src/cpu/x86/vm/interp_masm_x86_64.cpp

Fri, 11 Apr 2008 09:56:35 -0400

author
kamg
date
Fri, 11 Apr 2008 09:56:35 -0400
changeset 545
a49a647afe9a
parent 435
a61af66fc99e
child 548
ba764ed4b6f2
permissions
-rw-r--r--

Merge

     1 /*
     2  * Copyright 2003-2007 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
    20  * CA 95054 USA or visit www.sun.com if you need additional information or
    21  * have any questions.
    22  *
    23  */
    25 #include "incls/_precompiled.incl"
    26 #include "incls/_interp_masm_x86_64.cpp.incl"
    29 // Implementation of InterpreterMacroAssembler
    31 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
    32                                                   int number_of_arguments) {
    33   // interpreter specific
    34   //
    35   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
    36   //       since these are callee saved registers and no blocking/
    37   //       GC can happen in leaf calls.
    38 #ifdef ASSERT
    39   save_bcp();
    40   {
    41     Label L;
    42     cmpq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD);
    43     jcc(Assembler::equal, L);
    44     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    45          " last_sp != NULL");
    46     bind(L);
    47   }
    48 #endif
    49   // super call
    50   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
    51   // interpreter specific
    52 #ifdef ASSERT
    53   {
    54     Label L;
    55     cmpq(r13, Address(rbp, frame::interpreter_frame_bcx_offset * wordSize));
    56     jcc(Assembler::equal, L);
    57     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    58          " r13 not callee saved?");
    59     bind(L);
    60   }
    61   {
    62     Label L;
    63     cmpq(r14, Address(rbp, frame::interpreter_frame_locals_offset * wordSize));
    64     jcc(Assembler::equal, L);
    65     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    66          " r14 not callee saved?");
    67     bind(L);
    68   }
    69 #endif
    70 }
    72 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
    73                                              Register java_thread,
    74                                              Register last_java_sp,
    75                                              address  entry_point,
    76                                              int      number_of_arguments,
    77                                              bool     check_exceptions) {
    78   // interpreter specific
    79   //
    80   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
    81   //       really make a difference for these runtime calls, since they are
    82   //       slow anyway. Btw., bcp must be saved/restored since it may change
    83   //       due to GC.
    84   // assert(java_thread == noreg , "not expecting a precomputed java thread");
    85   save_bcp();
    86 #ifdef ASSERT
    87   {
    88     Label L;
    89     cmpq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD);
    90     jcc(Assembler::equal, L);
    91     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    92          " last_sp != NULL");
    93     bind(L);
    94   }
    95 #endif /* ASSERT */
    96   // super call
    97   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
    98                                entry_point, number_of_arguments,
    99                                check_exceptions);
   100   // interpreter specific
   101   restore_bcp();
   102   restore_locals();
   103 }
   106 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
   107   if (JvmtiExport::can_pop_frame()) {
   108     Label L;
   109     // Initiate popframe handling only if it is not already being
   110     // processed.  If the flag has the popframe_processing bit set, it
   111     // means that this code is called *during* popframe handling - we
   112     // don't want to reenter.
   113     // This method is only called just after the call into the vm in
   114     // call_VM_base, so the arg registers are available.
   115     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
   116     testl(c_rarg0, JavaThread::popframe_pending_bit);
   117     jcc(Assembler::zero, L);
   118     testl(c_rarg0, JavaThread::popframe_processing_bit);
   119     jcc(Assembler::notZero, L);
   120     // Call Interpreter::remove_activation_preserving_args_entry() to get the
   121     // address of the same-named entrypoint in the generated interpreter code.
   122     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
   123     jmp(rax);
   124     bind(L);
   125   }
   126 }
   129 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
   130   movq(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   131   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
   132   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
   133   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
   134   switch (state) {
   135     case atos: movq(rax, oop_addr);
   136                movptr(oop_addr, NULL_WORD);
   137                verify_oop(rax, state);              break;
   138     case ltos: movq(rax, val_addr);                 break;
   139     case btos:                                   // fall through
   140     case ctos:                                   // fall through
   141     case stos:                                   // fall through
   142     case itos: movl(rax, val_addr);                 break;
   143     case ftos: movflt(xmm0, val_addr);              break;
   144     case dtos: movdbl(xmm0, val_addr);              break;
   145     case vtos: /* nothing to do */                  break;
   146     default  : ShouldNotReachHere();
   147   }
   148   // Clean up tos value in the thread object
   149   movl(tos_addr,  (int) ilgl);
   150   movl(val_addr,  (int) NULL_WORD);
   151 }
   154 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
   155   if (JvmtiExport::can_force_early_return()) {
   156     Label L;
   157     movq(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   158     testq(c_rarg0, c_rarg0);
   159     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
   161     // Initiate earlyret handling only if it is not already being processed.
   162     // If the flag has the earlyret_processing bit set, it means that this code
   163     // is called *during* earlyret handling - we don't want to reenter.
   164     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
   165     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
   166     jcc(Assembler::notEqual, L);
   168     // Call Interpreter::remove_activation_early_entry() to get the address of the
   169     // same-named entrypoint in the generated interpreter code.
   170     movq(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   171     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
   172     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
   173     jmp(rax);
   174     bind(L);
   175   }
   176 }
   179 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
   180   Register reg,
   181   int bcp_offset) {
   182   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
   183   movl(reg, Address(r13, bcp_offset));
   184   bswapl(reg);
   185   shrl(reg, 16);
   186 }
   189 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
   190                                                            Register index,
   191                                                            int bcp_offset) {
   192   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
   193   assert(cache != index, "must use different registers");
   194   load_unsigned_word(index, Address(r13, bcp_offset));
   195   movq(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   196   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   197   // convert from field index to ConstantPoolCacheEntry index
   198   shll(index, 2);
   199 }
   202 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
   203                                                                Register tmp,
   204                                                                int bcp_offset) {
   205   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
   206   assert(cache != tmp, "must use different register");
   207   load_unsigned_word(tmp, Address(r13, bcp_offset));
   208   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   209   // convert from field index to ConstantPoolCacheEntry index
   210   // and from word offset to byte offset
   211   shll(tmp, 2 + LogBytesPerWord);
   212   movq(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   213   // skip past the header
   214   addq(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
   215   addq(cache, tmp);  // construct pointer to cache entry
   216 }
   219 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
   220 // subtype of super_klass.
