src/cpu/x86/vm/interp_masm_x86_64.cpp

Wed, 17 Jun 2015 17:48:25 -0700

author
ascarpino
date
Wed, 17 Jun 2015 17:48:25 -0700
changeset 9788
44ef77ad417c
parent 9604
da2e98c027fd
child 9637
eef07cd490d4
permissions
-rw-r--r--

8073108: Use x86 and SPARC CPU instructions for GHASH acceleration
Reviewed-by: kvn, jrose, phh

     1 /*
     2  * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "interp_masm_x86.hpp"
    27 #include "interpreter/interpreter.hpp"
    28 #include "interpreter/interpreterRuntime.hpp"
    29 #include "oops/arrayOop.hpp"
    30 #include "oops/markOop.hpp"
    31 #include "oops/methodData.hpp"
    32 #include "oops/method.hpp"
    33 #include "prims/jvmtiExport.hpp"
    34 #include "prims/jvmtiRedefineClassesTrace.hpp"
    35 #include "prims/jvmtiThreadState.hpp"
    36 #include "runtime/basicLock.hpp"
    37 #include "runtime/biasedLocking.hpp"
    38 #include "runtime/sharedRuntime.hpp"
    39 #include "runtime/thread.inline.hpp"
    42 // Implementation of InterpreterMacroAssembler
    44 #ifdef CC_INTERP
    45 void InterpreterMacroAssembler::get_method(Register reg) {
    46   movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
    47   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
    48 }
    49 #endif // CC_INTERP
    51 #ifndef CC_INTERP
    53 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
    54                                                   int number_of_arguments) {
    55   // interpreter specific
    56   //
    57   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
    58   //       since these are callee saved registers and no blocking/
    59   //       GC can happen in leaf calls.
    60   // Further Note: DO NOT save/restore bcp/locals. If a caller has
    61   // already saved them so that it can use esi/edi as temporaries
    62   // then a save/restore here will DESTROY the copy the caller
    63   // saved! There used to be a save_bcp() that only happened in
    64   // the ASSERT path (no restore_bcp). Which caused bizarre failures
    65   // when jvm built with ASSERTs.
    66 #ifdef ASSERT
    67   {
    68     Label L;
    69     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
    70     jcc(Assembler::equal, L);
    71     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    72          " last_sp != NULL");
    73     bind(L);
    74   }
    75 #endif
    76   // super call
    77   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
    78   // interpreter specific
    79   // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
    80   // but since they may not have been saved (and we don't want to
    81   // save thme here (see note above) the assert is invalid.
    82 }
    84 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
    85                                              Register java_thread,
    86                                              Register last_java_sp,
    87                                              address  entry_point,
    88                                              int      number_of_arguments,
    89                                              bool     check_exceptions) {
    90   // interpreter specific
    91   //
    92   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
    93   //       really make a difference for these runtime calls, since they are
    94   //       slow anyway. Btw., bcp must be saved/restored since it may change
    95   //       due to GC.
    96   // assert(java_thread == noreg , "not expecting a precomputed java thread");
    97   save_bcp();
    98 #ifdef ASSERT
    99   {
   100     Label L;
   101     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
   102     jcc(Assembler::equal, L);
   103     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
   104          " last_sp != NULL");
   105     bind(L);
   106   }
   107 #endif /* ASSERT */
   108   // super call
   109   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
   110                                entry_point, number_of_arguments,
   111                                check_exceptions);
   112   // interpreter specific
   113   restore_bcp();
   114   restore_locals();
   115 }
   118 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
   119   if (JvmtiExport::can_pop_frame()) {
   120     Label L;
   121     // Initiate popframe handling only if it is not already being
   122     // processed.  If the flag has the popframe_processing bit set, it
   123     // means that this code is called *during* popframe handling - we
   124     // don't want to reenter.
   125     // This method is only called just after the call into the vm in
   126     // call_VM_base, so the arg registers are available.
   127     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
   128     testl(c_rarg0, JavaThread::popframe_pending_bit);
   129     jcc(Assembler::zero, L);
   130     testl(c_rarg0, JavaThread::popframe_processing_bit);
   131     jcc(Assembler::notZero, L);
   132     // Call Interpreter::remove_activation_preserving_args_entry() to get the
   133     // address of the same-named entrypoint in the generated interpreter code.
   134     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
   135     jmp(rax);
   136     bind(L);
   137   }
   138 }
   141 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
   142   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   143   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
   144   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
   145   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
   146   switch (state) {
   147     case atos: movptr(rax, oop_addr);
   148                movptr(oop_addr, (int32_t)NULL_WORD);
   149                verify_oop(rax, state);              break;
   150     case ltos: movptr(rax, val_addr);                 break;
   151     case btos:                                   // fall through
   152     case ztos:                                   // fall through
   153     case ctos:                                   // fall through
   154     case stos:                                   // fall through
   155     case itos: movl(rax, val_addr);                 break;
   156     case ftos: movflt(xmm0, val_addr);              break;
   157     case dtos: movdbl(xmm0, val_addr);              break;
   158     case vtos: /* nothing to do */                  break;
   159     default  : ShouldNotReachHere();
   160   }
   161   // Clean up tos value in the thread object
   162   movl(tos_addr,  (int) ilgl);
   163   movl(val_addr,  (int32_t) NULL_WORD);
   164 }
   167 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
   168   if (JvmtiExport::can_force_early_return()) {
   169     Label L;
   170     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   171     testptr(c_rarg0, c_rarg0);
   172     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
   174     // Initiate earlyret handling only if it is not already being processed.
   175     // If the flag has the earlyret_processing bit set, it means that this code
   176     // is called *during* earlyret handling - we don't want to reenter.
   177     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
   178     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
   179     jcc(Assembler::notEqual, L);
   181     // Call Interpreter::remove_activation_early_entry() to get the address of the
   182     // same-named entrypoint in the generated interpreter code.
