Uses an object pool on x64

runtime/vm/assembler_x64.h

 // 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.
 
 #ifndef VM_ASSEMBLER_X64_H_
 #define VM_ASSEMBLER_X64_H_
 
 #ifndef VM_ASSEMBLER_H_
 #error Do not include assembler_x64.h directly; use assembler.h instead.
 #endif
@@ skipping 165 matching lines @@
       SetDisp8(disp);
     } else {
       SetModRM(2, base);
       if ((base & 7) == RSP) {
         SetSIB(TIMES_1, RSP, base);
       }
       SetDisp32(disp);
     }
   }
 
+  Address(Register base, int32_t disp, bool fixed) {

[Florian Schneider, 2013/09/05 10:57:45]

Please add a test to assembler_x64_test.cc for this encodings.

[Florian Schneider, 2013/09/05 10:57:45]

Consider making this private and add Assembler as a friend class of Address. I
think this should normally not be used by other classes.

[zra, 2013/09/05 20:29:26]

Done.

[zra, 2013/09/05 20:29:26]

Done.

+    ASSERT(fixed);
+    SetModRM(2, base);
+    if ((base & 7) == RSP) {
+      SetSIB(TIMES_1, RSP, base);
+    }
+    SetDisp32(disp);
+  }
+
   Address(Register index, ScaleFactor scale, int32_t disp) {
     ASSERT(index != RSP);  // Illegal addressing mode.
     SetModRM(0, RSP);
     SetSIB(scale, index, RBP);
     SetDisp32(disp);
   }
 
   Address(Register base, Register index, ScaleFactor scale, int32_t disp) {
     ASSERT(index != RSP);  // Illegal addressing mode.
     if ((disp == 0) && ((base & 7) != RBP)) {
@@ skipping 118 matching lines @@
 
   static bool sse4_1_supported_;
 #ifdef DEBUG
   static bool initialized_;
 #endif
 };
 
 
 class Assembler : public ValueObject {
  public:
-  explicit Assembler(bool use_far_branches = false)
-      : buffer_(),
-        object_pool_(GrowableObjectArray::Handle()),
-        prologue_offset_(-1),
-        comments_() {
-    // This mode is only needed and implemented for MIPS and ARM.
-    ASSERT(!use_far_branches);
-  }
+  explicit Assembler(bool use_far_branches = false);
+
   ~Assembler() { }
 
   static const bool kNearJump = true;
   static const bool kFarJump = false;
 
   /*
    * Emit Machine Instructions.
    */
   void call(Register reg);
   void call(const Address& address);
   void call(Label* label);
   void call(const ExternalLabel* label);
 
-  static const intptr_t kCallExternalLabelSize = 13;
+  static const intptr_t kCallExternalLabelSize = 10;
 
   void pushq(Register reg);
   void pushq(const Address& address);
   void pushq(const Immediate& imm);
 
   void popq(Register reg);
   void popq(const Address& address);
 
   void setcc(Condition condition, ByteRegister dst);
 
@@ skipping 289 matching lines @@
   void CompareRegisters(Register a, Register b);
 
   // Issues a move instruction if 'to' is not the same as 'from'.
   void MoveRegister(Register to, Register from);
   void PopRegister(Register r);
 
   void AddImmediate(Register reg, const Immediate& imm);
 
   void Drop(intptr_t stack_elements);
 
-  void LoadObject(Register dst, const Object& object);
+  enum Patchability {
+    kPatchable,
+    kNotPatchable,
+  };
+
+  void LoadObjectFromPool(Register dst, const Object& obj, Register pp);

[Florian Schneider, 2013/09/05 10:57:45]

You really only need 

LoadObject(Register dst, const Object& obj, Register pp);

This way, you also catch all remaining call-sites of LoadObject via compile-time
errors and can inspect them one-by-one. If all sites have PP as pool register,
then you could even remove the pp parameter.

[zra, 2013/09/05 20:29:26]

LoadObject is used in arch generic code. Changing its signature will require
adding an argument at all call-sites across all architectures. I opted to just
use LoadObject everwhere the pool-pointer is PP, and LoadObjectFromPool where
the pool-pointer isn't PP. 

[Florian Schneider, 2013/09/06 09:58:15]

Ok. But here is only one arch-independent place in
FlowGraphCompiler::EmitUnoptimizedStaticCall where LoadObject is used. You could
just move this function to the arch-specific files.

[zra, 2013/09/06 17:53:26]

Done with pool pointer argument added.

+  void LoadObject(Register dst, const Object& object) {
+    LoadObjectFromPool(dst, object, PP);
+  }
+  void JumpPatchable(const ExternalLabel* label, Register pp);
+  void JumpFromPool(const ExternalLabel* label, Register pp);
+  void ConditionalJumpFromPool(Condition condition, const ExternalLabel* label,
+                               Register pp);
+  void CallPatchable(const ExternalLabel* label);
+  void CallFromPool(const ExternalLabel* label);
   void StoreObject(const Address& dst, const Object& obj);
   void PushObject(const Object& object);
   void CompareObject(Register reg, const Object& object);
   void LoadDoubleConstant(XmmRegister dst, double value);
 
   // Destroys value.
   void StoreIntoObject(Register object,  // Object we are storing into.
                        const Address& dest,  // Where we are storing into.
                        Register value,  // Value we are storing.
                        bool can_value_be_smi = true);
 
   void StoreIntoObjectNoBarrier(Register object,
                                 const Address& dest,
                                 Register value);
 
   void DoubleNegate(XmmRegister d);
   void FloatNegate(XmmRegister f);
 
   void DoubleAbs(XmmRegister reg);
 
   void LockCmpxchgl(const Address& address, Register reg) {
     lock();
     cmpxchgl(address, reg);
   }
 
   void EnterFrame(intptr_t frame_space);
   void LeaveFrame();
+  void LeaveFrameWithPP();
   void ReserveAlignedFrameSpace(intptr_t frame_space);
 
