/*
 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
 * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 *
 */

#include "precompiled.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "gc/shared/barrierSet.hpp"
#include "gc/shared/barrierSetAssembler.hpp"
#include "interp_masm_riscv.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "logging/log.hpp"
#include "oops/arrayOop.hpp"
#include "oops/markWord.hpp"
#include "oops/method.hpp"
#include "oops/methodData.hpp"
#include "prims/jvmtiExport.hpp"
#include "prims/jvmtiThreadState.hpp"
#include "runtime/basicLock.hpp"
#include "runtime/frame.inline.hpp"
#include "runtime/javaThread.hpp"
#include "runtime/safepointMechanism.hpp"
#include "runtime/sharedRuntime.hpp"
#include "utilities/powerOfTwo.hpp"

void InterpreterMacroAssembler::narrow(Register result) {
  // Get method->_constMethod->_result_type
  ld(t0, Address(fp, frame::interpreter_frame_method_offset * wordSize));
  ld(t0, Address(t0, Method::const_offset()));
  lbu(t0, Address(t0, ConstMethod::result_type_offset()));

  Label done, notBool, notByte, notChar;

  // common case first
  mv(t1, T_INT);
  beq(t0, t1, done);

  // mask integer result to narrower return type.
  mv(t1, T_BOOLEAN);
  bne(t0, t1, notBool);

  andi(result, result, 0x1);
  j(done);

  bind(notBool);
  mv(t1, T_BYTE);
  bne(t0, t1, notByte);
  sign_extend(result, result, 8);
  j(done);

  bind(notByte);
  mv(t1, T_CHAR);
  bne(t0, t1, notChar);
  zero_extend(result, result, 16);
  j(done);

  bind(notChar);
  sign_extend(result, result, 16);

  // Nothing to do for T_INT
  bind(done);
  addw(result, result, zr);
}

void InterpreterMacroAssembler::jump_to_entry(address entry) {
  assert(entry != nullptr, "Entry must have been generated by now");
  j(entry);
}

void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
  if (JvmtiExport::can_pop_frame()) {
    Label L;
    // Initiate popframe handling only if it is not already being
    // processed. If the flag has the popframe_processing bit set,
    // it means that this code is called *during* popframe handling - we
    // don't want to reenter.
    // This method is only called just after the call into the vm in
    // call_VM_base, so the arg registers are available.
    lwu(t1, Address(xthread, JavaThread::popframe_condition_offset()));
    test_bit(t0, t1, exact_log2(JavaThread::popframe_pending_bit));
    beqz(t0, L);
    test_bit(t0, t1, exact_log2(JavaThread::popframe_processing_bit));
    bnez(t0, L);
    // Call Interpreter::remove_activation_preserving_args_entry() to get the
    // address of the same-named entrypoint in the generated interpreter code.
    call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
    jr(x10);
    bind(L);
  }
}


void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
  ld(x12, Address(xthread, JavaThread::jvmti_thread_state_offset()));
  const Address tos_addr(x12, JvmtiThreadState::earlyret_tos_offset());
  const Address oop_addr(x12, JvmtiThreadState::earlyret_oop_offset());
  const Address val_addr(x12, JvmtiThreadState::earlyret_value_offset());
  switch (state) {
    case atos:
      ld(x10, oop_addr);
      sd(zr, oop_addr);
      verify_oop(x10);
      break;
    case ltos:
      ld(x10, val_addr);
      break;
    case btos:  // fall through
    case ztos:  // fall through
    case ctos:  // fall through
    case stos:  // fall through
    case itos:
      lwu(x10, val_addr);
      break;
    case ftos:
      flw(f10, val_addr);
      break;
    case dtos:
      fld(f10, val_addr);
      break;
    case vtos:
      /* nothing to do */
      break;
    default:
      ShouldNotReachHere();
  }
  // Clean up tos value in the thread object
  mv(t0, (int)ilgl);
  sw(t0, tos_addr);
  sw(zr, val_addr);
}


void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
  if (JvmtiExport::can_force_early_return()) {
    Label L;
    ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
    beqz(t0, L);  // if thread->jvmti_thread_state() is null then exit

    // Initiate earlyret handling only if it is not already being processed.
    // If the flag has the earlyret_processing bit set, it means that this code
    // is called *during* earlyret handling - we don't want to reenter.
    lwu(t0, Address(t0, JvmtiThreadState::earlyret_state_offset()));
    mv(t1, JvmtiThreadState::earlyret_pending);
    bne(t0, t1, L);

    // Call Interpreter::remove_activation_early_entry() to get the address of the
    // same-named entrypoint in the generated interpreter code.
    ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
    lwu(t0, Address(t0, JvmtiThreadState::earlyret_tos_offset()));
    call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), t0);
    jr(x10);
    bind(L);
  }
}

void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
  assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
  if (AvoidUnalignedAccesses && (bcp_offset % 2)) {
    lbu(t1, Address(xbcp, bcp_offset));
    lbu(reg, Address(xbcp, bcp_offset + 1));
    slli(t1, t1, 8);
    add(reg, reg, t1);
  } else {
    lhu(reg, Address(xbcp, bcp_offset));
    revb_h_h_u(reg, reg);
  }
}

void InterpreterMacroAssembler::get_dispatch() {
  ExternalAddress target((address)Interpreter::dispatch_table());
  relocate(target.rspec(), [&] {
    int32_t offset;
    la_patchable(xdispatch, target, offset);
    addi(xdispatch, xdispatch, offset);
  });
}

void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
                                                       Register tmp,
                                                       int bcp_offset,
                                                       size_t index_size) {
  assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
  if (index_size == sizeof(u2)) {
    if (AvoidUnalignedAccesses) {
      assert_different_registers(index, tmp);
      load_unsigned_byte(index, Address(xbcp, bcp_offset));
      load_unsigned_byte(tmp, Address(xbcp, bcp_offset + 1));
      slli(tmp, tmp, 8);
      add(index, index, tmp);
    } else {
      load_unsigned_short(index, Address(xbcp, bcp_offset));
    }
  } else if (index_size == sizeof(u4)) {
    load_int_misaligned(index, Address(xbcp, bcp_offset), tmp, false);

    // Check if the secondary index definition is still ~x, otherwise
    // we have to change the following assembler code to calculate the
    // plain index.
    assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
    xori(index, index, -1);
    addw(index, index, zr);
  } else if (index_size == sizeof(u1)) {
    load_unsigned_byte(index, Address(xbcp, bcp_offset));
  } else {
    ShouldNotReachHere();
  }
}

