src/cpu/mips/vm/c1_LIRAssembler_mips.cpp

changeset 8865
ffcdff41a92f
parent 6880
52ea28d233d2
child 9126
bc5b8e3dcb6b
     1.1 --- a/src/cpu/mips/vm/c1_LIRAssembler_mips.cpp	Sat Jan 06 16:30:58 2018 +0800
     1.2 +++ b/src/cpu/mips/vm/c1_LIRAssembler_mips.cpp	Thu May 24 19:49:50 2018 +0800
     1.3 @@ -316,8 +316,8 @@
     1.4  
     1.5  void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
     1.6    jobject o = NULL;
     1.7 -  PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
     1.8    int oop_index = __ oop_recorder()->allocate_oop_index(o);
     1.9 +  PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
    1.10    RelocationHolder rspec = oop_Relocation::spec(oop_index);
    1.11    __ relocate(rspec);
    1.12  #ifndef _LP64
    1.13 @@ -326,57 +326,21 @@
    1.14    __ addiu(reg, reg, Assembler::split_low((int)o));
    1.15  #else
    1.16  //li may not pass NativeMovConstReg::verify. see nativeMovConstReg_at(pc_start()); in PatchingStub::install. by aoqi
    1.17 +//  __ li48(reg, (long)o);
    1.18    __ li48(reg, (long)o);
    1.19  #endif
    1.20    // patching_epilog(patch, LIR_Op1::patch_normal, noreg, info);
    1.21    patching_epilog(patch, lir_patch_normal, reg, info);
    1.22  }
    1.23  
    1.24 -
    1.25 -void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register unused, int monitor_no, Register exception) {
    1.26 -
    1.27 -  if (exception->is_valid()) {
    1.28 -    // preserve exception
    1.29 -    // note: the monitor_exit runtime call is a leaf routine
    1.30 -    //       and cannot block => no GC can happen
    1.31 -    // The slow case (MonitorAccessStub) uses the first two stack slots
    1.32 -    // ([SP+0] and [SP+4]), therefore we store the exception at [esp+8]
    1.33 -    __ st_ptr(exception, SP, 2 * wordSize);
    1.34 -  }
    1.35 -
    1.36 -  Register obj_reg  = obj_opr->as_register();
    1.37 -  Register lock_reg = lock_opr->as_register();
    1.38 -
    1.39 -  // compute pointer to BasicLock
    1.40 -  //Address lock_addr = frame_map()->address_for_monitor_lock_index(monitor_no);
    1.41 -  Address lock_addr = frame_map()->address_for_monitor_lock(monitor_no);
    1.42 -  __ lea(lock_reg, lock_addr);
    1.43 -  // unlock object
    1.44 -  MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, true, monitor_no);
    1.45 -  // temporary fix: must be created after exceptionhandler, therefore as call stub
    1.46 -  _slow_case_stubs->append(slow_case);
    1.47 -  if (UseFastLocking) {
    1.48 -    // try inlined fast unlocking first, revert to slow locking if it fails
    1.49 -    // note: lock_reg points to the displaced header since the displaced header offset is 0!
    1.50 -    assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
    1.51 -    __ unlock_object(NOREG, obj_reg, lock_reg, *slow_case->entry());
    1.52 -  } else {
    1.53 -    // always do slow unlocking
    1.54 -    // note: the slow unlocking code could be inlined here, however if we use
    1.55 -    //       slow unlocking, speed doesn't matter anyway and this solution is
    1.56 -    //       simpler and requires less duplicated code - additionally, the
    1.57 -    //       slow unlocking code is the same in either case which simplifies
    1.58 -    //       debugging
    1.59 -    __ b_far(*slow_case->entry());
    1.60 -    __ delayed()->nop();
    1.61 -  }
    1.62 -  // done
    1.63 -  __ bind(*slow_case->continuation());
    1.64 -
    1.65 -  if (exception->is_valid()) {
    1.66 -    // restore exception
    1.67 -    __ ld_ptr(exception, SP, 2 * wordSize);
    1.68 -  }
    1.69 +void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
    1.70 +  Metadata *o = NULL;
    1.71 +  int index = __ oop_recorder()->allocate_metadata_index(o);
    1.72 +  PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
    1.73 +  RelocationHolder rspec = metadata_Relocation::spec(index);
    1.74 +  __ relocate(rspec);
    1.75 +  __ li48(reg, (long)o);
    1.76 +  patching_epilog(patch, lir_patch_normal, reg, info);
    1.77  }
    1.78  
    1.79  // This specifies the esp decrement needed to build the frame
    1.80 @@ -405,61 +369,28 @@
    1.81    // generate code for exception handler
    1.82    address handler_base = __ start_a_stub(exception_handler_size);
    1.83    if (handler_base == NULL) {
    1.84 -    //no enough space
    1.85 +    // no enough space
    1.86      bailout("exception handler overflow");
    1.87      return -1;
    1.88    }
    1.89  
    1.90 -
    1.91 -
    1.92 -  //compilation()->offsets()->set_value(CodeOffsets::Exceptions, code_offset());
    1.93 -  // if the method does not have an exception handler, then there is
    1.94 -  // no reason to search for one
    1.95 -  //if (compilation()->has_exception_handlers() || JvmtiExport::can_post_exceptions()) {
    1.96 -           // the exception oop and pc are in V0 and V1
    1.97 -    // no other registers need to be preserved, so invalidate them
    1.98 -    // check that there is really an exception
    1.99 -//    __ verify_not_null_oop(V0);
   1.100 -
   1.101 -    // search an exception handler (V0: exception oop, V1: throwing pc)
   1.102 -//    __ call(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id),
   1.103 -//        relocInfo::runtime_call_type);
   1.104 -//    __ delayed()->nop();
   1.105 -    // if the call returns here, then the exception handler for particular
   1.106 -    // exception doesn't exist -> unwind activation and forward exception to caller
   1.107 - // }
   1.108 -       int offset = code_offset();
   1.109 -
   1.110 -  // the exception oop is in V0
   1.111 +  int offset = code_offset();
   1.112 +
   1.113 +  // the exception oop and pc are in V0, and V1
   1.114    // no other registers need to be preserved, so invalidate them
   1.115 +  //__ invalidate_registers(false, true, true, false, true, true);
   1.116 +
   1.117    // check that there is really an exception
   1.118    __ verify_not_null_oop(V0);
   1.119 -        //FIXME:wuhui??
   1.120 -        //__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
   1.121 -        //__ delayed()->nop();
   1.122 -        __ should_not_reach_here();
   1.123 -        guarantee(code_offset() - offset <= exception_handler_size, "overflow");
   1.124 -        __ end_a_stub();
   1.125 -        return offset;
   1.126 -
   1.127 -  // unlock the receiver/klass if necessary
   1.128 -  // V0: exception
   1.129 -//  ciMethod* method = compilation()->method();
   1.130 -//  if (method->is_synchronized() && GenerateSynchronizationCode) {
   1.131 -//#ifndef _LP64
   1.132 -//by_css
   1.133 -//    monitorexit(FrameMap::_t0_oop_opr, FrameMap::_t6_opr, NOREG, 0, V0);
   1.134 -//#else
   1.135 -//    monitorexit(FrameMap::_t0_oop_opr, FrameMap::_a6_opr, NOREG, 0, V0);
   1.136 -//#endif
   1.137 -//  }
   1.138 -
   1.139 -  // unwind activation and forward exception to caller
   1.140 -  // V0: exception
   1.141 -//  __ jmp(Runtime1::entry_for(Runtime1::unwind_exception_id),
   1.142 -//        relocInfo::runtime_call_type);
   1.143 -//  __ delayed()->nop();
   1.144 -//  __ end_a_stub();
   1.145 +
   1.146 +  // search an exception handler (V0: exception oop, V1: throwing pc)
   1.147 +  __ call(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id));
   1.148 +  __ delayed()->nop();
   1.149 +  __ should_not_reach_here();
   1.150 +  guarantee(code_offset() - offset <= exception_handler_size, "overflow");
   1.151 +  __ end_a_stub();
   1.152 +
   1.153 +  return offset;
   1.154  }
   1.155  
   1.156  // Emit the code to remove the frame from the stack in the exception
   1.157 @@ -472,46 +403,52 @@
   1.158  #endif
   1.159  
   1.160    int offset = code_offset();
   1.161 - /* // Fetch the exception from TLS and clear out exception related thread state
   1.162 -    __ get_thread(rsi);
   1.163 -  __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
   1.164 -  __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (intptr_t)NULL_WORD);
   1.165 -  __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (intptr_t)NULL_WORD);
   1.166 +  // Fetch the exception from TLS and clear out exception related thread state
   1.167 +  Register thread = TREG;
   1.168 +#ifndef OPT_THREAD
   1.169 +  __ get_thread(thread);
   1.170 +#endif
   1.171 +  __ ld_ptr(V0, Address(thread, JavaThread::exception_oop_offset()));
   1.172 +  __ st_ptr(R0, Address(thread, JavaThread::exception_oop_offset()));
   1.173 +  __ st_ptr(R0, Address(thread, JavaThread::exception_pc_offset()));
   1.174  
   1.175    __ bind(_unwind_handler_entry);
   1.176 -  __ verify_not_null_oop(rax);
   1.177 +  __ verify_not_null_oop(V0);
   1.178    if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
   1.179 -    __ mov(rsi, rax);  // Preserve the exception
   1.180 +    __ move(S0, V0);  // Preserve the exception (rbx is always callee-saved)
   1.181    }
   1.182 - // Preform needed unlocking
   1.183 -   MonitorExitStub* stub = NULL;
   1.184 +
   1.185 +  // Preform needed unlocking
   1.186 +  MonitorExitStub* stub = NULL;
   1.187    if (method()->is_synchronized()) {
   1.188 -    monitor_address(0, FrameMap::rax_opr);
   1.189 -    stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
   1.190 -    __ unlock_object(rdi, rbx, rax, *stub->entry());
   1.191 +    monitor_address(0, FrameMap::_v0_opr);
   1.192 +    stub = new MonitorExitStub(FrameMap::_v0_opr, true, 0);
