void C1_MacroAssembler::allocate_array(
  Register obj,                        // result: Pointer to array after successful allocation.
  Register len,                        // array length
  Register t1,                         // temp register
  Register t2,                         // temp register
  int      hdr_size,                   // object header size in words
  int      elt_size,                   // element size in bytes
  Register klass,                      // object klass
  Label&   slow_case                   // Continuation point if fast allocation fails.
) {
  assert_different_registers(obj, len, t1, t2, klass);

  // Determine alignment mask.
  assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");

  // Check for negative or excessive length.
  compareU64_and_branch(len, (int32_t)max_array_allocation_length, bcondHigh, slow_case);

  // Compute array size.
  // Note: If 0 <= len <= max_length, len*elt_size + header + alignment is
  // smaller or equal to the largest integer. Also, since top is always
  // aligned, we can do the alignment here instead of at the end address
  // computation.
  const Register arr_size = t2;
  switch (elt_size) {
    case  1: lgr_if_needed(arr_size, len); break;
    case  2: z_sllg(arr_size, len, 1); break;
    case  4: z_sllg(arr_size, len, 2); break;
    case  8: z_sllg(arr_size, len, 3); break;
    default: ShouldNotReachHere();
  }
  add2reg(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment.
  z_nill(arr_size, (~MinObjAlignmentInBytesMask) & 0xffff);            // Align array size.

  try_allocate(obj, arr_size, 0, t1, slow_case);

  initialize_header(obj, klass, len, noreg, t1);

  // Clear rest of allocated space.
  Label done;
  Register object_fields = t1;
  Register Rzero = Z_R1_scratch;
  z_aghi(arr_size, -(hdr_size * BytesPerWord));
  z_bre(done); // Jump if size of fields is zero.
  z_la(object_fields, hdr_size * BytesPerWord, obj);
  z_xgr(Rzero, Rzero);
  initialize_body(object_fields, arr_size, Rzero);
  bind(done);

  // Dtrace support is unimplemented.
  // if (CURRENT_ENV->dtrace_alloc_probes()) {
  //   assert(obj == rax, "must be");
  //   call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
  // }

  verify_oop(obj);
}
void C1_MacroAssembler::initialize_object(
  Register obj,                        // result: pointer to object after successful allocation
  Register klass,                      // object klass
  Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
  int      con_size_in_bytes,          // object size in bytes if   known at compile time
  Register t1,                         // temp register
  Register t2                          // temp register
  ) {
  const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;

  initialize_header(obj, klass, noreg, t1, t2);

#ifdef ASSERT
  {
    lwz(t1, in_bytes(Klass::layout_helper_offset()), klass);
    if (var_size_in_bytes != noreg) {
      cmpw(CCR0, t1, var_size_in_bytes);
    } else {
      cmpwi(CCR0, t1, con_size_in_bytes);
    }
    asm_assert_eq("bad size in initialize_object", 0x753);
  }
#endif

  // Initialize body.
  if (var_size_in_bytes != noreg) {
    // Use a loop.
    addi(t1, obj, hdr_size_in_bytes);                // Compute address of first element.
    addi(t2, var_size_in_bytes, -hdr_size_in_bytes); // Compute size of body.
    initialize_body(t1, t2);
  } else if (con_size_in_bytes > hdr_size_in_bytes) {
    // Use a loop.
    initialize_body(obj, t1, t2, con_size_in_bytes, hdr_size_in_bytes);
  }

  if (CURRENT_ENV->dtrace_alloc_probes()) {
    Unimplemented();
//    assert(obj == O0, "must be");
//    call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
//         relocInfo::runtime_call_type);
  }

  verify_oop(obj);
}
void C1_MacroAssembler::initialize_object(
  Register obj,                        // result: Pointer to object after successful allocation.
  Register klass,                      // object klass
  Register var_size_in_bytes,          // Object size in bytes if unknown at compile time; invalid otherwise.
  int      con_size_in_bytes,          // Object size in bytes if   known at compile time.
  Register t1,                         // temp register
  Register t2                          // temp register
 ) {
  assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
         "con_size_in_bytes is not multiple of alignment");
  assert(var_size_in_bytes == noreg, "not implemented");
  const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;

  const Register Rzero = t2;

  z_xgr(Rzero, Rzero);
  initialize_header(obj, klass, noreg, Rzero, t1);

