src/cpu/sparc/vm/templateTable_sparc.cpp

Sat, 30 Oct 2010 13:08:23 -0700

author
jrose
date
Sat, 30 Oct 2010 13:08:23 -0700
changeset 2268
3b2dea75431e
parent 2201
d55217dc206f
child 2314
f95d63e2154a
permissions
-rw-r--r--

6984311: JSR 292 needs optional bootstrap method parameters
Summary: Allow CONSTANT_InvokeDynamic nodes to have any number of extra operands.
Reviewed-by: twisti

     1 /*
     2  * Copyright (c) 1997, 2010, 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 "incls/_precompiled.incl"
    26 #include "incls/_templateTable_sparc.cpp.incl"
    28 #ifndef CC_INTERP
    29 #define __ _masm->
    31 // Misc helpers
    33 // Do an oop store like *(base + index + offset) = val
    34 // index can be noreg,
    35 static void do_oop_store(InterpreterMacroAssembler* _masm,
    36                          Register base,
    37                          Register index,
    38                          int offset,
    39                          Register val,
    40                          Register tmp,
    41                          BarrierSet::Name barrier,
    42                          bool precise) {
    43   assert(tmp != val && tmp != base && tmp != index, "register collision");
    44   assert(index == noreg || offset == 0, "only one offset");
    45   switch (barrier) {
    46 #ifndef SERIALGC
    47     case BarrierSet::G1SATBCT:
    48     case BarrierSet::G1SATBCTLogging:
    49       {
    50         __ g1_write_barrier_pre( base, index, offset, tmp, /*preserve_o_regs*/true);
    51         if (index == noreg ) {
    52           assert(Assembler::is_simm13(offset), "fix this code");
    53           __ store_heap_oop(val, base, offset);
    54         } else {
    55           __ store_heap_oop(val, base, index);
    56         }
    58         // No need for post barrier if storing NULL
    59         if (val != G0) {
    60           if (precise) {
    61             if (index == noreg) {
    62               __ add(base, offset, base);
    63             } else {
    64               __ add(base, index, base);
    65             }
    66           }
    67           __ g1_write_barrier_post(base, val, tmp);
    68         }
    69       }
    70       break;
    71 #endif // SERIALGC
    72     case BarrierSet::CardTableModRef:
    73     case BarrierSet::CardTableExtension:
    74       {
    75         if (index == noreg ) {
    76           assert(Assembler::is_simm13(offset), "fix this code");
    77           __ store_heap_oop(val, base, offset);
    78         } else {
    79           __ store_heap_oop(val, base, index);
    80         }
    81         // No need for post barrier if storing NULL
    82         if (val != G0) {
    83           if (precise) {
    84             if (index == noreg) {
    85               __ add(base, offset, base);
    86             } else {
    87               __ add(base, index, base);
    88             }
    89           }
    90           __ card_write_barrier_post(base, val, tmp);
    91         }
    92       }
    93       break;
    94     case BarrierSet::ModRef:
    95     case BarrierSet::Other:
    96       ShouldNotReachHere();
    97       break;
    98     default      :
    99       ShouldNotReachHere();
   101   }
   102 }
   105 //----------------------------------------------------------------------------------------------------
   106 // Platform-dependent initialization
   108 void TemplateTable::pd_initialize() {
   109   // (none)
   110 }
   113 //----------------------------------------------------------------------------------------------------
   114 // Condition conversion
   115 Assembler::Condition ccNot(TemplateTable::Condition cc) {
   116   switch (cc) {
   117     case TemplateTable::equal        : return Assembler::notEqual;
   118     case TemplateTable::not_equal    : return Assembler::equal;
   119     case TemplateTable::less         : return Assembler::greaterEqual;
   120     case TemplateTable::less_equal   : return Assembler::greater;
   121     case TemplateTable::greater      : return Assembler::lessEqual;
   122     case TemplateTable::greater_equal: return Assembler::less;
   123   }
   124   ShouldNotReachHere();
   125   return Assembler::zero;
   126 }
   128 //----------------------------------------------------------------------------------------------------
   129 // Miscelaneous helper routines
   132 Address TemplateTable::at_bcp(int offset) {
   133   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
   134   return Address(Lbcp, offset);
   135 }
   138 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register Rbyte_code,
   139                                    Register Rscratch,
   140                                    bool load_bc_into_scratch /*=true*/) {
   141   // With sharing on, may need to test methodOop flag.
   142   if (!RewriteBytecodes) return;
   143   if (load_bc_into_scratch) __ set(bc, Rbyte_code);
   144   Label patch_done;
   145   if (JvmtiExport::can_post_breakpoint()) {
   146     Label fast_patch;
   147     __ ldub(at_bcp(0), Rscratch);
   148     __ cmp(Rscratch, Bytecodes::_breakpoint);
   149     __ br(Assembler::notEqual, false, Assembler::pt, fast_patch);
   150     __ delayed()->nop();  // don't bother to hoist the stb here
   151     // perform the quickening, slowly, in the bowels of the breakpoint table
   152     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, Rbyte_code);
   153     __ ba(false, patch_done);
   154     __ delayed()->nop();
   155     __ bind(fast_patch);
   156   }
   157 #ifdef ASSERT
   158   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
   159   Label okay;
   160   __ ldub(at_bcp(0), Rscratch);
   161   __ cmp(Rscratch, orig_bytecode);
   162   __ br(Assembler::equal, false, Assembler::pt, okay);
   163   __ delayed() ->cmp(Rscratch, Rbyte_code);
   164   __ br(Assembler::equal, false, Assembler::pt, okay);
   165   __ delayed()->nop();
   166   __ stop("Rewriting wrong bytecode location");
   167   __ bind(okay);
   168 #endif
   169   __ stb(Rbyte_code, at_bcp(0));
   170   __ bind(patch_done);
   171 }
   173 //----------------------------------------------------------------------------------------------------
   174 // Individual instructions
   176 void TemplateTable::nop() {
   177   transition(vtos, vtos);
   178   // nothing to do
   179 }
   181 void TemplateTable::shouldnotreachhere() {
   182   transition(vtos, vtos);
   183   __ stop("shouldnotreachhere bytecode");
   184 }
   186 void TemplateTable::aconst_null() {
   187   transition(vtos, atos);
   188   __ clr(Otos_i);
   189 }
   192 void TemplateTable::iconst(int value) {
   193   transition(vtos, itos);
   194   __ set(value, Otos_i);
   195 }
   198 void TemplateTable::lconst(int value) {
   199   transition(vtos, ltos);
   200   assert(value >= 0, "check this code");
   201 #ifdef _LP64
   202   __ set(value, Otos_l);
   203 #else
   204   __ set(value, Otos_l2);
   205   __ clr( Otos_l1);
   206 #endif
   207 }
   210 void TemplateTable::fconst(int value) {
   211   transition(vtos, ftos);
   212   static float zero = 0.0, one = 1.0, two = 2.0;
   213   float* p;
   214   switch( value ) {
   215    default: ShouldNotReachHere();
   216    case 0:  p = &zero;  break;
   217    case 1:  p = &one;   break;
   218    case 2:  p = &two;   break;
   219   }
   220   AddressLiteral a(p);
   221   __ sethi(a, G3_scratch);
   222   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
   223 }
   226 void TemplateTable::dconst(int value) {
   227   transition(vtos, dtos);
   228   static double zero = 0.0, one = 1.0;
   229   double* p;
   230   switch( value ) {
   231    default: ShouldNotReachHere();
   232    case 0:  p = &zero;  break;
   233    case 1:  p = &one;   break;
   234   }
   235   AddressLiteral a(p);
   236   __ sethi(a, G3_scratch);
   237   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
   238 }
   241 // %%%%% Should factore most snippet templates across platforms
   243 void TemplateTable::bipush() {
   244   transition(vtos, itos);
   245   __ ldsb( at_bcp(1), Otos_i );
   246 }
   248 void TemplateTable::sipush() {
   249   transition(vtos, itos);
   250   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
   251 }
   253 void TemplateTable::ldc(bool wide) {
   254   transition(vtos, vtos);
   255   Label call_ldc, notInt, notString, notClass, exit;
   257   if (wide) {
   258     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   259   } else {
   260     __ ldub(Lbcp, 1, O1);
   261   }
   262   __ get_cpool_and_tags(O0, O2);
   264   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
   265   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
   267   // get type from tags
   268   __ add(O2, tags_offset, O2);
   269   __ ldub(O2, O1, O2);
   270   __ cmp(O2, JVM_CONSTANT_UnresolvedString);    // unresolved string? If so, must resolve
   271   __ brx(Assembler::equal, true, Assembler::pt, call_ldc);
   272   __ delayed()->nop();
   274   __ cmp(O2, JVM_CONSTANT_UnresolvedClass);     // unresolved class? If so, must resolve
   275   __ brx(Assembler::equal, true, Assembler::pt, call_ldc);
   276   __ delayed()->nop();
   278   __ cmp(O2, JVM_CONSTANT_UnresolvedClassInError);     // unresolved class in error state
   279   __ brx(Assembler::equal, true, Assembler::pn, call_ldc);
   280   __ delayed()->nop();
   282   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
   283   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
   284   __ delayed()->add(O0, base_offset, O0);
   286   __ bind(call_ldc);
   287   __ set(wide, O1);
   288   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
   289   __ push(atos);
   290   __ ba(false, exit);
   291   __ delayed()->nop();
   293   __ bind(notClass);
   294  // __ add(O0, base_offset, O0);
   295   __ sll(O1, LogBytesPerWord, O1);
   296   __ cmp(O2, JVM_CONSTANT_Integer);
   297   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
   298   __ delayed()->cmp(O2, JVM_CONSTANT_String);
   299   __ ld(O0, O1, Otos_i);
   300   __ push(itos);
   301   __ ba(false, exit);
   302   __ delayed()->nop();
   304   __ bind(notInt);
   305  // __ cmp(O2, JVM_CONSTANT_String);
   306   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
   307   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   308   __ ld_ptr(O0, O1, Otos_i);
   309   __ verify_oop(Otos_i);
   310   __ push(atos);
   311   __ ba(false, exit);
   312   __ delayed()->nop();
   314   __ bind(notString);
   315  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   316   __ push(ftos);
   318   __ bind(exit);
   319 }
   321 // Fast path for caching oop constants.