// Return
// Rindex: index into constant pool
// Rcache: address of cache entry - ConstantPoolCache::base_offset()
//
// A caller must add ConstantPoolCache::base_offset() to Rcache to get
// the true address of the cache entry.
//
void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
                                                           Register index,
                                                           int bcp_offset,
                                                           size_t index_size) {
  assert_different_registers(cache, index);
  assert_different_registers(cache, xcpool);
  // register "cache" is trashed in next shadd, so lets use it as a temporary register
  get_cache_index_at_bcp(index, cache, bcp_offset, index_size);
  assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
  // Convert from field index to ConstantPoolCacheEntry
  // riscv already has the cache in xcpool so there is no need to
  // install it in cache. Instead we pre-add the indexed offset to
  // xcpool and return it in cache. All clients of this method need to
  // be modified accordingly.
  shadd(cache, index, xcpool, cache, 5);
}


void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
                                                                        Register index,
                                                                        Register bytecode,
                                                                        int byte_no,
                                                                        int bcp_offset,
                                                                        size_t index_size) {
  get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
  // We use a 32-bit load here since the layout of 64-bit words on
  // little-endian machines allow us that.
  // n.b. unlike x86 cache already includes the index offset
  la(bytecode, Address(cache,
                       ConstantPoolCache::base_offset() +
                       ConstantPoolCacheEntry::indices_offset()));
  membar(MacroAssembler::AnyAny);
  lwu(bytecode, bytecode);
  membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
  const int shift_count = (1 + byte_no) * BitsPerByte;
  slli(bytecode, bytecode, XLEN - (shift_count + BitsPerByte));
  srli(bytecode, bytecode, XLEN - BitsPerByte);
}

void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
                                                               Register tmp,
                                                               int bcp_offset,
                                                               size_t index_size) {
  assert_different_registers(cache, tmp);
  // register "cache" is trashed in next ld, so lets use it as a temporary register
  get_cache_index_at_bcp(tmp, cache, bcp_offset, index_size);
  assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
  // Convert from field index to ConstantPoolCacheEntry index
  // and from word offset to byte offset
  assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord,
         "else change next line");
  ld(cache, Address(fp, frame::interpreter_frame_cache_offset * wordSize));
  // skip past the header
  add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
  // construct pointer to cache entry
  shadd(cache, tmp, cache, tmp, 2 + LogBytesPerWord);
}

// Load object from cpool->resolved_references(index)
void InterpreterMacroAssembler::load_resolved_reference_at_index(
                                Register result, Register index, Register tmp) {
  assert_different_registers(result, index);

  get_constant_pool(result);
  // Load pointer for resolved_references[] objArray
  ld(result, Address(result, ConstantPool::cache_offset()));
  ld(result, Address(result, ConstantPoolCache::resolved_references_offset()));
  resolve_oop_handle(result, tmp, t1);
  // Add in the index
  addi(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
  shadd(result, index, result, index, LogBytesPerHeapOop);
  load_heap_oop(result, Address(result, 0), tmp, t1);
}

void InterpreterMacroAssembler::load_resolved_klass_at_offset(
                                Register cpool, Register index, Register klass, Register temp) {
  shadd(temp, index, cpool, temp, LogBytesPerWord);
  lhu(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
  ld(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
  shadd(klass, temp, klass, temp, LogBytesPerWord);
  ld(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
}

void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
                                                              Register method,
                                                              Register cache) {
  const int method_offset = in_bytes(
    ConstantPoolCache::base_offset() +
      ((byte_no == TemplateTable::f2_byte)
       ? ConstantPoolCacheEntry::f2_offset()
       : ConstantPoolCacheEntry::f1_offset()));

  ld(method, Address(cache, method_offset)); // get f1 Method*
}

// Generate a subtype check: branch to ok_is_subtype if sub_klass is a
// subtype of super_klass.
//
// Args:
//      x10: superklass
//      Rsub_klass: subklass
//
// Kills:
//      x12, x15
void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
                                                  Label& ok_is_subtype) {
  assert(Rsub_klass != x10, "x10 holds superklass");
  assert(Rsub_klass != x12, "x12 holds 2ndary super array length");
  assert(Rsub_klass != x15, "x15 holds 2ndary super array scan ptr");

  // Profile the not-null value's klass.
  profile_typecheck(x12, Rsub_klass, x15); // blows x12, reloads x15

  // Do the check.
  check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12

  // Profile the failure of the check.
  profile_typecheck_failed(x12); // blows x12
}

// Java Expression Stack

void InterpreterMacroAssembler::pop_ptr(Register r) {
  ld(r, Address(esp, 0));
  addi(esp, esp, wordSize);
}

void InterpreterMacroAssembler::pop_i(Register r) {
  lw(r, Address(esp, 0)); // lw do signed extended
  addi(esp, esp, wordSize);
}

void InterpreterMacroAssembler::pop_l(Register r) {
  ld(r, Address(esp, 0));
  addi(esp, esp, 2 * Interpreter::stackElementSize);
}

void InterpreterMacroAssembler::push_ptr(Register r) {
  addi(esp, esp, -wordSize);
  sd(r, Address(esp, 0));
}

void InterpreterMacroAssembler::push_i(Register r) {
  addi(esp, esp, -wordSize);
  addw(r, r, zr); // signed extended
  sd(r, Address(esp, 0));
}

void InterpreterMacroAssembler::push_l(Register r) {
  addi(esp, esp, -2 * wordSize);
  sd(zr, Address(esp, wordSize));
  sd(r, Address(esp));
}

void InterpreterMacroAssembler::pop_f(FloatRegister r) {
  flw(r, Address(esp, 0));
  addi(esp, esp, wordSize);
}

void InterpreterMacroAssembler::pop_d(FloatRegister r) {
  fld(r, Address(esp, 0));
  addi(esp, esp, 2 * Interpreter::stackElementSize);
}

void InterpreterMacroAssembler::push_f(FloatRegister r) {
  addi(esp, esp, -wordSize);
  fsw(r, Address(esp, 0));
}

void InterpreterMacroAssembler::push_d(FloatRegister r) {
  addi(esp, esp, -2 * wordSize);
  fsd(r, Address(esp, 0));
}

void InterpreterMacroAssembler::pop(TosState state) {
  switch (state) {
    case atos:
      pop_ptr();
      verify_oop(x10);
      break;
    case btos:  // fall through
    case ztos:  // fall through
    case ctos:  // fall through
    case stos:  // fall through
    case itos:
      pop_i();
      break;
    case ltos:
      pop_l();
      break;
    case ftos:
      pop_f();
      break;
    case dtos:
      pop_d();
      break;
    case vtos:
      /* nothing to do */
      break;
    default:
      ShouldNotReachHere();
  }
}

void InterpreterMacroAssembler::push(TosState state) {
  switch (state) {
    case atos:
      verify_oop(x10);
      push_ptr();
      break;
    case btos:  // fall through
    case ztos:  // fall through
    case ctos:  // fall through
    case stos:  // fall through
    case itos:
      push_i();
      break;
    case ltos:
      push_l();
      break;
    case ftos:
      push_f();
      break;
    case dtos:
      push_d();
      break;
    case vtos:
      /* nothing to do */
      break;
    default:
      ShouldNotReachHere();
  }
}

// Helpers for swap and dup
void InterpreterMacroAssembler::load_ptr(int n, Register val) {
  ld(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
}

void InterpreterMacroAssembler::store_ptr(int n, Register val) {
  sd(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
}

void InterpreterMacroAssembler::load_float(Address src) {
  flw(f10, src);
}

void InterpreterMacroAssembler::load_double(Address src) {
  fld(f10, src);
}

void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
  // set sender sp
  mv(x19_sender_sp, sp);
  // record last_sp
  sd(esp, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
}