   183     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   184     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
   185     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
   186     jmp(rax);
   187     bind(L);
   188   }
   189 }
   192 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
   193   Register reg,
   194   int bcp_offset) {
   195   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
   196   load_unsigned_short(reg, Address(r13, bcp_offset));
   197   bswapl(reg);
   198   shrl(reg, 16);
   199 }
   202 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
   203                                                        int bcp_offset,
   204                                                        size_t index_size) {
   205   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
   206   if (index_size == sizeof(u2)) {
   207     load_unsigned_short(index, Address(r13, bcp_offset));
   208   } else if (index_size == sizeof(u4)) {
   209     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
   210     movl(index, Address(r13, bcp_offset));
   211     // Check if the secondary index definition is still ~x, otherwise
   212     // we have to change the following assembler code to calculate the
   213     // plain index.
   214     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
   215     notl(index);  // convert to plain index
   216   } else if (index_size == sizeof(u1)) {
   217     load_unsigned_byte(index, Address(r13, bcp_offset));
   218   } else {
   219     ShouldNotReachHere();
   220   }
   221 }
   224 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
   225                                                            Register index,
   226                                                            int bcp_offset,
   227                                                            size_t index_size) {
   228   assert_different_registers(cache, index);
   229   get_cache_index_at_bcp(index, bcp_offset, index_size);
   230   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   231   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   232   // convert from field index to ConstantPoolCacheEntry index
   233   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
   234   shll(index, 2);
   235 }
   238 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
   239                                                                         Register index,
   240                                                                         Register bytecode,
   241                                                                         int byte_no,
   242                                                                         int bcp_offset,
   243                                                                         size_t index_size) {
   244   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
   245   // We use a 32-bit load here since the layout of 64-bit words on
   246   // little-endian machines allow us that.
   247   movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
   248   const int shift_count = (1 + byte_no) * BitsPerByte;
   249   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
   250          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
   251          "correct shift count");
   252   shrl(bytecode, shift_count);
   253   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
   254   andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
   255 }
   258 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
   259                                                                Register tmp,
   260                                                                int bcp_offset,
   261                                                                size_t index_size) {
   262   assert(cache != tmp, "must use different register");
   263   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
   264   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   265   // convert from field index to ConstantPoolCacheEntry index
   266   // and from word offset to byte offset
   267   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
   268   shll(tmp, 2 + LogBytesPerWord);
   269   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   270   // skip past the header
   271   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
   272   addptr(cache, tmp);  // construct pointer to cache entry
   273 }
   275 void InterpreterMacroAssembler::get_method_counters(Register method,
   276                                                     Register mcs, Label& skip) {
   277   Label has_counters;
   278   movptr(mcs, Address(method, Method::method_counters_offset()));
   279   testptr(mcs, mcs);
   280   jcc(Assembler::notZero, has_counters);
   281   call_VM(noreg, CAST_FROM_FN_PTR(address,
   282           InterpreterRuntime::build_method_counters), method);
   283   movptr(mcs, Address(method,Method::method_counters_offset()));
   284   testptr(mcs, mcs);
   285   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
   286   bind(has_counters);
   287 }
   289 // Load object from cpool->resolved_references(index)
   290 void InterpreterMacroAssembler::load_resolved_reference_at_index(
   291                                            Register result, Register index) {
   292   assert_different_registers(result, index);
   293   // convert from field index to resolved_references() index and from
   294   // word index to byte offset. Since this is a java object, it can be compressed
   295   Register tmp = index;  // reuse
   296   shll(tmp, LogBytesPerHeapOop);
   298   get_constant_pool(result);
   299   // load pointer for resolved_references[] objArray
   300   movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
   301   // JNIHandles::resolve(obj);
   302   movptr(result, Address(result, 0));
   303   // Add in the index
   304   addptr(result, tmp);
   305   load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
   306 }
   308 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
   309 // subtype of super_klass.
   310 //
   311 // Args:
   312 //      rax: superklass
   313 //      Rsub_klass: subklass
   314 //
   315 // Kills:
   316 //      rcx, rdi
   317 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
   318                                                   Label& ok_is_subtype) {
   319   assert(Rsub_klass != rax, "rax holds superklass");
   320   assert(Rsub_klass != r14, "r14 holds locals");
   321   assert(Rsub_klass != r13, "r13 holds bcp");
   322   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
   323   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
   325   // Profile the not-null value's klass.
   326   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
   328   // Do the check.
   329   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
   331   // Profile the failure of the check.