   322 // %%% We should use this to handle Class and String constants also.
   323 // %%% It will simplify the ldc/primitive path considerably.
   324 void TemplateTable::fast_aldc(bool wide) {
   325   transition(vtos, atos);
   327   if (!EnableMethodHandles) {
   328     // We should not encounter this bytecode if !EnableMethodHandles.
   329     // The verifier will stop it.  However, if we get past the verifier,
   330     // this will stop the thread in a reasonable way, without crashing the JVM.
   331     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
   332                      InterpreterRuntime::throw_IncompatibleClassChangeError));
   333     // the call_VM checks for exception, so we should never return here.
   334     __ should_not_reach_here();
   335     return;
   336   }
   338   Register Rcache = G3_scratch;
   339   Register Rscratch = G4_scratch;
   341   resolve_cache_and_index(f1_oop, Otos_i, Rcache, Rscratch, wide ? sizeof(u2) : sizeof(u1));
   343   __ verify_oop(Otos_i);
   345   Label L_done;
   346   const Register Rcon_klass = G3_scratch;  // same as Rcache
   347   const Register Rarray_klass = G4_scratch;  // same as Rscratch
   348   __ load_klass(Otos_i, Rcon_klass);
   349   AddressLiteral array_klass_addr((address)Universe::systemObjArrayKlassObj_addr());
   350   __ load_contents(array_klass_addr, Rarray_klass);
   351   __ cmp(Rarray_klass, Rcon_klass);
   352   __ brx(Assembler::notEqual, false, Assembler::pt, L_done);
   353   __ delayed()->nop();
   354   __ ld(Address(Otos_i, arrayOopDesc::length_offset_in_bytes()), Rcon_klass);
   355   __ tst(Rcon_klass);
   356   __ brx(Assembler::zero, true, Assembler::pt, L_done);
   357   __ delayed()->clr(Otos_i);    // executed only if branch is taken
   359   // Load the exception from the system-array which wraps it:
   360   __ load_heap_oop(Otos_i, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   361   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
   363   __ bind(L_done);
   364 }
   366 void TemplateTable::ldc2_w() {
   367   transition(vtos, vtos);
   368   Label retry, resolved, Long, exit;
   370   __ bind(retry);
   371   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   372   __ get_cpool_and_tags(O0, O2);
   374   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
   375   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
   376   // get type from tags
   377   __ add(O2, tags_offset, O2);
   378   __ ldub(O2, O1, O2);
   380   __ sll(O1, LogBytesPerWord, O1);
   381   __ add(O0, O1, G3_scratch);
   383   __ cmp(O2, JVM_CONSTANT_Double);
   384   __ brx(Assembler::notEqual, false, Assembler::pt, Long);
   385   __ delayed()->nop();
   386   // A double can be placed at word-aligned locations in the constant pool.
   387   // Check out Conversions.java for an example.
   388   // Also constantPoolOopDesc::header_size() is 20, which makes it very difficult
   389   // to double-align double on the constant pool.  SG, 11/7/97
   390 #ifdef _LP64
   391   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
   392 #else
   393   FloatRegister f = Ftos_d;
   394   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
   395   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
   396          f->successor());
   397 #endif
   398   __ push(dtos);
   399   __ ba(false, exit);
   400   __ delayed()->nop();
   402   __ bind(Long);
   403 #ifdef _LP64
   404   __ ldx(G3_scratch, base_offset, Otos_l);
   405 #else
   406   __ ld(G3_scratch, base_offset, Otos_l);
   407   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
   408 #endif
   409   __ push(ltos);
   411   __ bind(exit);
   412 }
   415 void TemplateTable::locals_index(Register reg, int offset) {
   416   __ ldub( at_bcp(offset), reg );
   417 }
   420 void TemplateTable::locals_index_wide(Register reg) {
   421   // offset is 2, not 1, because Lbcp points to wide prefix code
   422   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
   423 }
   425 void TemplateTable::iload() {
   426   transition(vtos, itos);
   427   // Rewrite iload,iload  pair into fast_iload2
   428   //         iload,caload pair into fast_icaload
   429   if (RewriteFrequentPairs) {
   430     Label rewrite, done;
   432     // get next byte
   433     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
   435     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
   436     // last two iloads in a pair.  Comparing against fast_iload means that
   437     // the next bytecode is neither an iload or a caload, and therefore
   438     // an iload pair.
   439     __ cmp(G3_scratch, (int)Bytecodes::_iload);
   440     __ br(Assembler::equal, false, Assembler::pn, done);
   441     __ delayed()->nop();
   443     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
   444     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   445     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
   447     __ cmp(G3_scratch, (int)Bytecodes::_caload);
   448     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   449     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
   451     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
   452     // rewrite
   453     // G4_scratch: fast bytecode
   454     __ bind(rewrite);
   455     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
   456     __ bind(done);
   457   }
   459   // Get the local value into tos
   460   locals_index(G3_scratch);
   461   __ access_local_int( G3_scratch, Otos_i );
   462 }
   464 void TemplateTable::fast_iload2() {
   465   transition(vtos, itos);
   466   locals_index(G3_scratch);
   467   __ access_local_int( G3_scratch, Otos_i );
   468   __ push_i();
   469   locals_index(G3_scratch, 3);  // get next bytecode's local index.
   470   __ access_local_int( G3_scratch, Otos_i );
   471 }
   473 void TemplateTable::fast_iload() {
   474   transition(vtos, itos);
   475   locals_index(G3_scratch);
   476   __ access_local_int( G3_scratch, Otos_i );
   477 }
   479 void TemplateTable::lload() {
   480   transition(vtos, ltos);
   481   locals_index(G3_scratch);
   482   __ access_local_long( G3_scratch, Otos_l );
   483 }
   486 void TemplateTable::fload() {
   487   transition(vtos, ftos);
   488   locals_index(G3_scratch);
   489   __ access_local_float( G3_scratch, Ftos_f );
   490 }
   493 void TemplateTable::dload() {
   494   transition(vtos, dtos);
   495   locals_index(G3_scratch);
   496   __ access_local_double( G3_scratch, Ftos_d );
   497 }
   500 void TemplateTable::aload() {
   501   transition(vtos, atos);
   502   locals_index(G3_scratch);
   503   __ access_local_ptr( G3_scratch, Otos_i);
   504 }
   507 void TemplateTable::wide_iload() {
   508   transition(vtos, itos);
   509   locals_index_wide(G3_scratch);
   510   __ access_local_int( G3_scratch, Otos_i );
   511 }
   514 void TemplateTable::wide_lload() {
   515   transition(vtos, ltos);
   516   locals_index_wide(G3_scratch);
   517   __ access_local_long( G3_scratch, Otos_l );
   518 }
   521 void TemplateTable::wide_fload() {
   522   transition(vtos, ftos);
   523   locals_index_wide(G3_scratch);
   524   __ access_local_float( G3_scratch, Ftos_f );
   525 }
   528 void TemplateTable::wide_dload() {
   529   transition(vtos, dtos);
   530   locals_index_wide(G3_scratch);
   531   __ access_local_double( G3_scratch, Ftos_d );
   532 }
   535 void TemplateTable::wide_aload() {
   536   transition(vtos, atos);
   537   locals_index_wide(G3_scratch);
   538   __ access_local_ptr( G3_scratch, Otos_i );
   539   __ verify_oop(Otos_i);
   540 }
   543 void TemplateTable::iaload() {
   544   transition(itos, itos);
   545   // Otos_i: index
   546   // tos: array
   547   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   548   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
   549 }
   552 void TemplateTable::laload() {
   553   transition(itos, ltos);
   554   // Otos_i: index
   555   // O2: array
   556   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   557   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
   558 }
   561 void TemplateTable::faload() {
   562   transition(itos, ftos);
   563   // Otos_i: index
   564   // O2: array
   565   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   566   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
   567 }
   570 void TemplateTable::daload() {
   571   transition(itos, dtos);
   572   // Otos_i: index
   573   // O2: array
   574   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   575   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
   576 }
   579 void TemplateTable::aaload() {
   580   transition(itos, atos);
   581   // Otos_i: index
   582   // tos: array
   583   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
   584   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   585   __ verify_oop(Otos_i);
   586 }
   589 void TemplateTable::baload() {
   590   transition(itos, itos);
   591   // Otos_i: index
   592   // tos: array
   593   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
   594   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
   595 }
   598 void TemplateTable::caload() {
   599   transition(itos, itos);
   600   // Otos_i: index
   601   // tos: array
   602   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   603   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   604 }
   606 void TemplateTable::fast_icaload() {
   607   transition(vtos, itos);
   608   // Otos_i: index
   609   // tos: array
   610   locals_index(G3_scratch);
   611   __ access_local_int( G3_scratch, Otos_i );
   612   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   613   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   614 }
   617 void TemplateTable::saload() {
   618   transition(itos, itos);
   619   // Otos_i: index
   620   // tos: array
   621   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   622   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
   623 }
   626 void TemplateTable::iload(int n) {
   627   transition(vtos, itos);
   628   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   629 }
   632 void TemplateTable::lload(int n) {
   633   transition(vtos, ltos);
   634   assert(n+1 < Argument::n_register_parameters, "would need more code");
   635   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
   636 }
   639 void TemplateTable::fload(int n) {
   640   transition(vtos, ftos);
   641   assert(n < Argument::n_register_parameters, "would need more code");
   642   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
   643 }
   646 void TemplateTable::dload(int n) {
   647   transition(vtos, dtos);
   648   FloatRegister dst = Ftos_d;
   649   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
   650 }
   653 void TemplateTable::aload(int n) {
   654   transition(vtos, atos);
   655   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   656 }
   659 void TemplateTable::aload_0() {
   660   transition(vtos, atos);
   662   // According to bytecode histograms, the pairs:
   663   //
   664   // _aload_0, _fast_igetfield (itos)
   665   // _aload_0, _fast_agetfield (atos)
   666   // _aload_0, _fast_fgetfield (ftos)
   667   //
   668   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
   669   // bytecode checks the next bytecode and then rewrites the current
   670   // bytecode into a pair bytecode; otherwise it rewrites the current
   671   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
   672   //
   673   if (RewriteFrequentPairs) {
   674     Label rewrite, done;
   676     // get next byte
   677     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
   679     // do actual aload_0
   680     aload(0);
   682     // if _getfield then wait with rewrite
   683     __ cmp(G3_scratch, (int)Bytecodes::_getfield);
   684     __ br(Assembler::equal, false, Assembler::pn, done);
   685     __ delayed()->nop();
   687     // if _igetfield then rewrite to _fast_iaccess_0
   688     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   689     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