// Jump to from_interpreted entry of a call unless single stepping is possible
// in this thread in which case we must call the i2i entry
void InterpreterMacroAssembler::jump_from_interpreted(Register method) {
  prepare_to_jump_from_interpreted();
  if (JvmtiExport::can_post_interpreter_events()) {
    Label run_compiled_code;
    // JVMTI events, such as single-stepping, are implemented partly by avoiding running
    // compiled code in threads for which the event is enabled.  Check here for
    // interp_only_mode if these events CAN be enabled.
    lwu(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
    beqz(t0, run_compiled_code);
    ld(t0, Address(method, Method::interpreter_entry_offset()));
    jr(t0);
    bind(run_compiled_code);
  }

  ld(t0, Address(method, Method::from_interpreted_offset()));
  jr(t0);
}

// The following two routines provide a hook so that an implementation
// can schedule the dispatch in two parts.  amd64 does not do this.
void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
}

void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
  dispatch_next(state, step);
}

void InterpreterMacroAssembler::dispatch_base(TosState state,
                                              address* table,
                                              bool verifyoop,
                                              bool generate_poll,
                                              Register Rs) {
  // Pay attention to the argument Rs, which is acquiesce in t0.
  if (VerifyActivationFrameSize) {
    Unimplemented();
  }
  if (verifyoop && state == atos) {
    verify_oop(x10);
  }

  Label safepoint;
  address* const safepoint_table = Interpreter::safept_table(state);
  bool needs_thread_local_poll = generate_poll && table != safepoint_table;

  if (needs_thread_local_poll) {
    NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
    ld(t1, Address(xthread, JavaThread::polling_word_offset()));
    test_bit(t1, t1, exact_log2(SafepointMechanism::poll_bit()));
    bnez(t1, safepoint);
  }
  if (table == Interpreter::dispatch_table(state)) {
    mv(t1, Interpreter::distance_from_dispatch_table(state));
    add(t1, Rs, t1);
    shadd(t1, t1, xdispatch, t1, 3);
  } else {
    mv(t1, (address)table);
    shadd(t1, Rs, t1, Rs, 3);
  }
  ld(t1, Address(t1));
  jr(t1);

  if (needs_thread_local_poll) {
    bind(safepoint);
    la(t1, ExternalAddress((address)safepoint_table));
    shadd(t1, Rs, t1, Rs, 3);
    ld(t1, Address(t1));
    jr(t1);
  }
}

void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll, Register Rs) {
  dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll, Rs);
}

void InterpreterMacroAssembler::dispatch_only_normal(TosState state, Register Rs) {
  dispatch_base(state, Interpreter::normal_table(state), true, false, Rs);
}

void InterpreterMacroAssembler::dispatch_only_noverify(TosState state, Register Rs) {
  dispatch_base(state, Interpreter::normal_table(state), false, false, Rs);
}

void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
  // load next bytecode
  load_unsigned_byte(t0, Address(xbcp, step));
  add(xbcp, xbcp, step);
  dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
}

void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
  // load current bytecode
  lbu(t0, Address(xbcp, 0));
  dispatch_base(state, table);
}

// remove activation
//
// Apply stack watermark barrier.
// Unlock the receiver if this is a synchronized method.
// Unlock any Java monitors from synchronized blocks.
// Remove the activation from the stack.
//
// If there are locked Java monitors
//    If throw_monitor_exception
//       throws IllegalMonitorStateException
//    Else if install_monitor_exception
//       installs IllegalMonitorStateException
//    Else
//       no error processing
void InterpreterMacroAssembler::remove_activation(
                                TosState state,
                                bool throw_monitor_exception,
                                bool install_monitor_exception,
                                bool notify_jvmdi) {
  // Note: Registers x13 may be in use for the
  // result check if synchronized method
  Label unlocked, unlock, no_unlock;

  // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
  // that would normally not be safe to use. Such bad returns into unsafe territory of
  // the stack, will call InterpreterRuntime::at_unwind.
  Label slow_path;
  Label fast_path;
  safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
  j(fast_path);

  bind(slow_path);
  push(state);
  set_last_Java_frame(esp, fp, (address)pc(), t0);
  super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), xthread);
  reset_last_Java_frame(true);
  pop(state);

  bind(fast_path);

  // get the value of _do_not_unlock_if_synchronized into x13
  const Address do_not_unlock_if_synchronized(xthread,
    in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
  lbu(x13, do_not_unlock_if_synchronized);
  sb(zr, do_not_unlock_if_synchronized); // reset the flag

  // get method access flags
  ld(x11, Address(fp, frame::interpreter_frame_method_offset * wordSize));
  ld(x12, Address(x11, Method::access_flags_offset()));
  test_bit(t0, x12, exact_log2(JVM_ACC_SYNCHRONIZED));
  beqz(t0, unlocked);

  // Don't unlock anything if the _do_not_unlock_if_synchronized flag
  // is set.
  bnez(x13, no_unlock);

  // unlock monitor
  push(state); // save result

  // BasicObjectLock will be first in list, since this is a
  // synchronized method. However, need to check that the object has
  // not been unlocked by an explicit monitorexit bytecode.
  const Address monitor(fp, frame::interpreter_frame_initial_sp_offset *
                        wordSize - (int) sizeof(BasicObjectLock));
  // We use c_rarg1 so that if we go slow path it will be the correct
  // register for unlock_object to pass to VM directly
  la(c_rarg1, monitor); // address of first monitor

  ld(x10, Address(c_rarg1, BasicObjectLock::obj_offset()));
  bnez(x10, unlock);

  pop(state);
  if (throw_monitor_exception) {
    // Entry already unlocked, need to throw exception
    call_VM(noreg, CAST_FROM_FN_PTR(address,
                                    InterpreterRuntime::throw_illegal_monitor_state_exception));
    should_not_reach_here();
  } else {
    // Monitor already unlocked during a stack unroll. If requested,
    // install an illegal_monitor_state_exception.  Continue with
    // stack unrolling.
    if (install_monitor_exception) {
      call_VM(noreg, CAST_FROM_FN_PTR(address,
                                      InterpreterRuntime::new_illegal_monitor_state_exception));
    }
    j(unlocked);
  }

  bind(unlock);
  unlock_object(c_rarg1);
  pop(state);

  // Check that for block-structured locking (i.e., that all locked
  // objects has been unlocked)
  bind(unlocked);

  // x10: Might contain return value

  // Check that all monitors are unlocked
  {
    Label loop, exception, entry, restart;
    const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
    const Address monitor_block_top(
      fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
    const Address monitor_block_bot(
      fp, frame::interpreter_frame_initial_sp_offset * wordSize);

    bind(restart);
    // We use c_rarg1 so that if we go slow path it will be the correct
    // register for unlock_object to pass to VM directly
    ld(c_rarg1, monitor_block_top); // points to current entry, starting
                                     // with top-most entry
    la(x9, monitor_block_bot);  // points to word before bottom of
                                  // monitor block

    j(entry);

    // Entry already locked, need to throw exception
    bind(exception);

    if (throw_monitor_exception) {
      // Throw exception
      MacroAssembler::call_VM(noreg,
                              CAST_FROM_FN_PTR(address, InterpreterRuntime::
                                               throw_illegal_monitor_state_exception));

      should_not_reach_here();
    } else {
      // Stack unrolling. Unlock object and install illegal_monitor_exception.
      // Unlock does not block, so don't have to worry about the frame.
      // We don't have to preserve c_rarg1 since we are going to throw an exception.