   221 //
   222 // Args:
   223 //      rax: superklass
   224 //      Rsub_klass: subklass
   225 //
   226 // Kills:
   227 //      rcx, rdi
   228 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
   229                                                   Label& ok_is_subtype) {
   230   assert(Rsub_klass != rax, "rax holds superklass");
   231   assert(Rsub_klass != r14, "r14 holds locals");
   232   assert(Rsub_klass != r13, "r13 holds bcp");
   233   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
   234   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
   236   Label not_subtype, loop;
   238   // Profile the not-null value's klass.
   239   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, rdi
   241   // Load the super-klass's check offset into rcx
   242   movl(rcx, Address(rax, sizeof(oopDesc) +
   243                     Klass::super_check_offset_offset_in_bytes()));
   244   // Load from the sub-klass's super-class display list, or a 1-word
   245   // cache of the secondary superclass list, or a failing value with a
   246   // sentinel offset if the super-klass is an interface or
   247   // exceptionally deep in the Java hierarchy and we have to scan the
   248   // secondary superclass list the hard way.  See if we get an
   249   // immediate positive hit
   250   cmpq(rax, Address(Rsub_klass, rcx, Address::times_1));
   251   jcc(Assembler::equal,ok_is_subtype);
   253   // Check for immediate negative hit
   254   cmpl(rcx, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
   255   jcc( Assembler::notEqual, not_subtype );
   256   // Check for self
   257   cmpq(Rsub_klass, rax);
   258   jcc(Assembler::equal, ok_is_subtype);
   260   // Now do a linear scan of the secondary super-klass chain.
   261   movq(rdi, Address(Rsub_klass, sizeof(oopDesc) +
   262                     Klass::secondary_supers_offset_in_bytes()));
   263   // rdi holds the objArrayOop of secondary supers.
   264   // Load the array length
   265   movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
   266   // Skip to start of data; also clear Z flag incase rcx is zero
   267   addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
   268   // Scan rcx words at [rdi] for occurance of rax
   269   // Set NZ/Z based on last compare
   270   repne_scan();
   271   // Not equal?
   272   jcc(Assembler::notEqual, not_subtype);
   273   // Must be equal but missed in cache.  Update cache.
   274   movq(Address(Rsub_klass, sizeof(oopDesc) +
   275                Klass::secondary_super_cache_offset_in_bytes()), rax);
   276   jmp(ok_is_subtype);
   278   bind(not_subtype);
   279   profile_typecheck_failed(rcx); // blows rcx
   280 }
   283 // Java Expression Stack
   285 #ifdef ASSERT
   286 // Verifies that the stack tag matches.  Must be called before the stack
   287 // value is popped off the stack.
   288 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t) {
   289   if (TaggedStackInterpreter) {
   290     frame::Tag tag = t;
   291     if (t == frame::TagCategory2) {
   292       tag = frame::TagValue;
   293       Label hokay;
   294       cmpq(Address(rsp, 3*wordSize), (int)tag);
   295       jcc(Assembler::equal, hokay);
   296       stop("Java Expression stack tag high value is bad");
   297       bind(hokay);
   298     }
   299     Label okay;
   300     cmpq(Address(rsp, wordSize), (int)tag);
   301     jcc(Assembler::equal, okay);
   302     // Also compare if the stack value is zero, then the tag might
   303     // not have been set coming from deopt.
   304     cmpq(Address(rsp, 0), 0);
   305     jcc(Assembler::equal, okay);
   306     stop("Java Expression stack tag value is bad");
   307     bind(okay);
   308   }
   309 }
   310 #endif // ASSERT
   312 void InterpreterMacroAssembler::pop_ptr(Register r) {
   313   debug_only(verify_stack_tag(frame::TagReference));
   314   popq(r);
   315   if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
   316 }
   318 void InterpreterMacroAssembler::pop_ptr(Register r, Register tag) {
   319   popq(r);
   320   if (TaggedStackInterpreter) popq(tag);
   321 }
   323 void InterpreterMacroAssembler::pop_i(Register r) {
   324   // XXX can't use popq currently, upper half non clean
   325   debug_only(verify_stack_tag(frame::TagValue));
   326   movl(r, Address(rsp, 0));
   327   addq(rsp, wordSize);
   328   if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
   329 }
   331 void InterpreterMacroAssembler::pop_l(Register r) {
   332   debug_only(verify_stack_tag(frame::TagCategory2));
   333   movq(r, Address(rsp, 0));
   334   addq(rsp, 2 * Interpreter::stackElementSize());
   335 }
   337 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
   338   debug_only(verify_stack_tag(frame::TagValue));
   339   movflt(r, Address(rsp, 0));
   340   addq(rsp, wordSize);
   341   if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
   342 }
   344 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
   345   debug_only(verify_stack_tag(frame::TagCategory2));
   346   movdbl(r, Address(rsp, 0));
   347   addq(rsp, 2 * Interpreter::stackElementSize());
   348 }
   350 void InterpreterMacroAssembler::push_ptr(Register r) {
   351   if (TaggedStackInterpreter) pushq(frame::TagReference);
   352   pushq(r);
   353 }
   355 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) {
   356   if (TaggedStackInterpreter) pushq(tag);
   357   pushq(r);
   358 }
   360 void InterpreterMacroAssembler::push_i(Register r) {
   361   if (TaggedStackInterpreter) pushq(frame::TagValue);
   362   pushq(r);
   363 }
   365 void InterpreterMacroAssembler::push_l(Register r) {
   366   if (TaggedStackInterpreter) {
   367     pushq(frame::TagValue);
   368     subq(rsp, 1 * wordSize);
   369     pushq(frame::TagValue);
   370     subq(rsp, 1 * wordSize);
   371   } else {
   372     subq(rsp, 2 * wordSize);
   373   }
   374   movq(Address(rsp, 0), r);
   375 }
   377 void InterpreterMacroAssembler::push_f(XMMRegister r) {
   378   if (TaggedStackInterpreter) pushq(frame::TagValue);
   379   subq(rsp, wordSize);
   380   movflt(Address(rsp, 0), r);
   381 }
   383 void InterpreterMacroAssembler::push_d(XMMRegister r) {
   384   if (TaggedStackInterpreter) {
   385     pushq(frame::TagValue);
   386     subq(rsp, 1 * wordSize);
   387     pushq(frame::TagValue);
   388     subq(rsp, 1 * wordSize);
   389   } else {
   390     subq(rsp, 2 * wordSize);
   391   }
   392   movdbl(Address(rsp, 0), r);
   393 }
   395 void InterpreterMacroAssembler::pop(TosState state) {
   396   switch (state) {
   397   case atos: pop_ptr();                 break;
   398   case btos:
   399   case ctos:
   400   case stos:
   401   case itos: pop_i();                   break;
   402   case ltos: pop_l();                   break;
   403   case ftos: pop_f();                   break;
   404   case dtos: pop_d();                   break;
   405   case vtos: /* nothing to do */        break;
   406   default:   ShouldNotReachHere();
   407   }
   408   verify_oop(rax, state);
   409 }
   411 void InterpreterMacroAssembler::push(TosState state) {
   412   verify_oop(rax, state);
   413   switch (state) {
   414   case atos: push_ptr();                break;
   415   case btos:
   416   case ctos:
   417   case stos:
   418   case itos: push_i();                  break;
   419   case ltos: push_l();                  break;
   420   case ftos: push_f();                  break;
   421   case dtos: push_d();                  break;
   422   case vtos: /* nothing to do */        break;
   423   default  : ShouldNotReachHere();
   424   }
   425 }
   428 // Tagged stack helpers for swap and dup
   429 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val,
   430                                                  Register tag) {
   431   movq(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
   432   if (TaggedStackInterpreter) {
   433     movq(tag, Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)));
   434   }
   435 }
   437 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val,
   438                                                   Register tag) {
   439   movq(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
   440   if (TaggedStackInterpreter) {
   441     movq(Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)), tag);
   442   }
   443 }
   446 // Tagged local support
   447 void InterpreterMacroAssembler::tag_local(frame::Tag tag, int n) {
   448   if (TaggedStackInterpreter) {
   449     if (tag == frame::TagCategory2) {
   450       mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)),
   451            (intptr_t)frame::TagValue);
   452       mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)),
   453            (intptr_t)frame::TagValue);
   454     } else {
   455       mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (intptr_t)tag);
   456     }
   457   }
   458 }
   460 void InterpreterMacroAssembler::tag_local(frame::Tag tag, Register idx) {
   461   if (TaggedStackInterpreter) {
   462     if (tag == frame::TagCategory2) {
   463       mov64(Address(r14, idx, Address::times_8,
   464                   Interpreter::local_tag_offset_in_bytes(1)), (intptr_t)frame::TagValue);
   465       mov64(Address(r14, idx, Address::times_8,
   466                   Interpreter::local_tag_offset_in_bytes(0)), (intptr_t)frame::TagValue);
   467     } else {
   468       mov64(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)),
   469            (intptr_t)tag);
   470     }
   471   }
   472 }
   474 void InterpreterMacroAssembler::tag_local(Register tag, Register idx) {
   475   if (TaggedStackInterpreter) {
   476     // can only be TagValue or TagReference
   477     movq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), tag);
   478   }
   479 }
   482 void InterpreterMacroAssembler::tag_local(Register tag, int n) {
   483   if (TaggedStackInterpreter) {
   484     // can only be TagValue or TagReference
   485     movq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), tag);
   486   }
   487 }
   489 #ifdef ASSERT
   490 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, int n) {
   491   if (TaggedStackInterpreter) {
   492      frame::Tag t = tag;
   493     if (tag == frame::TagCategory2) {
   494       Label nbl;
   495       t = frame::TagValue;  // change to what is stored in locals
   496       cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)), (int)t);
   497       jcc(Assembler::equal, nbl);
   498       stop("Local tag is bad for long/double");
   499       bind(nbl);
   500     }
   501     Label notBad;
   502     cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (int)t);
   503     jcc(Assembler::equal, notBad);
   504     // Also compare if the local value is zero, then the tag might
   505     // not have been set coming from deopt.
   506     cmpq(Address(r14, Interpreter::local_offset_in_bytes(n)), 0);
   507     jcc(Assembler::equal, notBad);
   508     stop("Local tag is bad");
   509     bind(notBad);
   510   }
   511 }
   513 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, Register idx) {
   514   if (TaggedStackInterpreter) {
   515     frame::Tag t = tag;
   516     if (tag == frame::TagCategory2) {
   517       Label nbl;
   518       t = frame::TagValue;  // change to what is stored in locals
   519       cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(1)), (int)t);
   520       jcc(Assembler::equal, nbl);
   521       stop("Local tag is bad for long/double");
   522       bind(nbl);
   523     }
   524     Label notBad;
   525     cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), (int)t);
   526     jcc(Assembler::equal, notBad);
   527     // Also compare if the local value is zero, then the tag might
   528     // not have been set coming from deopt.
   529     cmpq(Address(r14, idx, Address::times_8, Interpreter::local_offset_in_bytes(0)), 0);
   530     jcc(Assembler::equal, notBad);
   531     stop("Local tag is bad");
   532     bind(notBad);
   533   }
   534 }
   535 #endif // ASSERT
   538 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
   539   MacroAssembler::call_VM_leaf_base(entry_point, 0);
   540 }
   543 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
   544                                                    Register arg_1) {
   545   if (c_rarg0 != arg_1) {
   546     movq(c_rarg0, arg_1);
   547   }
   548   MacroAssembler::call_VM_leaf_base(entry_point, 1);
   549 }
   552 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
   553                                                    Register arg_1,
   554                                                    Register arg_2) {
   555   assert(c_rarg0 != arg_2, "smashed argument");
   556   assert(c_rarg1 != arg_1, "smashed argument");
   557   if (c_rarg0 != arg_1) {
   558     movq(c_rarg0, arg_1);
   559   }
   560   if (c_rarg1 != arg_2) {
   561     movq(c_rarg1, arg_2);
   562   }
   563   MacroAssembler::call_VM_leaf_base(entry_point, 2);
   564 }
   566 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
   567                                                    Register arg_1,
   568                                                    Register arg_2,
   569                                                    Register arg_3) {
   570   assert(c_rarg0 != arg_2, "smashed argument");
   571   assert(c_rarg0 != arg_3, "smashed argument");
   572   assert(c_rarg1 != arg_1, "smashed argument");
   573   assert(c_rarg1 != arg_3, "smashed argument");
   574   assert(c_rarg2 != arg_1, "smashed argument");
   575   assert(c_rarg2 != arg_2, "smashed argument");
   576   if (c_rarg0 != arg_1) {
   577     movq(c_rarg0, arg_1);
   578   }
   579   if (c_rarg1 != arg_2) {
   580     movq(c_rarg1, arg_2);
   581   }
   582   if (c_rarg2 != arg_3) {
   583     movq(c_rarg2, arg_3);
   584   }
   585   MacroAssembler::call_VM_leaf_base(entry_point, 3);
   586 }
   588 // Jump to from_interpreted entry of a call unless single stepping is possible
   589 // in this thread in which case we must call the i2i entry
   590 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
   591   // set sender sp
   592   leaq(r13, Address(rsp, wordSize));
   593   // record last_sp
   594   movq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
   596   if (JvmtiExport::can_post_interpreter_events()) {
   597     Label run_compiled_code;
   598     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
   599     // compiled code in threads for which the event is enabled.  Check here for
   600     // interp_only_mode if these events CAN be enabled.