   332   profile_typecheck_failed(rcx); // blows rcx
   333 }
   337 // Java Expression Stack
   339 void InterpreterMacroAssembler::pop_ptr(Register r) {
   340   pop(r);
   341 }
   343 void InterpreterMacroAssembler::pop_i(Register r) {
   344   // XXX can't use pop currently, upper half non clean
   345   movl(r, Address(rsp, 0));
   346   addptr(rsp, wordSize);
   347 }
   349 void InterpreterMacroAssembler::pop_l(Register r) {
   350   movq(r, Address(rsp, 0));
   351   addptr(rsp, 2 * Interpreter::stackElementSize);
   352 }
   354 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
   355   movflt(r, Address(rsp, 0));
   356   addptr(rsp, wordSize);
   357 }
   359 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
   360   movdbl(r, Address(rsp, 0));
   361   addptr(rsp, 2 * Interpreter::stackElementSize);
   362 }
   364 void InterpreterMacroAssembler::push_ptr(Register r) {
   365   push(r);
   366 }
   368 void InterpreterMacroAssembler::push_i(Register r) {
   369   push(r);
   370 }
   372 void InterpreterMacroAssembler::push_l(Register r) {
   373   subptr(rsp, 2 * wordSize);
   374   movq(Address(rsp, 0), r);
   375 }
   377 void InterpreterMacroAssembler::push_f(XMMRegister r) {
   378   subptr(rsp, wordSize);
   379   movflt(Address(rsp, 0), r);
   380 }
   382 void InterpreterMacroAssembler::push_d(XMMRegister r) {
   383   subptr(rsp, 2 * wordSize);
   384   movdbl(Address(rsp, 0), r);
   385 }
   387 void InterpreterMacroAssembler::pop(TosState state) {
   388   switch (state) {
   389   case atos: pop_ptr();                 break;
   390   case btos:
   391   case ztos:
   392   case ctos:
   393   case stos:
   394   case itos: pop_i();                   break;
   395   case ltos: pop_l();                   break;
   396   case ftos: pop_f();                   break;
   397   case dtos: pop_d();                   break;
   398   case vtos: /* nothing to do */        break;
   399   default:   ShouldNotReachHere();
   400   }
   401   verify_oop(rax, state);
   402 }
   404 void InterpreterMacroAssembler::push(TosState state) {
   405   verify_oop(rax, state);
   406   switch (state) {
   407   case atos: push_ptr();                break;
   408   case btos:
   409   case ztos:
   410   case ctos:
   411   case stos:
   412   case itos: push_i();                  break;
   413   case ltos: push_l();                  break;
   414   case ftos: push_f();                  break;
   415   case dtos: push_d();                  break;
   416   case vtos: /* nothing to do */        break;
   417   default  : ShouldNotReachHere();
   418   }
   419 }
   422 // Helpers for swap and dup
   423 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
   424   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
   425 }
   427 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
   428   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
   429 }
   432 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   433   // set sender sp
   434   lea(r13, Address(rsp, wordSize));
   435   // record last_sp
   436   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
   437 }
   440 // Jump to from_interpreted entry of a call unless single stepping is possible
   441 // in this thread in which case we must call the i2i entry
   442 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
   443   prepare_to_jump_from_interpreted();
   445   if (JvmtiExport::can_post_interpreter_events()) {
   446     Label run_compiled_code;
   447     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
   448     // compiled code in threads for which the event is enabled.  Check here for
   449     // interp_only_mode if these events CAN be enabled.
   450     // interp_only is an int, on little endian it is sufficient to test the byte only
   451     // Is a cmpl faster?
   452     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
   453     jccb(Assembler::zero, run_compiled_code);
   454     jmp(Address(method, Method::interpreter_entry_offset()));
   455     bind(run_compiled_code);
   456   }
   458   jmp(Address(method, Method::from_interpreted_offset()));
   460 }
   463 // The following two routines provide a hook so that an implementation
   464 // can schedule the dispatch in two parts.  amd64 does not do this.
   465 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
   466   // Nothing amd64 specific to be done here
   467 }
   469 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
   470   dispatch_next(state, step);
   471 }
   473 void InterpreterMacroAssembler::dispatch_base(TosState state,
   474                                               address* table,
   475                                               bool verifyoop) {
   476   verify_FPU(1, state);
   477   if (VerifyActivationFrameSize) {
   478     Label L;
   479     mov(rcx, rbp);
   480     subptr(rcx, rsp);
   481     int32_t min_frame_size =
   482       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
   483       wordSize;
   484     cmpptr(rcx, (int32_t)min_frame_size);
   485     jcc(Assembler::greaterEqual, L);
   486     stop("broken stack frame");
   487     bind(L);
   488   }
   489   if (verifyoop) {
   490     verify_oop(rax, state);
   491   }
   492   lea(rscratch1, ExternalAddress((address)table));
   493   jmp(Address(rscratch1, rbx, Address::times_8));
   494 }
   496 void InterpreterMacroAssembler::dispatch_only(TosState state) {
   497   dispatch_base(state, Interpreter::dispatch_table(state));
   498 }
   500 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
   501   dispatch_base(state, Interpreter::normal_table(state));
   502 }
   504 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
   505   dispatch_base(state, Interpreter::normal_table(state), false);
   506 }
   509 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
   510   // load next bytecode (load before advancing r13 to prevent AGI)
   511   load_unsigned_byte(rbx, Address(r13, step));
   512   // advance r13
   513   increment(r13, step);
   514   dispatch_base(state, Interpreter::dispatch_table(state));
   515 }
   517 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
   518   // load current bytecode
   519   load_unsigned_byte(rbx, Address(r13, 0));
   520   dispatch_base(state, table);
   521 }
   523 // remove activation
   524 //
   525 // Unlock the receiver if this is a synchronized method.
   526 // Unlock any Java monitors from syncronized blocks.
   527 // Remove the activation from the stack.
   528 //
   529 // If there are locked Java monitors
   530 //    If throw_monitor_exception
   531 //       throws IllegalMonitorStateException
   532 //    Else if install_monitor_exception
   533 //       installs IllegalMonitorStateException
   534 //    Else
   535 //       no error processing
   536 void InterpreterMacroAssembler::remove_activation(
   537         TosState state,
   538         Register ret_addr,
   539         bool throw_monitor_exception,
   540         bool install_monitor_exception,
   541         bool notify_jvmdi) {
   542   // Note: Registers rdx xmm0 may be in use for the
   543   // result check if synchronized method
   544   Label unlocked, unlock, no_unlock;
   546   // get the value of _do_not_unlock_if_synchronized into rdx
   547   const Address do_not_unlock_if_synchronized(r15_thread,
   548     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
   549   movbool(rdx, do_not_unlock_if_synchronized);
   550   movbool(do_not_unlock_if_synchronized, false); // reset the flag
   552  // get method access flags
   553   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
   554   movl(rcx, Address(rbx, Method::access_flags_offset()));
   555   testl(rcx, JVM_ACC_SYNCHRONIZED);
   556   jcc(Assembler::zero, unlocked);
   558   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
   559   // is set.