   1.193 +    __ unlock_object(A0, A1, V0, *stub->entry());
   1.194      __ bind(*stub->continuation());
   1.195    }
   1.196  
   1.197    if (compilation()->env()->dtrace_method_probes()) {
   1.198 -    __ get_thread(rax);
   1.199 -    __ movptr(Address(rsp, 0), rax);
   1.200 -    __ mov_metadata(Address(rsp, sizeof(void*)), method()->constant_encoding());
   1.201 -    __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
   1.202 +    __ move(A0, thread);
   1.203 +    __ mov_metadata(A1, method()->constant_encoding());
   1.204 +    __ patchable_call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit));
   1.205    }
   1.206  
   1.207    if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
   1.208 -    __ mov(rax, rsi);  // Restore the exception
   1.209 +    __ move(V0, S0);  // Restore the exception
   1.210    }
   1.211  
   1.212    // remove the activation and dispatch to the unwind handler
   1.213 -    __ remove_frame(initial_frame_size_in_bytes());
   1.214 -  __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
   1.215 +  // leave activation of nmethod
   1.216 +  __ remove_frame(initial_frame_size_in_bytes());
   1.217 +
   1.218 +  __ jmp(Runtime1::entry_for(Runtime1::unwind_exception_id));
   1.219 +  __ delayed()->nop();
   1.220  
   1.221    // Emit the slow path assembly
   1.222 -    if (stub != NULL) {
   1.223 +  if (stub != NULL) {
   1.224      stub->emit_code(this);
   1.225    }
   1.226 -*/
   1.227 +
   1.228    return offset;
   1.229  }
   1.230  
   1.231 @@ -653,7 +590,7 @@
   1.232  void LIR_Assembler::return_op(LIR_Opr result) {
   1.233    assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == V0, "word returns are in V0");
   1.234    // Pop the stack before the safepoint code
   1.235 -  __ leave();
   1.236 +  __ remove_frame(initial_frame_size_in_bytes());
   1.237  #ifndef _LP64
   1.238    //by aoqi
   1.239    __ lui(AT, Assembler::split_high((intptr_t)os::get_polling_page()
   1.240 @@ -674,6 +611,7 @@
   1.241    #endif
   1.242  #endif
   1.243  
   1.244 +  __ pop(RA);
   1.245    __ jr(RA);
   1.246    __ delayed()->nop();
   1.247  }
   1.248 @@ -724,16 +662,15 @@
   1.249    assert(dest->is_register(), "should not call otherwise");
   1.250    LIR_Const* c = src->as_constant_ptr();
   1.251    switch (c->type()) {
   1.252 +    case T_ADDRESS: {
   1.253 +      assert(patch_code == lir_patch_none, "no patching handled here");
   1.254 +      __ move(dest->as_register(), c->as_jint()); // FIXME
   1.255 +      break;
   1.256 +    }
   1.257 +
   1.258      case T_INT: {
   1.259 -      jint con = c->as_jint();
   1.260 -      if (dest->is_single_cpu()) {
   1.261 -        assert(patch_code == lir_patch_none, "no patching handled here");
   1.262 -        __ move(dest->as_register(), con);
   1.263 -      } else {
   1.264 -        assert(dest->is_single_fpu(), "wrong register kind");
   1.265 -        __ move(AT, con);
   1.266 -        __ mtc1(AT, dest->as_float_reg());
   1.267 -      }
   1.268 +      assert(patch_code == lir_patch_none, "no patching handled here");
   1.269 +      __ move(dest->as_register(), c->as_jint());
   1.270        break;
   1.271      }
   1.272  
   1.273 @@ -772,6 +709,15 @@
   1.274        break;
   1.275      }
   1.276  
   1.277 +    case T_METADATA: {
   1.278 +      if (patch_code != lir_patch_none) {
   1.279 +        klass2reg_with_patching(dest->as_register(), info);
   1.280 +      } else {
   1.281 +        __ mov_metadata(dest->as_register(), c->as_metadata());
   1.282 +      }
   1.283 +      break;
   1.284 +    }
   1.285 +
   1.286      case T_FLOAT: {
   1.287        address const_addr = float_constant(c->as_jfloat());
   1.288        assert (const_addr != NULL, "must create float constant in the constant table");
   1.289 @@ -855,6 +801,11 @@
   1.290        __ sw(AT, frame_map()->address_for_slot(dest->single_stack_ix()));
   1.291        break;
   1.292  
   1.293 +    case T_ADDRESS:
   1.294 +      __ move(AT, c->as_jint_bits());
   1.295 +      __ st_ptr(AT, frame_map()->address_for_slot(dest->single_stack_ix()));
   1.296 +      break;
   1.297 +
   1.298      case T_OBJECT:
   1.299        if (c->as_jobject() == NULL) {
   1.300          __ st_ptr(R0, frame_map()->address_for_slot(dest->single_stack_ix()));
   1.301 @@ -957,6 +908,10 @@
   1.302        } else
   1.303          __ sw(R0, as_Address(addr));
   1.304        break;
   1.305 +    case T_ADDRESS:
   1.306 +      __ move(AT, c->as_jint_bits());
   1.307 +      __ st_ptr(AT, as_Address(addr));
   1.308 +      break;
   1.309      case T_BOOLEAN: // fall through
   1.310      case T_BYTE:
   1.311        if(c->as_jint() != 0) {
   1.312 @@ -1217,7 +1172,7 @@
   1.313        break;
   1.314  
   1.315      case T_LONG: {
   1.316 -       Register from_lo = src->as_register_lo();
   1.317 +      Register from_lo = src->as_register_lo();
   1.318        Register from_hi = src->as_register_hi();
   1.319  #ifdef _LP64
   1.320        if (needs_patching) {
   1.321 @@ -1535,42 +1490,55 @@
   1.322      case T_ARRAY:
   1.323        if (UseCompressedOops && !wide) {
   1.324            if (disp_reg == noreg) {
   1.325 -        __ lw(dest->as_register(), src_reg, disp_value);
   1.326 +            __ lwu(dest->as_register(), src_reg, disp_value);
   1.327            } else if (needs_patching) {
   1.328 -        __ dadd(AT, src_reg, disp_reg);
   1.329 -        offset = code_offset();
   1.330 -        __ lw(dest->as_register(), AT, 0);
   1.331 +            __ dadd(AT, src_reg, disp_reg);
   1.332 +            offset = code_offset();
   1.333 +            __ lwu(dest->as_register(), AT, 0);
   1.334            } else {
   1.335 -        __ dadd(AT, src_reg, disp_reg);
   1.336 -        offset = code_offset();
   1.337 -        __ lw(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.338 +            __ dadd(AT, src_reg, disp_reg);
   1.339 +            offset = code_offset();
   1.340 +            __ lwu(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.341            }
   1.342 -
   1.343        } else {
   1.344            if (disp_reg == noreg) {
   1.345 -        __ ld_ptr(dest->as_register(), src_reg, disp_value);
   1.346 +            __ ld_ptr(dest->as_register(), src_reg, disp_value);
   1.347            } else if (needs_patching) {
   1.348 -        __ dadd(AT, src_reg, disp_reg);
   1.349 -        offset = code_offset();
   1.350 -        __ ld_ptr(dest->as_register(), AT, 0);
   1.351 +            __ dadd(AT, src_reg, disp_reg);
   1.352 +            offset = code_offset();
   1.353 +            __ ld_ptr(dest->as_register(), AT, 0);
   1.354            } else {
   1.355 -        __ dadd(AT, src_reg, disp_reg);
   1.356 -        offset = code_offset();
   1.357 -        __ ld_ptr(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.358 +            __ dadd(AT, src_reg, disp_reg);
   1.359 +            offset = code_offset();
   1.360 +            __ ld_ptr(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.361            }
   1.362        }
   1.363        break;
   1.364      case T_ADDRESS:
   1.365 -      if (disp_reg == noreg) {
   1.366 +      if (UseCompressedClassPointers && addr->disp() == oopDesc::klass_offset_in_bytes()) {
   1.367 +        if (disp_reg == noreg) {
   1.368 +          __ lwu(dest->as_register(), src_reg, disp_value);
   1.369 +        } else if (needs_patching) {
   1.370 +          __ dadd(AT, src_reg, disp_reg);
   1.371 +          offset = code_offset();
   1.372 +          __ lwu(dest->as_register(), AT, 0);
   1.373 +        } else {
   1.374 +          __ dadd(AT, src_reg, disp_reg);
   1.375 +          offset = code_offset();
   1.376 +          __ lwu(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.377 +        }
   1.378 +      } else {
   1.379 +        if (disp_reg == noreg) {
   1.380            __ ld_ptr(dest->as_register(), src_reg, disp_value);
   1.381 -      } else if (needs_patching) {
   1.382 +        } else if (needs_patching) {
   1.383            __ dadd(AT, src_reg, disp_reg);
   1.384            offset = code_offset();
   1.385            __ ld_ptr(dest->as_register(), AT, 0);
   1.386 -      } else {
   1.387 +        } else {
   1.388            __ dadd(AT, src_reg, disp_reg);
   1.389            offset = code_offset();
   1.390            __ ld_ptr(dest->as_register(), AT, Assembler::split_low(disp_value));
   1.391 +        }
   1.392        }
   1.393        break;
   1.394      case T_INT:  {
   1.395 @@ -1704,6 +1672,10 @@
   1.396        }
   1.397  #endif
   1.398        __ verify_oop(dest->as_register());
   1.399 +  } else if (type == T_ADDRESS && addr->disp() == oopDesc::klass_offset_in_bytes()) {
   1.400 +    if (UseCompressedClassPointers) {
   1.401 +      __ decode_klass_not_null(dest->as_register());
   1.402 +    }
   1.403    }
   1.404    if (info != NULL) add_debug_info_for_null_check(offset, info);
   1.405  }
   1.406 @@ -3088,11 +3060,197 @@
   1.407    __ bind(*op->stub()->continuation());
   1.408  }
   1.409  
   1.410 +void LIR_Assembler::type_profile_helper(Register mdo,
   1.411 +                                        ciMethodData *md, ciProfileData *data,
   1.412 +                                        Register recv, Label* update_done) {
   1.413 +  for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
   1.414 +    Label next_test;
   1.415 +    // See if the receiver is receiver[n].