   601     get_thread(temp);
   602     // interp_only is an int, on little endian it is sufficient to test the byte only
   603     // Is a cmpl faster (ce
   604     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
   605     jcc(Assembler::zero, run_compiled_code);
   606     jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
   607     bind(run_compiled_code);
   608   }
   610   jmp(Address(method, methodOopDesc::from_interpreted_offset()));
   612 }
   615 // The following two routines provide a hook so that an implementation
   616 // can schedule the dispatch in two parts.  amd64 does not do this.
   617 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
   618   // Nothing amd64 specific to be done here
   619 }
   621 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
   622   dispatch_next(state, step);
   623 }
   625 void InterpreterMacroAssembler::dispatch_base(TosState state,
   626                                               address* table,
   627                                               bool verifyoop) {
   628   verify_FPU(1, state);
   629   if (VerifyActivationFrameSize) {
   630     Label L;
   631     movq(rcx, rbp);
   632     subq(rcx, rsp);
   633     int min_frame_size =
   634       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
   635       wordSize;
   636     cmpq(rcx, min_frame_size);
   637     jcc(Assembler::greaterEqual, L);
   638     stop("broken stack frame");
   639     bind(L);
   640   }
   641   if (verifyoop) {
   642     verify_oop(rax, state);
   643   }
   644   lea(rscratch1, ExternalAddress((address)table));
   645   jmp(Address(rscratch1, rbx, Address::times_8));
   646 }
   648 void InterpreterMacroAssembler::dispatch_only(TosState state) {
   649   dispatch_base(state, Interpreter::dispatch_table(state));
   650 }
   652 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
   653   dispatch_base(state, Interpreter::normal_table(state));
   654 }
   656 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
   657   dispatch_base(state, Interpreter::normal_table(state), false);
   658 }
   661 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
   662   // load next bytecode (load before advancing r13 to prevent AGI)
   663   load_unsigned_byte(rbx, Address(r13, step));
   664   // advance r13
   665   incrementq(r13, step);
   666   dispatch_base(state, Interpreter::dispatch_table(state));
   667 }
   669 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
   670   // load current bytecode
   671   load_unsigned_byte(rbx, Address(r13, 0));
   672   dispatch_base(state, table);
   673 }
   675 // remove activation
   676 //
   677 // Unlock the receiver if this is a synchronized method.
   678 // Unlock any Java monitors from syncronized blocks.
   679 // Remove the activation from the stack.
   680 //
   681 // If there are locked Java monitors
   682 //    If throw_monitor_exception
   683 //       throws IllegalMonitorStateException
   684 //    Else if install_monitor_exception
   685 //       installs IllegalMonitorStateException
   686 //    Else
   687 //       no error processing
   688 void InterpreterMacroAssembler::remove_activation(
   689         TosState state,
   690         Register ret_addr,
   691         bool throw_monitor_exception,
   692         bool install_monitor_exception,
   693         bool notify_jvmdi) {
   694   // Note: Registers rdx xmm0 may be in use for the
   695   // result check if synchronized method
   696   Label unlocked, unlock, no_unlock;
   698   // get the value of _do_not_unlock_if_synchronized into rdx
   699   const Address do_not_unlock_if_synchronized(r15_thread,
   700     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
   701   movbool(rdx, do_not_unlock_if_synchronized);
   702   movbool(do_not_unlock_if_synchronized, false); // reset the flag
   704  // get method access flags
   705   movq(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
   706   movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
   707   testl(rcx, JVM_ACC_SYNCHRONIZED);
   708   jcc(Assembler::zero, unlocked);
   710   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
   711   // is set.
   712   testbool(rdx);
   713   jcc(Assembler::notZero, no_unlock);
   715   // unlock monitor
   716   push(state); // save result
   718   // BasicObjectLock will be first in list, since this is a
   719   // synchronized method. However, need to check that the object has
   720   // not been unlocked by an explicit monitorexit bytecode.
   721   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
   722                         wordSize - (int) sizeof(BasicObjectLock));
   723   // We use c_rarg1 so that if we go slow path it will be the correct
   724   // register for unlock_object to pass to VM directly
   725   leaq(c_rarg1, monitor); // address of first monitor
   727   movq(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
   728   testq(rax, rax);
   729   jcc(Assembler::notZero, unlock);
   731   pop(state);
   732   if (throw_monitor_exception) {
   733     // Entry already unlocked, need to throw exception
   734     call_VM(noreg, CAST_FROM_FN_PTR(address,
   735                    InterpreterRuntime::throw_illegal_monitor_state_exception));
   736     should_not_reach_here();
   737   } else {
   738     // Monitor already unlocked during a stack unroll. If requested,
   739     // install an illegal_monitor_state_exception.  Continue with
   740     // stack unrolling.
