void G1PreBarrierStub::emit_code(LIR_Assembler* ce) {
    // At this point we know that marking is in progress.
    // If do_load() is true then we have to emit the
    // load of the previous value; otherwise it has already
    // been loaded into _pre_val.

    __ bind(_entry);

    assert(pre_val()->is_register(), "Precondition.");
    Register pre_val_reg = pre_val()->as_register();

    if (do_load()) {
        ce->mem2reg(addr(), pre_val(), T_OBJECT, patch_code(), info(), false /*wide*/, false /*unaligned*/);
    }

    if (__ is_in_wdisp16_range(_continuation)) {
        __ br_null(pre_val_reg, /*annul*/false, Assembler::pt, _continuation);
    } else {
        __ cmp(pre_val_reg, G0);
        __ brx(Assembler::equal, false, Assembler::pn, _continuation);
    }
    __ delayed()->nop();

    __ call(Runtime1::entry_for(Runtime1::Runtime1::g1_pre_barrier_slow_id));
    __ delayed()->mov(pre_val_reg, G4);
    __ br(Assembler::always, false, Assembler::pt, _continuation);
    __ delayed()->nop();

}

address AbstractInterpreterGenerator::generate_slow_signature_handler() {
  address entry = __ pc();
  Argument argv(0, true);

  // We are in the jni transition frame. Save the last_java_frame corresponding to the
  // outer interpreter frame
  //
  __ set_last_Java_frame(FP, noreg);
  // make sure the interpreter frame we've pushed has a valid return pc
  __ mov(O7, I7);
  __ mov(Lmethod, G3_scratch);
  __ mov(Llocals, G4_scratch);
  __ save_frame(0);
  __ mov(G2_thread, L7_thread_cache);
  __ add(argv.address_in_frame(), O3);
  __ mov(G2_thread, O0);
  __ mov(G3_scratch, O1);
  __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
  __ delayed()->mov(G4_scratch, O2);
  __ mov(L7_thread_cache, G2_thread);
  __ reset_last_Java_frame();

  // load the register arguments (the C code packed them as varargs)
  for (Argument ldarg = argv.successor(); ldarg.is_register(); ldarg = ldarg.successor()) {
      __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
  }
  __ ret();
  __ delayed()->
     restore(O0, 0, Lscratch);  // caller's Lscratch gets the result handler
  return entry;
}

void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
  //---------------slow case: call to native-----------------
  __ bind(_entry);
  __ mov(src()->as_register(),     O0);
  __ mov(src_pos()->as_register(), O1);
  __ mov(dst()->as_register(),     O2);
  __ mov(dst_pos()->as_register(), O3);
  __ mov(length()->as_register(),  O4);

  ce->emit_static_call_stub();

  __ call(SharedRuntime::get_resolve_static_call_stub(), relocInfo::static_call_type);
  __ delayed()->nop();
  ce->add_call_info_here(info());
  ce->verify_oop_map(info());

#ifndef PRODUCT
  __ set((intptr_t)&Runtime1::_arraycopy_slowcase_cnt, O0);
  __ ld(O0, 0, O1);
  __ inc(O1);
  __ st(O1, 0, O0);
#endif

  __ br(Assembler::always, false, Assembler::pt, _continuation);
  __ delayed()->nop();
}

OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
  // make a frame and preserve the caller's caller-save registers
  OopMap* oop_map = save_live_registers(sasm);

  // call the runtime patching routine, returns non-zero if nmethod got deopted.
  int call_offset = __ call_RT(noreg, noreg, target);
  OopMapSet* oop_maps = new OopMapSet();
  oop_maps->add_gc_map(call_offset, oop_map);

  // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
  // deoptimization handler entry that will cause re-execution of the current bytecode
  DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
  assert(deopt_blob != NULL, "deoptimization blob must have been created");

  Label no_deopt;
  __ br_null_short(O0, Assembler::pt, no_deopt);

  // return to the deoptimization handler entry for unpacking and rexecute
  // if we simply returned the we'd deopt as if any call we patched had just
  // returned.