   1.416 +    __ ld_ptr(AT, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
   1.417 +    __ bne(AT, recv, next_test);
   1.418 +    __ delayed()->nop();
   1.419 +    Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
   1.420 +    __ ld_ptr(AT, data_addr);
   1.421 +    __ addi(AT, AT, DataLayout::counter_increment);
   1.422 +    __ st_ptr(AT, data_addr);
   1.423 +    __ b(*update_done);
   1.424 +    __ delayed()->nop();
   1.425 +    __ bind(next_test);
   1.426 +  }
   1.427 +
   1.428 +  // Didn't find receiver; find next empty slot and fill it in
   1.429 +  for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
   1.430 +    Label next_test;
   1.431 +    Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
   1.432 +    __ ld_ptr(AT, recv_addr);
   1.433 +    __ bne(AT, R0, next_test);
   1.434 +    __ delayed()->nop();
   1.435 +    __ st_ptr(recv, recv_addr);
   1.436 +    __ move(AT, DataLayout::counter_increment);
   1.437 +    __ st_ptr(AT, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))));
   1.438 +    __ b(*update_done);
   1.439 +    __ delayed()->nop();
   1.440 +    __ bind(next_test);
   1.441 +  }
   1.442 +}
   1.443 +
   1.444 +void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
   1.445 +  // we always need a stub for the failure case.
   1.446 +  CodeStub* stub = op->stub();
   1.447 +  Register obj = op->object()->as_register();
   1.448 +  Register k_RInfo = op->tmp1()->as_register();
   1.449 +  Register klass_RInfo = op->tmp2()->as_register();
   1.450 +  Register dst = op->result_opr()->as_register();
   1.451 +  ciKlass* k = op->klass();
   1.452 +  Register Rtmp1 = noreg;
   1.453 +
   1.454 +  // check if it needs to be profiled
   1.455 +  ciMethodData* md;
   1.456 +  ciProfileData* data;
   1.457 +
   1.458 +  if (op->should_profile()) {
   1.459 +    ciMethod* method = op->profiled_method();
   1.460 +    assert(method != NULL, "Should have method");
   1.461 +    int bci = op->profiled_bci();
   1.462 +    md = method->method_data_or_null();
   1.463 +    assert(md != NULL, "Sanity");
   1.464 +    data = md->bci_to_data(bci);
   1.465 +    assert(data != NULL,                "need data for type check");
   1.466 +    assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
   1.467 +  }
   1.468 +  Label profile_cast_success, profile_cast_failure;
   1.469 +  Label *success_target = op->should_profile() ? &profile_cast_success : success;
   1.470 +  Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
   1.471 +
   1.472 +  if (obj == k_RInfo) {
   1.473 +    k_RInfo = dst;
   1.474 +  } else if (obj == klass_RInfo) {
   1.475 +    klass_RInfo = dst;
   1.476 +  }
   1.477 +  if (k->is_loaded() && !UseCompressedClassPointers) {
   1.478 +    select_different_registers(obj, dst, k_RInfo, klass_RInfo);
   1.479 +  } else {
   1.480 +    Rtmp1 = op->tmp3()->as_register();
   1.481 +    select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
   1.482 +  }
   1.483 +
   1.484 +  assert_different_registers(obj, k_RInfo, klass_RInfo);
   1.485 +
   1.486 +  if (op->should_profile()) {
   1.487 +    Label not_null;
   1.488 +    __ bne(obj, R0, not_null);
   1.489 +    __ delayed()->nop();
   1.490 +    // Object is null; update MDO and exit
   1.491 +    Register mdo  = klass_RInfo;
   1.492 +    __ mov_metadata(mdo, md->constant_encoding());
   1.493 +    Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
   1.494 +    int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
   1.495 +    __ lw(AT, data_addr);
   1.496 +    __ ori(AT, AT, header_bits);
   1.497 +    __ sw(AT,data_addr);
   1.498 +    __ b(*obj_is_null);
   1.499 +    __ delayed()->nop();
   1.500 +    __ bind(not_null);
   1.501 +  } else {
   1.502 +    __ beq(obj, R0, *obj_is_null);
   1.503 +    __ delayed()->nop();
   1.504 +  }
   1.505 +
   1.506 +  if (!k->is_loaded()) {
   1.507 +    klass2reg_with_patching(k_RInfo, op->info_for_patch());
   1.508 +  } else {
   1.509 +#ifdef _LP64
   1.510 +    __ mov_metadata(k_RInfo, k->constant_encoding());
   1.511 +#endif // _LP64
   1.512 +  }
   1.513 +  __ verify_oop(obj);
   1.514 +
   1.515 +  if (op->fast_check()) {
   1.516 +    // get object class
   1.517 +    // not a safepoint as obj null check happens earlier
   1.518 +    if (UseCompressedClassPointers) {
   1.519 +      __ load_klass(Rtmp1, obj);
   1.520 +      __ bne(k_RInfo, Rtmp1, *failure_target);
   1.521 +      __ delayed()->nop();
   1.522 +    } else {
   1.523 +      __ ld(AT, Address(obj, oopDesc::klass_offset_in_bytes()));
   1.524 +      __ bne(k_RInfo, AT, *failure_target);
   1.525 +      __ delayed()->nop();
   1.526 +    }
   1.527 +    // successful cast, fall through to profile or jump
   1.528 +  } else {
   1.529 +    // get object class
   1.530 +    // not a safepoint as obj null check happens earlier
   1.531 +    __ load_klass(klass_RInfo, obj);
   1.532 +    if (k->is_loaded()) {
   1.533 +      // See if we get an immediate positive hit
   1.534 +      __ ld(AT, Address(klass_RInfo, k->super_check_offset()));
   1.535 +      if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
   1.536 +        __ bne(k_RInfo, AT, *failure_target);
   1.537 +        __ delayed()->nop();
   1.538 +        // successful cast, fall through to profile or jump
   1.539 +      } else {
   1.540 +        // See if we get an immediate positive hit
   1.541 +        __ beq(k_RInfo, AT, *success_target);
   1.542 +        __ delayed()->nop();
   1.543 +        // check for self
   1.544 +        __ beq(k_RInfo, klass_RInfo, *success_target);
   1.545 +        __ delayed()->nop();
   1.546 +
   1.547 +        __ push(klass_RInfo);
   1.548 +        __ push(k_RInfo);
   1.549 +        __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
   1.550 +        __ pop(klass_RInfo);
   1.551 +        __ pop(klass_RInfo);
   1.552 +        // result is a boolean
   1.553 +        __ beq(klass_RInfo, R0, *failure_target);
   1.554 +        __ delayed()->nop();
   1.555 +        // successful cast, fall through to profile or jump
   1.556 +      }
   1.557 +    } else {
   1.558 +      // perform the fast part of the checking logic
   1.559 +      __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
   1.560 +      // call out-of-line instance of __ check_klass_subtype_slow_path(...):
   1.561 +      __ push(klass_RInfo);
   1.562 +      __ push(k_RInfo);
   1.563 +      __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
   1.564 +      __ pop(klass_RInfo);
   1.565 +      __ pop(k_RInfo);
   1.566 +      // result is a boolean
   1.567 +      __ beq(k_RInfo, R0, *failure_target);
   1.568 +      __ delayed()->nop();
   1.569 +      // successful cast, fall through to profile or jump
   1.570 +    }
   1.571 +  }
   1.572 +  if (op->should_profile()) {
   1.573 +    Register mdo  = klass_RInfo, recv = k_RInfo;
   1.574 +    __ bind(profile_cast_success);
   1.575 +    __ mov_metadata(mdo, md->constant_encoding());
   1.576 +    __ load_klass(recv, obj);
   1.577 +    Label update_done;
   1.578 +    type_profile_helper(mdo, md, data, recv, success);
   1.579 +    __ b(*success);
   1.580 +    __ delayed()->nop();
   1.581 +
   1.582 +    __ bind(profile_cast_failure);
   1.583 +    __ mov_metadata(mdo, md->constant_encoding());
   1.584 +    Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
   1.585 +    __ ld_ptr(AT, counter_addr);
   1.586 +    __ addi(AT, AT, -DataLayout::counter_increment);
   1.587 +    __ st_ptr(AT, counter_addr);
   1.588 +
   1.589 +    __ b(*failure);
   1.590 +    __ delayed()->nop();
   1.591 +  }
   1.592 +  __ b(*success);
   1.593 +  __ delayed()->nop();
   1.594 +}