   741     if (install_monitor_exception) {
   742       call_VM(noreg, CAST_FROM_FN_PTR(address,
   743                      InterpreterRuntime::new_illegal_monitor_state_exception));
   744     }
   745     jmp(unlocked);
   746   }
   748   bind(unlock);
   749   unlock_object(c_rarg1);
   750   pop(state);
   752   // Check that for block-structured locking (i.e., that all locked
   753   // objects has been unlocked)
   754   bind(unlocked);
   756   // rax: Might contain return value
   758   // Check that all monitors are unlocked
   759   {
   760     Label loop, exception, entry, restart;
   761     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
   762     const Address monitor_block_top(
   763         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
   764     const Address monitor_block_bot(
   765         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
   767     bind(restart);
   768     // We use c_rarg1 so that if we go slow path it will be the correct
   769     // register for unlock_object to pass to VM directly
   770     movq(c_rarg1, monitor_block_top); // points to current entry, starting
   771                                   // with top-most entry
   772     leaq(rbx, monitor_block_bot); // points to word before bottom of
   773                                   // monitor block
   774     jmp(entry);
   776     // Entry already locked, need to throw exception
   777     bind(exception);
   779     if (throw_monitor_exception) {
   780       // Throw exception
   781       MacroAssembler::call_VM(noreg,
   782                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
   783                                    throw_illegal_monitor_state_exception));
   784       should_not_reach_here();
   785     } else {
   786       // Stack unrolling. Unlock object and install illegal_monitor_exception.
   787       // Unlock does not block, so don't have to worry about the frame.
   788       // We don't have to preserve c_rarg1 since we are going to throw an exception.
   790       push(state);
   791       unlock_object(c_rarg1);
   792       pop(state);
   794       if (install_monitor_exception) {
   795         call_VM(noreg, CAST_FROM_FN_PTR(address,
   796                                         InterpreterRuntime::
   797                                         new_illegal_monitor_state_exception));
   798       }
   800       jmp(restart);
   801     }
   803     bind(loop);
   804     // check if current entry is used
   805     cmpq(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int) NULL);
   806     jcc(Assembler::notEqual, exception);
   808     addq(c_rarg1, entry_size); // otherwise advance to next entry
   809     bind(entry);
   810     cmpq(c_rarg1, rbx); // check if bottom reached
   811     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
   812   }
   814   bind(no_unlock);
   816   // jvmti support
   817   if (notify_jvmdi) {
   818     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
   819   } else {
   820     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
   821   }
   823   // remove activation
   824   // get sender sp
   825   movq(rbx,
   826        Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
   827   leave();                           // remove frame anchor
   828   popq(ret_addr);                    // get return address
   829   movq(rsp, rbx);                    // set sp to sender sp
   830 }
   832 // Lock object
   833 //
   834 // Args:
   835 //      c_rarg1: BasicObjectLock to be used for locking
   836 //
   837 // Kills:
   838 //      rax
   839 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
   840 //      rscratch1, rscratch2 (scratch regs)
   841 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
   842   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
   844   if (UseHeavyMonitors) {
   845     call_VM(noreg,
   846             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   847             lock_reg);
   848   } else {
   849     Label done;
   851     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
   852     const Register obj_reg = c_rarg3; // Will contain the oop
   854     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
   855     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
   856     const int mark_offset = lock_offset +
   857                             BasicLock::displaced_header_offset_in_bytes();
   859     Label slow_case;
   861     // Load object pointer into obj_reg %c_rarg3
   862     movq(obj_reg, Address(lock_reg, obj_offset));
   864     if (UseBiasedLocking) {
   865       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
   866     }
   868     // Load immediate 1 into swap_reg %rax
   869     movl(swap_reg, 1);
   871     // Load (object->mark() | 1) into swap_reg %rax
   872     orq(swap_reg, Address(obj_reg, 0));
   874     // Save (object->mark() | 1) into BasicLock's displaced header
   875     movq(Address(lock_reg, mark_offset), swap_reg);
   877     assert(lock_offset == 0,
   878            "displached header must be first word in BasicObjectLock");
   880     if (os::is_MP()) lock();
   881     cmpxchgq(lock_reg, Address(obj_reg, 0));
   882     if (PrintBiasedLockingStatistics) {
   883       cond_inc32(Assembler::zero,
   884                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   885     }
   886     jcc(Assembler::zero, done);
   888     // Test if the oopMark is an obvious stack pointer, i.e.,
   889     //  1) (mark & 7) == 0, and
   890     //  2) rsp <= mark < mark + os::pagesize()
   891     //
   892     // These 3 tests can be done by evaluating the following
   893     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
   894     // assuming both stack pointer and pagesize have their
   895     // least significant 3 bits clear.
   896     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
   897     subq(swap_reg, rsp);
   898     andq(swap_reg, 7 - os::vm_page_size());
   900     // Save the test result, for recursive case, the result is zero
   901     movq(Address(lock_reg, mark_offset), swap_reg);
   903     if (PrintBiasedLockingStatistics) {
   904       cond_inc32(Assembler::zero,
   905                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   906     }
   907     jcc(Assembler::zero, done);
   909     bind(slow_case);
   911     // Call the runtime routine for slow case
   912     call_VM(noreg,
   913             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   914             lock_reg);
   916     bind(done);
   917   }
   918 }
   921 // Unlocks an object. Used in monitorexit bytecode and
   922 // remove_activation.  Throws an IllegalMonitorException if object is
   923 // not locked by current thread.
   924 //
   925 // Args:
   926 //      c_rarg1: BasicObjectLock for lock
   927 //
   928 // Kills:
   929 //      rax
   930 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
   931 //      rscratch1, rscratch2 (scratch regs)
   932 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
   933   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
   935   if (UseHeavyMonitors) {
   936     call_VM(noreg,
   937             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   938             lock_reg);
   939   } else {
   940     Label done;
   942     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
   943     const Register header_reg = c_rarg2;  // Will contain the old oopMark
   944     const Register obj_reg    = c_rarg3;  // Will contain the oop
   946     save_bcp(); // Save in case of exception
   948     // Convert from BasicObjectLock structure to object and BasicLock
   949     // structure Store the BasicLock address into %rax
   950     leaq(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
   952     // Load oop into obj_reg(%c_rarg3)
   953     movq(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
   955     // Free entry
   956     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
   958     if (UseBiasedLocking) {
   959       biased_locking_exit(obj_reg, header_reg, done);
   960     }
   962     // Load the old header from BasicLock structure
   963     movq(header_reg, Address(swap_reg,
   964                              BasicLock::displaced_header_offset_in_bytes()));
   966     // Test for recursion
   967     testq(header_reg, header_reg);
   969     // zero for recursive case
   970     jcc(Assembler::zero, done);
   972     // Atomic swap back the old header
   973     if (os::is_MP()) lock();
   974     cmpxchgq(header_reg, Address(obj_reg, 0));
   976     // zero for recursive case
   977     jcc(Assembler::zero, done);
   979     // Call the runtime routine for slow case.