      push(state);
      unlock_object(c_rarg1);
      pop(state);

      if (install_monitor_exception) {
        call_VM(noreg, CAST_FROM_FN_PTR(address,
                                        InterpreterRuntime::
                                        new_illegal_monitor_state_exception));
      }

      j(restart);
    }

    bind(loop);
    // check if current entry is used
    add(t0, c_rarg1, in_bytes(BasicObjectLock::obj_offset()));
    ld(t0, Address(t0, 0));
    bnez(t0, exception);

    add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
    bind(entry);
    bne(c_rarg1, x9, loop); // check if bottom reached if not at bottom then check this entry
  }

  bind(no_unlock);

  // jvmti support
  if (notify_jvmdi) {
    notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA

  } else {
    notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
  }

  // remove activation
  // get sender esp
  ld(t1,
     Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize));
  if (StackReservedPages > 0) {
    // testing if reserved zone needs to be re-enabled
    Label no_reserved_zone_enabling;

    ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset()));
    ble(t1, t0, no_reserved_zone_enabling);

    call_VM_leaf(
      CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), xthread);
    call_VM(noreg, CAST_FROM_FN_PTR(address,
                                    InterpreterRuntime::throw_delayed_StackOverflowError));
    should_not_reach_here();

    bind(no_reserved_zone_enabling);
  }

  // restore sender esp
  mv(esp, t1);

  // remove frame anchor
  leave();
  // If we're returning to interpreted code we will shortly be
  // adjusting SP to allow some space for ESP.  If we're returning to
  // compiled code the saved sender SP was saved in sender_sp, so this
  // restores it.
  andi(sp, esp, -16);
}

// Lock object
//
// Args:
//      c_rarg1: BasicObjectLock to be used for locking
//
// Kills:
//      x10
//      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
//      t0, t1 (temp regs)
void InterpreterMacroAssembler::lock_object(Register lock_reg)
{
  assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
  if (LockingMode == LM_MONITOR) {
    call_VM(noreg,
            CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
            lock_reg);
  } else {
    Label count, done;

    const Register swap_reg = x10;
    const Register tmp = c_rarg2;
    const Register obj_reg = c_rarg3; // Will contain the oop

    const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
    const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
    const int mark_offset = lock_offset +
                            BasicLock::displaced_header_offset_in_bytes();

    Label slow_case;

    // Load object pointer into obj_reg c_rarg3
    ld(obj_reg, Address(lock_reg, obj_offset));

    if (DiagnoseSyncOnValueBasedClasses != 0) {
      load_klass(tmp, obj_reg);
      lwu(tmp, Address(tmp, Klass::access_flags_offset()));
      test_bit(tmp, tmp, exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
      bnez(tmp, slow_case);
    }

    if (LockingMode == LM_LIGHTWEIGHT) {
      ld(tmp, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
      fast_lock(obj_reg, tmp, t0, t1, slow_case);
      j(count);
    } else if (LockingMode == LM_LEGACY) {
      // Load (object->mark() | 1) into swap_reg
      ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
      ori(swap_reg, t0, 1);

      // Save (object->mark() | 1) into BasicLock's displaced header
      sd(swap_reg, Address(lock_reg, mark_offset));

      assert(lock_offset == 0,
             "displached header must be first word in BasicObjectLock");

      cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, t0, count, /*fallthrough*/nullptr);

      // Test if the oopMark is an obvious stack pointer, i.e.,
      //  1) (mark & 7) == 0, and
      //  2) sp <= mark < mark + os::pagesize()
      //
      // These 3 tests can be done by evaluating the following
      // expression: ((mark - sp) & (7 - os::vm_page_size())),
      // assuming both stack pointer and pagesize have their
      // least significant 3 bits clear.
      // NOTE: the oopMark is in swap_reg x10 as the result of cmpxchg
      sub(swap_reg, swap_reg, sp);
      mv(t0, (int64_t)(7 - (int)os::vm_page_size()));
      andr(swap_reg, swap_reg, t0);

      // Save the test result, for recursive case, the result is zero
      sd(swap_reg, Address(lock_reg, mark_offset));
      beqz(swap_reg, count);
    }

    bind(slow_case);

    // Call the runtime routine for slow case
    if (LockingMode == LM_LIGHTWEIGHT) {
      call_VM(noreg,
              CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
              obj_reg);
    } else {
      call_VM(noreg,
              CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
              lock_reg);
    }
    j(done);

    bind(count);
    increment(Address(xthread, JavaThread::held_monitor_count_offset()));

    bind(done);
  }
}


// Unlocks an object. Used in monitorexit bytecode and
// remove_activation.  Throws an IllegalMonitorException if object is
// not locked by current thread.
//
// Args:
//      c_rarg1: BasicObjectLock for lock
//
// Kills:
//      x10
//      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
//      t0, t1 (temp regs)
void InterpreterMacroAssembler::unlock_object(Register lock_reg)
{
  assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");

  if (LockingMode == LM_MONITOR) {
    call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
  } else {
    Label count, done;

    const Register swap_reg   = x10;
    const Register header_reg = c_rarg2;  // Will contain the old oopMark
    const Register obj_reg    = c_rarg3;  // Will contain the oop

    save_bcp(); // Save in case of exception

    if (LockingMode != LM_LIGHTWEIGHT) {
      // Convert from BasicObjectLock structure to object and BasicLock
      // structure Store the BasicLock address into x10
      la(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
    }

    // Load oop into obj_reg(c_rarg3)
    ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));

    // Free entry
    sd(zr, Address(lock_reg, BasicObjectLock::obj_offset()));

    if (LockingMode == LM_LIGHTWEIGHT) {
      Label slow_case;

      // Check for non-symmetric locking. This is allowed by the spec and the interpreter
      // must handle it.
      Register tmp1 = t0;
      Register tmp2 = header_reg;
      // First check for lock-stack underflow.
      lwu(tmp1, Address(xthread, JavaThread::lock_stack_top_offset()));
      mv(tmp2, (unsigned)LockStack::start_offset());
      ble(tmp1, tmp2, slow_case);
      // Then check if the top of the lock-stack matches the unlocked object.
      subw(tmp1, tmp1, oopSize);
      add(tmp1, xthread, tmp1);
      ld(tmp1, Address(tmp1, 0));
      bne(tmp1, obj_reg, slow_case);

      ld(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
      test_bit(t0, header_reg, exact_log2(markWord::monitor_value));
      bnez(t0, slow_case);
      fast_unlock(obj_reg, header_reg, swap_reg, t0, slow_case);
      j(count);

      bind(slow_case);
    } else if (LockingMode == LM_LEGACY) {
      // Load the old header from BasicLock structure
      ld(header_reg, Address(swap_reg,
                             BasicLock::displaced_header_offset_in_bytes()));

      // Test for recursion
      beqz(header_reg, count);

      // Atomic swap back the old header
      cmpxchg_obj_header(swap_reg, header_reg, obj_reg, t0, count, /*fallthrough*/nullptr);
    }