   1.595 +
   1.596  
   1.597  
   1.598  void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
   1.599    LIR_Code code = op->code();
   1.600 -//  if (code == lir_store_check) {
   1.601 +  if (code == lir_store_check) {
   1.602      Register value = op->object()->as_register();
   1.603      Register array = op->array()->as_register();
   1.604      Register k_RInfo = op->tmp1()->as_register();
   1.605 @@ -3100,133 +3258,109 @@
   1.606      Register tmp = op->tmp3()->as_register();
   1.607  
   1.608      CodeStub* stub = op->stub();
   1.609 +
   1.610      //check if it needs to be profiled
   1.611      ciMethodData* md;
   1.612      ciProfileData* data;
   1.613 +
   1.614      if (op->should_profile()) {
   1.615        ciMethod* method = op->profiled_method();
   1.616        assert(method != NULL, "Should have method");
   1.617 -      int bci          = op->profiled_bci();
   1.618 +      int bci = op->profiled_bci();
   1.619        md = method->method_data_or_null();
   1.620        assert(md != NULL, "Sanity");
   1.621        data = md->bci_to_data(bci);
   1.622 -      assert(data != NULL, "need data for type check");
   1.623 +      assert(data != NULL,                "need data for type check");
   1.624        assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
   1.625        }
   1.626        Label profile_cast_success, profile_cast_failure, done;
   1.627        Label *success_target = op->should_profile() ? &profile_cast_success : &done;
   1.628 -      Label *failure_target = op->should_profile() ? &profile_cast_failure : &done;
   1.629 -      //__ cmpptr(value, (int32_t)NULL_WORD);
   1.630 +      Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
   1.631 +
   1.632        if(op->should_profile()) {
   1.633 -      Label not_null;
   1.634 -      __ bne(value, R0, not_null);
   1.635 -      __ delayed()->nop();
   1.636 -
   1.637 -      //  __ jcc(Assembler::notEqual, profile_done);
   1.638 -      // __ bne(obj, R0, profile_done);
   1.639 -      //__ delayed()->nop();
   1.640 -
   1.641 -      // Object is null; update methodDataOop
   1.642 -      //ciMethodData* md = method->method_data();
   1.643 -      //if (md == NULL) {
   1.644 -//  bailout("out of memory building methodDataOop");
   1.645 -//  return;
   1.646 -  //    }
   1.647 -    //  ciProfileData* data = md->bci_to_data(bci);
   1.648 -      //assert(data != NULL,       "need data for checkcast");
   1.649 -     // assert(data->is_BitData(), "need BitData for checkcast");
   1.650 -      Register mdo  = klass_RInfo;
   1.651 -      int oop_index = __ oop_recorder()->find_index(md->constant_encoding());
   1.652 -      RelocationHolder rspec = oop_Relocation::spec(oop_index);
   1.653 -      __ relocate(rspec);
   1.654 -#ifndef _LP64
   1.655 -      //by_css
   1.656 -      __ lui(mdo, Assembler::split_high((int)md->constant_encoding()));
   1.657 -      __ addiu(mdo, mdo, Assembler::split_low((int)md->consant_encoding()));
   1.658 -#else
   1.659 -      __ li48(mdo, (long)md->constant_encoding());
   1.660 -#endif
   1.661 -
   1.662 -      Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
   1.663 -      //FIXME, it very ineffictive to replace orl with 3 mips instruction @jerome, 12/27,06
   1.664 -      //__ orl(data_addr, BitData::null_flag_constant());
   1.665 -      int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
   1.666 -      __ lw(AT, data_addr);
   1.667 -      __ ori(AT, AT, header_bits);
   1.668 -      __ sw(AT,data_addr);
   1.669 -      __ b(done);
   1.670 -      __ delayed()->nop();
   1.671 -      __ bind(not_null);
   1.672 -    } else {
   1.673 -      __ beq(value, R0, done);
   1.674 -      __ delayed()->nop();
   1.675 -    }
   1.676 -    //__ verify_oop(obj);
   1.677 +        Label not_null;
   1.678 +        __ bne(value, R0, not_null);
   1.679 +        __ delayed()->nop();
   1.680 +
   1.681 +        Register mdo  = klass_RInfo;
   1.682 +        __ mov_metadata(mdo, md->constant_encoding());
   1.683 +        Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
   1.684 +        int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
   1.685 +        __ lw(AT, data_addr);
   1.686 +        __ ori(AT, AT, header_bits);
   1.687 +        __ sw(AT,data_addr);
   1.688 +        __ b(done);
   1.689 +        __ delayed()->nop();
   1.690 +        __ bind(not_null);
   1.691 +      } else {
   1.692 +        __ beq(value, R0, done);
   1.693 +        __ delayed()->nop();
   1.694 +      }
   1.695 +
   1.696      add_debug_info_for_null_check_here(op->info_for_exception());
   1.697      __ load_klass(k_RInfo, array);
   1.698      __ load_klass(klass_RInfo, value);
   1.699 -     // get instance klass (it's already uncompressed)
   1.700 -    //__ movptr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
   1.701 -    __ daddi (k_RInfo, k_RInfo, in_bytes(ObjArrayKlass::element_klass_offset()));
   1.702 -     // perform the fast part of the checking logic
   1.703 -    //__ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
   1.704 -     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
   1.705 -//1899     __ push(klass_RInfo);
   1.706 -//1900     __ push(k_RInfo);
   1.707 -//1901     __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
   1.708 -//1902     __ pop(klass_RInfo);
   1.709 -//1903     __ pop(k_RInfo);
   1.710 -//1904     // result is a boolean
   1.711 -///1905     __ cmpl(k_RInfo, 0);
   1.712 -//1906     __ jcc(Assembler::equal, *failure_target);
   1.713 -//1907     // fall through to the success case
   1.714 -//1908
   1.715 -//1909     if (op->should_profile()) {
   1.716 -//1910       Register mdo  = klass_RInfo, recv = k_RInfo;
   1.717 -//1911       __ bind(profile_cast_success);
   1.718 -//1912       __ mov_metadata(mdo, md->constant_encoding());
   1.719 -//1913       __ load_klass(recv, value);
   1.720 -//1914       Label update_done;
   1.721 -//1915       type_profile_helper(mdo, md, data, recv, &done);
   1.722 -//1916       __ jmpb(done);
   1.723 -//1917
   1.724 -//1918       __ bind(profile_cast_failure);
   1.725 -//1919       __ mov_metadata(mdo, md->constant_encoding());
   1.726 -//1920       Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
   1.727 -//1921       __ subptr(counter_addr, DataLayout::counter_increment);
   1.728 -//1922       __ jmp(*stub->entry());
   1.729 -//1923     }
   1.730 -//1925     __ bind(done);
   1.731 -//1926   } else
   1.732 -//1927     if (code == lir_checkcast) {
   1.733 -//1928       Register obj = op->object()->as_register();
   1.734 -//1929       Register dst = op->result_opr()->as_register();
   1.735 -//1930       Label success;
   1.736 -//1931       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
   1.737 -//1932       __ bind(success);
   1.738 -//1933       if (dst != obj) {
   1.739 -//1934         __ mov(dst, obj);
   1.740 -//1935       }
   1.741 -//1936     } else
   1.742 -//1937       if (code == lir_instanceof) {
   1.743 -//1938         Register obj = op->object()->as_register();
   1.744 -///1939         Register dst = op->result_opr()->as_register();
   1.745 -//1940         Label success, failure, done;
   1.746 -//1941         emit_typecheck_helper(op, &success, &failure, &failure);
   1.747 -///1942         __ bind(failure);
   1.748 -//1943         __ xorptr(dst, dst);
   1.749 -//1944         __ jmpb(done);
   1.750 -//1945         __ bind(success);
   1.751 -//1946         __ movptr(dst, 1);
   1.752 -//1947         __ bind(done);
   1.753 -//1948       } else {
   1.754 -//1949         ShouldNotReachHere();
   1.755 -//1950       }
   1.756 -//FIXME:wuhui.