   980     movq(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
   981          obj_reg); // restore obj
   982     call_VM(noreg,
   983             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   984             lock_reg);
   986     bind(done);
   988     restore_bcp();
   989   }
   990 }
   993 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
   994                                                          Label& zero_continue) {
   995   assert(ProfileInterpreter, "must be profiling interpreter");
   996   movq(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
   997   testq(mdp, mdp);
   998   jcc(Assembler::zero, zero_continue);
   999 }
  1002 // Set the method data pointer for the current bcp.
  1003 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
  1004   assert(ProfileInterpreter, "must be profiling interpreter");
  1005   Label zero_continue;
  1006   pushq(rax);
  1007   pushq(rbx);
  1009   get_method(rbx);
  1010   // Test MDO to avoid the call if it is NULL.
  1011   movq(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
  1012   testq(rax, rax);
  1013   jcc(Assembler::zero, zero_continue);
  1015   // rbx: method
  1016   // r13: bcp
  1017   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
  1018   // rax: mdi
  1020   movq(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
  1021   testq(rbx, rbx);
  1022   jcc(Assembler::zero, zero_continue);
  1023   addq(rbx, in_bytes(methodDataOopDesc::data_offset()));
  1024   addq(rbx, rax);
  1025   movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
  1027   bind(zero_continue);
  1028   popq(rbx);
  1029   popq(rax);
  1032 void InterpreterMacroAssembler::verify_method_data_pointer() {
  1033   assert(ProfileInterpreter, "must be profiling interpreter");
  1034 #ifdef ASSERT
  1035   Label verify_continue;
  1036   pushq(rax);
  1037   pushq(rbx);
  1038   pushq(c_rarg3);
  1039   pushq(c_rarg2);
  1040   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
  1041   get_method(rbx);
  1043   // If the mdp is valid, it will point to a DataLayout header which is
  1044   // consistent with the bcp.  The converse is highly probable also.
  1045   load_unsigned_word(c_rarg2,
  1046                      Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
  1047   addq(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
  1048   leaq(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
  1049   cmpq(c_rarg2, r13);
  1050   jcc(Assembler::equal, verify_continue);
  1051   // rbx: method
  1052   // r13: bcp
  1053   // c_rarg3: mdp
  1054   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
  1055                rbx, r13, c_rarg3);
  1056   bind(verify_continue);
  1057   popq(c_rarg2);
  1058   popq(c_rarg3);
  1059   popq(rbx);
  1060   popq(rax);
  1061 #endif // ASSERT
  1065 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
  1066                                                 int constant,
  1067                                                 Register value) {
  1068   assert(ProfileInterpreter, "must be profiling interpreter");
  1069   Address data(mdp_in, constant);
  1070   movq(data, value);
  1074 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
  1075                                                       int constant,
  1076                                                       bool decrement) {
  1077   // Counter address
  1078   Address data(mdp_in, constant);
  1080   increment_mdp_data_at(data, decrement);
  1083 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
  1084                                                       bool decrement) {
  1085   assert(ProfileInterpreter, "must be profiling interpreter");
  1087   if (decrement) {
  1088     // Decrement the register.  Set condition codes.
  1089     addq(data, -DataLayout::counter_increment);
  1090     // If the decrement causes the counter to overflow, stay negative
  1091     Label L;
  1092     jcc(Assembler::negative, L);
  1093     addq(data, DataLayout::counter_increment);
  1094     bind(L);
  1095   } else {
  1096     assert(DataLayout::counter_increment == 1,
  1097            "flow-free idiom only works with 1");
  1098     // Increment the register.  Set carry flag.
  1099     addq(data, DataLayout::counter_increment);
  1100     // If the increment causes the counter to overflow, pull back by 1.
  1101     sbbq(data, 0);
  1106 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
  1107                                                       Register reg,
  1108                                                       int constant,
  1109                                                       bool decrement) {
  1110   Address data(mdp_in, reg, Address::times_1, constant);
  1112   increment_mdp_data_at(data, decrement);
  1115 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
  1116                                                 int flag_byte_constant) {
  1117   assert(ProfileInterpreter, "must be profiling interpreter");
  1118   int header_offset = in_bytes(DataLayout::header_offset());
  1119   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
  1120   // Set the flag
  1121   orl(Address(mdp_in, header_offset), header_bits);
  1126 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
  1127                                                  int offset,
  1128                                                  Register value,
  1129                                                  Register test_value_out,
  1130                                                  Label& not_equal_continue) {
  1131   assert(ProfileInterpreter, "must be profiling interpreter");
  1132   if (test_value_out == noreg) {
  1133     cmpq(value, Address(mdp_in, offset));
  1134   } else {
  1135     // Put the test value into a register, so caller can use it:
  1136     movq(test_value_out, Address(mdp_in, offset));
  1137     cmpq(test_value_out, value);
  1139   jcc(Assembler::notEqual, not_equal_continue);
  1143 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
  1144                                                      int offset_of_disp) {
  1145   assert(ProfileInterpreter, "must be profiling interpreter");
  1146   Address disp_address(mdp_in, offset_of_disp);
  1147   addq(mdp_in, disp_address);
  1148   movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1152 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
  1153                                                      Register reg,
  1154                                                      int offset_of_disp) {
  1155   assert(ProfileInterpreter, "must be profiling interpreter");
  1156   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
  1157   addq(mdp_in, disp_address);
  1158   movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1162 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
  1163                                                        int constant) {
  1164   assert(ProfileInterpreter, "must be profiling interpreter");
  1165   addq(mdp_in, constant);
  1166   movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1170 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
  1171   assert(ProfileInterpreter, "must be profiling interpreter");
  1172   pushq(return_bci); // save/restore across call_VM
  1173   call_VM(noreg,
  1174           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
  1175           return_bci);
  1176   popq(return_bci);
  1180 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
  1181                                                      Register bumped_count) {
  1182   if (ProfileInterpreter) {
  1183     Label profile_continue;
  1185     // If no method data exists, go to profile_continue.
  1186     // Otherwise, assign to mdp
  1187     test_method_data_pointer(mdp, profile_continue);
  1189     // We are taking a branch.  Increment the taken count.