   690     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   691     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
   693     // if _agetfield then rewrite to _fast_aaccess_0
   694     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   695     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
   696     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   697     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
   699     // if _fgetfield then rewrite to _fast_faccess_0
   700     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   701     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
   702     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   703     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
   705     // else rewrite to _fast_aload0
   706     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   707     __ set(Bytecodes::_fast_aload_0, G4_scratch);
   709     // rewrite
   710     // G4_scratch: fast bytecode
   711     __ bind(rewrite);
   712     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
   713     __ bind(done);
   714   } else {
   715     aload(0);
   716   }
   717 }
   720 void TemplateTable::istore() {
   721   transition(itos, vtos);
   722   locals_index(G3_scratch);
   723   __ store_local_int( G3_scratch, Otos_i );
   724 }
   727 void TemplateTable::lstore() {
   728   transition(ltos, vtos);
   729   locals_index(G3_scratch);
   730   __ store_local_long( G3_scratch, Otos_l );
   731 }
   734 void TemplateTable::fstore() {
   735   transition(ftos, vtos);
   736   locals_index(G3_scratch);
   737   __ store_local_float( G3_scratch, Ftos_f );
   738 }
   741 void TemplateTable::dstore() {
   742   transition(dtos, vtos);
   743   locals_index(G3_scratch);
   744   __ store_local_double( G3_scratch, Ftos_d );
   745 }
   748 void TemplateTable::astore() {
   749   transition(vtos, vtos);
   750   __ load_ptr(0, Otos_i);
   751   __ inc(Lesp, Interpreter::stackElementSize);
   752   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   753   locals_index(G3_scratch);
   754   __ store_local_ptr(G3_scratch, Otos_i);
   755 }
   758 void TemplateTable::wide_istore() {
   759   transition(vtos, vtos);
   760   __ pop_i();
   761   locals_index_wide(G3_scratch);
   762   __ store_local_int( G3_scratch, Otos_i );
   763 }
   766 void TemplateTable::wide_lstore() {
   767   transition(vtos, vtos);
   768   __ pop_l();
   769   locals_index_wide(G3_scratch);
   770   __ store_local_long( G3_scratch, Otos_l );
   771 }
   774 void TemplateTable::wide_fstore() {
   775   transition(vtos, vtos);
   776   __ pop_f();
   777   locals_index_wide(G3_scratch);
   778   __ store_local_float( G3_scratch, Ftos_f );
   779 }
   782 void TemplateTable::wide_dstore() {
   783   transition(vtos, vtos);
   784   __ pop_d();
   785   locals_index_wide(G3_scratch);
   786   __ store_local_double( G3_scratch, Ftos_d );
   787 }
   790 void TemplateTable::wide_astore() {
   791   transition(vtos, vtos);
   792   __ load_ptr(0, Otos_i);
   793   __ inc(Lesp, Interpreter::stackElementSize);
   794   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   795   locals_index_wide(G3_scratch);
   796   __ store_local_ptr(G3_scratch, Otos_i);
   797 }
   800 void TemplateTable::iastore() {
   801   transition(itos, vtos);
   802   __ pop_i(O2); // index
   803   // Otos_i: val
   804   // O3: array
   805   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   806   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
   807 }
   810 void TemplateTable::lastore() {
   811   transition(ltos, vtos);
   812   __ pop_i(O2); // index
   813   // Otos_l: val
   814   // O3: array
   815   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   816   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
   817 }
   820 void TemplateTable::fastore() {
   821   transition(ftos, vtos);
   822   __ pop_i(O2); // index
   823   // Ftos_f: val
   824   // O3: array
   825   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   826   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
   827 }
   830 void TemplateTable::dastore() {
   831   transition(dtos, vtos);
   832   __ pop_i(O2); // index
   833   // Fos_d: val
   834   // O3: array
   835   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   836   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
   837 }
   840 void TemplateTable::aastore() {
   841   Label store_ok, is_null, done;
   842   transition(vtos, vtos);
   843   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
   844   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
   845   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
   846   // Otos_i: val
   847   // O2: index
   848   // O3: array
   849   __ verify_oop(Otos_i);
   850   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
   852   // do array store check - check for NULL value first
   853   __ br_null( Otos_i, false, Assembler::pn, is_null );
   854   __ delayed()->nop();
   856   __ load_klass(O3, O4); // get array klass
   857   __ load_klass(Otos_i, O5); // get value klass
   859   // do fast instanceof cache test
   861   __ ld_ptr(O4,     sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes(),  O4);
   863   assert(Otos_i == O0, "just checking");
   865   // Otos_i:    value
   866   // O1:        addr - offset
   867   // O2:        index
   868   // O3:        array
   869   // O4:        array element klass
   870   // O5:        value klass
   872   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
   874   // Generate a fast subtype check.  Branch to store_ok if no
   875   // failure.  Throw if failure.
   876   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
   878   // Not a subtype; so must throw exception
   879   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
   881   // Store is OK.
   882   __ bind(store_ok);
   883   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
   885   __ ba(false,done);
   886   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
   888   __ bind(is_null);
   889   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
   891   __ profile_null_seen(G3_scratch);
   892   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
   893   __ bind(done);
   894 }
   897 void TemplateTable::bastore() {
   898   transition(itos, vtos);
   899   __ pop_i(O2); // index
   900   // Otos_i: val
   901   // O3: array
   902   __ index_check(O3, O2, 0, G3_scratch, O2);
   903   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
   904 }
   907 void TemplateTable::castore() {
   908   transition(itos, vtos);
   909   __ pop_i(O2); // index
   910   // Otos_i: val
   911   // O3: array
   912   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
   913   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
   914 }
   917 void TemplateTable::sastore() {
   918   // %%%%% Factor across platform
   919   castore();
   920 }
   923 void TemplateTable::istore(int n) {
   924   transition(itos, vtos);
   925   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
   926 }
   929 void TemplateTable::lstore(int n) {
   930   transition(ltos, vtos);
   931   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
   932   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
   934 }
   937 void TemplateTable::fstore(int n) {
   938   transition(ftos, vtos);
   939   assert(n < Argument::n_register_parameters, "only handle register cases");
   940   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
   941 }
   944 void TemplateTable::dstore(int n) {
   945   transition(dtos, vtos);
   946   FloatRegister src = Ftos_d;
   947   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
   948 }
   951 void TemplateTable::astore(int n) {
   952   transition(vtos, vtos);
   953   __ load_ptr(0, Otos_i);
   954   __ inc(Lesp, Interpreter::stackElementSize);
   955   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   956   __ store_local_ptr(n, Otos_i);
   957 }
   960 void TemplateTable::pop() {
   961   transition(vtos, vtos);
   962   __ inc(Lesp, Interpreter::stackElementSize);
   963 }
   966 void TemplateTable::pop2() {
   967   transition(vtos, vtos);
   968   __ inc(Lesp, 2 * Interpreter::stackElementSize);
   969 }
   972 void TemplateTable::dup() {
   973   transition(vtos, vtos);
   974   // stack: ..., a
   975   // load a and tag
   976   __ load_ptr(0, Otos_i);
   977   __ push_ptr(Otos_i);
   978   // stack: ..., a, a
   979 }
   982 void TemplateTable::dup_x1() {
   983   transition(vtos, vtos);
   984   // stack: ..., a, b
   985   __ load_ptr( 1, G3_scratch);  // get a
   986   __ load_ptr( 0, Otos_l1);     // get b
   987   __ store_ptr(1, Otos_l1);     // put b
   988   __ store_ptr(0, G3_scratch);  // put a - like swap
   989   __ push_ptr(Otos_l1);         // push b
   990   // stack: ..., b, a, b
   991 }
   994 void TemplateTable::dup_x2() {
   995   transition(vtos, vtos);
   996   // stack: ..., a, b, c
   997   // get c and push on stack, reuse registers
   998   __ load_ptr( 0, G3_scratch);  // get c
   999   __ push_ptr(G3_scratch);      // push c with tag
  1000   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
  1001   // (stack offsets n+1 now)
  1002   __ load_ptr( 3, Otos_l1);     // get a
  1003   __ store_ptr(3, G3_scratch);  // put c at 3
  1004   // stack: ..., c, b, c, c  (a in reg)
  1005   __ load_ptr( 2, G3_scratch);  // get b
  1006   __ store_ptr(2, Otos_l1);     // put a at 2
  1007   // stack: ..., c, a, c, c  (b in reg)
  1008   __ store_ptr(1, G3_scratch);  // put b at 1
  1009   // stack: ..., c, a, b, c
  1013 void TemplateTable::dup2() {
  1014   transition(vtos, vtos);
  1015   __ load_ptr(1, G3_scratch);  // get a
  1016   __ load_ptr(0, Otos_l1);     // get b
  1017   __ push_ptr(G3_scratch);     // push a
  1018   __ push_ptr(Otos_l1);        // push b
  1019   // stack: ..., a, b, a, b
  1023 void TemplateTable::dup2_x1() {
  1024   transition(vtos, vtos);
  1025   // stack: ..., a, b, c
  1026   __ load_ptr( 1, Lscratch);    // get b
  1027   __ load_ptr( 2, Otos_l1);     // get a
  1028   __ store_ptr(2, Lscratch);    // put b at a
  1029   // stack: ..., b, b, c
  1030   __ load_ptr( 0, G3_scratch);  // get c
  1031   __ store_ptr(1, G3_scratch);  // put c at b
  1032   // stack: ..., b, c, c
  1033   __ store_ptr(0, Otos_l1);     // put a at c
  1034   // stack: ..., b, c, a
  1035   __ push_ptr(Lscratch);        // push b
  1036   __ push_ptr(G3_scratch);      // push c
  1037   // stack: ..., b, c, a, b, c
  1041 // The spec says that these types can be a mixture of category 1 (1 word)
  1042 // types and/or category 2 types (long and doubles)
  1043 void TemplateTable::dup2_x2() {
  1044   transition(vtos, vtos);
  1045   // stack: ..., a, b, c, d
  1046   __ load_ptr( 1, Lscratch);    // get c
  1047   __ load_ptr( 3, Otos_l1);     // get a
  1048   __ store_ptr(3, Lscratch);    // put c at 3
  1049   __ store_ptr(1, Otos_l1);     // put a at 1
  1050   // stack: ..., c, b, a, d
  1051   __ load_ptr( 2, G3_scratch);  // get b
  1052   __ load_ptr( 0, Otos_l1);     // get d
  1053   __ store_ptr(0, G3_scratch);  // put b at 0
  1054   __ store_ptr(2, Otos_l1);     // put d at 2
  1055   // stack: ..., c, d, a, b
  1056   __ push_ptr(Lscratch);        // push c
  1057   __ push_ptr(Otos_l1);         // push d
  1058   // stack: ..., c, d, a, b, c, d
  1062 void TemplateTable::swap() {
  1063   transition(vtos, vtos);
  1064   // stack: ..., a, b
  1065   __ load_ptr( 1, G3_scratch);  // get a
  1066   __ load_ptr( 0, Otos_l1);     // get b
  1067   __ store_ptr(0, G3_scratch);  // put b
  1068   __ store_ptr(1, Otos_l1);     // put a
  1069   // stack: ..., b, a
  1073 void TemplateTable::iop2(Operation op) {
  1074   transition(itos, itos);
  1075   __ pop_i(O1);
  1076   switch (op) {
  1077    case  add:  __  add(O1, Otos_i, Otos_i);  break;
  1078    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
  1079      // %%%%% Mul may not exist: better to call .mul?