  // Clear rest of allocated space.
  const int threshold = 4 * BytesPerWord;
  if (con_size_in_bytes <= threshold) {
    // Use explicit null stores.
    // code size = 6*n bytes (n = number of fields to clear)
    for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
      z_stg(Rzero, Address(obj, i));
  } else {
    // Code size generated by initialize_body() is 16.
    Register object_fields = Z_R0_scratch;
    Register len_in_bytes  = Z_R1_scratch;
    z_la(object_fields, hdr_size_in_bytes, obj);
    load_const_optimized(len_in_bytes, con_size_in_bytes - hdr_size_in_bytes);
    initialize_body(object_fields, len_in_bytes, Rzero);
  }

  // Dtrace support is unimplemented.
  //  if (CURRENT_ENV->dtrace_alloc_probes()) {
  //    assert(obj == rax, "must be");
  //    call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id)));
  //  }

  verify_oop(obj);
}
void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
    assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
           "con_size_in_bytes is not multiple of alignment");
    const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;

    initialize_header(obj, klass, noreg, t1, t2);

    if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
        // clear rest of allocated space
        const Register t1_zero = t1;
        const Register index = t2;
        const int threshold = 6 * BytesPerWord;   // approximate break even point for code size (see comments below)
        if (var_size_in_bytes != noreg) {
            mov(index, var_size_in_bytes);
            initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
        } else if (con_size_in_bytes <= threshold) {
            // use explicit null stores
            // code size = 2 + 3*n bytes (n = number of fields to clear)
            xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
            for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
                movptr(Address(obj, i), t1_zero);
        } else if (con_size_in_bytes > hdr_size_in_bytes) {
            // use loop to null out the fields
            // code size = 16 bytes for even n (n = number of fields to clear)
            // initialize last object field first if odd number of fields
            xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
            movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
            // initialize last object field if constant size is odd
            if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
                movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
            // initialize remaining object fields: rdx is a multiple of 2
            {   Label loop;
                bind(loop);
                movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
                       t1_zero);
                NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
                                t1_zero);)
                decrement(index);
                jcc(Assembler::notZero, loop);
            }
        }
void C1_MacroAssembler::allocate_array(
    Register obj,                        // result: pointer to array after successful allocation
    Register len,                        // array length
    Register t1,                         // temp register
    Register t2,                         // temp register
    Register t3,                         // temp register
    int      hdr_size,                   // object header size in words
    int      elt_size,                   // element size in bytes
    Register klass,                      // object klass
    Label&   slow_case                   // continuation point if fast allocation fails
) {
    assert_different_registers(obj, len, t1, t2, t3, klass);
    assert(klass == G5, "must be G5");
    assert(t1 == G1, "must be G1");

    // determine alignment mask
    assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");

    // check for negative or excessive length
    // note: the maximum length allowed is chosen so that arrays of any
    //       element size with this length are always smaller or equal
    //       to the largest integer (i.e., array size computation will
    //       not overflow)
    set(max_array_allocation_length, t1);
    cmp(len, t1);
    br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);

    // compute array size
    // note: if 0 <= len <= max_length, len*elt_size + header + alignment is
    //       smaller or equal to the largest integer; also, since top is always
    //       aligned, we can do the alignment here instead of at the end address
    //       computation
    const Register arr_size = t1;
    switch (elt_size) {
    case  1:
        delayed()->mov(len,    arr_size);
        break;
    case  2:
        delayed()->sll(len, 1, arr_size);
        break;
    case  4:
        delayed()->sll(len, 2, arr_size);
        break;
    case  8:
        delayed()->sll(len, 3, arr_size);
        break;
    default:
        ShouldNotReachHere();
    }
    add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment
    and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size);                     // align array size

    // allocate space & initialize header
    if (UseTLAB) {
        tlab_allocate(obj, arr_size, 0, t2, slow_case);
    } else {
        eden_allocate(obj, arr_size, 0, t2, t3, slow_case);
    }
    initialize_header(obj, klass, len, t2, t3);

    // initialize body
    const Register base  = t2;
    const Register index = t3;
    add(obj, hdr_size * wordSize, base);               // compute address of first element
    sub(arr_size, hdr_size * wordSize, index);         // compute index = number of words to clear
    initialize_body(base, index);

    if (CURRENT_ENV->dtrace_alloc_probes()) {
        assert(obj == O0, "must be");
        call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
             relocInfo::runtime_call_type);
        delayed()->nop();
    }

    verify_oop(obj);
}
void C1_MacroAssembler::initialize_object(
    Register obj,                        // result: pointer to object after successful allocation
    Register klass,                      // object klass
    Register var_size_in_bytes,          // object size in bytes if unknown at compile time; invalid otherwise
    int      con_size_in_bytes,          // object size in bytes if   known at compile time
    Register t1,                         // temp register
    Register t2                          // temp register
) {
    const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;

    initialize_header(obj, klass, noreg, t1, t2);