   1.757 -
   1.758 +    // get instance klass (it's already uncompressed)
   1.759 +    __ ld_ptr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
   1.760 +    // perform the fast part of the checking logic
   1.761 +    __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, tmp, success_target, failure_target, NULL);
   1.762 +    __ push(klass_RInfo);
   1.763 +    __ push(k_RInfo);
   1.764 +    __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
   1.765 +    __ pop(klass_RInfo);
   1.766 +    __ pop(k_RInfo);
   1.767 +    // result is a boolean
   1.768 +    __ beq(k_RInfo, R0, *failure_target);
   1.769 +    __ delayed()->nop();
   1.770 +    // fall through to the success case
   1.771 +
   1.772 +    if (op->should_profile()) {
   1.773 +      Register mdo = klass_RInfo, recv = k_RInfo;
   1.774 +      __ bind(profile_cast_success);
   1.775 +      __ mov_metadata(mdo, md->constant_encoding());
   1.776 +      __ load_klass(recv, value);
   1.777 +      Label update_done;
   1.778 +      type_profile_helper(mdo, md, data, recv, &done);
   1.779 +      __ b(done);
   1.780 +      __ delayed()->nop();
   1.781 +
   1.782 +      __ bind(profile_cast_failure);
   1.783 +      __ mov_metadata(mdo, md->constant_encoding());
   1.784 +      Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
   1.785 +      __ ld_ptr(AT, counter_addr);
   1.786 +      __ addi(AT, AT, -DataLayout::counter_increment);
   1.787 +      __ st_ptr(AT, counter_addr);
   1.788 +      __ b(*stub->entry());
   1.789 +      __ delayed()->nop();
   1.790 +    }
   1.791 +
   1.792 +    __ bind(done);
   1.793 +  } else if (code == lir_checkcast) {
   1.794 +    Register obj = op->object()->as_register();
   1.795 +    Register dst = op->result_opr()->as_register();
   1.796 +    Label success;
   1.797 +    emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
   1.798 +    __ bind(success);
   1.799 +    if (dst != obj) {
   1.800 +      __ move(dst, obj);
   1.801 +    }
   1.802 +  } else if (code == lir_instanceof) {
   1.803 +    Register obj = op->object()->as_register();
   1.804 +    Register dst = op->result_opr()->as_register();
   1.805 +    Label success, failure, done;
   1.806 +    emit_typecheck_helper(op, &success, &failure, &failure);
   1.807 +    __ bind(failure);
   1.808 +    __ move(dst, R0);
   1.809 +    __ b(done);
   1.810 +    __ delayed()->nop();
   1.811 +    __ bind(success);
   1.812 +    __ addi(dst, R0, 1);
   1.813 +    __ bind(done);
   1.814 +  } else {
   1.815 +    ShouldNotReachHere();
   1.816 +  }
   1.817  }
   1.818  
   1.819 -
   1.820  void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
   1.821    if (op->code() == lir_cas_long) {
   1.822  #ifdef _LP64
   1.823 @@ -3370,8 +3504,18 @@
   1.824        Unimplemented();
   1.825      } else if (right->is_constant()) {
   1.826        // cpu register - constant
   1.827 -      Register res = dest->as_register();
   1.828 -      jint c = right->as_constant_ptr()->as_jint();
   1.829 +      Register res;
   1.830 +      if (dest->is_double_cpu()) {
   1.831 +        res = dest->as_register_lo();
   1.832 +      } else {
   1.833 +        res = dest->as_register();
   1.834 +      }
   1.835 +      jint c;
   1.836 +      if (right->type() == T_INT) {
   1.837 +        c = right->as_constant_ptr()->as_jint();
   1.838 +      } else {
   1.839 +        c = right->as_constant_ptr()->as_jlong();
   1.840 +      }
   1.841  
   1.842        switch (code) {
   1.843    case lir_mul_strictfp:
   1.844 @@ -3763,43 +3907,43 @@
   1.845        int val = right->as_constant_ptr()->as_jint();
   1.846        __ move(AT, val);
   1.847        switch (code) {
   1.848 -  case lir_logic_and:
   1.849 -    __ andr (dstreg, reg, AT);
   1.850 -    break;
   1.851 -  case lir_logic_or:
   1.852 -    __ orr(dstreg, reg, AT);
   1.853 -    break;
   1.854 -  case lir_logic_xor:
   1.855 -    __ xorr(dstreg, reg, AT);
   1.856 -    break;
   1.857 -  default: ShouldNotReachHere();
   1.858 +        case lir_logic_and:
   1.859 +          __ andr (dstreg, reg, AT);
   1.860 +          break;
   1.861 +        case lir_logic_or:
   1.862 +          __ orr(dstreg, reg, AT);
   1.863 +          break;
   1.864 +        case lir_logic_xor:
   1.865 +          __ xorr(dstreg, reg, AT);
   1.866 +          break;
   1.867 +        default: ShouldNotReachHere();
   1.868        }
   1.869      } else if (right->is_stack()) {
   1.870        // added support for stack operands
   1.871        Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
   1.872        switch (code) {
   1.873 -  case lir_logic_and:
   1.874 -    //FIXME. lw or ld_ptr?
   1.875 -    __ lw(AT, raddr);
   1.876 -    __ andr(reg, reg,AT);
   1.877 -    break;
   1.878 -  case lir_logic_or:
   1.879 -    __ lw(AT, raddr);
   1.880 -    __ orr(reg, reg, AT);
   1.881 -    break;
   1.882 -  case lir_logic_xor:
   1.883 -    __ lw(AT, raddr);
   1.884 -    __ xorr(reg, reg, AT);
   1.885 -    break;
   1.886 -  default: ShouldNotReachHere();
   1.887 +        case lir_logic_and:
   1.888 +          //FIXME. lw or ld_ptr?
   1.889 +          __ lw(AT, raddr);
   1.890 +          __ andr(reg, reg,AT);
   1.891 +          break;
   1.892 +        case lir_logic_or:
   1.893 +          __ lw(AT, raddr);
   1.894 +          __ orr(reg, reg, AT);
   1.895 +          break;
   1.896 +        case lir_logic_xor:
   1.897 +          __ lw(AT, raddr);
   1.898 +          __ xorr(reg, reg, AT);
   1.899 +          break;
   1.900 +        default: ShouldNotReachHere();
   1.901        }
   1.902      } else {
   1.903        Register rright = right->as_register();
   1.904        switch (code) {
   1.905 -  case lir_logic_and: __ andr (dstreg, reg, rright); break;
   1.906 -  case lir_logic_or : __ orr  (dstreg, reg, rright); break;
   1.907 -  case lir_logic_xor: __ xorr (dstreg, reg, rright); break;
   1.908 -  default: ShouldNotReachHere();
   1.909 +        case lir_logic_and: __ andr (dstreg, reg, rright); break;
   1.910 +        case lir_logic_or : __ orr  (dstreg, reg, rright); break;
   1.911 +        case lir_logic_xor: __ xorr (dstreg, reg, rright); break;
   1.912 +        default: ShouldNotReachHere();
   1.913        }
   1.914      }
   1.915    } else {
   1.916 @@ -3817,46 +3961,46 @@
   1.917        int r_hi = right->as_constant_ptr()->as_jint_hi();
   1.918  
   1.919        switch (code) {
   1.920 -  case lir_logic_and:
   1.921 -    __ move(AT, r_lo);
   1.922 -    __ andr(dst_lo, l_lo, AT);
   1.923 -    __ move(AT, r_hi);
   1.924 -    __ andr(dst_hi, l_hi, AT);
   1.925 -    break;
   1.926 -
   1.927 -  case lir_logic_or:
   1.928 -    __ move(AT, r_lo);
   1.929 -    __ orr(dst_lo, l_lo, AT);
   1.930 -    __ move(AT, r_hi);
   1.931 -    __ orr(dst_hi, l_hi, AT);
   1.932 -    break;
   1.933 -
   1.934 -  case lir_logic_xor:
   1.935 -    __ move(AT, r_lo);
   1.936 -    __ xorr(dst_lo, l_lo, AT);
   1.937 -    __ move(AT, r_hi);
   1.938 -    __ xorr(dst_hi, l_hi, AT);
   1.939 -    break;
   1.940 -
   1.941 -  default: ShouldNotReachHere();
   1.942 +        case lir_logic_and:
   1.943 +          __ move(AT, r_lo);
   1.944 +          __ andr(dst_lo, l_lo, AT);
   1.945 +          __ move(AT, r_hi);
   1.946 +          __ andr(dst_hi, l_hi, AT);
   1.947 +          break;
   1.948 +
   1.949 +        case lir_logic_or:
   1.950 +          __ move(AT, r_lo);
   1.951 +          __ orr(dst_lo, l_lo, AT);
   1.952 +          __ move(AT, r_hi);
   1.953 +          __ orr(dst_hi, l_hi, AT);
   1.954 +          break;
   1.955 +
   1.956 +        case lir_logic_xor:
   1.957 +          __ move(AT, r_lo);
   1.958 +          __ xorr(dst_lo, l_lo, AT);
   1.959 +          __ move(AT, r_hi);
   1.960 +          __ xorr(dst_hi, l_hi, AT);
   1.961 +          break;
   1.962 +
   1.963 +        default: ShouldNotReachHere();
   1.964        }
   1.965  #else
   1.966        __ li(AT, right->as_constant_ptr()->as_jlong());
   1.967  
   1.968        switch (code) {
   1.969 -  case lir_logic_and:
   1.970 -    __ andr(dst_lo, l_lo, AT);
   1.971 -    break;
   1.972 -
   1.973 -  case lir_logic_or:
   1.974 -    __ orr(dst_lo, l_lo, AT);
   1.975 -    break;
   1.976 -
   1.977 -  case lir_logic_xor:
   1.978 -    __ xorr(dst_lo, l_lo, AT);
   1.979 -    break;
   1.980 -
   1.981 -  default: ShouldNotReachHere();
   1.982 +        case lir_logic_and:
   1.983 +          __ andr(dst_lo, l_lo, AT);
   1.984 +          break;
   1.985 +
   1.986 +        case lir_logic_or:
   1.987 +          __ orr(dst_lo, l_lo, AT);
   1.988 +          break;
   1.989 +
   1.990 +        case lir_logic_xor:
   1.991 +          __ xorr(dst_lo, l_lo, AT);
   1.992 +          break;
   1.993 +
   1.994 +        default: ShouldNotReachHere();
   1.995        }
   1.996  #endif
   1.997  
   1.998 @@ -3865,19 +4009,19 @@
   1.999        Register r_hi = right->as_register_hi();
  1.1000  
  1.1001        switch (code) {
  1.1002 -  case lir_logic_and:
  1.1003 -    __ andr(dst_lo, l_lo, r_lo);
  1.1004 -    NOT_LP64(__ andr(dst_hi, l_hi, r_hi);)
  1.1005 -      break;
  1.1006 -  case lir_logic_or:
  1.1007 -    __ orr(dst_lo, l_lo, r_lo);
  1.1008 -    NOT_LP64(__ orr(dst_hi, l_hi, r_hi);)
  1.1009 -      break;
  1.1010 -  case lir_logic_xor:
  1.1011 -    __ xorr(dst_lo, l_lo, r_lo);
  1.1012 -    NOT_LP64(__ xorr(dst_hi, l_hi, r_hi);)
  1.1013 -      break;
  1.1014 -  default: ShouldNotReachHere();
  1.1015 +        case lir_logic_and:
  1.1016 +          __ andr(dst_lo, l_lo, r_lo);
  1.1017 +          NOT_LP64(__ andr(dst_hi, l_hi, r_hi);)
  1.1018 +            break;
  1.1019 +        case lir_logic_or:
  1.1020 +          __ orr(dst_lo, l_lo, r_lo);
  1.1021 +          NOT_LP64(__ orr(dst_hi, l_hi, r_hi);)
  1.1022 +            break;
  1.1023 +        case lir_logic_xor:
  1.1024 +          __ xorr(dst_lo, l_lo, r_lo);
  1.1025 +          NOT_LP64(__ xorr(dst_hi, l_hi, r_hi);)
  1.1026 +            break;
  1.1027 +        default: ShouldNotReachHere();
  1.1028        }
  1.1029      }
  1.1030    }
  1.1031 @@ -4111,28 +4255,6 @@
  1.1032  
  1.1033  
  1.1034  void LIR_Assembler::align_call(LIR_Code code) {
  1.1035 -//FIXME. aoqi, this right?