    // Call the runtime routine for slow case.
    sd(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
    call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);

    j(done);

    bind(count);
    decrement(Address(xthread, JavaThread::held_monitor_count_offset()));

    bind(done);

    restore_bcp();
  }
}


void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
                                                         Label& zero_continue) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  ld(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
  beqz(mdp, zero_continue);
}

// Set the method data pointer for the current bcp.
void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
  assert(ProfileInterpreter, "must be profiling interpreter");
  Label set_mdp;
  push_reg(RegSet::of(x10, x11), sp); // save x10, x11

  // Test MDO to avoid the call if it is null.
  ld(x10, Address(xmethod, in_bytes(Method::method_data_offset())));
  beqz(x10, set_mdp);
  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), xmethod, xbcp);
  // x10: mdi
  // mdo is guaranteed to be non-zero here, we checked for it before the call.
  ld(x11, Address(xmethod, in_bytes(Method::method_data_offset())));
  la(x11, Address(x11, in_bytes(MethodData::data_offset())));
  add(x10, x11, x10);
  sd(x10, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
  bind(set_mdp);
  pop_reg(RegSet::of(x10, x11), sp);
}

void InterpreterMacroAssembler::verify_method_data_pointer() {
  assert(ProfileInterpreter, "must be profiling interpreter");
#ifdef ASSERT
  Label verify_continue;
  add(sp, sp, -4 * wordSize);
  sd(x10, Address(sp, 0));
  sd(x11, Address(sp, wordSize));
  sd(x12, Address(sp, 2 * wordSize));
  sd(x13, Address(sp, 3 * wordSize));
  test_method_data_pointer(x13, verify_continue); // If mdp is zero, continue
  get_method(x11);

  // If the mdp is valid, it will point to a DataLayout header which is
  // consistent with the bcp.  The converse is highly probable also.
  lh(x12, Address(x13, in_bytes(DataLayout::bci_offset())));
  ld(t0, Address(x11, Method::const_offset()));
  add(x12, x12, t0);
  la(x12, Address(x12, ConstMethod::codes_offset()));
  beq(x12, xbcp, verify_continue);
  // x10: method
  // xbcp: bcp // xbcp == 22
  // x13: mdp
  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
               x11, xbcp, x13);
  bind(verify_continue);
  ld(x10, Address(sp, 0));
  ld(x11, Address(sp, wordSize));
  ld(x12, Address(sp, 2 * wordSize));
  ld(x13, Address(sp, 3 * wordSize));
  add(sp, sp, 4 * wordSize);
#endif // ASSERT
}


void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
                                                int constant,
                                                Register value) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  Address data(mdp_in, constant);
  sd(value, data);
}


void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
                                                      int constant,
                                                      bool decrement) {
  increment_mdp_data_at(mdp_in, noreg, constant, decrement);
}

void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
                                                      Register reg,
                                                      int constant,
                                                      bool decrement) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  // %%% this does 64bit counters at best it is wasting space
  // at worst it is a rare bug when counters overflow

  assert_different_registers(t1, t0, mdp_in, reg);

  Address addr1(mdp_in, constant);
  Address addr2(t1, 0);
  Address &addr = addr1;
  if (reg != noreg) {
    la(t1, addr1);
    add(t1, t1, reg);
    addr = addr2;
  }

  if (decrement) {
    ld(t0, addr);
    addi(t0, t0, -DataLayout::counter_increment);
    Label L;
    bltz(t0, L);      // skip store if counter underflow
    sd(t0, addr);
    bind(L);
  } else {
    assert(DataLayout::counter_increment == 1,
           "flow-free idiom only works with 1");
    ld(t0, addr);
    addi(t0, t0, DataLayout::counter_increment);
    Label L;
    blez(t0, L);       // skip store if counter overflow
    sd(t0, addr);
    bind(L);
  }
}

void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
                                                int flag_byte_constant) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  int flags_offset = in_bytes(DataLayout::flags_offset());
  // Set the flag
  lbu(t1, Address(mdp_in, flags_offset));
  ori(t1, t1, flag_byte_constant);
  sb(t1, Address(mdp_in, flags_offset));
}


void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
                                                 int offset,
                                                 Register value,
                                                 Register test_value_out,
                                                 Label& not_equal_continue) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  if (test_value_out == noreg) {
    ld(t1, Address(mdp_in, offset));
    bne(value, t1, not_equal_continue);
  } else {
    // Put the test value into a register, so caller can use it:
    ld(test_value_out, Address(mdp_in, offset));
    bne(value, test_value_out, not_equal_continue);
  }
}


void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
                                                     int offset_of_disp) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  ld(t1, Address(mdp_in, offset_of_disp));
  add(mdp_in, mdp_in, t1);
  sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
}

void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
                                                     Register reg,
                                                     int offset_of_disp) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  add(t1, mdp_in, reg);
  ld(t1, Address(t1, offset_of_disp));
  add(mdp_in, mdp_in, t1);
  sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
}


void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
                                                       int constant) {
  assert(ProfileInterpreter, "must be profiling interpreter");
  addi(mdp_in, mdp_in, (unsigned)constant);
  sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
}


void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
  assert(ProfileInterpreter, "must be profiling interpreter");

  // save/restore across call_VM
  addi(sp, sp, -2 * wordSize);
  sd(zr, Address(sp, 0));
  sd(return_bci, Address(sp, wordSize));
  call_VM(noreg,
          CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
          return_bci);
  ld(zr, Address(sp, 0));
  ld(return_bci, Address(sp, wordSize));
  addi(sp, sp, 2 * wordSize);
}

void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
                                                     Register bumped_count) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    // Otherwise, assign to mdp
    test_method_data_pointer(mdp, profile_continue);

    // We are taking a branch.  Increment the taken count.
    Address data(mdp, in_bytes(JumpData::taken_offset()));
    ld(bumped_count, data);
    assert(DataLayout::counter_increment == 1,
            "flow-free idiom only works with 1");
    addi(bumped_count, bumped_count, DataLayout::counter_increment);
    Label L;
    // eg: bumped_count=0x7fff ffff ffff ffff  + 1 < 0. so we use <= 0;
    blez(bumped_count, L);       // skip store if counter overflow,
    sd(bumped_count, data);
    bind(L);
    // The method data pointer needs to be updated to reflect the new target.
    update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // We are taking a branch.  Increment the not taken count.
    increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));

    // The method data pointer needs to be updated to correspond to
    // the next bytecode
    update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_call(Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // We are making a call.  Increment the count.
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));

    // The method data pointer needs to be updated to reflect the new target.
    update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_final_call(Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // We are making a call.  Increment the count.
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));

    // The method data pointer needs to be updated to reflect the new target.
    update_mdp_by_constant(mdp,
                           in_bytes(VirtualCallData::
                                    virtual_call_data_size()));
    bind(profile_continue);
  }
}


void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
                                                     Register mdp,
                                                     Register reg2,
                                                     bool receiver_can_be_null) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    Label skip_receiver_profile;
    if (receiver_can_be_null) {
      Label not_null;
      // We are making a call.  Increment the count for null receiver.
      increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
      j(skip_receiver_profile);
      bind(not_null);
    }

    // Record the receiver type.
    record_klass_in_profile(receiver, mdp, reg2, true);
    bind(skip_receiver_profile);

    // The method data pointer needs to be updated to reflect the new target.