  restore_live_registers(sasm);

  AddressLiteral dest(deopt_blob->unpack_with_reexecution());
  __ jump_to(dest, O0);
  __ delayed()->restore();

  __ bind(no_deopt);
  restore_live_registers(sasm);
  __ ret();
  __ delayed()->restore();

  return oop_maps;
}

void ArrayCopyStub::emit_code(LIR_Assembler* ce) {
  //---------------slow case: call to native-----------------
  __ bind(_entry);
  __ mov(src().as_register(),     O0);
  __ mov(src_pos().as_register(), O1);
  __ mov(dst().as_register(),     O2);
  __ mov(dst_pos().as_register(), O3);
  __ mov(length().as_register(),  O4);

  address call_pc = __ pc();
  ce->emit_code_stub(_call_stub);
  __ call(Runtime1::entry_for(Runtime1::resolve_invokestatic_id), relocInfo::static_call_type);
  __ delayed()->nop();
  ce->add_call_info_here(_info);

#ifndef PRODUCT
  __ set((intptr_t)&Runtime1::_arraycopy_slowcase_cnt, O0);
  __ ld(O0, 0, O1);
  __ inc(O1);
  __ st(O1, 0, O0);
#endif

  __ br(Assembler::always, false, Assembler::pt, _continuation);
  __ delayed()->nop();

  _call_stub->set_code_pc(call_pc);
}

void RangeCheckStub::emit_code(LIR_Assembler* ce) {
  __ bind(_entry);

  if (_info->deoptimize_on_exception()) {
    address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id);
    __ call(a, relocInfo::runtime_call_type);
    __ delayed()->nop();
    ce->add_call_info_here(_info);
    ce->verify_oop_map(_info);
    debug_only(__ should_not_reach_here());
    return;
  }

  if (_index->is_register()) {
    __ mov(_index->as_register(), G4);
  } else {
    __ set(_index->as_jint(), G4);
  }
  if (_throw_index_out_of_bounds_exception) {
    __ call(Runtime1::entry_for(Runtime1::throw_index_exception_id), relocInfo::runtime_call_type);
  } else {
    __ call(Runtime1::entry_for(Runtime1::throw_range_check_failed_id), relocInfo::runtime_call_type);
  }
  __ delayed()->nop();
  ce->add_call_info_here(_info);
  ce->verify_oop_map(_info);
  debug_only(__ should_not_reach_here());
}

address JNI_FastGetField::generate_fast_get_float_field0(BasicType type) {
  const char *name;
  switch (type) {
    case T_FLOAT:  name = "jni_fast_GetFloatField";  break;
    case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
    default:       ShouldNotReachHere();
  }
  ResourceMark rm;
  BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE*wordSize);
  CodeBuffer cbuf(blob);
  MacroAssembler* masm = new MacroAssembler(&cbuf);
  address fast_entry = __ pc();

  Label label1, label2;

  AddressLiteral cnt_addrlit(SafepointSynchronize::safepoint_counter_addr());
  __ sethi (cnt_addrlit, O3);
  Address cnt_addr(O3, cnt_addrlit.low10());
  __ ld (cnt_addr, G4);
  __ andcc (G4, 1, G0);
  __ br (Assembler::notZero, false, Assembler::pn, label1);
  __ delayed()->srl (O2, 2, O4);
  __ ld_ptr (O1, 0, O5);

  assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
  speculative_load_pclist[count] = __ pc();
  switch (type) {
    case T_FLOAT:  __ ldf (FloatRegisterImpl::S, O5, O4, F0); break;
    case T_DOUBLE: __ ldf (FloatRegisterImpl::D, O5, O4, F0); break;
    default:       ShouldNotReachHere();
  }

  __ ld (cnt_addr, O5);
  __ cmp (O5, G4);
  __ br (Assembler::notEqual, false, Assembler::pn, label2);
  __ delayed()->mov (O7, G1);