   560   testbool(rdx);
   561   jcc(Assembler::notZero, no_unlock);
   563   // unlock monitor
   564   push(state); // save result
   566   // BasicObjectLock will be first in list, since this is a
   567   // synchronized method. However, need to check that the object has
   568   // not been unlocked by an explicit monitorexit bytecode.
   569   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
   570                         wordSize - (int) sizeof(BasicObjectLock));
   571   // We use c_rarg1 so that if we go slow path it will be the correct
   572   // register for unlock_object to pass to VM directly
   573   lea(c_rarg1, monitor); // address of first monitor
   575   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
   576   testptr(rax, rax);
   577   jcc(Assembler::notZero, unlock);
   579   pop(state);
   580   if (throw_monitor_exception) {
   581     // Entry already unlocked, need to throw exception
   582     call_VM(noreg, CAST_FROM_FN_PTR(address,
   583                    InterpreterRuntime::throw_illegal_monitor_state_exception));
   584     should_not_reach_here();
   585   } else {
   586     // Monitor already unlocked during a stack unroll. If requested,
   587     // install an illegal_monitor_state_exception.  Continue with
   588     // stack unrolling.
   589     if (install_monitor_exception) {
   590       call_VM(noreg, CAST_FROM_FN_PTR(address,
   591                      InterpreterRuntime::new_illegal_monitor_state_exception));
   592     }
   593     jmp(unlocked);
   594   }
   596   bind(unlock);
   597   unlock_object(c_rarg1);
   598   pop(state);
   600   // Check that for block-structured locking (i.e., that all locked
   601   // objects has been unlocked)
   602   bind(unlocked);
   604   // rax: Might contain return value
   606   // Check that all monitors are unlocked
   607   {
   608     Label loop, exception, entry, restart;
   609     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
   610     const Address monitor_block_top(
   611         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
   612     const Address monitor_block_bot(
   613         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
   615     bind(restart);
   616     // We use c_rarg1 so that if we go slow path it will be the correct
   617     // register for unlock_object to pass to VM directly
   618     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
   619                                   // with top-most entry
   620     lea(rbx, monitor_block_bot);  // points to word before bottom of
   621                                   // monitor block
   622     jmp(entry);
   624     // Entry already locked, need to throw exception
   625     bind(exception);
   627     if (throw_monitor_exception) {
   628       // Throw exception
   629       MacroAssembler::call_VM(noreg,
   630                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
   631                                    throw_illegal_monitor_state_exception));
   632       should_not_reach_here();
   633     } else {
   634       // Stack unrolling. Unlock object and install illegal_monitor_exception.
   635       // Unlock does not block, so don't have to worry about the frame.
   636       // We don't have to preserve c_rarg1 since we are going to throw an exception.
   638       push(state);
   639       unlock_object(c_rarg1);
   640       pop(state);
   642       if (install_monitor_exception) {
   643         call_VM(noreg, CAST_FROM_FN_PTR(address,
   644                                         InterpreterRuntime::
   645                                         new_illegal_monitor_state_exception));
   646       }
   648       jmp(restart);
   649     }
   651     bind(loop);
   652     // check if current entry is used
   653     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
   654     jcc(Assembler::notEqual, exception);
   656     addptr(c_rarg1, entry_size); // otherwise advance to next entry
   657     bind(entry);
   658     cmpptr(c_rarg1, rbx); // check if bottom reached
   659     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
   660   }
   662   bind(no_unlock);
   664   // jvmti support
   665   if (notify_jvmdi) {
   666     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
   667   } else {
   668     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
   669   }
   671   // remove activation
   672   // get sender sp
   673   movptr(rbx,
   674          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
   675   leave();                           // remove frame anchor
   676   pop(ret_addr);                     // get return address
   677   mov(rsp, rbx);                     // set sp to sender sp
   678 }
   680 #endif // C_INTERP
   682 // Lock object
   683 //
   684 // Args:
   685 //      c_rarg1: BasicObjectLock to be used for locking
   686 //
   687 // Kills:
   688 //      rax
   689 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
   690 //      rscratch1, rscratch2 (scratch regs)
   691 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
   692   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
   694   if (UseHeavyMonitors) {
   695     call_VM(noreg,
   696             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   697             lock_reg);
   698   } else {
   699     Label done;
   701     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
   702     const Register obj_reg = c_rarg3; // Will contain the oop
   704     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
   705     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
   706     const int mark_offset = lock_offset +
   707                             BasicLock::displaced_header_offset_in_bytes();
   709     Label slow_case;
   711     // Load object pointer into obj_reg %c_rarg3
   712     movptr(obj_reg, Address(lock_reg, obj_offset));
   714     if (UseBiasedLocking) {
   715       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
   716     }
   718     // Load immediate 1 into swap_reg %rax
   719     movl(swap_reg, 1);
   721     // Load (object->mark() | 1) into swap_reg %rax
   722     orptr(swap_reg, Address(obj_reg, 0));
   724     // Save (object->mark() | 1) into BasicLock's displaced header
   725     movptr(Address(lock_reg, mark_offset), swap_reg);
   727     assert(lock_offset == 0,
   728            "displached header must be first word in BasicObjectLock");
   730     if (os::is_MP()) lock();
   731     cmpxchgptr(lock_reg, Address(obj_reg, 0));
   732     if (PrintBiasedLockingStatistics) {
   733       cond_inc32(Assembler::zero,
   734                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   735     }
   736     jcc(Assembler::zero, done);
   738     // Test if the oopMark is an obvious stack pointer, i.e.,
   739     //  1) (mark & 7) == 0, and
   740     //  2) rsp <= mark < mark + os::pagesize()
   741     //
   742     // These 3 tests can be done by evaluating the following
   743     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
   744     // assuming both stack pointer and pagesize have their
   745     // least significant 3 bits clear.