  1080    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
  1081    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
  1082    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
  1083    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
  1084    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
  1085    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
  1086    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
  1087    default: ShouldNotReachHere();
  1092 void TemplateTable::lop2(Operation op) {
  1093   transition(ltos, ltos);
  1094   __ pop_l(O2);
  1095   switch (op) {
  1096 #ifdef _LP64
  1097    case  add:  __  add(O2, Otos_l, Otos_l);  break;
  1098    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
  1099    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
  1100    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
  1101    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
  1102 #else
  1103    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
  1104    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
  1105    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
  1106    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
  1107    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
  1108 #endif
  1109    default: ShouldNotReachHere();
  1114 void TemplateTable::idiv() {
  1115   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
  1116   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
  1118   transition(itos, itos);
  1119   __ pop_i(O1); // get 1st op
  1121   // Y contains upper 32 bits of result, set it to 0 or all ones
  1122   __ wry(G0);
  1123   __ mov(~0, G3_scratch);
  1125   __ tst(O1);
  1126      Label neg;
  1127   __ br(Assembler::negative, true, Assembler::pn, neg);
  1128   __ delayed()->wry(G3_scratch);
  1129   __ bind(neg);
  1131      Label ok;
  1132   __ tst(Otos_i);
  1133   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
  1135   const int min_int = 0x80000000;
  1136   Label regular;
  1137   __ cmp(Otos_i, -1);
  1138   __ br(Assembler::notEqual, false, Assembler::pt, regular);
  1139 #ifdef _LP64
  1140   // Don't put set in delay slot
  1141   // Set will turn into multiple instructions in 64 bit mode
  1142   __ delayed()->nop();
  1143   __ set(min_int, G4_scratch);
  1144 #else
  1145   __ delayed()->set(min_int, G4_scratch);
  1146 #endif
  1147   Label done;
  1148   __ cmp(O1, G4_scratch);
  1149   __ br(Assembler::equal, true, Assembler::pt, done);
  1150   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
  1152   __ bind(regular);
  1153   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
  1154   __ bind(done);
  1158 void TemplateTable::irem() {
  1159   transition(itos, itos);
  1160   __ mov(Otos_i, O2); // save divisor
  1161   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
  1162   __ smul(Otos_i, O2, Otos_i);
  1163   __ sub(O1, Otos_i, Otos_i);
  1167 void TemplateTable::lmul() {
  1168   transition(ltos, ltos);
  1169   __ pop_l(O2);
  1170 #ifdef _LP64
  1171   __ mulx(Otos_l, O2, Otos_l);
  1172 #else
  1173   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
  1174 #endif
  1179 void TemplateTable::ldiv() {
  1180   transition(ltos, ltos);
  1182   // check for zero
  1183   __ pop_l(O2);
  1184 #ifdef _LP64
  1185   __ tst(Otos_l);
  1186   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1187   __ sdivx(O2, Otos_l, Otos_l);
  1188 #else
  1189   __ orcc(Otos_l1, Otos_l2, G0);
  1190   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1191   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
  1192 #endif
  1196 void TemplateTable::lrem() {
  1197   transition(ltos, ltos);
  1199   // check for zero
  1200   __ pop_l(O2);
  1201 #ifdef _LP64
  1202   __ tst(Otos_l);
  1203   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1204   __ sdivx(O2, Otos_l, Otos_l2);
  1205   __ mulx (Otos_l2, Otos_l, Otos_l2);
  1206   __ sub  (O2, Otos_l2, Otos_l);
  1207 #else
  1208   __ orcc(Otos_l1, Otos_l2, G0);
  1209   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1210   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
  1211 #endif
  1215 void TemplateTable::lshl() {
  1216   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
  1218   __ pop_l(O2);                          // shift value in O2, O3
  1219 #ifdef _LP64
  1220   __ sllx(O2, Otos_i, Otos_l);
  1221 #else
  1222   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1223 #endif
  1227 void TemplateTable::lshr() {
  1228   transition(itos, ltos); // %%%% see lshl comment
  1230   __ pop_l(O2);                          // shift value in O2, O3
  1231 #ifdef _LP64
  1232   __ srax(O2, Otos_i, Otos_l);
  1233 #else
  1234   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1235 #endif
  1240 void TemplateTable::lushr() {
  1241   transition(itos, ltos); // %%%% see lshl comment
  1243   __ pop_l(O2);                          // shift value in O2, O3
  1244 #ifdef _LP64
  1245   __ srlx(O2, Otos_i, Otos_l);
  1246 #else
  1247   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1248 #endif
  1252 void TemplateTable::fop2(Operation op) {
  1253   transition(ftos, ftos);
  1254   switch (op) {
  1255    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1256    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1257    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1258    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1259    case  rem:
  1260      assert(Ftos_f == F0, "just checking");
  1261 #ifdef _LP64
  1262      // LP64 calling conventions use F1, F3 for passing 2 floats
  1263      __ pop_f(F1);
  1264      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
  1265 #else
  1266      __ pop_i(O0);
  1267      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
  1268      __ ld( __ d_tmp, O1 );
  1269 #endif
  1270      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
  1271      assert( Ftos_f == F0, "fix this code" );
  1272      break;
  1274    default: ShouldNotReachHere();
  1279 void TemplateTable::dop2(Operation op) {
  1280   transition(dtos, dtos);
  1281   switch (op) {
  1282    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1283    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1284    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1285    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1286    case  rem:
  1287 #ifdef _LP64
  1288      // Pass arguments in D0, D2
  1289      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
  1290      __ pop_d( F0 );
  1291 #else
  1292      // Pass arguments in O0O1, O2O3
  1293      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1294      __ ldd( __ d_tmp, O2 );
  1295      __ pop_d(Ftos_f);
  1296      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1297      __ ldd( __ d_tmp, O0 );
  1298 #endif
  1299      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
  1300      assert( Ftos_d == F0, "fix this code" );
  1301      break;
  1303    default: ShouldNotReachHere();
  1308 void TemplateTable::ineg() {
  1309   transition(itos, itos);
  1310   __ neg(Otos_i);
  1314 void TemplateTable::lneg() {
  1315   transition(ltos, ltos);
  1316 #ifdef _LP64
  1317   __ sub(G0, Otos_l, Otos_l);
  1318 #else
  1319   __ lneg(Otos_l1, Otos_l2);
  1320 #endif
  1324 void TemplateTable::fneg() {
  1325   transition(ftos, ftos);
  1326   __ fneg(FloatRegisterImpl::S, Ftos_f);
  1330 void TemplateTable::dneg() {
  1331   transition(dtos, dtos);
  1332   // v8 has fnegd if source and dest are the same
  1333   __ fneg(FloatRegisterImpl::D, Ftos_f);
  1337 void TemplateTable::iinc() {
  1338   transition(vtos, vtos);
  1339   locals_index(G3_scratch);
  1340   __ ldsb(Lbcp, 2, O2);  // load constant
  1341   __ access_local_int(G3_scratch, Otos_i);
  1342   __ add(Otos_i, O2, Otos_i);
  1343   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1347 void TemplateTable::wide_iinc() {
  1348   transition(vtos, vtos);
  1349   locals_index_wide(G3_scratch);
  1350   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
  1351   __ access_local_int(G3_scratch, Otos_i);
  1352   __ add(Otos_i, O3, Otos_i);
  1353   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1357 void TemplateTable::convert() {
  1358 // %%%%% Factor this first part accross platforms
  1359   #ifdef ASSERT
  1360     TosState tos_in  = ilgl;
  1361     TosState tos_out = ilgl;
  1362     switch (bytecode()) {
  1363       case Bytecodes::_i2l: // fall through
  1364       case Bytecodes::_i2f: // fall through
  1365       case Bytecodes::_i2d: // fall through
  1366       case Bytecodes::_i2b: // fall through
  1367       case Bytecodes::_i2c: // fall through
  1368       case Bytecodes::_i2s: tos_in = itos; break;
  1369       case Bytecodes::_l2i: // fall through
  1370       case Bytecodes::_l2f: // fall through
  1371       case Bytecodes::_l2d: tos_in = ltos; break;
  1372       case Bytecodes::_f2i: // fall through
  1373       case Bytecodes::_f2l: // fall through
  1374       case Bytecodes::_f2d: tos_in = ftos; break;
  1375       case Bytecodes::_d2i: // fall through
  1376       case Bytecodes::_d2l: // fall through
  1377       case Bytecodes::_d2f: tos_in = dtos; break;
  1378       default             : ShouldNotReachHere();
  1380     switch (bytecode()) {
  1381       case Bytecodes::_l2i: // fall through
  1382       case Bytecodes::_f2i: // fall through
  1383       case Bytecodes::_d2i: // fall through
  1384       case Bytecodes::_i2b: // fall through
  1385       case Bytecodes::_i2c: // fall through
  1386       case Bytecodes::_i2s: tos_out = itos; break;
  1387       case Bytecodes::_i2l: // fall through
  1388       case Bytecodes::_f2l: // fall through
  1389       case Bytecodes::_d2l: tos_out = ltos; break;
  1390       case Bytecodes::_i2f: // fall through
  1391       case Bytecodes::_l2f: // fall through
  1392       case Bytecodes::_d2f: tos_out = ftos; break;
  1393       case Bytecodes::_i2d: // fall through
  1394       case Bytecodes::_l2d: // fall through
  1395       case Bytecodes::_f2d: tos_out = dtos; break;
  1396       default             : ShouldNotReachHere();
  1398     transition(tos_in, tos_out);
  1399   #endif
  1402   // Conversion
  1403   Label done;
  1404   switch (bytecode()) {
  1405    case Bytecodes::_i2l:
  1406 #ifdef _LP64
  1407     // Sign extend the 32 bits
  1408     __ sra ( Otos_i, 0, Otos_l );