   // Create a frame for calling into runtime that preserves all volatile
   // registers.  Frame's RSP is guaranteed to be correctly aligned and
   // frame_space bytes are reserved under it.
   void EnterCallRuntimeFrame(intptr_t frame_space);
   void LeaveCallRuntimeFrame();
 
 
   void CallRuntime(const RuntimeEntry& entry, intptr_t argument_count);
@@ skipping 31 matching lines @@
   int prologue_offset() const { return prologue_offset_; }
   const ZoneGrowableArray<int>& GetPointerOffsets() const {
     return buffer_.pointer_offsets();
   }
   const GrowableObjectArray& object_pool() const { return object_pool_; }
 
   void FinalizeInstructions(const MemoryRegion& region) {
     buffer_.FinalizeInstructions(region);
   }
 
+  void LoadPoolPointer(Register pp);
+
   // Set up a Dart frame on entry with a frame pointer and PC information to
   // enable easy access to the RawInstruction object of code corresponding
   // to this frame.
   // The dart frame layout is as follows:
   //   ....
   //   ret PC
   //   saved RBP     <=== RBP
   //   pc (used to derive the RawInstruction Object of the dart code)
+  //   saved PP
   //   locals space  <=== RSP
   //   .....
   // This code sets this up with the sequence:
   //   pushq rbp
   //   movq rbp, rsp
   //   call L
   //   L: <code to adjust saved pc if there is any intrinsification code>
+  //   ...
+  //   pushq r15
   //   .....
-  void EnterDartFrame(intptr_t frame_size);
+  void EnterDartFrame(intptr_t frame_size, Register new_pp, Register new_pc);
 
   // Set up a Dart frame for a function compiled for on-stack replacement.
   // The frame layout is a normal Dart frame, but the frame is partially set
   // up on entry (it is the frame of the unoptimized code).
-  void EnterOsrFrame(intptr_t extra_size);
+  void EnterOsrFrame(intptr_t extra_size, Register new_pp, Register new_pc);
 
   // Set up a stub frame so that the stack traversal code can easily identify
   // a stub frame.
   // The stub frame layout is as follows:
   //   ....
   //   ret PC
   //   saved RBP
   //   pc (used to derive the RawInstruction Object of the stub)
   //   .....
   // This code sets this up with the sequence:
   //   pushq rbp
   //   movq rbp, rsp
   //   pushq immediate(0)
   //   .....
   void EnterStubFrame();
+  void EnterStubFrameWithPP();
 
-  // Instruction pattern from entrypoint is used in dart frame prologs
+  // Instruction pattern from entrypoint is used in dart frame prologues
   // to set up the frame and save a PC which can be used to figure out the
   // RawInstruction object corresponding to the code running in the frame.
   // entrypoint:
   //   pushq rbp          (size is 1 byte)
   //   movq rbp, rsp      (size is 3 bytes)
   //   call L             (size is 5 bytes)
   //   L:
   static const intptr_t kEntryPointToPcMarkerOffset = 9;
 
   // Inlined allocation of an instance of class 'cls', code has no runtime
@@ skipping 12 matching lines @@
   void Unreachable(const char* message);
 
   static void InitializeMemoryWithBreakpoints(uword data, int length);
 
   static const char* RegisterName(Register reg);
 
   static const char* FpuRegisterName(FpuRegister reg);
 
  private:
   AssemblerBuffer buffer_;
-  GrowableObjectArray& object_pool_;  // Object pool is not used on x64.
+  GrowableObjectArray& object_pool_;  // Objects and patchable jump targets.
   int prologue_offset_;
 
   class CodeComment : public ZoneAllocated {
    public:
     CodeComment(intptr_t pc_offset, const String& comment)
         : pc_offset_(pc_offset), comment_(comment) { }
 
     intptr_t pc_offset() const { return pc_offset_; }
     const String& comment() const { return comment_; }
 
    private:
     intptr_t pc_offset_;
     const String& comment_;
 
     DISALLOW_COPY_AND_ASSIGN(CodeComment);
   };
 
   GrowableArray<CodeComment*> comments_;
 
+  intptr_t AddObject(const Object& obj);
+  intptr_t AddExternalLabel(const ExternalLabel* label, Patchability patchable);
+  void LoadExternalLabel(Register dst,
+                         const ExternalLabel* label,
+                         Patchability patchabe,

[Florian Schneider, 2013/09/05 10:57:45]

s/patchabe/patchable/

[zra, 2013/09/05 20:29:26]

Done.

+                         Register pp);
+  bool CanLoadFromObjectPool(const Object& object);
+  void LoadWordFromPoolOffset(Register dst, Register pp, int32_t offset);
+
   inline void EmitUint8(uint8_t value);
   inline void EmitInt32(int32_t value);
   inline void EmitInt64(int64_t value);
 
   inline void EmitRegisterREX(Register reg, uint8_t rex);
   inline void EmitRegisterOperand(int rm, int reg);
   inline void EmitOperandREX(int rm, const Operand& operand, uint8_t rex);
   inline void EmitXmmRegisterOperand(int rm, XmmRegister reg);
   inline void EmitFixup(AssemblerFixup* fixup);
   inline void EmitOperandSizeOverride();
@@ skipping 102 matching lines @@
 }
 
 
 inline void Assembler::EmitOperandSizeOverride() {
   EmitUint8(0x66);
 }
 
 }  // namespace dart
 
 #endif  // VM_ASSEMBLER_X64_H_