  1.1036 -// do nothing since all instructions are word aligned on sparc
  1.1037 -/*
  1.1038 -  if (os::is_MP()) {
  1.1039 -    // make sure that the displacement word of the call ends up word aligned
  1.1040 -    int offset = __ offset();
  1.1041 -    switch (code) {
  1.1042 -      case lir_static_call:
  1.1043 -      case lir_optvirtual_call:
  1.1044 -        offset += NativeCall::displacement_offset;
  1.1045 -        break;
  1.1046 -      case lir_icvirtual_call:
  1.1047 -        offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
  1.1048 -        break;
  1.1049 -      case lir_virtual_call:  // currently, sparc-specific for niagara
  1.1050 -      default: ShouldNotReachHere();
  1.1051 -    }
  1.1052 -    while (offset++ % BytesPerWord != 0) {
  1.1053 -      __ nop();
  1.1054 -    }
  1.1055 -  }
  1.1056 -*/
  1.1057  }
  1.1058  
  1.1059  
  1.1060 @@ -4145,26 +4267,8 @@
  1.1061  
  1.1062  
  1.1063  void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
  1.1064 -  RelocationHolder rh = virtual_call_Relocation::spec(pc());
  1.1065 -//  int oop_index = __ oop_recorder()->allocate_oop_index((jobject)Universe::non_oop_word());
  1.1066 -//  RelocationHolder rspec = oop_Relocation::spec(oop_index);
  1.1067 -///  __ relocate(rspec);
  1.1068 -#ifndef _LP64
  1.1069 -//by_css
  1.1070 -  __ lui(IC_Klass, Assembler::split_high((int)Universe::non_oop_word()));
  1.1071 -  __ addiu(IC_Klass, IC_Klass, Assembler::split_low((int)Universe::non_oop_word()));
  1.1072 -#else
  1.1073 -  __ li48(IC_Klass, (long)Universe::non_oop_word());
  1.1074 -#endif
  1.1075 -  __ call(op->addr(), rh);
  1.1076 -  __ delayed()->nop();
  1.1077 -//  add_call_info(code_offset(), info);
  1.1078 -
  1.1079 -   add_call_info(code_offset(), op->info());
  1.1080 -   assert(!os::is_MP() ||
  1.1081 -          (__ offset() - NativeCall::instruction_size + NativeCall::displacement_offset) % BytesPerWord == 0,
  1.1082 -          "must be aligned");
  1.1083 -
  1.1084 +  __ ic_call(op->addr());
  1.1085 +  add_call_info(code_offset(), op->info());
  1.1086  }
  1.1087  
  1.1088  
  1.1089 @@ -4182,36 +4286,17 @@
  1.1090      bailout("static call stub overflow");
  1.1091      return;
  1.1092    }
  1.1093 -
  1.1094    int start = __ offset();
  1.1095 -  /*
  1.1096 -  if (os::is_MP()) {
  1.1097 -    // make sure that the displacement word of the call ends up word aligned
  1.1098 -    int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
  1.1099 -    while (offset++ % BytesPerWord != 0) {
  1.1100 -      __ nop();
  1.1101 -    }
  1.1102 -  }
  1.1103 -  */
  1.1104    __ relocate(static_stub_Relocation::spec(call_pc));
  1.1105 -  jobject o=NULL;
  1.1106 -  int oop_index = __ oop_recorder()->allocate_oop_index((jobject)o);
  1.1107 -  RelocationHolder rspec = oop_Relocation::spec(oop_index);
  1.1108 +
  1.1109 +  Metadata *o = NULL;
  1.1110 +  int index = __ oop_recorder()->allocate_metadata_index(o);
  1.1111 +  RelocationHolder rspec = metadata_Relocation::spec(index);
  1.1112    __ relocate(rspec);
  1.1113  //see set_to_interpreted
  1.1114 -#ifndef _LP64
  1.1115 -  __ lui(T7, Assembler::split_high((int)o));
  1.1116 -  __ addiu(T7, T7, Assembler::split_low((int)o));
  1.1117 -#else
  1.1118 -  __ li48(Rmethod, (long)o);
  1.1119 -#endif
  1.1120 -#ifndef _LP64
  1.1121 -  __ lui(AT, Assembler::split_high((int)-1));
  1.1122 -  __ addiu(AT, AT, Assembler::split_low((int)-1));
  1.1123 -#else
  1.1124 -  __ li48(AT, (long)-1);
  1.1125 -#endif
  1.1126 -  //assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
  1.1127 +  __ patchable_set48(Rmethod, (long)o);
  1.1128 +
  1.1129 +  __ patchable_set48(AT, (long)-1);
  1.1130    __ jr(AT);
  1.1131    __ delayed()->nop();
  1.1132    assert(__ offset() - start <= call_stub_size, "stub too big");
  1.1133 @@ -4227,52 +4312,29 @@
  1.1134    // (LinearScan assumes that no oops are in fixed registers)
  1.1135  
  1.1136    info->add_register_oop(exceptionOop);
  1.1137 -  //if (!unwind) {
  1.1138 -    // get current pc information
  1.1139 -    // pc is only needed if the method has an exception handler, the unwind code does not need it.
  1.1140 -#ifndef _LP64
  1.1141 -//by_css
  1.1142 -    int pc_for_athrow  = (int)__ pc();
  1.1143 -    int pc_for_athrow_offset = __ offset();
  1.1144 -    Register epc = exceptionPC->as_register();
  1.1145 -    //__ nop();
  1.1146 -    // pc_for_athrow can not point to itself (relocInfo restriction), no need now
  1.1147 -    __ relocate(relocInfo::internal_pc_type);
  1.1148 -    __ lui(epc, Assembler::split_high(pc_for_athrow));
  1.1149 -    __ addiu(epc, epc, Assembler::split_low(pc_for_athrow));
  1.1150 -#else
  1.1151 -    long pc_for_athrow  = (long)__ pc();
  1.1152 -    int pc_for_athrow_offset = __ offset();
  1.1153 -    Register epc = exceptionPC->as_register();
  1.1154 -    //__ nop();
  1.1155 -    // pc_for_athrow can not point to itself (relocInfo restriction), no need now
  1.1156 -    __ relocate(relocInfo::internal_pc_type);
  1.1157 -    __ li48(epc, pc_for_athrow);
  1.1158 -#endif
  1.1159 -    add_call_info(pc_for_athrow_offset, info); // for exception handler
  1.1160 -    __ verify_not_null_oop(V0);
  1.1161 -    // search an exception handler (eax: exception oop, edx: throwing pc)
  1.1162 -    if (compilation()->has_fpu_code()) {
  1.1163 -      __ call(Runtime1::entry_for(Runtime1::handle_exception_id),
  1.1164 -        relocInfo::runtime_call_type);
  1.1165 -    } else {
  1.1166 -      __ call(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id),
  1.1167 -        relocInfo::runtime_call_type);
  1.1168 -    }
  1.1169 -//  } else {
  1.1170 -//    __ call(Runtime1::entry_for(Runtime1::unwind_exception_id),
  1.1171 -//        relocInfo::runtime_call_type);
  1.1172 -//  }
  1.1173 -
  1.1174 -  // enough room for two byte trap
  1.1175 +  long pc_for_athrow  = (long)__ pc();
  1.1176 +  int pc_for_athrow_offset = __ offset();
  1.1177 +  Register epc = exceptionPC->as_register();
  1.1178 +  __ relocate(relocInfo::internal_pc_type);
  1.1179 +  __ li48(epc, pc_for_athrow);
  1.1180 +  add_call_info(pc_for_athrow_offset, info); // for exception handler
  1.1181 +  __ verify_not_null_oop(V0);
  1.1182 +  // search an exception handler (eax: exception oop, edx: throwing pc)
  1.1183 +  if (compilation()->has_fpu_code()) {
  1.1184 +    __ call(Runtime1::entry_for(Runtime1::handle_exception_id),
  1.1185 +      relocInfo::runtime_call_type);
  1.1186 +  } else {
  1.1187 +    __ call(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id),
  1.1188 +      relocInfo::runtime_call_type);
  1.1189 +  }
  1.1190    __ delayed()->nop();
  1.1191  }
  1.1192  
  1.1193 -void LIR_Assembler::unwind_op(LIR_Opr exceptionOop){
  1.1194 - assert(exceptionOop->as_register()== FSR, "must match");
  1.1195 - __ b(_unwind_handler_entry);
  1.1196 - __ delayed()->nop();
  1.1197 - }
  1.1198 +void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
  1.1199 +  assert(exceptionOop->as_register()== FSR, "must match");
  1.1200 +  __ b(_unwind_handler_entry);
  1.1201 +  __ delayed()->nop();
  1.1202 +}
  1.1203  
  1.1204  void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
  1.1205    // optimized version for linear scan:
  1.1206 @@ -4296,12 +4358,11 @@
  1.1207    assert_different_registers(count_reg, value_reg);
  1.1208    switch (code) {
  1.1209      case lir_shl:
  1.1210 -        if (dest->type() == T_INT)
  1.1211 -          __ sllv(dest_reg, value_reg, count_reg);
  1.1212 -        else
  1.1213 -          __ dsllv(dest_reg, value_reg, count_reg);
  1.1214 -        break;
  1.1215 -//__ dsllv(dest_reg, value_reg, count_reg); break;
  1.1216 +      if (dest->type() == T_INT)
  1.1217 +        __ sllv(dest_reg, value_reg, count_reg);
  1.1218 +      else
  1.1219 +        __ dsllv(dest_reg, value_reg, count_reg);
  1.1220 +      break;
  1.1221      case lir_shr:  __ dsrav(dest_reg, value_reg, count_reg); break;
  1.1222      case lir_ushr:
  1.1223  #if 1
  1.1224 @@ -4325,15 +4386,14 @@
  1.1225  //
  1.1226  //  108 ushift_right [a6|I] [a4|I] [a4|I]
  1.1227  //   0x00000055646d2f70: dsll32 a4, a6, 0  \
  1.1228 -//   0x00000055646d2f74: dsrl32 a4, a4, 0  |- error!