   746     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
   747     subptr(swap_reg, rsp);
   748     andptr(swap_reg, 7 - os::vm_page_size());
   750     // Save the test result, for recursive case, the result is zero
   751     movptr(Address(lock_reg, mark_offset), swap_reg);
   753     if (PrintBiasedLockingStatistics) {
   754       cond_inc32(Assembler::zero,
   755                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   756     }
   757     jcc(Assembler::zero, done);
   759     bind(slow_case);
   761     // Call the runtime routine for slow case
   762     call_VM(noreg,
   763             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   764             lock_reg);
   766     bind(done);
   767   }
   768 }
   771 // Unlocks an object. Used in monitorexit bytecode and
   772 // remove_activation.  Throws an IllegalMonitorException if object is
   773 // not locked by current thread.
   774 //
   775 // Args:
   776 //      c_rarg1: BasicObjectLock for lock
   777 //
   778 // Kills:
   779 //      rax
   780 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
   781 //      rscratch1, rscratch2 (scratch regs)
   782 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
   783   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
   785   if (UseHeavyMonitors) {
   786     call_VM(noreg,
   787             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   788             lock_reg);
   789   } else {
   790     Label done;
   792     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
   793     const Register header_reg = c_rarg2;  // Will contain the old oopMark
   794     const Register obj_reg    = c_rarg3;  // Will contain the oop
   796     save_bcp(); // Save in case of exception
   798     // Convert from BasicObjectLock structure to object and BasicLock
   799     // structure Store the BasicLock address into %rax
   800     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
   802     // Load oop into obj_reg(%c_rarg3)
   803     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
   805     // Free entry
   806     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
   808     if (UseBiasedLocking) {
   809       biased_locking_exit(obj_reg, header_reg, done);
   810     }
   812     // Load the old header from BasicLock structure
   813     movptr(header_reg, Address(swap_reg,
   814                                BasicLock::displaced_header_offset_in_bytes()));
   816     // Test for recursion
   817     testptr(header_reg, header_reg);
   819     // zero for recursive case
   820     jcc(Assembler::zero, done);
   822     // Atomic swap back the old header
   823     if (os::is_MP()) lock();
   824     cmpxchgptr(header_reg, Address(obj_reg, 0));
   826     // zero for recursive case
   827     jcc(Assembler::zero, done);
   829     // Call the runtime routine for slow case.
   830     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
   831          obj_reg); // restore obj
   832     call_VM(noreg,
   833             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   834             lock_reg);
   836     bind(done);
   838     restore_bcp();
   839   }
   840 }
   842 #ifndef CC_INTERP
   844 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
   845                                                          Label& zero_continue) {
   846   assert(ProfileInterpreter, "must be profiling interpreter");
   847   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
   848   testptr(mdp, mdp);
   849   jcc(Assembler::zero, zero_continue);
   850 }
   853 // Set the method data pointer for the current bcp.
   854 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
   855   assert(ProfileInterpreter, "must be profiling interpreter");
   856   Label set_mdp;
   857   push(rax);
   858   push(rbx);
   860   get_method(rbx);
   861   // Test MDO to avoid the call if it is NULL.
   862   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
   863   testptr(rax, rax);
   864   jcc(Assembler::zero, set_mdp);
   865   // rbx: method
   866   // r13: bcp
   867   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
   868   // rax: mdi
   869   // mdo is guaranteed to be non-zero here, we checked for it before the call.
   870   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
   871   addptr(rbx, in_bytes(MethodData::data_offset()));
   872   addptr(rax, rbx);
   873   bind(set_mdp);
   874   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
   875   pop(rbx);
   876   pop(rax);
   877 }
   879 void InterpreterMacroAssembler::verify_method_data_pointer() {
   880   assert(ProfileInterpreter, "must be profiling interpreter");
   881 #ifdef ASSERT
   882   Label verify_continue;
   883   push(rax);
   884   push(rbx);
   885   push(c_rarg3);
   886   push(c_rarg2);
   887   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
   888   get_method(rbx);
   890   // If the mdp is valid, it will point to a DataLayout header which is
   891   // consistent with the bcp.  The converse is highly probable also.
   892   load_unsigned_short(c_rarg2,
   893                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
   894   addptr(c_rarg2, Address(rbx, Method::const_offset()));
   895   lea(c_rarg2, Address(c_rarg2, ConstMethod::codes_offset()));
   896   cmpptr(c_rarg2, r13);
   897   jcc(Assembler::equal, verify_continue);
   898   // rbx: method
   899   // r13: bcp
   900   // c_rarg3: mdp
   901   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
   902                rbx, r13, c_rarg3);
   903   bind(verify_continue);
   904   pop(c_rarg2);
   905   pop(c_rarg3);
   906   pop(rbx);
   907   pop(rax);
   908 #endif // ASSERT
   909 }
   912 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
   913                                                 int constant,
   914                                                 Register value) {
   915   assert(ProfileInterpreter, "must be profiling interpreter");
   916   Address data(mdp_in, constant);
   917   movptr(data, value);
   918 }
   921 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   922                                                       int constant,
   923                                                       bool decrement) {
   924   // Counter address
   925   Address data(mdp_in, constant);
   927   increment_mdp_data_at(data, decrement);
   928 }
   930 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
   931                                                       bool decrement) {
   932   assert(ProfileInterpreter, "must be profiling interpreter");
   933   // %%% this does 64bit counters at best it is wasting space
   934   // at worst it is a rare bug when counters overflow
   936   if (decrement) {
   937     // Decrement the register.  Set condition codes.
   938     addptr(data, (int32_t) -DataLayout::counter_increment);
   939     // If the decrement causes the counter to overflow, stay negative
   940     Label L;
   941     jcc(Assembler::negative, L);
   942     addptr(data, (int32_t) DataLayout::counter_increment);
   943     bind(L);
   944   } else {
   945     assert(DataLayout::counter_increment == 1,
   946            "flow-free idiom only works with 1");
   947     // Increment the register.  Set carry flag.
