __ cmpq(FieldAddress(result, Field::guarded_cid_offset()),
+            Immediate(kDoubleCid));
+    __ j(EQUAL, &load_double);
+
+    // Fall through.
+    __ jmp(&load_pointer);
+
+    __ Bind(&load_double);
+
     BoxDoubleSlowPath* slow_path = new BoxDoubleSlowPath(this);
     compiler->AddSlowPathCode(slow_path);
 
     if (!compiler->is_optimizing()) {
       locs()->live_registers()->Add(locs()->in(0));
     }
 
     __ TryAllocate(compiler->double_class(),
                    slow_path->entry_label(),
                    Assembler::kFarJump,
                    result,
                    PP);
     __ Bind(slow_path->exit_label());
     __ movq(temp, FieldAddress(instance_reg, offset_in_bytes()));
     __ movsd(value, FieldAddress(temp, Double::value_offset()));
     __ movsd(FieldAddress(result, Double::value_offset()), value);
     __ jmp(&done);
+
+    // TODO(johnmccutchan): Add Float32x4 path here.
+
     __ Bind(&load_pointer);
   }
   __ movq(result, FieldAddress(instance_reg, offset_in_bytes()));
   __ Bind(&done);
 }
 
 
 LocationSummary* InstantiateTypeInstr::MakeLocationSummary(bool opt) const {
   const intptr_t kNumInputs = 1;
   const intptr_t kNumTemps = 0;
@@ skipping 3071 matching lines @@
                          PcDescriptors::kOther,
                          locs());
   __ Drop(2);  // Discard type arguments and receiver.
 }
 
 }  // namespace dart
 
 #undef __
 
 #endif  // defined TARGET_ARCH_X64