  1190     // We inline increment_mdp_data_at to return bumped_count in a register
  1191     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
  1192     Address data(mdp, in_bytes(JumpData::taken_offset()));
  1193     movq(bumped_count, data);
  1194     assert(DataLayout::counter_increment == 1,
  1195             "flow-free idiom only works with 1");
  1196     addq(bumped_count, DataLayout::counter_increment);
  1197     sbbq(bumped_count, 0);
  1198     movq(data, bumped_count); // Store back out
  1200     // The method data pointer needs to be updated to reflect the new target.
  1201     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
  1202     bind(profile_continue);
  1207 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
  1208   if (ProfileInterpreter) {
  1209     Label profile_continue;
  1211     // If no method data exists, go to profile_continue.
  1212     test_method_data_pointer(mdp, profile_continue);
  1214     // We are taking a branch.  Increment the not taken count.
  1215     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
  1217     // The method data pointer needs to be updated to correspond to
  1218     // the next bytecode
  1219     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
  1220     bind(profile_continue);
  1225 void InterpreterMacroAssembler::profile_call(Register mdp) {
  1226   if (ProfileInterpreter) {
  1227     Label profile_continue;
  1229     // If no method data exists, go to profile_continue.
  1230     test_method_data_pointer(mdp, profile_continue);
  1232     // We are making a call.  Increment the count.
  1233     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1235     // The method data pointer needs to be updated to reflect the new target.
  1236     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
  1237     bind(profile_continue);
  1242 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
  1243   if (ProfileInterpreter) {
  1244     Label profile_continue;
  1246     // If no method data exists, go to profile_continue.
  1247     test_method_data_pointer(mdp, profile_continue);
  1249     // We are making a call.  Increment the count.
  1250     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1252     // The method data pointer needs to be updated to reflect the new target.
  1253     update_mdp_by_constant(mdp,
  1254                            in_bytes(VirtualCallData::
  1255                                     virtual_call_data_size()));
  1256     bind(profile_continue);
  1261 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
  1262                                                      Register mdp,
  1263                                                      Register reg2) {
  1264   if (ProfileInterpreter) {
  1265     Label profile_continue;
  1267     // If no method data exists, go to profile_continue.
  1268     test_method_data_pointer(mdp, profile_continue);
  1270     // We are making a call.  Increment the count.
  1271     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1273     // Record the receiver type.
  1274     record_klass_in_profile(receiver, mdp, reg2);
  1276     // The method data pointer needs to be updated to reflect the new target.
  1277     update_mdp_by_constant(mdp,
  1278                            in_bytes(VirtualCallData::
  1279                                     virtual_call_data_size()));
  1280     bind(profile_continue);
  1284 // This routine creates a state machine for updating the multi-row
  1285 // type profile at a virtual call site (or other type-sensitive bytecode).
  1286 // The machine visits each row (of receiver/count) until the receiver type
  1287 // is found, or until it runs out of rows.  At the same time, it remembers
  1288 // the location of the first empty row.  (An empty row records null for its
  1289 // receiver, and can be allocated for a newly-observed receiver type.)
  1290 // Because there are two degrees of freedom in the state, a simple linear
  1291 // search will not work; it must be a decision tree.  Hence this helper
  1292 // function is recursive, to generate the required tree structured code.
  1293 // It's the interpreter, so we are trading off code space for speed.
  1294 // See below for example code.
  1295 void InterpreterMacroAssembler::record_klass_in_profile_helper(
  1296                                         Register receiver, Register mdp,
  1297                                         Register reg2,
  1298                                         int start_row, Label& done) {
  1299   int last_row = VirtualCallData::row_limit() - 1;
  1300   assert(start_row <= last_row, "must be work left to do");
  1301   // Test this row for both the receiver and for null.
  1302   // Take any of three different outcomes:
  1303   //   1. found receiver => increment count and goto done
  1304   //   2. found null => keep looking for case 1, maybe allocate this cell
  1305   //   3. found something else => keep looking for cases 1 and 2
  1306   // Case 3 is handled by a recursive call.
  1307   for (int row = start_row; row <= last_row; row++) {
  1308     Label next_test;
  1309     bool test_for_null_also = (row == start_row);
  1311     // See if the receiver is receiver[n].
  1312     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
  1313     test_mdp_data_at(mdp, recvr_offset, receiver,
  1314                      (test_for_null_also ? reg2 : noreg),
  1315                      next_test);
  1316     // (Reg2 now contains the receiver from the CallData.)
  1318     // The receiver is receiver[n].  Increment count[n].
  1319     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
  1320     increment_mdp_data_at(mdp, count_offset);
  1321     jmp(done);
  1322     bind(next_test);
  1324     if (test_for_null_also) {
  1325       // Failed the equality check on receiver[n]...  Test for null.
  1326       testq(reg2, reg2);
  1327       if (start_row == last_row) {
  1328         // The only thing left to do is handle the null case.
  1329         jcc(Assembler::notZero, done);
  1330         break;
  1332       // Since null is rare, make it be the branch-taken case.
  1333       Label found_null;
  1334       jcc(Assembler::zero, found_null);
  1336       // Put all the "Case 3" tests here.
  1337       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done);
  1339       // Found a null.  Keep searching for a matching receiver,
  1340       // but remember that this is an empty (unused) slot.
  1341       bind(found_null);
  1345   // In the fall-through case, we found no matching receiver, but we
  1346   // observed the receiver[start_row] is NULL.
  1348   // Fill in the receiver field and increment the count.
  1349   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
  1350   set_mdp_data_at(mdp, recvr_offset, receiver);
  1351   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
  1352   movl(reg2, DataLayout::counter_increment);
  1353   set_mdp_data_at(mdp, count_offset, reg2);
  1354   jmp(done);
  1357 // Example state machine code for three profile rows:
  1358 //   // main copy of decision tree, rooted at row[1]
  1359 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
  1360 //   if (row[0].rec != NULL) {
  1361 //     // inner copy of decision tree, rooted at row[1]
  1362 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1363 //     if (row[1].rec != NULL) {
  1364 //       // degenerate decision tree, rooted at row[2]
  1365 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1366 //       if (row[2].rec != NULL) { goto done; } // overflow
  1367 //       row[2].init(rec); goto done;
  1368 //     } else {
  1369 //       // remember row[1] is empty
  1370 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1371 //       row[1].init(rec); goto done;
  1372 //     }
  1373 //   } else {
  1374 //     // remember row[0] is empty
  1375 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1376 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
  1377 //     row[0].init(rec); goto done;
  1378 //   }
  1380 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
  1381                                                         Register mdp,
  1382                                                         Register reg2) {
  1383   assert(ProfileInterpreter, "must be profiling");
  1384   Label done;
  1386   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
  1388   bind (done);
  1391 void InterpreterMacroAssembler::profile_ret(Register return_bci,
  1392                                             Register mdp) {
  1393   if (ProfileInterpreter) {
  1394     Label profile_continue;
  1395     uint row;
  1397     // If no method data exists, go to profile_continue.