   948     addptr(data, DataLayout::counter_increment);
   949     // If the increment causes the counter to overflow, pull back by 1.
   950     sbbptr(data, (int32_t)0);
   951   }
   952 }
   955 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   956                                                       Register reg,
   957                                                       int constant,
   958                                                       bool decrement) {
   959   Address data(mdp_in, reg, Address::times_1, constant);
   961   increment_mdp_data_at(data, decrement);
   962 }
   964 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
   965                                                 int flag_byte_constant) {
   966   assert(ProfileInterpreter, "must be profiling interpreter");
   967   int header_offset = in_bytes(DataLayout::header_offset());
   968   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
   969   // Set the flag
   970   orl(Address(mdp_in, header_offset), header_bits);
   971 }
   975 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
   976                                                  int offset,
   977                                                  Register value,
   978                                                  Register test_value_out,
   979                                                  Label& not_equal_continue) {
   980   assert(ProfileInterpreter, "must be profiling interpreter");
   981   if (test_value_out == noreg) {
   982     cmpptr(value, Address(mdp_in, offset));
   983   } else {
   984     // Put the test value into a register, so caller can use it:
   985     movptr(test_value_out, Address(mdp_in, offset));
   986     cmpptr(test_value_out, value);
   987   }
   988   jcc(Assembler::notEqual, not_equal_continue);
   989 }
   992 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
   993                                                      int offset_of_disp) {
   994   assert(ProfileInterpreter, "must be profiling interpreter");
   995   Address disp_address(mdp_in, offset_of_disp);
   996   addptr(mdp_in, disp_address);
   997   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
   998 }
  1001 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
  1002                                                      Register reg,
  1003                                                      int offset_of_disp) {
  1004   assert(ProfileInterpreter, "must be profiling interpreter");
  1005   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
  1006   addptr(mdp_in, disp_address);
  1007   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1011 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
  1012                                                        int constant) {
  1013   assert(ProfileInterpreter, "must be profiling interpreter");
  1014   addptr(mdp_in, constant);
  1015   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1019 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
  1020   assert(ProfileInterpreter, "must be profiling interpreter");
  1021   push(return_bci); // save/restore across call_VM
  1022   call_VM(noreg,
  1023           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
  1024           return_bci);
  1025   pop(return_bci);
  1029 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
  1030                                                      Register bumped_count) {
  1031   if (ProfileInterpreter) {
  1032     Label profile_continue;
  1034     // If no method data exists, go to profile_continue.
  1035     // Otherwise, assign to mdp
  1036     test_method_data_pointer(mdp, profile_continue);
  1038     // We are taking a branch.  Increment the taken count.
  1039     // We inline increment_mdp_data_at to return bumped_count in a register
  1040     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
  1041     Address data(mdp, in_bytes(JumpData::taken_offset()));
  1042     movptr(bumped_count, data);
  1043     assert(DataLayout::counter_increment == 1,
  1044             "flow-free idiom only works with 1");
  1045     addptr(bumped_count, DataLayout::counter_increment);
  1046     sbbptr(bumped_count, 0);
  1047     movptr(data, bumped_count); // Store back out
  1049     // The method data pointer needs to be updated to reflect the new target.
  1050     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
  1051     bind(profile_continue);
  1056 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
  1057   if (ProfileInterpreter) {
  1058     Label profile_continue;
  1060     // If no method data exists, go to profile_continue.
  1061     test_method_data_pointer(mdp, profile_continue);
  1063     // We are taking a branch.  Increment the not taken count.
  1064     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
  1066     // The method data pointer needs to be updated to correspond to
  1067     // the next bytecode
  1068     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
  1069     bind(profile_continue);
  1073 void InterpreterMacroAssembler::profile_call(Register mdp) {
  1074   if (ProfileInterpreter) {
  1075     Label profile_continue;
  1077     // If no method data exists, go to profile_continue.
  1078     test_method_data_pointer(mdp, profile_continue);
  1080     // We are making a call.  Increment the count.
  1081     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1083     // The method data pointer needs to be updated to reflect the new target.
  1084     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
  1085     bind(profile_continue);
  1090 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
  1091   if (ProfileInterpreter) {
  1092     Label profile_continue;
  1094     // If no method data exists, go to profile_continue.
  1095     test_method_data_pointer(mdp, profile_continue);
  1097     // We are making a call.  Increment the count.
  1098     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1100     // The method data pointer needs to be updated to reflect the new target.
  1101     update_mdp_by_constant(mdp,
  1102                            in_bytes(VirtualCallData::
  1103                                     virtual_call_data_size()));
  1104     bind(profile_continue);
  1109 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
  1110                                                      Register mdp,
  1111                                                      Register reg2,
  1112                                                      bool receiver_can_be_null) {
  1113   if (ProfileInterpreter) {
  1114     Label profile_continue;
  1116     // If no method data exists, go to profile_continue.
  1117     test_method_data_pointer(mdp, profile_continue);
  1119     Label skip_receiver_profile;
  1120     if (receiver_can_be_null) {
  1121       Label not_null;
  1122       testptr(receiver, receiver);
  1123       jccb(Assembler::notZero, not_null);
  1124       // We are making a call.  Increment the count for null receiver.
  1125       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1126       jmp(skip_receiver_profile);
  1127       bind(not_null);
  1130     // Record the receiver type.
  1131     record_klass_in_profile(receiver, mdp, reg2, true);
  1132     bind(skip_receiver_profile);
  1134     // The method data pointer needs to be updated to reflect the new target.