    update_mdp_by_constant(mdp,
                           in_bytes(VirtualCallData::
                                    virtual_call_data_size()));
    bind(profile_continue);
  }
}

// This routine creates a state machine for updating the multi-row
// type profile at a virtual call site (or other type-sensitive bytecode).
// The machine visits each row (of receiver/count) until the receiver type
// is found, or until it runs out of rows.  At the same time, it remembers
// the location of the first empty row.  (An empty row records null for its
// receiver, and can be allocated for a newly-observed receiver type.)
// Because there are two degrees of freedom in the state, a simple linear
// search will not work; it must be a decision tree.  Hence this helper
// function is recursive, to generate the required tree structured code.
// It's the interpreter, so we are trading off code space for speed.
// See below for example code.
void InterpreterMacroAssembler::record_klass_in_profile_helper(
                                Register receiver, Register mdp,
                                Register reg2,
                                Label& done, bool is_virtual_call) {
  if (TypeProfileWidth == 0) {
    if (is_virtual_call) {
      increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
    }

  } else {
    int non_profiled_offset = -1;
    if (is_virtual_call) {
      non_profiled_offset = in_bytes(CounterData::count_offset());
    }

    record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
      &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
  }
}

void InterpreterMacroAssembler::record_item_in_profile_helper(
  Register item, Register mdp, Register reg2, int start_row, Label& done, int total_rows,
  OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, int non_profiled_offset) {
  int last_row = total_rows - 1;
  assert(start_row <= last_row, "must be work left to do");
  // Test this row for both the item and for null.
  // Take any of three different outcomes:
  //   1. found item => increment count and goto done
  //   2. found null => keep looking for case 1, maybe allocate this cell
  //   3. found something else => keep looking for cases 1 and 2
  // Case 3 is handled by a recursive call.
  for (int row = start_row; row <= last_row; row++) {
    Label next_test;
    bool test_for_null_also = (row == start_row);

    // See if the item is item[n].
    int item_offset = in_bytes(item_offset_fn(row));
    test_mdp_data_at(mdp, item_offset, item,
                     (test_for_null_also ? reg2 : noreg),
                     next_test);
    // (Reg2 now contains the item from the CallData.)

    // The item is item[n].  Increment count[n].
    int count_offset = in_bytes(item_count_offset_fn(row));
    increment_mdp_data_at(mdp, count_offset);
    j(done);
    bind(next_test);

    if (test_for_null_also) {
      Label found_null;
      // Failed the equality check on item[n]...  Test for null.
      if (start_row == last_row) {
        // The only thing left to do is handle the null case.
        if (non_profiled_offset >= 0) {
          beqz(reg2, found_null);
          // Item did not match any saved item and there is no empty row for it.
          // Increment total counter to indicate polymorphic case.
          increment_mdp_data_at(mdp, non_profiled_offset);
          j(done);
          bind(found_null);
        } else {
          bnez(reg2, done);
        }
        break;
      }
      // Since null is rare, make it be the branch-taken case.
      beqz(reg2, found_null);

      // Put all the "Case 3" tests here.
      record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
        item_offset_fn, item_count_offset_fn, non_profiled_offset);

      // Found a null.  Keep searching for a matching item,
      // but remember that this is an empty (unused) slot.
      bind(found_null);
    }
  }

  // In the fall-through case, we found no matching item, but we
  // observed the item[start_row] is null.
  // Fill in the item field and increment the count.
  int item_offset = in_bytes(item_offset_fn(start_row));
  set_mdp_data_at(mdp, item_offset, item);
  int count_offset = in_bytes(item_count_offset_fn(start_row));
  mv(reg2, DataLayout::counter_increment);
  set_mdp_data_at(mdp, count_offset, reg2);
  if (start_row > 0) {
    j(done);
  }
}

// Example state machine code for three profile rows:
//   # main copy of decision tree, rooted at row[1]
//   if (row[0].rec == rec) then [
//     row[0].incr()
//     goto done
//   ]
//   if (row[0].rec != nullptr) then [
//     # inner copy of decision tree, rooted at row[1]
//     if (row[1].rec == rec) then [
//       row[1].incr()
//       goto done
//     ]
//     if (row[1].rec != nullptr) then [
//       # degenerate decision tree, rooted at row[2]
//       if (row[2].rec == rec) then [
//         row[2].incr()
//         goto done
//       ]
//       if (row[2].rec != nullptr) then [
//         count.incr()
//         goto done
//       ] # overflow
//       row[2].init(rec)
//       goto done
//     ] else [
//       # remember row[1] is empty
//       if (row[2].rec == rec) then [
//         row[2].incr()
//         goto done
//       ]
//       row[1].init(rec)
//       goto done
//     ]
//   else [
//     # remember row[0] is empty
//     if (row[1].rec == rec) then [
//       row[1].incr()
//       goto done
//     ]
//     if (row[2].rec == rec) then [
//       row[2].incr()
//       goto done
//     ]
//     row[0].init(rec)
//     goto done
//   ]
//   done:

void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
                                                        Register mdp, Register reg2,
                                                        bool is_virtual_call) {
  assert(ProfileInterpreter, "must be profiling");
  Label done;

  record_klass_in_profile_helper(receiver, mdp, reg2, done, is_virtual_call);

  bind(done);
}

void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // Update the total ret count.
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));

    for (uint row = 0; row < RetData::row_limit(); row++) {
      Label next_test;

      // See if return_bci is equal to bci[n]:
      test_mdp_data_at(mdp,
                       in_bytes(RetData::bci_offset(row)),
                       return_bci, noreg,
                       next_test);

      // return_bci is equal to bci[n].  Increment the count.
      increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));

      // The method data pointer needs to be updated to reflect the new target.
      update_mdp_by_offset(mdp,
                           in_bytes(RetData::bci_displacement_offset(row)));
      j(profile_continue);
      bind(next_test);
    }

    update_mdp_for_ret(return_bci);

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());

    // The method data pointer needs to be updated.
    int mdp_delta = in_bytes(BitData::bit_data_size());
    if (TypeProfileCasts) {
      mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
    }
    update_mdp_by_constant(mdp, mdp_delta);

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
    if (ProfileInterpreter && TypeProfileCasts) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    int count_offset = in_bytes(CounterData::count_offset());
    // Back up the address, since we have already bumped the mdp.
    count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());

    // *Decrement* the counter.  We expect to see zero or small negatives.
    increment_mdp_data_at(mdp, count_offset, true);

    bind (profile_continue);
  }
}

void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // The method data pointer needs to be updated.
    int mdp_delta = in_bytes(BitData::bit_data_size());
    if (TypeProfileCasts) {
      mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());

      // Record the object type.
      record_klass_in_profile(klass, mdp, reg2, false);
    }
    update_mdp_by_constant(mdp, mdp_delta);

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // Update the default case count
    increment_mdp_data_at(mdp,
                          in_bytes(MultiBranchData::default_count_offset()));

    // The method data pointer needs to be updated.
    update_mdp_by_offset(mdp,
                         in_bytes(MultiBranchData::
                                  default_displacement_offset()));

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_switch_case(Register index,
                                                    Register mdp,
                                                    Register reg2) {
  if (ProfileInterpreter) {
    Label profile_continue;