#ifdef ASSERT
    {
        Label ok;
        ld(klass, in_bytes(Klass::layout_helper_offset()), t1);
        if (var_size_in_bytes != noreg) {
            cmp_and_brx_short(t1, var_size_in_bytes, Assembler::equal, Assembler::pt, ok);
        } else {
            cmp_and_brx_short(t1, con_size_in_bytes, Assembler::equal, Assembler::pt, ok);
        }
        stop("bad size in initialize_object");
        should_not_reach_here();

        bind(ok);
    }

#endif

    // initialize body
    const int threshold = 5 * HeapWordSize;              // approximate break even point for code size
    if (var_size_in_bytes != noreg) {
        // use a loop
        add(obj, hdr_size_in_bytes, t1);               // compute address of first element
        sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body
        initialize_body(t1, t2);
#ifndef _LP64
    } else if (con_size_in_bytes < threshold * 2) {
        // on v9 we can do double word stores to fill twice as much space.
        assert(hdr_size_in_bytes % 8 == 0, "double word aligned");
        assert(con_size_in_bytes % 8 == 0, "double word aligned");
        for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i);
#endif
    } else if (con_size_in_bytes <= threshold) {
        // use explicit NULL stores
        for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize)     st_ptr(G0, obj, i);
    } else if (con_size_in_bytes > hdr_size_in_bytes) {
        // use a loop
        const Register base  = t1;
        const Register index = t2;
        add(obj, hdr_size_in_bytes, base);               // compute address of first element
        // compute index = number of words to clear
        set(con_size_in_bytes - hdr_size_in_bytes, index);
        initialize_body(base, index);
    }

    if (CURRENT_ENV->dtrace_alloc_probes()) {
        assert(obj == O0, "must be");
        call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
             relocInfo::runtime_call_type);
        delayed()->nop();
    }

    verify_oop(obj);
}
Exemple #7
0
OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {

  OopMapSet* oop_maps = NULL;
  // for better readability
  const bool must_gc_arguments = true;
  const bool dont_gc_arguments = false;

  // stub code & info for the different stubs
  switch (id) {
    case forward_exception_id:
      {
        oop_maps = generate_handle_exception(id, sasm);
      }
      break;

    case new_instance_id:
    case fast_new_instance_id:
    case fast_new_instance_init_check_id:
      {
        Register G5_klass = G5; // Incoming
        Register O0_obj   = O0; // Outgoing

        if (id == new_instance_id) {
          __ set_info("new_instance", dont_gc_arguments);
        } else if (id == fast_new_instance_id) {
          __ set_info("fast new_instance", dont_gc_arguments);
        } else {
          assert(id == fast_new_instance_init_check_id, "bad StubID");
          __ set_info("fast new_instance init check", dont_gc_arguments);
        }

        if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
            UseTLAB && FastTLABRefill) {
          Label slow_path;
          Register G1_obj_size = G1;
          Register G3_t1 = G3;
          Register G4_t2 = G4;
          assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);

          // Push a frame since we may do dtrace notification for the
          // allocation which requires calling out and we don't want
          // to stomp the real return address.
          __ save_frame(0);

          if (id == fast_new_instance_init_check_id) {
            // make sure the klass is initialized
            __ ldub(G5_klass, in_bytes(InstanceKlass::init_state_offset()), G3_t1);
            __ cmp_and_br_short(G3_t1, InstanceKlass::fully_initialized, Assembler::notEqual, Assembler::pn, slow_path);
          }
#ifdef ASSERT
          // assert object can be fast path allocated
          {
            Label ok, not_ok;
          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
          // make sure it's an instance (LH > 0)
          __ cmp_and_br_short(G1_obj_size, 0, Assembler::lessEqual, Assembler::pn, not_ok);
          __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
          __ br(Assembler::zero, false, Assembler::pn, ok);
          __ delayed()->nop();
          __ bind(not_ok);
          __ stop("assert(can be fast path allocated)");
          __ should_not_reach_here();
          __ bind(ok);
          }
#endif // ASSERT
          // if we got here then the TLAB allocation failed, so try
          // refilling the TLAB or allocating directly from eden.
          Label retry_tlab, try_eden;
          __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass

          __ bind(retry_tlab);

          // get the instance size
          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);

          __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);

          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
          __ verify_oop(O0_obj);
          __ mov(O0, I0);
          __ ret();
          __ delayed()->restore();

          __ bind(try_eden);
          // get the instance size
          __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
          __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
          __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);

          __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
          __ verify_oop(O0_obj);
          __ mov(O0, I0);
          __ ret();
          __ delayed()->restore();

          __ bind(slow_path);

          // pop this frame so generate_stub_call can push it's own
          __ restore();
        }

        oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
        // I0->O0: new instance
      }

      break;

    case counter_overflow_id:
        // G4 contains bci, G5 contains method
      oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
      break;

    case new_type_array_id:
    case new_object_array_id:
      {
        Register G5_klass = G5; // Incoming
        Register G4_length = G4; // Incoming
        Register O0_obj   = O0; // Outgoing

        Address klass_lh(G5_klass, Klass::layout_helper_offset());
        assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
        assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
        // Use this offset to pick out an individual byte of the layout_helper:
        const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
                                                 - Klass::_lh_header_size_shift / BitsPerByte);

        if (id == new_type_array_id) {
          __ set_info("new_type_array", dont_gc_arguments);
        } else {
          __ set_info("new_object_array", dont_gc_arguments);
        }

#ifdef ASSERT
        // assert object type is really an array of the proper kind
        {
          Label ok;
          Register G3_t1 = G3;
          __ ld(klass_lh, G3_t1);
          __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
          int tag = ((id == new_type_array_id)
                     ? Klass::_lh_array_tag_type_value
                     : Klass::_lh_array_tag_obj_value);
          __ cmp_and_brx_short(G3_t1, tag, Assembler::equal, Assembler::pt, ok);
          __ stop("assert(is an array klass)");
          __ should_not_reach_here();
          __ bind(ok);
        }
#endif // ASSERT

        if (UseTLAB && FastTLABRefill) {
          Label slow_path;
          Register G1_arr_size = G1;
          Register G3_t1 = G3;
          Register O1_t2 = O1;
          assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);

          // check that array length is small enough for fast path
          __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
          __ cmp_and_br_short(G4_length, G3_t1, Assembler::greaterUnsigned, Assembler::pn, slow_path);

          // if we got here then the TLAB allocation failed, so try
          // refilling the TLAB or allocating directly from eden.
          Label retry_tlab, try_eden;
          __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass

          __ bind(retry_tlab);

          // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
          __ ld(klass_lh, G3_t1);
          __ sll(G4_length, G3_t1, G1_arr_size);
          __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
          __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
          __ add(G1_arr_size, G3_t1, G1_arr_size);
          __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
          __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);

          __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size

          __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
          __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
          __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
          __ add(O0_obj, G3_t1, G3_t1);       // body start
          __ initialize_body(G3_t1, O1_t2);
          __ verify_oop(O0_obj);
          __ retl();
          __ delayed()->nop();

          __ bind(try_eden);
          // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
          __ ld(klass_lh, G3_t1);
          __ sll(G4_length, G3_t1, G1_arr_size);
          __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
          __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
          __ add(G1_arr_size, G3_t1, G1_arr_size);
          __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
          __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);

          __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
          __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);

          __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
          __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
          __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
          __ add(O0_obj, G3_t1, G3_t1);       // body start
          __ initialize_body(G3_t1, O1_t2);
          __ verify_oop(O0_obj);
          __ retl();
          __ delayed()->nop();

          __ bind(slow_path);
        }

        if (id == new_type_array_id) {
          oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
        } else {
          oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
        }
        // I0 -> O0: new array
      }
      break;

    case new_multi_array_id:
      { // O0: klass
        // O1: rank
        // O2: address of 1st dimension
        __ set_info("new_multi_array", dont_gc_arguments);
        oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
        // I0 -> O0: new multi array
      }
      break;

    case register_finalizer_id:
      {
        __ set_info("register_finalizer", dont_gc_arguments);

        // load the klass and check the has finalizer flag
        Label register_finalizer;
        Register t = O1;
        __ load_klass(O0, t);
        __ ld(t, in_bytes(Klass::access_flags_offset()), t);
        __ set(JVM_ACC_HAS_FINALIZER, G3);
        __ andcc(G3, t, G0);
        __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
        __ delayed()->nop();