  1135     update_mdp_by_constant(mdp,
  1136                            in_bytes(VirtualCallData::
  1137                                     virtual_call_data_size()));
  1138     bind(profile_continue);
  1142 // This routine creates a state machine for updating the multi-row
  1143 // type profile at a virtual call site (or other type-sensitive bytecode).
  1144 // The machine visits each row (of receiver/count) until the receiver type
  1145 // is found, or until it runs out of rows.  At the same time, it remembers
  1146 // the location of the first empty row.  (An empty row records null for its
  1147 // receiver, and can be allocated for a newly-observed receiver type.)
  1148 // Because there are two degrees of freedom in the state, a simple linear
  1149 // search will not work; it must be a decision tree.  Hence this helper
  1150 // function is recursive, to generate the required tree structured code.
  1151 // It's the interpreter, so we are trading off code space for speed.
  1152 // See below for example code.
  1153 void InterpreterMacroAssembler::record_klass_in_profile_helper(
  1154                                         Register receiver, Register mdp,
  1155                                         Register reg2, int start_row,
  1156                                         Label& done, bool is_virtual_call) {
  1157   if (TypeProfileWidth == 0) {
  1158     if (is_virtual_call) {
  1159       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1161     return;
  1164   int last_row = VirtualCallData::row_limit() - 1;
  1165   assert(start_row <= last_row, "must be work left to do");
  1166   // Test this row for both the receiver and for null.
  1167   // Take any of three different outcomes:
  1168   //   1. found receiver => increment count and goto done
  1169   //   2. found null => keep looking for case 1, maybe allocate this cell
  1170   //   3. found something else => keep looking for cases 1 and 2
  1171   // Case 3 is handled by a recursive call.
  1172   for (int row = start_row; row <= last_row; row++) {
  1173     Label next_test;
  1174     bool test_for_null_also = (row == start_row);
  1176     // See if the receiver is receiver[n].
  1177     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
  1178     test_mdp_data_at(mdp, recvr_offset, receiver,
  1179                      (test_for_null_also ? reg2 : noreg),
  1180                      next_test);
  1181     // (Reg2 now contains the receiver from the CallData.)
  1183     // The receiver is receiver[n].  Increment count[n].
  1184     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
  1185     increment_mdp_data_at(mdp, count_offset);
  1186     jmp(done);
  1187     bind(next_test);
  1189     if (test_for_null_also) {
  1190       Label found_null;
  1191       // Failed the equality check on receiver[n]...  Test for null.
  1192       testptr(reg2, reg2);
  1193       if (start_row == last_row) {
  1194         // The only thing left to do is handle the null case.
  1195         if (is_virtual_call) {
  1196           jccb(Assembler::zero, found_null);
  1197           // Receiver did not match any saved receiver and there is no empty row for it.
  1198           // Increment total counter to indicate polymorphic case.
  1199           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1200           jmp(done);
  1201           bind(found_null);
  1202         } else {
  1203           jcc(Assembler::notZero, done);
  1205         break;
  1207       // Since null is rare, make it be the branch-taken case.
  1208       jcc(Assembler::zero, found_null);
  1210       // Put all the "Case 3" tests here.
  1211       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
  1213       // Found a null.  Keep searching for a matching receiver,
  1214       // but remember that this is an empty (unused) slot.
  1215       bind(found_null);
  1219   // In the fall-through case, we found no matching receiver, but we
  1220   // observed the receiver[start_row] is NULL.
  1222   // Fill in the receiver field and increment the count.
  1223   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
  1224   set_mdp_data_at(mdp, recvr_offset, receiver);
  1225   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
  1226   movl(reg2, DataLayout::counter_increment);
  1227   set_mdp_data_at(mdp, count_offset, reg2);
  1228   if (start_row > 0) {
  1229     jmp(done);
  1233 // Example state machine code for three profile rows:
  1234 //   // main copy of decision tree, rooted at row[1]
  1235 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
  1236 //   if (row[0].rec != NULL) {
  1237 //     // inner copy of decision tree, rooted at row[1]
  1238 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1239 //     if (row[1].rec != NULL) {
  1240 //       // degenerate decision tree, rooted at row[2]
  1241 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1242 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
  1243 //       row[2].init(rec); goto done;
  1244 //     } else {
  1245 //       // remember row[1] is empty
  1246 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1247 //       row[1].init(rec); goto done;
  1248 //     }
  1249 //   } else {
  1250 //     // remember row[0] is empty
  1251 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1252 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
  1253 //     row[0].init(rec); goto done;
  1254 //   }
  1255 //   done:
  1257 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
  1258                                                         Register mdp, Register reg2,
  1259                                                         bool is_virtual_call) {
  1260   assert(ProfileInterpreter, "must be profiling");
  1261   Label done;
  1263   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
  1265   bind (done);
  1268 void InterpreterMacroAssembler::profile_ret(Register return_bci,
  1269                                             Register mdp) {
  1270   if (ProfileInterpreter) {
  1271     Label profile_continue;
  1272     uint row;
  1274     // If no method data exists, go to profile_continue.
  1275     test_method_data_pointer(mdp, profile_continue);
  1277     // Update the total ret count.
  1278     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1280     for (row = 0; row < RetData::row_limit(); row++) {
  1281       Label next_test;
  1283       // See if return_bci is equal to bci[n]:
  1284       test_mdp_data_at(mdp,
  1285                        in_bytes(RetData::bci_offset(row)),
  1286                        return_bci, noreg,
  1287                        next_test);
  1289       // return_bci is equal to bci[n].  Increment the count.
  1290       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
  1292       // The method data pointer needs to be updated to reflect the new target.
  1293       update_mdp_by_offset(mdp,
  1294                            in_bytes(RetData::bci_displacement_offset(row)));
  1295       jmp(profile_continue);
  1296       bind(next_test);
  1299     update_mdp_for_ret(return_bci);
  1301     bind(profile_continue);
  1306 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
  1307   if (ProfileInterpreter) {
  1308     Label profile_continue;
  1310     // If no method data exists, go to profile_continue.