  1398     test_method_data_pointer(mdp, profile_continue);
  1400     // Update the total ret count.
  1401     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1403     for (row = 0; row < RetData::row_limit(); row++) {
  1404       Label next_test;
  1406       // See if return_bci is equal to bci[n]:
  1407       test_mdp_data_at(mdp,
  1408                        in_bytes(RetData::bci_offset(row)),
  1409                        return_bci, noreg,
  1410                        next_test);
  1412       // return_bci is equal to bci[n].  Increment the count.
  1413       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
  1415       // The method data pointer needs to be updated to reflect the new target.
  1416       update_mdp_by_offset(mdp,
  1417                            in_bytes(RetData::bci_displacement_offset(row)));
  1418       jmp(profile_continue);
  1419       bind(next_test);
  1422     update_mdp_for_ret(return_bci);
  1424     bind(profile_continue);
  1429 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
  1430   if (ProfileInterpreter) {
  1431     Label profile_continue;
  1433     // If no method data exists, go to profile_continue.
  1434     test_method_data_pointer(mdp, profile_continue);
  1436     // The method data pointer needs to be updated.
  1437     int mdp_delta = in_bytes(BitData::bit_data_size());
  1438     if (TypeProfileCasts) {
  1439       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1441     update_mdp_by_constant(mdp, mdp_delta);
  1443     bind(profile_continue);
  1448 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
  1449   if (ProfileInterpreter && TypeProfileCasts) {
  1450     Label profile_continue;
  1452     // If no method data exists, go to profile_continue.
  1453     test_method_data_pointer(mdp, profile_continue);
  1455     int count_offset = in_bytes(CounterData::count_offset());
  1456     // Back up the address, since we have already bumped the mdp.
  1457     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
  1459     // *Decrement* the counter.  We expect to see zero or small negatives.
  1460     increment_mdp_data_at(mdp, count_offset, true);
  1462     bind (profile_continue);
  1467 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
  1468   if (ProfileInterpreter) {
  1469     Label profile_continue;
  1471     // If no method data exists, go to profile_continue.
  1472     test_method_data_pointer(mdp, profile_continue);
  1474     // The method data pointer needs to be updated.
  1475     int mdp_delta = in_bytes(BitData::bit_data_size());
  1476     if (TypeProfileCasts) {
  1477       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1479       // Record the object type.
  1480       record_klass_in_profile(klass, mdp, reg2);
  1482     update_mdp_by_constant(mdp, mdp_delta);
  1484     bind(profile_continue);
  1489 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
  1490   if (ProfileInterpreter) {
  1491     Label profile_continue;
  1493     // If no method data exists, go to profile_continue.
  1494     test_method_data_pointer(mdp, profile_continue);
  1496     // Update the default case count
  1497     increment_mdp_data_at(mdp,
  1498                           in_bytes(MultiBranchData::default_count_offset()));
  1500     // The method data pointer needs to be updated.
  1501     update_mdp_by_offset(mdp,
  1502                          in_bytes(MultiBranchData::
  1503                                   default_displacement_offset()));
  1505     bind(profile_continue);
  1510 void InterpreterMacroAssembler::profile_switch_case(Register index,
  1511                                                     Register mdp,
  1512                                                     Register reg2) {
  1513   if (ProfileInterpreter) {
  1514     Label profile_continue;
  1516     // If no method data exists, go to profile_continue.
  1517     test_method_data_pointer(mdp, profile_continue);
  1519     // Build the base (index * per_case_size_in_bytes()) +
  1520     // case_array_offset_in_bytes()
  1521     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
  1522     imulq(index, reg2); // XXX l ?
  1523     addq(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
  1525     // Update the case count
  1526     increment_mdp_data_at(mdp,
  1527                           index,
  1528                           in_bytes(MultiBranchData::relative_count_offset()));
  1530     // The method data pointer needs to be updated.
  1531     update_mdp_by_offset(mdp,
  1532                          index,
  1533                          in_bytes(MultiBranchData::
  1534                                   relative_displacement_offset()));
  1536     bind(profile_continue);
  1541 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
  1542   if (state == atos) {
  1543     MacroAssembler::verify_oop(reg);
  1547 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
  1551 void InterpreterMacroAssembler::notify_method_entry() {
  1552   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1553   // track stack depth.  If it is possible to enter interp_only_mode we add
  1554   // the code to check if the event should be sent.
  1555   if (JvmtiExport::can_post_interpreter_events()) {
  1556     Label L;
  1557     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1558     testl(rdx, rdx);
  1559     jcc(Assembler::zero, L);
  1560     call_VM(noreg, CAST_FROM_FN_PTR(address,
  1561                                     InterpreterRuntime::post_method_entry));
  1562     bind(L);
  1566     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1567     get_method(c_rarg1);
  1568     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
  1569                  r15_thread, c_rarg1);
  1574 void InterpreterMacroAssembler::notify_method_exit(
  1575     TosState state, NotifyMethodExitMode mode) {
  1576   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1577   // track stack depth.  If it is possible to enter interp_only_mode we add
  1578   // the code to check if the event should be sent.
  1579   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
  1580     Label L;
  1581     // Note: frame::interpreter_frame_result has a dependency on how the
  1582     // method result is saved across the call to post_method_exit. If this
  1583     // is changed then the interpreter_frame_result implementation will
  1584     // need to be updated too.
  1585     push(state);
  1586     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1587     testl(rdx, rdx);
  1588     jcc(Assembler::zero, L);
  1589     call_VM(noreg,
  1590             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
  1591     bind(L);
  1592     pop(state);
  1596     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1597     push(state);
  1598     get_method(c_rarg1);
  1599     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
  1600                  r15_thread, c_rarg1);
  1601     pop(state);

mercurial