        // do a leaf return
        __ retl();
        __ delayed()->nop();

        __ bind(register_finalizer);
        OopMap* oop_map = save_live_registers(sasm);
        int call_offset = __ call_RT(noreg, noreg,
                                     CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
        oop_maps = new OopMapSet();
        oop_maps->add_gc_map(call_offset, oop_map);

        // Now restore all the live registers
        restore_live_registers(sasm);

        __ ret();
        __ delayed()->restore();
      }
      break;

    case throw_range_check_failed_id:
      { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
        // G4: index
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
      }
      break;

    case throw_index_exception_id:
      { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
        // G4: index
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
      }
      break;

    case throw_div0_exception_id:
      { __ set_info("throw_div0_exception", dont_gc_arguments);
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
      }
      break;

    case throw_null_pointer_exception_id:
      { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
      }
      break;

    case handle_exception_id:
      { __ set_info("handle_exception", dont_gc_arguments);
        oop_maps = generate_handle_exception(id, sasm);
      }
      break;

    case handle_exception_from_callee_id:
      { __ set_info("handle_exception_from_callee", dont_gc_arguments);
        oop_maps = generate_handle_exception(id, sasm);
      }
      break;

    case unwind_exception_id:
      {
        // O0: exception
        // I7: address of call to this method

        __ set_info("unwind_exception", dont_gc_arguments);
        __ mov(Oexception, Oexception->after_save());
        __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());

        __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
                        G2_thread, Oissuing_pc->after_save());
        __ verify_not_null_oop(Oexception->after_save());

        // Restore SP from L7 if the exception PC is a method handle call site.
        __ mov(O0, G5);  // Save the target address.
        __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
        __ tst(L0);  // Condition codes are preserved over the restore.
        __ restore();

        __ jmp(G5, 0);
        __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
      }
      break;

    case throw_array_store_exception_id:
      {
        __ set_info("throw_array_store_exception", dont_gc_arguments);
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
      }
      break;

    case throw_class_cast_exception_id:
      {
        // G4: object
        __ set_info("throw_class_cast_exception", dont_gc_arguments);
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
      }
      break;

    case throw_incompatible_class_change_error_id:
      {
        __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
        oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
      }
      break;

    case slow_subtype_check_id:
      { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
        // Arguments :
        //
        //      ret  : G3
        //      sub  : G3, argument, destroyed
        //      super: G1, argument, not changed
        //      raddr: O7, blown by call
        Label miss;

        __ save_frame(0);               // Blow no registers!

        __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);

        __ mov(1, G3);
        __ ret();                       // Result in G5 is 'true'
        __ delayed()->restore();        // free copy or add can go here

        __ bind(miss);
        __ mov(0, G3);
        __ ret();                       // Result in G5 is 'false'
        __ delayed()->restore();        // free copy or add can go here
      }

    case monitorenter_nofpu_id:
    case monitorenter_id:
      { // G4: object
        // G5: lock address
        __ set_info("monitorenter", dont_gc_arguments);

        int save_fpu_registers = (id == monitorenter_id);
        // make a frame and preserve the caller's caller-save registers
        OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);

        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);

        oop_maps = new OopMapSet();
        oop_maps->add_gc_map(call_offset, oop_map);
        restore_live_registers(sasm, save_fpu_registers);

        __ ret();
        __ delayed()->restore();
      }
      break;

    case monitorexit_nofpu_id:
    case monitorexit_id:
      { // G4: lock address
        // note: really a leaf routine but must setup last java sp
        //       => use call_RT for now (speed can be improved by
        //       doing last java sp setup manually)
        __ set_info("monitorexit", dont_gc_arguments);

        int save_fpu_registers = (id == monitorexit_id);
        // make a frame and preserve the caller's caller-save registers
        OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);

        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);

        oop_maps = new OopMapSet();
        oop_maps->add_gc_map(call_offset, oop_map);
        restore_live_registers(sasm, save_fpu_registers);