  1.1229 -//   0x00000055646d2f78: dsrl a4, a4, a4  /
  1.1230 -        if (left->type() == T_INT && dest->type() == T_INT)
  1.1231 -        {
  1.1232 -          __ dsll32(AT, value_reg, 0);   // Omit the high 32 bits
  1.1233 -          __ dsrl32(AT, AT, 0);
  1.1234 -          __ dsrlv(dest_reg, AT, count_reg); // Unsigned right shift
  1.1235 -          break;
  1.1236 -        }
  1.1237 +//   0x00000055646d2f74: dsrl32 a4, a4, 0   |- error!
  1.1238 +//   0x00000055646d2f78: dsrl a4, a4, a4   /
  1.1239 +     if (left->type() == T_INT && dest->type() == T_INT) {
  1.1240 +       __ dsll32(AT, value_reg, 0);   // Omit the high 32 bits
  1.1241 +       __ dsrl32(AT, AT, 0);
  1.1242 +       __ dsrlv(dest_reg, AT, count_reg); // Unsigned right shift
  1.1243 +       break;
  1.1244 +     }
  1.1245  #endif
  1.1246        __ dsrlv(dest_reg, value_reg, count_reg); break;
  1.1247      default: ShouldNotReachHere();
  1.1248 @@ -4494,8 +4554,7 @@
  1.1249        case lir_shr:  __ dsra(dest_reg, value_reg, count); break;
  1.1250        case lir_ushr:
  1.1251  #if 1
  1.1252 -        if (left->type() == T_INT && dest->type() == T_INT)
  1.1253 -        {
  1.1254 +        if (left->type() == T_INT && dest->type() == T_INT) {
  1.1255          /* Jin: in java, ushift_right requires 32-bit UNSIGNED operation!
  1.1256               However, dsrl will shift in company with the highest 32 bits.
  1.1257               Thus, if the source register contains a negative value,
  1.1258 @@ -4790,13 +4849,12 @@
  1.1259    }
  1.1260  
  1.1261    if (flags & LIR_OpArrayCopy::type_check) {
  1.1262 -      if (UseCompressedOops) {
  1.1263 -    __ lw(AT, src_klass_addr);
  1.1264 -    __ lw(tmp, dst_klass_addr);
  1.1265 -      } else {
  1.1266 -    __ ld(AT, src_klass_addr);
  1.1267 -    __ ld(tmp, dst_klass_addr);
  1.1268 -      }
  1.1269 +    if (UseCompressedClassPointers) {
  1.1270 +      __ lw(AT, src_klass_addr);
  1.1271 +      __ lw(tmp, dst_klass_addr);
  1.1272 +    } else {
  1.1273 +      __ ld(AT, src_klass_addr); __ ld(tmp, dst_klass_addr);
  1.1274 +    }
  1.1275      __ bne_far(AT, tmp, *stub->entry());
  1.1276      __ delayed()->nop();
  1.1277    }
  1.1278 @@ -4810,36 +4868,37 @@
  1.1279      // a type check i needed then at this point the classes are known to be
  1.1280      // the same but again which don't know which type so we can't check them.
  1.1281      Label known_ok, halt;
  1.1282 -//FIXME:wuhui. not finished.    __ mov_metadata(tmp, default_type->constant_encoding());
  1.1283 +    __ mov_metadata(tmp, default_type->constant_encoding());
  1.1284  #ifdef _LP64
  1.1285 -    if (UseCompressedOops) {
  1.1286 -  __ encode_heap_oop(AT);
  1.1287 -  __ lw(tmp, dst_klass_addr);
  1.1288 -    } else
  1.1289 +    if (UseCompressedClassPointers) {
  1.1290 +      __ encode_klass_not_null(tmp);
  1.1291 +    }
  1.1292  #endif
  1.1293 -    {
  1.1294 -  __ ld(tmp, dst_klass_addr);
  1.1295 -    }
  1.1296      if (basic_type != T_OBJECT) {
  1.1297 +      if (UseCompressedClassPointers) {
  1.1298 +        __ lw(AT, dst_klass_addr);
  1.1299 +      } else {
  1.1300 +        __ ld(AT, dst_klass_addr);
  1.1301 +      }
  1.1302        __ bne(AT, tmp, halt);
  1.1303        __ delayed()->nop();
  1.1304 -      if (UseCompressedOops) {
  1.1305 -    __ lw(tmp, src_klass_addr);
  1.1306 +      if (UseCompressedClassPointers) {
  1.1307 +        __ lw(AT, src_klass_addr);
  1.1308        }  else {
  1.1309 -    __ ld(tmp, src_klass_addr);
  1.1310 +        __ ld(AT, src_klass_addr);
  1.1311        }
  1.1312        __ beq(AT, tmp, known_ok);
  1.1313        __ delayed()->nop();
  1.1314      } else {
  1.1315 -  if (UseCompressedOops) {
  1.1316 -      __ lw(tmp, dst_klass_addr);
  1.1317 -  } else {
  1.1318 -      __ ld(tmp, dst_klass_addr);
  1.1319 -  }
  1.1320 -  __ beq(AT, tmp, known_ok);
  1.1321 -  __ delayed()->nop();
  1.1322 -  __ beq(src, dst, known_ok);
  1.1323 -  __ delayed()->nop();
  1.1324 +      if (UseCompressedClassPointers) {
  1.1325 +        __ lw(AT, dst_klass_addr);
  1.1326 +      } else {
  1.1327 +        __ ld(AT, dst_klass_addr);
  1.1328 +      }
  1.1329 +      __ beq(AT, tmp, known_ok);
  1.1330 +      __ delayed()->nop();
  1.1331 +      __ beq(src, dst, known_ok);
  1.1332 +      __ delayed()->nop();
  1.1333      }
  1.1334      __ bind(halt);
  1.1335      __ stop("incorrect type information in arraycopy");
  1.1336 @@ -4916,6 +4975,7 @@
  1.1337    Register lock = op->lock_opr()->is_single_cpu() ? op->lock_opr()->as_register(): op->lock_opr()->as_register_lo();
  1.1338    if (!UseFastLocking) {
  1.1339      __ b_far(*op->stub()->entry());
  1.1340 +    __ delayed()->nop();
  1.1341    } else if (op->code() == lir_lock) {
  1.1342      Register scratch = noreg;
  1.1343      if (UseBiasedLocking) {
  1.1344 @@ -4957,21 +5017,12 @@
  1.1345    assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
  1.1346    Register mdo  = op->mdo()->as_register();
  1.1347  
  1.1348 -  int oop_index = __ oop_recorder()->find_index(md->constant_encoding());
  1.1349 -  RelocationHolder rspec = oop_Relocation::spec(oop_index);
  1.1350 -  __ relocate(rspec);
  1.1351 -#ifndef _LP64
  1.1352 -  //by_css
  1.1353 -  __ lui(mdo, Assembler::split_high((int)md->constant_encoding()));
  1.1354 -  __ addiu(mdo, mdo, Assembler::split_low((int)md->constant_encoding()));
  1.1355 -#else
  1.1356 -  __ li48(mdo, (long)md->constant_encoding());
  1.1357 -#endif
  1.1358 +  __ mov_metadata(mdo, md->constant_encoding());
  1.1359  
  1.1360    Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
  1.1361 -  __ lw(AT, counter_addr);
  1.1362 -  __ addi(AT,AT, DataLayout::counter_increment);
  1.1363 -  __ sw(AT,counter_addr);
  1.1364 +  __ ld_ptr(AT, counter_addr);
  1.1365 +  __ addi(AT, AT, DataLayout::counter_increment);
  1.1366 +  __ st_ptr(AT, counter_addr);
  1.1367  
  1.1368    Bytecodes::Code bc = method->java_code_at_bci(bci);
  1.1369    const bool callee_is_static = callee->is_loaded() && callee->is_static();
  1.1370 @@ -4980,11 +5031,11 @@
  1.1371    if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
  1.1372        !callee_is_static && //required for optimized MH invokes
  1.1373        C1ProfileVirtualCalls) {
  1.1374 -    assert(op->recv()->is_single_cpu(), "recv must be allocated");
  1.1375 -    Register recv = op->recv()->as_register();
  1.1376 -    assert_different_registers(mdo, recv);
  1.1377 -    assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
  1.1378 -    ciKlass* known_klass = op->known_holder();
  1.1379 +      assert(op->recv()->is_single_cpu(), "recv must be allocated");
  1.1380 +      Register recv = op->recv()->as_register();
  1.1381 +      assert_different_registers(mdo, recv);
  1.1382 +      assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
  1.1383 +      ciKlass* known_klass = op->known_holder();
  1.1384      if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
  1.1385        // We know the type that will be seen at this call site; we can
  1.1386        // statically update the methodDataOop rather than needing to do
  1.1387 @@ -4995,14 +5046,14 @@
  1.1388        ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
  1.1389        uint i;
  1.1390        for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1391 -  ciKlass* receiver = vc_data->receiver(i);
  1.1392 -  if (known_klass->equals(receiver)) {