  1409 #else
  1410     __ addcc(Otos_i, 0, Otos_l2);
  1411     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
  1412     __ delayed()->clr(Otos_l1);
  1413     __ set(~0, Otos_l1);
  1414 #endif
  1415     break;
  1417    case Bytecodes::_i2f:
  1418     __ st(Otos_i, __ d_tmp );
  1419     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1420     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
  1421     break;
  1423    case Bytecodes::_i2d:
  1424     __ st(Otos_i, __ d_tmp);
  1425     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1426     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
  1427     break;
  1429    case Bytecodes::_i2b:
  1430     __ sll(Otos_i, 24, Otos_i);
  1431     __ sra(Otos_i, 24, Otos_i);
  1432     break;
  1434    case Bytecodes::_i2c:
  1435     __ sll(Otos_i, 16, Otos_i);
  1436     __ srl(Otos_i, 16, Otos_i);
  1437     break;
  1439    case Bytecodes::_i2s:
  1440     __ sll(Otos_i, 16, Otos_i);
  1441     __ sra(Otos_i, 16, Otos_i);
  1442     break;
  1444    case Bytecodes::_l2i:
  1445 #ifndef _LP64
  1446     __ mov(Otos_l2, Otos_i);
  1447 #else
  1448     // Sign-extend into the high 32 bits
  1449     __ sra(Otos_l, 0, Otos_i);
  1450 #endif
  1451     break;
  1453    case Bytecodes::_l2f:
  1454    case Bytecodes::_l2d:
  1455     __ st_long(Otos_l, __ d_tmp);
  1456     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
  1458     if (VM_Version::v9_instructions_work()) {
  1459       if (bytecode() == Bytecodes::_l2f) {
  1460         __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1461       } else {
  1462         __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
  1464     } else {
  1465       __ call_VM_leaf(
  1466         Lscratch,
  1467         bytecode() == Bytecodes::_l2f
  1468           ? CAST_FROM_FN_PTR(address, SharedRuntime::l2f)
  1469           : CAST_FROM_FN_PTR(address, SharedRuntime::l2d)
  1470       );
  1472     break;
  1474   case Bytecodes::_f2i:  {
  1475       Label isNaN;
  1476       // result must be 0 if value is NaN; test by comparing value to itself
  1477       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
  1478       // According to the v8 manual, you have to have a non-fp instruction
  1479       // between fcmp and fb.
  1480       if (!VM_Version::v9_instructions_work()) {
  1481         __ nop();
  1483       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
  1484       __ delayed()->clr(Otos_i);                                     // NaN
  1485       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
  1486       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
  1487       __ ld(__ d_tmp, Otos_i);
  1488       __ bind(isNaN);
  1490     break;
  1492    case Bytecodes::_f2l:
  1493     // must uncache tos
  1494     __ push_f();
  1495 #ifdef _LP64
  1496     __ pop_f(F1);
  1497 #else
  1498     __ pop_i(O0);
  1499 #endif
  1500     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
  1501     break;
  1503    case Bytecodes::_f2d:
  1504     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
  1505     break;
  1507    case Bytecodes::_d2i:
  1508    case Bytecodes::_d2l:
  1509     // must uncache tos
  1510     __ push_d();
  1511 #ifdef _LP64
  1512     // LP64 calling conventions pass first double arg in D0
  1513     __ pop_d( Ftos_d );
  1514 #else
  1515     __ pop_i( O0 );
  1516     __ pop_i( O1 );
  1517 #endif
  1518     __ call_VM_leaf(Lscratch,
  1519         bytecode() == Bytecodes::_d2i
  1520           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
  1521           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
  1522     break;
  1524     case Bytecodes::_d2f:
  1525     if (VM_Version::v9_instructions_work()) {
  1526       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1528     else {
  1529       // must uncache tos
  1530       __ push_d();
  1531       __ pop_i(O0);
  1532       __ pop_i(O1);
  1533       __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::d2f));
  1535     break;
  1537     default: ShouldNotReachHere();
  1539   __ bind(done);
  1543 void TemplateTable::lcmp() {
  1544   transition(ltos, itos);
  1546 #ifdef _LP64
  1547   __ pop_l(O1); // pop off value 1, value 2 is in O0
  1548   __ lcmp( O1, Otos_l, Otos_i );
  1549 #else
  1550   __ pop_l(O2); // cmp O2,3 to O0,1
  1551   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
  1552 #endif
  1556 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
  1558   if (is_float) __ pop_f(F2);
  1559   else          __ pop_d(F2);
  1561   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
  1563   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
  1566 void TemplateTable::branch(bool is_jsr, bool is_wide) {
  1567   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
  1568   __ verify_oop(Lmethod);
  1569   __ verify_thread();
  1571   const Register O2_bumped_count = O2;
  1572   __ profile_taken_branch(G3_scratch, O2_bumped_count);
  1574   // get (wide) offset to O1_disp
  1575   const Register O1_disp = O1;
  1576   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
  1577   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
  1579   // Handle all the JSR stuff here, then exit.
  1580   // It's much shorter and cleaner than intermingling with the
  1581   // non-JSR normal-branch stuff occurring below.
  1582   if( is_jsr ) {
  1583     // compute return address as bci in Otos_i
  1584     __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1585     __ sub(Lbcp, G3_scratch, G3_scratch);
  1586     __ sub(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
  1588     // Bump Lbcp to target of JSR
  1589     __ add(Lbcp, O1_disp, Lbcp);
  1590     // Push returnAddress for "ret" on stack
  1591     __ push_ptr(Otos_i);
  1592     // And away we go!
  1593     __ dispatch_next(vtos);
  1594     return;
  1597   // Normal (non-jsr) branch handling
  1599   // Save the current Lbcp
  1600   const Register O0_cur_bcp = O0;
  1601   __ mov( Lbcp, O0_cur_bcp );
  1604   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
  1605   if ( increment_invocation_counter_for_backward_branches ) {
  1606     Label Lforward;
  1607     // check branch direction
  1608     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
  1609     // Bump bytecode pointer by displacement (take the branch)
  1610     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
  1612     if (TieredCompilation) {
  1613       Label Lno_mdo, Loverflow;
  1614       int increment = InvocationCounter::count_increment;
  1615       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
  1616       if (ProfileInterpreter) {
  1617         // If no method data exists, go to profile_continue.
  1618         __ ld_ptr(Lmethod, methodOopDesc::method_data_offset(), G4_scratch);
  1619         __ br_null(G4_scratch, false, Assembler::pn, Lno_mdo);
  1620         __ delayed()->nop();
  1622         // Increment backedge counter in the MDO
  1623         Address mdo_backedge_counter(G4_scratch, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
  1624                                                  in_bytes(InvocationCounter::counter_offset()));
  1625         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, Lscratch,
  1626                                    Assembler::notZero, &Lforward);
  1627         __ ba(false, Loverflow);
  1628         __ delayed()->nop();
  1631       // If there's no MDO, increment counter in methodOop
  1632       __ bind(Lno_mdo);
  1633       Address backedge_counter(Lmethod, in_bytes(methodOopDesc::backedge_counter_offset()) +
  1634                                         in_bytes(InvocationCounter::counter_offset()));
  1635       __ increment_mask_and_jump(backedge_counter, increment, mask, G3_scratch, Lscratch,
  1636                                  Assembler::notZero, &Lforward);
  1637       __ bind(Loverflow);
  1639       // notify point for loop, pass branch bytecode
  1640       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O0_cur_bcp);
  1642       // Was an OSR adapter generated?
  1643       // O0 = osr nmethod
  1644       __ br_null(O0, false, Assembler::pn, Lforward);
  1645       __ delayed()->nop();
  1647       // Has the nmethod been invalidated already?
  1648       __ ld(O0, nmethod::entry_bci_offset(), O2);
  1649       __ cmp(O2, InvalidOSREntryBci);
  1650       __ br(Assembler::equal, false, Assembler::pn, Lforward);
  1651       __ delayed()->nop();
  1653       // migrate the interpreter frame off of the stack
  1655       __ mov(G2_thread, L7);
  1656       // save nmethod
  1657       __ mov(O0, L6);
  1658       __ set_last_Java_frame(SP, noreg);
  1659       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
  1660       __ reset_last_Java_frame();
  1661       __ mov(L7, G2_thread);
  1663       // move OSR nmethod to I1
  1664       __ mov(L6, I1);
  1666       // OSR buffer to I0
  1667       __ mov(O0, I0);
  1669       // remove the interpreter frame
  1670       __ restore(I5_savedSP, 0, SP);
  1672       // Jump to the osr code.
  1673       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
  1674       __ jmp(O2, G0);
  1675       __ delayed()->nop();
  1677     } else {
  1678       // Update Backedge branch separately from invocations
  1679       const Register G4_invoke_ctr = G4;
  1680       __ increment_backedge_counter(G4_invoke_ctr, G1_scratch);
  1681       if (ProfileInterpreter) {
  1682         __ test_invocation_counter_for_mdp(G4_invoke_ctr, Lbcp, G3_scratch, Lforward);
  1683         if (UseOnStackReplacement) {
  1684           __ test_backedge_count_for_osr(O2_bumped_count, O0_cur_bcp, G3_scratch);
  1686       } else {
  1687         if (UseOnStackReplacement) {
  1688           __ test_backedge_count_for_osr(G4_invoke_ctr, O0_cur_bcp, G3_scratch);
  1693     __ bind(Lforward);
  1694   } else
  1695     // Bump bytecode pointer by displacement (take the branch)
  1696     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
  1698   // continue with bytecode @ target
  1699   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
  1700   // %%%%% and changing dispatch_next to dispatch_only
  1701   __ dispatch_next(vtos);
  1705 // Note Condition in argument is TemplateTable::Condition
  1706 // arg scope is within class scope
  1708 void TemplateTable::if_0cmp(Condition cc) {
  1709   // no pointers, integer only!