        __ ret();
        __ delayed()->restore();
      }
      break;

    case deoptimize_id:
      {
        __ set_info("deoptimize", dont_gc_arguments);
        OopMap* oop_map = save_live_registers(sasm);
        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
        oop_maps = new OopMapSet();
        oop_maps->add_gc_map(call_offset, oop_map);
        restore_live_registers(sasm);
        DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
        assert(deopt_blob != NULL, "deoptimization blob must have been created");
        AddressLiteral dest(deopt_blob->unpack_with_reexecution());
        __ jump_to(dest, O0);
        __ delayed()->restore();
      }
      break;

    case access_field_patching_id:
      { __ set_info("access_field_patching", dont_gc_arguments);
        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
      }
      break;

    case load_klass_patching_id:
      { __ set_info("load_klass_patching", dont_gc_arguments);
        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
      }
      break;

    case load_mirror_patching_id:
      { __ set_info("load_mirror_patching", dont_gc_arguments);
        oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
      }
      break;

    case dtrace_object_alloc_id:
      { // O0: object
        __ set_info("dtrace_object_alloc", dont_gc_arguments);
        // we can't gc here so skip the oopmap but make sure that all
        // the live registers get saved.
        save_live_registers(sasm);

        __ save_thread(L7_thread_cache);
        __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
                relocInfo::runtime_call_type);
        __ delayed()->mov(I0, O0);
        __ restore_thread(L7_thread_cache);

        restore_live_registers(sasm);
        __ ret();
        __ delayed()->restore();
      }
      break;

#if INCLUDE_ALL_GCS
    case g1_pre_barrier_slow_id:
      { // G4: previous value of memory
        BarrierSet* bs = Universe::heap()->barrier_set();
        if (bs->kind() != BarrierSet::G1SATBCTLogging) {
          __ save_frame(0);
          __ set((int)id, O1);
          __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
          __ should_not_reach_here();
          break;
        }

        __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);

        Register pre_val = G4;
        Register tmp  = G1_scratch;
        Register tmp2 = G3_scratch;

        Label refill, restart;
        bool with_frame = false; // I don't know if we can do with-frame.
        int satb_q_index_byte_offset =
          in_bytes(JavaThread::satb_mark_queue_offset() +
                   PtrQueue::byte_offset_of_index());
        int satb_q_buf_byte_offset =
          in_bytes(JavaThread::satb_mark_queue_offset() +
                   PtrQueue::byte_offset_of_buf());

        __ bind(restart);
        // Load the index into the SATB buffer. PtrQueue::_index is a
        // size_t so ld_ptr is appropriate
        __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);

        // index == 0?
        __ cmp_and_brx_short(tmp, G0, Assembler::equal, Assembler::pn, refill);

        __ ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
        __ sub(tmp, oopSize, tmp);

        __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
        // Use return-from-leaf
        __ retl();
        __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);

        __ bind(refill);
        __ save_frame(0);

        __ mov(pre_val, L0);
        __ mov(tmp,     L1);
        __ mov(tmp2,    L2);

        __ call_VM_leaf(L7_thread_cache,
                        CAST_FROM_FN_PTR(address,
                                         SATBMarkQueueSet::handle_zero_index_for_thread),
                                         G2_thread);

        __ mov(L0, pre_val);
        __ mov(L1, tmp);
        __ mov(L2, tmp2);

        __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
        __ delayed()->restore();
      }
      break;

    case g1_post_barrier_slow_id:
      {
        BarrierSet* bs = Universe::heap()->barrier_set();
        if (bs->kind() != BarrierSet::G1SATBCTLogging) {
          __ save_frame(0);
          __ set((int)id, O1);
          __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
          __ should_not_reach_here();
          break;
        }

        __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);

        Register addr = G4;
        Register cardtable = G5;
        Register tmp  = G1_scratch;
        Register tmp2 = G3_scratch;
        jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;

        Label not_already_dirty, restart, refill;

#ifdef _LP64
        __ srlx(addr, CardTableModRefBS::card_shift, addr);
#else
        __ srl(addr, CardTableModRefBS::card_shift, addr);
#endif

        AddressLiteral rs(byte_map_base);
        __ set(rs, cardtable);         // cardtable := <card table base>
        __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]

        assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
        __ cmp_and_br_short(tmp, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);

        // We didn't take the branch, so we're already dirty: return.
        // Use return-from-leaf
        __ retl();
        __ delayed()->nop();

        // Not dirty.
        __ bind(not_already_dirty);

        // Get cardtable + tmp into a reg by itself
        __ add(addr, cardtable, tmp2);

        // First, dirty it.
        __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).