  1311     test_method_data_pointer(mdp, profile_continue);
  1313     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
  1315     // The method data pointer needs to be updated.
  1316     int mdp_delta = in_bytes(BitData::bit_data_size());
  1317     if (TypeProfileCasts) {
  1318       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1320     update_mdp_by_constant(mdp, mdp_delta);
  1322     bind(profile_continue);
  1327 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
  1328   if (ProfileInterpreter && TypeProfileCasts) {
  1329     Label profile_continue;
  1331     // If no method data exists, go to profile_continue.
  1332     test_method_data_pointer(mdp, profile_continue);
  1334     int count_offset = in_bytes(CounterData::count_offset());
  1335     // Back up the address, since we have already bumped the mdp.
  1336     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
  1338     // *Decrement* the counter.  We expect to see zero or small negatives.
  1339     increment_mdp_data_at(mdp, count_offset, true);
  1341     bind (profile_continue);
  1346 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
  1347   if (ProfileInterpreter) {
  1348     Label profile_continue;
  1350     // If no method data exists, go to profile_continue.
  1351     test_method_data_pointer(mdp, profile_continue);
  1353     // The method data pointer needs to be updated.
  1354     int mdp_delta = in_bytes(BitData::bit_data_size());
  1355     if (TypeProfileCasts) {
  1356       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1358       // Record the object type.
  1359       record_klass_in_profile(klass, mdp, reg2, false);
  1361     update_mdp_by_constant(mdp, mdp_delta);
  1363     bind(profile_continue);
  1368 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
  1369   if (ProfileInterpreter) {
  1370     Label profile_continue;
  1372     // If no method data exists, go to profile_continue.
  1373     test_method_data_pointer(mdp, profile_continue);
  1375     // Update the default case count
  1376     increment_mdp_data_at(mdp,
  1377                           in_bytes(MultiBranchData::default_count_offset()));
  1379     // The method data pointer needs to be updated.
  1380     update_mdp_by_offset(mdp,
  1381                          in_bytes(MultiBranchData::
  1382                                   default_displacement_offset()));
  1384     bind(profile_continue);
  1389 void InterpreterMacroAssembler::profile_switch_case(Register index,
  1390                                                     Register mdp,
  1391                                                     Register reg2) {
  1392   if (ProfileInterpreter) {
  1393     Label profile_continue;
  1395     // If no method data exists, go to profile_continue.
  1396     test_method_data_pointer(mdp, profile_continue);
  1398     // Build the base (index * per_case_size_in_bytes()) +
  1399     // case_array_offset_in_bytes()
  1400     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
  1401     imulptr(index, reg2); // XXX l ?
  1402     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
  1404     // Update the case count
  1405     increment_mdp_data_at(mdp,
  1406                           index,
  1407                           in_bytes(MultiBranchData::relative_count_offset()));
  1409     // The method data pointer needs to be updated.
  1410     update_mdp_by_offset(mdp,
  1411                          index,
  1412                          in_bytes(MultiBranchData::
  1413                                   relative_displacement_offset()));
  1415     bind(profile_continue);
  1421 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
  1422   if (state == atos) {
  1423     MacroAssembler::verify_oop(reg);
  1427 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
  1429 #endif // !CC_INTERP
  1432 void InterpreterMacroAssembler::notify_method_entry() {
  1433   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1434   // track stack depth.  If it is possible to enter interp_only_mode we add
  1435   // the code to check if the event should be sent.
  1436   if (JvmtiExport::can_post_interpreter_events()) {
  1437     Label L;
  1438     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1439     testl(rdx, rdx);
  1440     jcc(Assembler::zero, L);
  1441     call_VM(noreg, CAST_FROM_FN_PTR(address,
  1442                                     InterpreterRuntime::post_method_entry));
  1443     bind(L);
  1447     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1448     get_method(c_rarg1);
  1449     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
  1450                  r15_thread, c_rarg1);
  1453   // RedefineClasses() tracing support for obsolete method entry
  1454   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
  1455     get_method(c_rarg1);
  1456     call_VM_leaf(
  1457       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
  1458       r15_thread, c_rarg1);
  1463 void InterpreterMacroAssembler::notify_method_exit(
  1464     TosState state, NotifyMethodExitMode mode) {
  1465   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1466   // track stack depth.  If it is possible to enter interp_only_mode we add
  1467   // the code to check if the event should be sent.
  1468   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
  1469     Label L;
  1470     // Note: frame::interpreter_frame_result has a dependency on how the
  1471     // method result is saved across the call to post_method_exit. If this
  1472     // is changed then the interpreter_frame_result implementation will
  1473     // need to be updated too.
  1475     // For c++ interpreter the result is always stored at a known location in the frame
  1476     // template interpreter will leave it on the top of the stack.
  1477     NOT_CC_INTERP(push(state);)
  1478     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1479     testl(rdx, rdx);
  1480     jcc(Assembler::zero, L);
  1481     call_VM(noreg,
  1482             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
  1483     bind(L);
  1484     NOT_CC_INTERP(pop(state));
  1488     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1489     NOT_CC_INTERP(push(state));
  1490     get_method(c_rarg1);
  1491     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
  1492                  r15_thread, c_rarg1);
  1493     NOT_CC_INTERP(pop(state));
  1497 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
  1498 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
  1499                                                         int increment, int mask,
  1500                                                         Register scratch, bool preloaded,
  1501                                                         Condition cond, Label* where) {
  1502   if (!preloaded) {
  1503     movl(scratch, counter_addr);
  1505   incrementl(scratch, increment);
  1506   movl(counter_addr, scratch);
  1507   andl(scratch, mask);
  1508   if (where != NULL) {
  1509     jcc(cond, *where);

mercurial