  1710   transition(itos, vtos);
  1711   // assume branch is more often taken than not (loops use backward branches)
  1712   __ cmp( Otos_i, 0);
  1713   __ if_cmp(ccNot(cc), false);
  1717 void TemplateTable::if_icmp(Condition cc) {
  1718   transition(itos, vtos);
  1719   __ pop_i(O1);
  1720   __ cmp(O1, Otos_i);
  1721   __ if_cmp(ccNot(cc), false);
  1725 void TemplateTable::if_nullcmp(Condition cc) {
  1726   transition(atos, vtos);
  1727   __ tst(Otos_i);
  1728   __ if_cmp(ccNot(cc), true);
  1732 void TemplateTable::if_acmp(Condition cc) {
  1733   transition(atos, vtos);
  1734   __ pop_ptr(O1);
  1735   __ verify_oop(O1);
  1736   __ verify_oop(Otos_i);
  1737   __ cmp(O1, Otos_i);
  1738   __ if_cmp(ccNot(cc), true);
  1743 void TemplateTable::ret() {
  1744   transition(vtos, vtos);
  1745   locals_index(G3_scratch);
  1746   __ access_local_returnAddress(G3_scratch, Otos_i);
  1747   // Otos_i contains the bci, compute the bcp from that
  1749 #ifdef _LP64
  1750 #ifdef ASSERT
  1751   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
  1752   // the result.  The return address (really a BCI) was stored with an
  1753   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
  1754   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
  1755   // loaded value.
  1756   { Label zzz ;
  1757      __ set (65536, G3_scratch) ;
  1758      __ cmp (Otos_i, G3_scratch) ;
  1759      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
  1760      __ delayed()->nop();
  1761      __ stop("BCI is in the wrong register half?");
  1762      __ bind (zzz) ;
  1764 #endif
  1765 #endif
  1767   __ profile_ret(vtos, Otos_i, G4_scratch);
  1769   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1770   __ add(G3_scratch, Otos_i, G3_scratch);
  1771   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1772   __ dispatch_next(vtos);
  1776 void TemplateTable::wide_ret() {
  1777   transition(vtos, vtos);
  1778   locals_index_wide(G3_scratch);
  1779   __ access_local_returnAddress(G3_scratch, Otos_i);
  1780   // Otos_i contains the bci, compute the bcp from that
  1782   __ profile_ret(vtos, Otos_i, G4_scratch);
  1784   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1785   __ add(G3_scratch, Otos_i, G3_scratch);
  1786   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1787   __ dispatch_next(vtos);
  1791 void TemplateTable::tableswitch() {
  1792   transition(itos, vtos);
  1793   Label default_case, continue_execution;
  1795   // align bcp
  1796   __ add(Lbcp, BytesPerInt, O1);
  1797   __ and3(O1, -BytesPerInt, O1);
  1798   // load lo, hi
  1799   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
  1800   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
  1801 #ifdef _LP64
  1802   // Sign extend the 32 bits
  1803   __ sra ( Otos_i, 0, Otos_i );
  1804 #endif /* _LP64 */
  1806   // check against lo & hi
  1807   __ cmp( Otos_i, O2);
  1808   __ br( Assembler::less, false, Assembler::pn, default_case);
  1809   __ delayed()->cmp( Otos_i, O3 );
  1810   __ br( Assembler::greater, false, Assembler::pn, default_case);
  1811   // lookup dispatch offset
  1812   __ delayed()->sub(Otos_i, O2, O2);
  1813   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
  1814   __ sll(O2, LogBytesPerInt, O2);
  1815   __ add(O2, 3 * BytesPerInt, O2);
  1816   __ ba(false, continue_execution);
  1817   __ delayed()->ld(O1, O2, O2);
  1818   // handle default
  1819   __ bind(default_case);
  1820   __ profile_switch_default(O3);
  1821   __ ld(O1, 0, O2); // get default offset
  1822   // continue execution
  1823   __ bind(continue_execution);
  1824   __ add(Lbcp, O2, Lbcp);
  1825   __ dispatch_next(vtos);
  1829 void TemplateTable::lookupswitch() {
  1830   transition(itos, itos);
  1831   __ stop("lookupswitch bytecode should have been rewritten");
  1834 void TemplateTable::fast_linearswitch() {
  1835   transition(itos, vtos);
  1836     Label loop_entry, loop, found, continue_execution;
  1837   // align bcp
  1838   __ add(Lbcp, BytesPerInt, O1);
  1839   __ and3(O1, -BytesPerInt, O1);
  1840  // set counter
  1841   __ ld(O1, BytesPerInt, O2);
  1842   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
  1843   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
  1844   __ ba(false, loop_entry);
  1845   __ delayed()->add(O3, O2, O2); // counter now points past last pair
  1847   // table search
  1848   __ bind(loop);
  1849   __ cmp(O4, Otos_i);
  1850   __ br(Assembler::equal, true, Assembler::pn, found);
  1851   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
  1852   __ inc(O3, 2 * BytesPerInt);
  1854   __ bind(loop_entry);
  1855   __ cmp(O2, O3);
  1856   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
  1857   __ delayed()->ld(O3, 0, O4);
  1859   // default case
  1860   __ ld(O1, 0, O4); // get default offset
  1861   if (ProfileInterpreter) {
  1862     __ profile_switch_default(O3);
  1863     __ ba(false, continue_execution);
  1864     __ delayed()->nop();
  1867   // entry found -> get offset
  1868   __ bind(found);
  1869   if (ProfileInterpreter) {
  1870     __ sub(O3, O1, O3);
  1871     __ sub(O3, 2*BytesPerInt, O3);
  1872     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
  1873     __ profile_switch_case(O3, O1, O2, G3_scratch);
  1875     __ bind(continue_execution);
  1877   __ add(Lbcp, O4, Lbcp);
  1878   __ dispatch_next(vtos);
  1882 void TemplateTable::fast_binaryswitch() {
  1883   transition(itos, vtos);
  1884   // Implementation using the following core algorithm: (copied from Intel)
  1885   //
  1886   // int binary_search(int key, LookupswitchPair* array, int n) {
  1887   //   // Binary search according to "Methodik des Programmierens" by
  1888   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
  1889   //   int i = 0;
  1890   //   int j = n;
  1891   //   while (i+1 < j) {
  1892   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
  1893   //     // with      Q: for all i: 0 <= i < n: key < a[i]
  1894   //     // where a stands for the array and assuming that the (inexisting)
  1895   //     // element a[n] is infinitely big.
  1896   //     int h = (i + j) >> 1;
  1897   //     // i < h < j
  1898   //     if (key < array[h].fast_match()) {
  1899   //       j = h;
  1900   //     } else {
  1901   //       i = h;
  1902   //     }
  1903   //   }
  1904   //   // R: a[i] <= key < a[i+1] or Q
  1905   //   // (i.e., if key is within array, i is the correct index)
  1906   //   return i;
  1907   // }
  1909   // register allocation
  1910   assert(Otos_i == O0, "alias checking");
  1911   const Register Rkey     = Otos_i;                    // already set (tosca)
  1912   const Register Rarray   = O1;
  1913   const Register Ri       = O2;
  1914   const Register Rj       = O3;
  1915   const Register Rh       = O4;
  1916   const Register Rscratch = O5;
  1918   const int log_entry_size = 3;
  1919   const int entry_size = 1 << log_entry_size;
  1921   Label found;
  1922   // Find Array start
  1923   __ add(Lbcp, 3 * BytesPerInt, Rarray);
  1924   __ and3(Rarray, -BytesPerInt, Rarray);
  1925   // initialize i & j (in delay slot)
  1926   __ clr( Ri );
  1928   // and start
  1929   Label entry;
  1930   __ ba(false, entry);
  1931   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
  1932   // (Rj is already in the native byte-ordering.)
  1934   // binary search loop
  1935   { Label loop;
  1936     __ bind( loop );
  1937     // int h = (i + j) >> 1;
  1938     __ sra( Rh, 1, Rh );
  1939     // if (key < array[h].fast_match()) {
  1940     //   j = h;
  1941     // } else {
  1942     //   i = h;
  1943     // }
  1944     __ sll( Rh, log_entry_size, Rscratch );
  1945     __ ld( Rarray, Rscratch, Rscratch );
  1946     // (Rscratch is already in the native byte-ordering.)
  1947     __ cmp( Rkey, Rscratch );
  1948     if ( VM_Version::v9_instructions_work() ) {
  1949       __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
  1950       __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
  1952     else {
  1953       Label end_of_if;
  1954       __ br( Assembler::less, true, Assembler::pt, end_of_if );
  1955       __ delayed()->mov( Rh, Rj ); // if (<) Rj = Rh
  1956       __ mov( Rh, Ri );            // else i = h
  1957       __ bind(end_of_if);          // }
  1960     // while (i+1 < j)
  1961     __ bind( entry );
  1962     __ add( Ri, 1, Rscratch );
  1963     __ cmp(Rscratch, Rj);
  1964     __ br( Assembler::less, true, Assembler::pt, loop );
  1965     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
  1968   // end of binary search, result index is i (must check again!)
  1969   Label default_case;
  1970   Label continue_execution;
  1971   if (ProfileInterpreter) {
  1972     __ mov( Ri, Rh );              // Save index in i for profiling
  1974   __ sll( Ri, log_entry_size, Ri );
  1975   __ ld( Rarray, Ri, Rscratch );
  1976   // (Rscratch is already in the native byte-ordering.)
  1977   __ cmp( Rkey, Rscratch );
  1978   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
  1979   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
  1981   // entry found -> j = offset
  1982   __ inc( Ri, BytesPerInt );
  1983   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
  1984   __ ld( Rarray, Ri, Rj );
  1985   // (Rj is already in the native byte-ordering.)