    // If no method data exists, go to profile_continue.
    test_method_data_pointer(mdp, profile_continue);

    // Build the base (index * per_case_size_in_bytes()) +
    // case_array_offset_in_bytes()
    mv(reg2, in_bytes(MultiBranchData::per_case_size()));
    mv(t0, in_bytes(MultiBranchData::case_array_offset()));
    Assembler::mul(index, index, reg2);
    Assembler::add(index, index, t0);

    // Update the case count
    increment_mdp_data_at(mdp,
                          index,
                          in_bytes(MultiBranchData::relative_count_offset()));

    // The method data pointer need to be updated.
    update_mdp_by_offset(mdp,
                         index,
                         in_bytes(MultiBranchData::
                                  relative_displacement_offset()));

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }

void InterpreterMacroAssembler::notify_method_entry() {
  // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  // track stack depth.  If it is possible to enter interp_only_mode we add
  // the code to check if the event should be sent.
  if (JvmtiExport::can_post_interpreter_events()) {
    Label L;
    lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
    beqz(x13, L);
    call_VM(noreg, CAST_FROM_FN_PTR(address,
                                    InterpreterRuntime::post_method_entry));
    bind(L);
  }

  {
    SkipIfEqual skip(this, &DTraceMethodProbes, false);
    get_method(c_rarg1);
    call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
                 xthread, c_rarg1);
  }

  // RedefineClasses() tracing support for obsolete method entry
  if (log_is_enabled(Trace, redefine, class, obsolete)) {
    get_method(c_rarg1);
    call_VM_leaf(
      CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
      xthread, c_rarg1);
  }
}


void InterpreterMacroAssembler::notify_method_exit(
    TosState state, NotifyMethodExitMode mode) {
  // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  // track stack depth.  If it is possible to enter interp_only_mode we add
  // the code to check if the event should be sent.
  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
    Label L;
    // Note: frame::interpreter_frame_result has a dependency on how the
    // method result is saved across the call to post_method_exit. If this
    // is changed then the interpreter_frame_result implementation will
    // need to be updated too.

    // template interpreter will leave the result on the top of the stack.
    push(state);
    lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
    beqz(x13, L);
    call_VM(noreg,
            CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
    bind(L);
    pop(state);
  }

  {
    SkipIfEqual skip(this, &DTraceMethodProbes, false);
    push(state);
    get_method(c_rarg1);
    call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
                 xthread, c_rarg1);
    pop(state);
  }
}


// Jump if ((*counter_addr += increment) & mask) satisfies the condition.
void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
                                                        int increment, Address mask,
                                                        Register tmp1, Register tmp2,
                                                        bool preloaded, Label* where) {
  Label done;
  if (!preloaded) {
    lwu(tmp1, counter_addr);
  }
  add(tmp1, tmp1, increment);
  sw(tmp1, counter_addr);
  lwu(tmp2, mask);
  andr(tmp1, tmp1, tmp2);
  bnez(tmp1, done);
  j(*where); // offset is too large so we have to use j instead of beqz here
  bind(done);
}

void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
                                                  int number_of_arguments) {
  // interpreter specific
  //
  // Note: No need to save/restore rbcp & rlocals pointer since these
  //       are callee saved registers and no blocking/ GC can happen
  //       in leaf calls.
#ifdef ASSERT
  {
   Label L;
   ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
   beqz(t0, L);
   stop("InterpreterMacroAssembler::call_VM_leaf_base:"
        " last_sp isn't null");
   bind(L);
  }
#endif /* ASSERT */
  // super call
  MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
}

void InterpreterMacroAssembler::call_VM_base(Register oop_result,
                                             Register java_thread,
                                             Register last_java_sp,
                                             address  entry_point,
                                             int      number_of_arguments,
                                             bool     check_exceptions) {
  // interpreter specific
  //
  // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
  //       really make a difference for these runtime calls, since they are
  //       slow anyway. Btw., bcp must be saved/restored since it may change
  //       due to GC.
  save_bcp();
#ifdef ASSERT
  {
    Label L;
    ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
    beqz(t0, L);
    stop("InterpreterMacroAssembler::call_VM_base:"
         " last_sp isn't null");
    bind(L);
  }
#endif /* ASSERT */
  // super call
  MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
                               entry_point, number_of_arguments,
                               check_exceptions);
// interpreter specific
  restore_bcp();
  restore_locals();
}

void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
  assert_different_registers(obj, tmp, t0, mdo_addr.base());
  Label update, next, none;

  verify_oop(obj);

  bnez(obj, update);
  orptr(mdo_addr, TypeEntries::null_seen, t0, tmp);
  j(next);

  bind(update);
  load_klass(obj, obj);

  ld(t0, mdo_addr);
  xorr(obj, obj, t0);
  andi(t0, obj, TypeEntries::type_klass_mask);
  beqz(t0, next); // klass seen before, nothing to
                  // do. The unknown bit may have been
                  // set already but no need to check.

  test_bit(t0, obj, exact_log2(TypeEntries::type_unknown));
  bnez(t0, next);
  // already unknown. Nothing to do anymore.

  ld(t0, mdo_addr);
  beqz(t0, none);
  mv(tmp, (u1)TypeEntries::null_seen);
  beq(t0, tmp, none);
  // There is a chance that the checks above (re-reading profiling
  // data from memory) fail if another thread has just set the
  // profiling to this obj's klass
  ld(t0, mdo_addr);
  xorr(obj, obj, t0);
  andi(t0, obj, TypeEntries::type_klass_mask);
  beqz(t0, next);

  // different than before. Cannot keep accurate profile.
  orptr(mdo_addr, TypeEntries::type_unknown, t0, tmp);
  j(next);

  bind(none);
  // first time here. Set profile type.
  sd(obj, mdo_addr);

  bind(next);
}

void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
  if (!ProfileInterpreter) {
    return;
  }

  if (MethodData::profile_arguments() || MethodData::profile_return()) {
    Label profile_continue;

    test_method_data_pointer(mdp, profile_continue);

    int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());

    lbu(t0, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
    if (is_virtual) {
      mv(tmp, (u1)DataLayout::virtual_call_type_data_tag);
      bne(t0, tmp, profile_continue);
    } else {
      mv(tmp, (u1)DataLayout::call_type_data_tag);
      bne(t0, tmp, profile_continue);
    }

    // calculate slot step
    static int stack_slot_offset0 = in_bytes(TypeEntriesAtCall::stack_slot_offset(0));
    static int slot_step = in_bytes(TypeEntriesAtCall::stack_slot_offset(1)) - stack_slot_offset0;