  1.1393 -    Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1394 -    __ lw(AT, data_addr);
  1.1395 -    __ addi(AT, AT, DataLayout::counter_increment);
  1.1396 -    __ sw(AT, data_addr);
  1.1397 -    return;
  1.1398 -  }
  1.1399 +        ciKlass* receiver = vc_data->receiver(i);
  1.1400 +        if (known_klass->equals(receiver)) {
  1.1401 +          Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1402 +          __ ld_ptr(AT, data_addr);
  1.1403 +          __ addi(AT, AT, DataLayout::counter_increment);
  1.1404 +          __ st_ptr(AT, data_addr);
  1.1405 +          return;
  1.1406 +        }
  1.1407        }
  1.1408  
  1.1409        // Receiver type not found in profile data; select an empty slot
  1.1410 @@ -5011,65 +5062,55 @@
  1.1411        // always does a write to the receiver part of the
  1.1412        // VirtualCallData rather than just the first time
  1.1413        for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1414 -  ciKlass* receiver = vc_data->receiver(i);
  1.1415 -  if (receiver == NULL) {
  1.1416 -    Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  1.1417 -    int oop_index = __ oop_recorder()->find_index(known_klass->constant_encoding());
  1.1418 -    RelocationHolder rspec = oop_Relocation::spec(oop_index);
  1.1419 -    __ relocate(rspec);
  1.1420 -#ifndef _LP64
  1.1421 -    //by_css
  1.1422 -    __ lui(AT, Assembler::split_high((int)known_klass->constant_encoding()));
  1.1423 -    __ addiu(AT, AT, Assembler::split_low((int)known_klass->constant_encoding()));
  1.1424 -#else
  1.1425 -    __ li48(AT, (long)known_klass->constant_encoding());
  1.1426 -#endif
  1.1427 -    __ st_ptr(AT,recv_addr);
  1.1428 -    Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1429 -    __ lw(AT, data_addr);
  1.1430 -    __ addi(AT, AT, DataLayout::counter_increment);
  1.1431 -    __ sw(AT, data_addr);
  1.1432 -    return;
  1.1433 -  }
  1.1434 +        ciKlass* receiver = vc_data->receiver(i);
  1.1435 +        if (receiver == NULL) {
  1.1436 +          Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  1.1437 +          __ mov_metadata(AT, known_klass->constant_encoding());
  1.1438 +          __ st_ptr(AT,recv_addr);
  1.1439 +          Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1440 +          __ ld_ptr(AT, data_addr);
  1.1441 +          __ addi(AT, AT, DataLayout::counter_increment);
  1.1442 +          __ st_ptr(AT, data_addr);
  1.1443 +          return;
  1.1444 +        }
  1.1445        }
  1.1446      } else {
  1.1447        //__ ld_ptr(recv, Address(recv, oopDesc::klass_offset_in_bytes()));
  1.1448 -      __ load_klass(recv, recv);
  1.1449 -      Label update_done;
  1.1450 -      uint i;
  1.1451 -      for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1452 -  Label next_test;
  1.1453 +        __ load_klass(recv, recv);
  1.1454 +        Label update_done;
  1.1455 +        uint i;
  1.1456 +        for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1457 +          Label next_test;
  1.1458    // See if the receiver is receiver[n].
  1.1459 -  __ ld_ptr(AT, Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i))));
  1.1460 -  __ bne(recv,AT,next_test);
  1.1461 -  __ delayed()->nop();
  1.1462 -  Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1463 -  __ lw(AT, data_addr);
  1.1464 -  __ addi(AT, AT, DataLayout::counter_increment);
  1.1465 -  __ sw(AT, data_addr);
  1.1466 -  __ b(update_done);
  1.1467 -  __ delayed()->nop();
  1.1468 -  __ bind(next_test);
  1.1469 -      }
  1.1470 +          __ ld_ptr(AT, Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i))));
  1.1471 +          __ bne(recv,AT,next_test);
  1.1472 +          __ delayed()->nop();
  1.1473 +          Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
  1.1474 +          __ ld_ptr(AT, data_addr);
  1.1475 +          __ addi(AT, AT, DataLayout::counter_increment);
  1.1476 +          __ st_ptr(AT, data_addr);
  1.1477 +          __ b(update_done);
  1.1478 +          __ delayed()->nop();
  1.1479 +          __ bind(next_test);
  1.1480 +        }
  1.1481  
  1.1482        // Didn't find receiver; find next empty slot and fill it in
  1.1483 -      for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1484 -  Label next_test;
  1.1485 -  Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  1.1486 -  __ ld_ptr(AT, recv_addr);
  1.1487 -  __ bne(AT, R0, next_test);
  1.1488 -  __ delayed()->nop();
  1.1489 -  __ st_ptr(recv, recv_addr);
  1.1490 -  __ move(AT,DataLayout::counter_increment);
  1.1491 -  __ sw(AT,Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i))));
  1.1492 -  if (i < (VirtualCallData::row_limit() - 1)) {
  1.1493 -    __ b(update_done);
  1.1494 -    __ delayed()->nop();
  1.1495 -  }
  1.1496 -  __ bind(next_test);
  1.1497 -      }
  1.1498 -
  1.1499 -      __ bind(update_done);
  1.1500 +        for (i = 0; i < VirtualCallData::row_limit(); i++) {
  1.1501 +          Label next_test;
  1.1502 +          Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
  1.1503 +          __ ld_ptr(AT, recv_addr);
  1.1504 +          __ bne(AT, R0, next_test);
  1.1505 +          __ delayed()->nop();
  1.1506 +          __ st_ptr(recv, recv_addr);
  1.1507 +          __ move(AT, DataLayout::counter_increment);
  1.1508 +          __ st_ptr(AT, Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i))));
  1.1509 +          if (i < (VirtualCallData::row_limit() - 1)) {
  1.1510 +            __ b(update_done);
  1.1511 +            __ delayed()->nop();
  1.1512 +          }
  1.1513 +          __ bind(next_test);
  1.1514 +        }
  1.1515 +        __ bind(update_done);
  1.1516      }
  1.1517    }
  1.1518  }
  1.1519 @@ -5092,6 +5133,7 @@
  1.1520  }
  1.1521  
  1.1522  void LIR_Assembler::align_backward_branch_target() {
  1.1523 +  __ align(BytesPerWord);
  1.1524  }
  1.1525  
  1.1526  
  1.1527 @@ -5124,11 +5166,11 @@
  1.1528    }
  1.1529  }
  1.1530  
  1.1531 -
  1.1532  void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
  1.1533    assert(addr->is_address() && dest->is_register(), "check");
  1.1534 -  Register reg = dest->as_register();
  1.1535 -  __ lea(dest->as_register(), as_Address(addr->as_address_ptr()));
  1.1536 +  Register reg;
  1.1537 +  reg = dest->as_pointer_register();
  1.1538 +  __ lea(reg, as_Address(addr->as_address_ptr()));
  1.1539  }
  1.1540  
  1.1541  
  1.1542 @@ -5142,8 +5184,8 @@
  1.1543      __ lui(reg, Assembler::split_high((int)o));
  1.1544      __ addiu(reg, reg, Assembler::split_low((int)o));
  1.1545  #else
  1.1546 -    //__ li48(reg, (long)o);
  1.1547 -    __ li(reg, (long)o);
  1.1548 +    __ li48(reg, (long)o);
  1.1549 +    //__ patchable_set48(reg, (long)o);
  1.1550  #endif
  1.1551    } else {
  1.1552      int oop_index = __ oop_recorder()->find_index(o);
  1.1553 @@ -5154,8 +5196,8 @@
  1.1554      __ lui(reg, Assembler::split_high((int)o));
  1.1555      __ addiu(reg, reg, Assembler::split_low((int)o));
  1.1556  #else
  1.1557 -    //__ li48(reg, (long)o);
  1.1558 -    __ li(reg, (long)o);
  1.1559 +    __ li48(reg, (long)o);
  1.1560 +    //__ patchable_set48(reg, (long)o);
  1.1561  #endif
  1.1562    }
  1.1563  }
  1.1564 @@ -5321,6 +5363,7 @@
  1.1565    } else {
  1.1566      ShouldNotReachHere();
  1.1567    }*/
  1.1568 +  ShouldNotReachHere();
  1.1569  }
  1.1570  
  1.1571  #undef __

mercurial