  __ retl ();
  __ delayed()-> nop ();

  slowcase_entry_pclist[count++] = __ pc();
  __ bind (label1);
  __ mov (O7, G1);

  address slow_case_addr;
  switch (type) {
    case T_FLOAT:  slow_case_addr = jni_GetFloatField_addr();  break;
    case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr(); break;
    default:       ShouldNotReachHere();
  }
  __ bind (label2);
  __ call (slow_case_addr, relocInfo::none);
  __ delayed()->mov (G1, O7);

  __ flush ();

  return fast_entry;
}

//
// Generate the on-stack replacement stub, that is used to replace the
// interpreter frame
//
OSRAdapter* SharedRuntime::generate_osr_blob(int frame_size) {
  ResourceMark rm;
  
  // setup code generation tools
  CodeBuffer*       cb = new CodeBuffer(128, 128, 0, 0, 0, false);
  MacroAssembler* masm = new MacroAssembler(cb);    
  
  OopMapSet *oop_maps = new OopMapSet();
  // frame_size is in words, Oopmap want slots
  OopMap* map =  new OopMap(frame_size * (wordSize / sizeof(jint)), 0 );
  oop_maps->add_gc_map(0, true, map);
  
  // Continuation point after returning from osr compiled method.
  // Position a potential integer result for returning from the original interpreted activation.
  __ mov(O0, I0);
  __ mov(O1, I1);
  const Register Gtmp1 = G3_scratch ;

  // Return from the current method
  // The caller's SP was adjusted upon method entry to accomodate
  // the callee's non-argument locals. Undo that adjustment.
  __ ret();
  __ delayed()->restore(IsavedSP, G0, SP);

  return OSRAdapter::new_osr_adapter(cb, oop_maps, frame_size, 0);
}

void CompiledStaticCall::emit_to_interp_stub(CodeBuffer &cbuf) {
#ifdef COMPILER2
  // Stub is fixed up when the corresponding call is converted from calling
  // compiled code to calling interpreted code.
  // set (empty), G5
  // jmp -1

  address mark = cbuf.insts_mark();  // Get mark within main instrs section.

  MacroAssembler _masm(&cbuf);

  address base =
  __ start_a_stub(to_interp_stub_size()*2);
  if (base == NULL) return;  // CodeBuffer::expand failed.

  // Static stub relocation stores the instruction address of the call.
  __ relocate(static_stub_Relocation::spec(mark));

  __ set_metadata(NULL, as_Register(Matcher::inline_cache_reg_encode()));

  __ set_inst_mark();
  AddressLiteral addrlit(-1);
  __ JUMP(addrlit, G3, 0);

  __ delayed()->nop();

  // Update current stubs pointer and restore code_end.
  __ end_a_stub();
#else
  ShouldNotReachHere();
#endif
}

address CompiledStaticCall::emit_to_interp_stub(CodeBuffer &cbuf, address mark) {
  // Stub is fixed up when the corresponding call is converted from calling
  // compiled code to calling interpreted code.
  // set (empty), G5
  // jmp -1

  if (mark == NULL) {
    mark = cbuf.insts_mark();  // Get mark within main instrs section.
  }

  MacroAssembler _masm(&cbuf);

  address base = __ start_a_stub(to_interp_stub_size());
  if (base == NULL) {
    return NULL;  // CodeBuffer::expand failed.
  }

  // Static stub relocation stores the instruction address of the call.
  __ relocate(static_stub_Relocation::spec(mark));

  __ set_metadata(NULL, as_Register(Matcher::inline_cache_reg_encode()));

  __ set_inst_mark();
  AddressLiteral addrlit(-1);
  __ JUMP(addrlit, G3, 0);

  __ delayed()->nop();

  assert(__ pc() - base <= to_interp_stub_size(), "wrong stub size");

  // Update current stubs pointer and restore code_end.
  __ end_a_stub();
  return base;
}

void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
    assert_different_registers(Rmark, Roop, Rbox);

    Label done;

    Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
    assert(mark_addr.disp() == 0, "cas must take a zero displacement");

    if (UseBiasedLocking) {
        // load the object out of the BasicObjectLock
        ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
        verify_oop(Roop);
        biased_locking_exit(mark_addr, Rmark, done);
    }
    // Test first it it is a fast recursive unlock
    ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
    br_null_short(Rmark, Assembler::pt, done);
    if (!UseBiasedLocking) {
        // load object
        ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
        verify_oop(Roop);
    }

    // Check if it is still a light weight lock, this is is true if we see
    // the stack address of the basicLock in the markOop of the object
    cas_ptr(mark_addr.base(), Rbox, Rmark);
    cmp(Rbox, Rmark);

    brx(Assembler::notEqual, false, Assembler::pn, slow_case);
    delayed()->nop();
    // Done
    bind(done);
}

void DeoptimizeStub::emit_code(LIR_Assembler* ce) {
  __ bind(_entry);
  __ call(SharedRuntime::deopt_blob()->unpack_with_reexecution());
  __ delayed()->nop();
  ce->add_call_info_here(_info);
  debug_only(__ should_not_reach_here());
}

void InterpreterRuntime::SignatureHandlerGenerator::pass_object() {
  Argument  jni_arg(jni_offset(), false);
  Argument java_arg(    offset(), true);
  Register    Rtmp1 = O0;
  Register    Rtmp2 =  jni_arg.is_register() ?  jni_arg.as_register() : O0;
  Register    Rtmp3 =  G3_scratch;

  // the handle for a receiver will never be null
  bool do_NULL_check = offset() != 0 || is_static();

  Address     h_arg = Address(Llocals, Interpreter::local_offset_in_bytes(offset()));
  __ ld_ptr(h_arg, Rtmp1);
  if (!do_NULL_check) {
    __ add(h_arg.base(), h_arg.disp(), Rtmp2);
  } else {
    if (Rtmp1 == Rtmp2)
          __ tst(Rtmp1);
    else  __ addcc(G0, Rtmp1, Rtmp2); // optimize mov/test pair
    Label L;
    __ brx(Assembler::notZero, true, Assembler::pt, L);
    __ delayed()->add(h_arg.base(), h_arg.disp(), Rtmp2);
    __ bind(L);
  }
  __ store_ptr_argument(Rtmp2, jni_arg);    // this is often a no-op
}

void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) {
  ce->compilation()->null_check_table()->append(_offset, __ offset());
  __ bind(_entry);
  __ call(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id),
          relocInfo::runtime_call_type);
  __ delayed()->nop();
  ce->add_call_info_here(_info);
}

void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) {
  Label no_deopt;
  Label no_handler;

  __ verify_not_null_oop(Oexception);

  // save the exception and issuing pc in the thread
  __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
  __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));

  // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map)
  __ mov(I7, L0);
  __ mov(Oissuing_pc, I7);
  __ sub(I7, frame::pc_return_offset, I7);
  int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));

  // Note: if nmethod has been deoptimized then regardless of
  // whether it had a handler or not we will deoptimize
  // by entering the deopt blob with a pending exception.

  __ tst(O0);
  __ br(Assembler::zero, false, Assembler::pn, no_handler);
  __ delayed()->nop();

  // restore the registers that were saved at the beginning and jump to the exception handler.
  restore_live_registers(sasm);

  __ jmp(O0, 0);
  __ delayed()->restore();

  __ bind(no_handler);
  __ mov(L0, I7); // restore return address

  // restore exception oop
  __ ld_ptr(G2_thread, in_bytes(JavaThread::exception_oop_offset()), Oexception->after_save());
  __ st_ptr(G0, G2_thread, in_bytes(JavaThread::exception_oop_offset()));

  __ restore();

  AddressLiteral exc(Runtime1::entry_for(Runtime1::unwind_exception_id));
  __ jump_to(exc, G4);
  __ delayed()->nop();


  oop_maps->add_gc_map(call_offset, oop_map);
}

/*
 * implementation for context_switchback
 */
void context_switchback() {
  if (top_context != NULL) {
    if (current_context != top_context && delayed(top_context)) {
      context_switch(top_context);
    }
  }
  return;
}