  1987   if (ProfileInterpreter) {
  1988     __ ba(false, continue_execution);
  1989     __ delayed()->nop();
  1992   __ bind(default_case); // fall through (if not profiling)
  1993   __ profile_switch_default(Ri);
  1995   __ bind(continue_execution);
  1996   __ add( Lbcp, Rj, Lbcp );
  1997   __ dispatch_next( vtos );
  2001 void TemplateTable::_return(TosState state) {
  2002   transition(state, state);
  2003   assert(_desc->calls_vm(), "inconsistent calls_vm information");
  2005   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
  2006     assert(state == vtos, "only valid state");
  2007     __ mov(G0, G3_scratch);
  2008     __ access_local_ptr(G3_scratch, Otos_i);
  2009     __ load_klass(Otos_i, O2);
  2010     __ set(JVM_ACC_HAS_FINALIZER, G3);
  2011     __ ld(O2, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), O2);
  2012     __ andcc(G3, O2, G0);
  2013     Label skip_register_finalizer;
  2014     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
  2015     __ delayed()->nop();
  2017     // Call out to do finalizer registration
  2018     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
  2020     __ bind(skip_register_finalizer);
  2023   __ remove_activation(state, /* throw_monitor_exception */ true);
  2025   // The caller's SP was adjusted upon method entry to accomodate
  2026   // the callee's non-argument locals. Undo that adjustment.
  2027   __ ret();                             // return to caller
  2028   __ delayed()->restore(I5_savedSP, G0, SP);
  2032 // ----------------------------------------------------------------------------
  2033 // Volatile variables demand their effects be made known to all CPU's in
  2034 // order.  Store buffers on most chips allow reads & writes to reorder; the
  2035 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
  2036 // memory barrier (i.e., it's not sufficient that the interpreter does not
  2037 // reorder volatile references, the hardware also must not reorder them).
  2038 //
  2039 // According to the new Java Memory Model (JMM):
  2040 // (1) All volatiles are serialized wrt to each other.
  2041 // ALSO reads & writes act as aquire & release, so:
  2042 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
  2043 // the read float up to before the read.  It's OK for non-volatile memory refs
  2044 // that happen before the volatile read to float down below it.
  2045 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
  2046 // that happen BEFORE the write float down to after the write.  It's OK for
  2047 // non-volatile memory refs that happen after the volatile write to float up
  2048 // before it.
  2049 //
  2050 // We only put in barriers around volatile refs (they are expensive), not
  2051 // _between_ memory refs (that would require us to track the flavor of the
  2052 // previous memory refs).  Requirements (2) and (3) require some barriers
  2053 // before volatile stores and after volatile loads.  These nearly cover
  2054 // requirement (1) but miss the volatile-store-volatile-load case.  This final
  2055 // case is placed after volatile-stores although it could just as well go
  2056 // before volatile-loads.
  2057 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
  2058   // Helper function to insert a is-volatile test and memory barrier
  2059   // All current sparc implementations run in TSO, needing only StoreLoad
  2060   if ((order_constraint & Assembler::StoreLoad) == 0) return;
  2061   __ membar( order_constraint );
  2064 // ----------------------------------------------------------------------------
  2065 void TemplateTable::resolve_cache_and_index(int byte_no,
  2066                                             Register result,
  2067                                             Register Rcache,
  2068                                             Register index,
  2069                                             size_t index_size) {
  2070   // Depends on cpCacheOop layout!
  2071   Label resolved;
  2073   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2074   if (byte_no == f1_oop) {
  2075     // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
  2076     // This kind of CP cache entry does not need to match the flags byte, because
  2077     // there is a 1-1 relation between bytecode type and CP entry type.
  2078     assert_different_registers(result, Rcache);
  2079     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2080               ConstantPoolCacheEntry::f1_offset(), result);
  2081     __ tst(result);
  2082     __ br(Assembler::notEqual, false, Assembler::pt, resolved);
  2083     __ delayed()->set((int)bytecode(), O1);
  2084   } else {
  2085     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
  2086     assert(result == noreg, "");  //else change code for setting result
  2087     const int shift_count = (1 + byte_no)*BitsPerByte;
  2089     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2090               ConstantPoolCacheEntry::indices_offset(), Lbyte_code);
  2092     __ srl(  Lbyte_code, shift_count, Lbyte_code );
  2093     __ and3( Lbyte_code,        0xFF, Lbyte_code );
  2094     __ cmp(  Lbyte_code, (int)bytecode());
  2095     __ br(   Assembler::equal, false, Assembler::pt, resolved);
  2096     __ delayed()->set((int)bytecode(), O1);
  2099   address entry;
  2100   switch (bytecode()) {
  2101     case Bytecodes::_getstatic      : // fall through
  2102     case Bytecodes::_putstatic      : // fall through
  2103     case Bytecodes::_getfield       : // fall through
  2104     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
  2105     case Bytecodes::_invokevirtual  : // fall through
  2106     case Bytecodes::_invokespecial  : // fall through
  2107     case Bytecodes::_invokestatic   : // fall through
  2108     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
  2109     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
  2110     case Bytecodes::_fast_aldc      : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2111     case Bytecodes::_fast_aldc_w    : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2112     default                         : ShouldNotReachHere();                                 break;
  2114   // first time invocation - must resolve first
  2115   __ call_VM(noreg, entry, O1);
  2116   // Update registers with resolved info
  2117   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2118   if (result != noreg)
  2119     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2120               ConstantPoolCacheEntry::f1_offset(), result);
  2121   __ bind(resolved);
  2124 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
  2125                                                Register Rmethod,
  2126                                                Register Ritable_index,
  2127                                                Register Rflags,
  2128                                                bool is_invokevirtual,
  2129                                                bool is_invokevfinal,
  2130                                                bool is_invokedynamic) {
  2131   // Uses both G3_scratch and G4_scratch
  2132   Register Rcache = G3_scratch;
  2133   Register Rscratch = G4_scratch;
  2134   assert_different_registers(Rcache, Rmethod, Ritable_index);
  2136   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2138   // determine constant pool cache field offsets
  2139   const int method_offset = in_bytes(
  2140     cp_base_offset +
  2141       (is_invokevirtual
  2142        ? ConstantPoolCacheEntry::f2_offset()
  2143        : ConstantPoolCacheEntry::f1_offset()
  2145     );
  2146   const int flags_offset = in_bytes(cp_base_offset +
  2147                                     ConstantPoolCacheEntry::flags_offset());
  2148   // access constant pool cache fields
  2149   const int index_offset = in_bytes(cp_base_offset +
  2150                                     ConstantPoolCacheEntry::f2_offset());
  2152   if (is_invokevfinal) {
  2153     __ get_cache_and_index_at_bcp(Rcache, Rscratch, 1);
  2154     __ ld_ptr(Rcache, method_offset, Rmethod);
  2155   } else if (byte_no == f1_oop) {
  2156     // Resolved f1_oop goes directly into 'method' register.
  2157     resolve_cache_and_index(byte_no, Rmethod, Rcache, Rscratch, sizeof(u4));
  2158   } else {
  2159     resolve_cache_and_index(byte_no, noreg, Rcache, Rscratch, sizeof(u2));
  2160     __ ld_ptr(Rcache, method_offset, Rmethod);
  2163   if (Ritable_index != noreg) {
  2164     __ ld_ptr(Rcache, index_offset, Ritable_index);
  2166   __ ld_ptr(Rcache, flags_offset, Rflags);
  2169 // The Rcache register must be set before call
  2170 void TemplateTable::load_field_cp_cache_entry(Register Robj,
  2171                                               Register Rcache,
  2172                                               Register index,
  2173                                               Register Roffset,
  2174                                               Register Rflags,
  2175                                               bool is_static) {
  2176   assert_different_registers(Rcache, Rflags, Roffset);
  2178   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2180   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2181   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2182   if (is_static) {
  2183     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
  2187 // The registers Rcache and index expected to be set before call.
  2188 // Correct values of the Rcache and index registers are preserved.
  2189 void TemplateTable::jvmti_post_field_access(Register Rcache,
  2190                                             Register index,
  2191                                             bool is_static,
  2192                                             bool has_tos) {
  2193   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2195   if (JvmtiExport::can_post_field_access()) {
  2196     // Check to see if a field access watch has been set before we take
  2197     // the time to call into the VM.