        Register tmp3 = cardtable;
        Register tmp4 = tmp;

        // these registers are now dead
        addr = cardtable = tmp = noreg;

        int dirty_card_q_index_byte_offset =
          in_bytes(JavaThread::dirty_card_queue_offset() +
                   PtrQueue::byte_offset_of_index());
        int dirty_card_q_buf_byte_offset =
          in_bytes(JavaThread::dirty_card_queue_offset() +
                   PtrQueue::byte_offset_of_buf());

        __ bind(restart);

        // Get the index into the update buffer. PtrQueue::_index is
        // a size_t so ld_ptr is appropriate here.
        __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);

        // index == 0?
        __ cmp_and_brx_short(tmp3, G0, Assembler::equal,  Assembler::pn, refill);

        __ ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
        __ sub(tmp3, oopSize, tmp3);

        __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
        // Use return-from-leaf
        __ retl();
        __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);

        __ bind(refill);
        __ save_frame(0);

        __ mov(tmp2, L0);
        __ mov(tmp3, L1);
        __ mov(tmp4, L2);

        __ call_VM_leaf(L7_thread_cache,
                        CAST_FROM_FN_PTR(address,
                                         DirtyCardQueueSet::handle_zero_index_for_thread),
                                         G2_thread);

        __ mov(L0, tmp2);
        __ mov(L1, tmp3);
        __ mov(L2, tmp4);

        __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
        __ delayed()->restore();
      }
      break;
#endif // INCLUDE_ALL_GCS

    case predicate_failed_trap_id:
      {
        __ set_info("predicate_failed_trap", dont_gc_arguments);
        OopMap* oop_map = save_live_registers(sasm);

        int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));

        oop_maps = new OopMapSet();
        oop_maps->add_gc_map(call_offset, oop_map);

        DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
        assert(deopt_blob != NULL, "deoptimization blob must have been created");
        restore_live_registers(sasm);

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

    default:
      { __ set_info("unimplemented entry", dont_gc_arguments);
        __ save_frame(0);
        __ set((int)id, O1);
        __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
        __ should_not_reach_here();
      }
      break;
  }
  return oop_maps;
}
void C1_MacroAssembler::allocate_array(
  Register obj,                        // result: pointer to array after successful allocation
  Register len,                        // array length
  Register t1,                         // temp register
  Register t2,                         // temp register
  Register t3,                         // temp register
  int      hdr_size,                   // object header size in words
  int      elt_size,                   // element size in bytes
  Register klass,                      // object klass
  Label&   slow_case                   // continuation point if fast allocation fails
) {
  assert_different_registers(obj, len, t1, t2, t3, klass);

  // Determine alignment mask.
  assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
  int log2_elt_size = exact_log2(elt_size);

  // Check for negative or excessive length.
  size_t max_length = max_array_allocation_length >> log2_elt_size;
  if (UseTLAB) {
    size_t max_tlab = align_size_up(ThreadLocalAllocBuffer::max_size() >> log2_elt_size, 64*K);
    if (max_tlab < max_length) { max_length = max_tlab; }
  }
  load_const_optimized(t1, max_length);
  cmpld(CCR0, len, t1);
  bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::greater), slow_case);

  // compute array size
  // note: If 0 <= len <= max_length, len*elt_size + header + alignment is
  //       smaller or equal to the largest integer; also, since top is always
  //       aligned, we can do the alignment here instead of at the end address
  //       computation.
  const Register arr_size = t1;
  Register arr_len_in_bytes = len;
  if (elt_size != 1) {
    sldi(t1, len, log2_elt_size);
    arr_len_in_bytes = t1;
  }
  addi(arr_size, arr_len_in_bytes, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment.
  clrrdi(arr_size, arr_size, LogMinObjAlignmentInBytes);                              // Align array size.

  // Allocate space & initialize header.
  if (UseTLAB) {
    tlab_allocate(obj, arr_size, 0, t2, slow_case);
  } else {
    eden_allocate(obj, arr_size, 0, t2, t3, slow_case);
  }
  initialize_header(obj, klass, len, t2, t3);

  // Initialize body.
  const Register base  = t2;
  const Register index = t3;
  addi(base, obj, hdr_size * wordSize);               // compute address of first element
  addi(index, arr_size, -(hdr_size * wordSize));      // compute index = number of bytes to clear
  initialize_body(base, index);

  if (CURRENT_ENV->dtrace_alloc_probes()) {
    Unimplemented();
    //assert(obj == O0, "must be");
    //call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
    //     relocInfo::runtime_call_type);
  }

  verify_oop(obj);
}