    // calculate type step
    static int argument_type_offset0 = in_bytes(TypeEntriesAtCall::argument_type_offset(0));
    static int type_step = in_bytes(TypeEntriesAtCall::argument_type_offset(1)) - argument_type_offset0;

    if (MethodData::profile_arguments()) {
      Label done, loop, loopEnd, profileArgument, profileReturnType;
      RegSet pushed_registers;
      pushed_registers += x15;
      pushed_registers += x16;
      pushed_registers += x17;
      Register mdo_addr = x15;
      Register index = x16;
      Register off_to_args = x17;
      push_reg(pushed_registers, sp);

      mv(off_to_args, in_bytes(TypeEntriesAtCall::args_data_offset()));
      mv(t0, TypeProfileArgsLimit);
      beqz(t0, loopEnd);

      mv(index, zr); // index < TypeProfileArgsLimit
      bind(loop);
      bgtz(index, profileReturnType);
      mv(t0, (int)MethodData::profile_return());
      beqz(t0, profileArgument); // (index > 0 || MethodData::profile_return()) == false
      bind(profileReturnType);
      // If return value type is profiled we may have no argument to profile
      ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
      mv(t1, - TypeStackSlotEntries::per_arg_count());
      mul(t1, index, t1);
      add(tmp, tmp, t1);
      mv(t1, TypeStackSlotEntries::per_arg_count());
      add(t0, mdp, off_to_args);
      blt(tmp, t1, done);

      bind(profileArgument);

      ld(tmp, Address(callee, Method::const_offset()));
      load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
      // stack offset o (zero based) from the start of the argument
      // list, for n arguments translates into offset n - o - 1 from
      // the end of the argument list
      mv(t0, stack_slot_offset0);
      mv(t1, slot_step);
      mul(t1, index, t1);
      add(t0, t0, t1);
      add(t0, mdp, t0);
      ld(t0, Address(t0));
      sub(tmp, tmp, t0);
      addi(tmp, tmp, -1);
      Address arg_addr = argument_address(tmp);
      ld(tmp, arg_addr);

      mv(t0, argument_type_offset0);
      mv(t1, type_step);
      mul(t1, index, t1);
      add(t0, t0, t1);
      add(mdo_addr, mdp, t0);
      Address mdo_arg_addr(mdo_addr, 0);
      profile_obj_type(tmp, mdo_arg_addr, t1);

      int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
      addi(off_to_args, off_to_args, to_add);

      // increment index by 1
      addi(index, index, 1);
      mv(t1, TypeProfileArgsLimit);
      blt(index, t1, loop);
      bind(loopEnd);

      if (MethodData::profile_return()) {
        ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
        addi(tmp, tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
      }

      add(t0, mdp, off_to_args);
      bind(done);
      mv(mdp, t0);

      // unspill the clobbered registers
      pop_reg(pushed_registers, sp);

      if (MethodData::profile_return()) {
        // We're right after the type profile for the last
        // argument. tmp is the number of cells left in the
        // CallTypeData/VirtualCallTypeData to reach its end. Non null
        // if there's a return to profile.
        assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
        shadd(mdp, tmp, mdp, tmp, exact_log2(DataLayout::cell_size));
      }
      sd(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
    } else {
      assert(MethodData::profile_return(), "either profile call args or call ret");
      update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
    }

    // mdp points right after the end of the
    // CallTypeData/VirtualCallTypeData, right after the cells for the
    // return value type if there's one

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
  assert_different_registers(mdp, ret, tmp, xbcp, t0, t1);
  if (ProfileInterpreter && MethodData::profile_return()) {
    Label profile_continue, done;

    test_method_data_pointer(mdp, profile_continue);

    if (MethodData::profile_return_jsr292_only()) {
      assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");

      // If we don't profile all invoke bytecodes we must make sure
      // it's a bytecode we indeed profile. We can't go back to the
      // beginning of the ProfileData we intend to update to check its
      // type because we're right after it and we don't known its
      // length
      Label do_profile;
      lbu(t0, Address(xbcp, 0));
      mv(tmp, (u1)Bytecodes::_invokedynamic);
      beq(t0, tmp, do_profile);
      mv(tmp, (u1)Bytecodes::_invokehandle);
      beq(t0, tmp, do_profile);
      get_method(tmp);
      lhu(t0, Address(tmp, Method::intrinsic_id_offset()));
      mv(t1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
      bne(t0, t1, profile_continue);
      bind(do_profile);
    }

    Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
    mv(tmp, ret);
    profile_obj_type(tmp, mdo_ret_addr, t1);

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2, Register tmp3) {
  assert_different_registers(t0, t1, mdp, tmp1, tmp2, tmp3);
  if (ProfileInterpreter && MethodData::profile_parameters()) {
    Label profile_continue, done;

    test_method_data_pointer(mdp, profile_continue);

    // Load the offset of the area within the MDO used for
    // parameters. If it's negative we're not profiling any parameters
    lwu(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
    srli(tmp2, tmp1, 31);
    bnez(tmp2, profile_continue);  // i.e. sign bit set

    // Compute a pointer to the area for parameters from the offset
    // and move the pointer to the slot for the last
    // parameters. Collect profiling from last parameter down.
    // mdo start + parameters offset + array length - 1
    add(mdp, mdp, tmp1);
    ld(tmp1, Address(mdp, ArrayData::array_len_offset()));
    add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());

    Label loop;
    bind(loop);

    int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
    int type_base = in_bytes(ParametersTypeData::type_offset(0));
    int per_arg_scale = exact_log2(DataLayout::cell_size);
    add(t0, mdp, off_base);
    add(t1, mdp, type_base);

    shadd(tmp2, tmp1, t0, tmp2, per_arg_scale);
    // load offset on the stack from the slot for this parameter
    ld(tmp2, Address(tmp2, 0));
    neg(tmp2, tmp2);

    // read the parameter from the local area
    shadd(tmp2, tmp2, xlocals, tmp2, Interpreter::logStackElementSize);
    ld(tmp2, Address(tmp2, 0));

    // profile the parameter
    shadd(t1, tmp1, t1, t0, per_arg_scale);
    Address arg_type(t1, 0);
    profile_obj_type(tmp2, arg_type, tmp3);

    // go to next parameter
    add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
    bgez(tmp1, loop);

    bind(profile_continue);
  }
}

void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
  // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
  // register "cache" is trashed in next ld, so lets use it as a temporary register
  get_cache_index_at_bcp(index, cache, 1, sizeof(u4));
  // Get address of invokedynamic array
  ld(cache, Address(xcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
  // Scale the index to be the entry index * sizeof(ResolvedInvokeDynamicInfo)
  slli(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
  add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
  add(cache, cache, index);
  la(cache, Address(cache, 0));
}

void InterpreterMacroAssembler::get_method_counters(Register method,
                                                    Register mcs, Label& skip) {
  Label has_counters;
  ld(mcs, Address(method, Method::method_counters_offset()));
  bnez(mcs, has_counters);
  call_VM(noreg, CAST_FROM_FN_PTR(address,
          InterpreterRuntime::build_method_counters), method);
  ld(mcs, Address(method, Method::method_counters_offset()));
  beqz(mcs, skip); // No MethodCounters allocated, OutOfMemory
  bind(has_counters);
}

#ifdef ASSERT
void InterpreterMacroAssembler::verify_access_flags(Register access_flags, uint32_t flag,
                                                    const char* msg, bool stop_by_hit) {
  Label L;
  test_bit(t0, access_flags, exact_log2(flag));
  if (stop_by_hit) {
    beqz(t0, L);
  } else {
    bnez(t0, L);
  }
  stop(msg);
  bind(L);
}

void InterpreterMacroAssembler::verify_frame_setup() {
  Label L;
  const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
  ld(t0, monitor_block_top);
  beq(esp, t0, L);
  stop("broken stack frame setup in interpreter");
  bind(L);
}
#endif