  2198     Label Label1;
  2199     assert_different_registers(Rcache, index, G1_scratch);
  2200     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
  2201     __ load_contents(get_field_access_count_addr, G1_scratch);
  2202     __ tst(G1_scratch);
  2203     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2204     __ delayed()->nop();
  2206     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
  2208     if (is_static) {
  2209       __ clr(Otos_i);
  2210     } else {
  2211       if (has_tos) {
  2212       // save object pointer before call_VM() clobbers it
  2213         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
  2214       } else {
  2215         // Load top of stack (do not pop the value off the stack);
  2216         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
  2218       __ verify_oop(Otos_i);
  2220     // Otos_i: object pointer or NULL if static
  2221     // Rcache: cache entry pointer
  2222     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
  2223                Otos_i, Rcache);
  2224     if (!is_static && has_tos) {
  2225       __ pop_ptr(Otos_i);  // restore object pointer
  2226       __ verify_oop(Otos_i);
  2228     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2229     __ bind(Label1);
  2233 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
  2234   transition(vtos, vtos);
  2236   Register Rcache = G3_scratch;
  2237   Register index  = G4_scratch;
  2238   Register Rclass = Rcache;
  2239   Register Roffset= G4_scratch;
  2240   Register Rflags = G1_scratch;
  2241   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2243   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2244   jvmti_post_field_access(Rcache, index, is_static, false);
  2245   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2247   if (!is_static) {
  2248     pop_and_check_object(Rclass);
  2249   } else {
  2250     __ verify_oop(Rclass);
  2253   Label exit;
  2255   Assembler::Membar_mask_bits membar_bits =
  2256     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2258   if (__ membar_has_effect(membar_bits)) {
  2259     // Get volatile flag
  2260     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2261     __ and3(Rflags, Lscratch, Lscratch);
  2264   Label checkVolatile;
  2266   // compute field type
  2267   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
  2268   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2269   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2270   ConstantPoolCacheEntry::verify_tosBits();
  2272   // Check atos before itos for getstatic, more likely (in Queens at least)
  2273   __ cmp(Rflags, atos);
  2274   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2275   __ delayed() ->cmp(Rflags, itos);
  2277   // atos
  2278   __ load_heap_oop(Rclass, Roffset, Otos_i);
  2279   __ verify_oop(Otos_i);
  2280   __ push(atos);
  2281   if (!is_static) {
  2282     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
  2284   __ ba(false, checkVolatile);
  2285   __ delayed()->tst(Lscratch);
  2287   __ bind(notObj);
  2289   // cmp(Rflags, itos);
  2290   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2291   __ delayed() ->cmp(Rflags, ltos);
  2293   // itos
  2294   __ ld(Rclass, Roffset, Otos_i);
  2295   __ push(itos);
  2296   if (!is_static) {
  2297     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
  2299   __ ba(false, checkVolatile);
  2300   __ delayed()->tst(Lscratch);
  2302   __ bind(notInt);
  2304   // cmp(Rflags, ltos);
  2305   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2306   __ delayed() ->cmp(Rflags, btos);
  2308   // ltos
  2309   // load must be atomic
  2310   __ ld_long(Rclass, Roffset, Otos_l);
  2311   __ push(ltos);
  2312   if (!is_static) {
  2313     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
  2315   __ ba(false, checkVolatile);
  2316   __ delayed()->tst(Lscratch);
  2318   __ bind(notLong);
  2320   // cmp(Rflags, btos);
  2321   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2322   __ delayed() ->cmp(Rflags, ctos);
  2324   // btos
  2325   __ ldsb(Rclass, Roffset, Otos_i);
  2326   __ push(itos);
  2327   if (!is_static) {
  2328     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
  2330   __ ba(false, checkVolatile);
  2331   __ delayed()->tst(Lscratch);
  2333   __ bind(notByte);
  2335   // cmp(Rflags, ctos);
  2336   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2337   __ delayed() ->cmp(Rflags, stos);
  2339   // ctos
  2340   __ lduh(Rclass, Roffset, Otos_i);
  2341   __ push(itos);
  2342   if (!is_static) {
  2343     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
  2345   __ ba(false, checkVolatile);
  2346   __ delayed()->tst(Lscratch);
  2348   __ bind(notChar);
  2350   // cmp(Rflags, stos);
  2351   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2352   __ delayed() ->cmp(Rflags, ftos);
  2354   // stos
  2355   __ ldsh(Rclass, Roffset, Otos_i);
  2356   __ push(itos);
  2357   if (!is_static) {
  2358     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
  2360   __ ba(false, checkVolatile);
  2361   __ delayed()->tst(Lscratch);
  2363   __ bind(notShort);
  2366   // cmp(Rflags, ftos);
  2367   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
  2368   __ delayed() ->tst(Lscratch);
  2370   // ftos
  2371   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
  2372   __ push(ftos);
  2373   if (!is_static) {
  2374     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
  2376   __ ba(false, checkVolatile);
  2377   __ delayed()->tst(Lscratch);
  2379   __ bind(notFloat);
  2382   // dtos
  2383   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
  2384   __ push(dtos);
  2385   if (!is_static) {
  2386     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
  2389   __ bind(checkVolatile);
  2390   if (__ membar_has_effect(membar_bits)) {
  2391     // __ tst(Lscratch); executed in delay slot
  2392     __ br(Assembler::zero, false, Assembler::pt, exit);
  2393     __ delayed()->nop();
  2394     volatile_barrier(membar_bits);
  2397   __ bind(exit);
  2401 void TemplateTable::getfield(int byte_no) {
  2402   getfield_or_static(byte_no, false);
  2405 void TemplateTable::getstatic(int byte_no) {
  2406   getfield_or_static(byte_no, true);
  2410 void TemplateTable::fast_accessfield(TosState state) {
  2411   transition(atos, state);
  2412   Register Rcache  = G3_scratch;
  2413   Register index   = G4_scratch;
  2414   Register Roffset = G4_scratch;
  2415   Register Rflags  = Rcache;
  2416   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2418   __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2419   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
  2421   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2423   __ null_check(Otos_i);
  2424   __ verify_oop(Otos_i);
  2426   Label exit;
  2428   Assembler::Membar_mask_bits membar_bits =
  2429     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2430   if (__ membar_has_effect(membar_bits)) {
  2431     // Get volatile flag
  2432     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
  2433     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2436   switch (bytecode()) {
  2437     case Bytecodes::_fast_bgetfield:
  2438       __ ldsb(Otos_i, Roffset, Otos_i);
  2439       break;
  2440     case Bytecodes::_fast_cgetfield:
  2441       __ lduh(Otos_i, Roffset, Otos_i);
  2442       break;
  2443     case Bytecodes::_fast_sgetfield:
  2444       __ ldsh(Otos_i, Roffset, Otos_i);
  2445       break;
  2446     case Bytecodes::_fast_igetfield:
  2447       __ ld(Otos_i, Roffset, Otos_i);
  2448       break;
  2449     case Bytecodes::_fast_lgetfield:
  2450       __ ld_long(Otos_i, Roffset, Otos_l);
  2451       break;
  2452     case Bytecodes::_fast_fgetfield:
  2453       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
  2454       break;
  2455     case Bytecodes::_fast_dgetfield:
  2456       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
  2457       break;
  2458     case Bytecodes::_fast_agetfield:
  2459       __ load_heap_oop(Otos_i, Roffset, Otos_i);
  2460       break;
  2461     default:
  2462       ShouldNotReachHere();
  2465   if (__ membar_has_effect(membar_bits)) {
  2466     __ btst(Lscratch, Rflags);
  2467     __ br(Assembler::zero, false, Assembler::pt, exit);
  2468     __ delayed()->nop();
  2469     volatile_barrier(membar_bits);
  2470     __ bind(exit);
  2473   if (state == atos) {
  2474     __ verify_oop(Otos_i);    // does not blow flags!
  2478 void TemplateTable::jvmti_post_fast_field_mod() {
  2479   if (JvmtiExport::can_post_field_modification()) {
  2480     // Check to see if a field modification watch has been set before we take
  2481     // the time to call into the VM.
  2482     Label done;
  2483     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2484     __ load_contents(get_field_modification_count_addr, G4_scratch);
  2485     __ tst(G4_scratch);
  2486     __ br(Assembler::zero, false, Assembler::pt, done);
  2487     __ delayed()->nop();
  2488     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
  2489     __ verify_oop(G4_scratch);
  2490     __ push_ptr(G4_scratch);    // put the object pointer back on tos
  2491     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
  2492     // Save tos values before call_VM() clobbers them. Since we have
  2493     // to do it for every data type, we use the saved values as the
  2494     // jvalue object.
  2495     switch (bytecode()) {  // save tos values before call_VM() clobbers them
  2496     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
  2497     case Bytecodes::_fast_bputfield: // fall through
  2498     case Bytecodes::_fast_sputfield: // fall through
  2499     case Bytecodes::_fast_cputfield: // fall through
  2500     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
  2501     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
  2502     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
  2503     // get words in right order for use as jvalue object
  2504     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
  2506     // setup pointer to jvalue object
  2507     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
  2508     // G4_scratch:  object pointer
  2509     // G1_scratch: cache entry pointer
  2510     // G3_scratch: jvalue object on the stack
  2511     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
  2512     switch (bytecode()) {             // restore tos values
  2513     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
  2514     case Bytecodes::_fast_bputfield: // fall through
  2515     case Bytecodes::_fast_sputfield: // fall through
  2516     case Bytecodes::_fast_cputfield: // fall through
  2517     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
  2518     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
  2519     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
  2520     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
  2522     __ bind(done);
  2526 // The registers Rcache and index expected to be set before call.
  2527 // The function may destroy various registers, just not the Rcache and index registers.
  2528 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
  2529   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2531   if (JvmtiExport::can_post_field_modification()) {
  2532     // Check to see if a field modification watch has been set before we take
  2533     // the time to call into the VM.
  2534     Label Label1;
  2535     assert_different_registers(Rcache, index, G1_scratch);
  2536     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2537     __ load_contents(get_field_modification_count_addr, G1_scratch);
  2538     __ tst(G1_scratch);
  2539     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2540     __ delayed()->nop();
  2542     // The Rcache and index registers have been already set.
  2543     // This allows to eliminate this call but the Rcache and index
  2544     // registers must be correspondingly used after this line.
  2545     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
  2547     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
  2548     if (is_static) {
  2549       // Life is simple.  Null out the object pointer.
  2550       __ clr(G4_scratch);
  2551     } else {
  2552       Register Rflags = G1_scratch;
  2553       // Life is harder. The stack holds the value on top, followed by the
  2554       // object.  We don't know the size of the value, though; it could be
  2555       // one or two words depending on its type. As a result, we must find
  2556       // the type to determine where the object is.
  2558       Label two_word, valsizeknown;
  2559       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2560       __ mov(Lesp, G4_scratch);
  2561       __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2562       // Make sure we don't need to mask Rflags for tosBits after the above shift
  2563       ConstantPoolCacheEntry::verify_tosBits();
  2564       __ cmp(Rflags, ltos);
  2565       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2566       __ delayed()->cmp(Rflags, dtos);
  2567       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2568       __ delayed()->nop();
  2569       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
  2570       __ br(Assembler::always, false, Assembler::pt, valsizeknown);
  2571       __ delayed()->nop();
  2572       __ bind(two_word);
  2574       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
  2576       __ bind(valsizeknown);
  2577       // setup object pointer
  2578       __ ld_ptr(G4_scratch, 0, G4_scratch);
  2579       __ verify_oop(G4_scratch);
  2581     // setup pointer to jvalue object
  2582     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
  2583     // G4_scratch:  object pointer or NULL if static
  2584     // G3_scratch: cache entry pointer
  2585     // G1_scratch: jvalue object on the stack
  2586     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
  2587                G4_scratch, G3_scratch, G1_scratch);
  2588     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2589     __ bind(Label1);
  2593 void TemplateTable::pop_and_check_object(Register r) {
  2594   __ pop_ptr(r);
  2595   __ null_check(r);  // for field access must check obj.
  2596   __ verify_oop(r);
  2599 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
  2600   transition(vtos, vtos);
  2601   Register Rcache = G3_scratch;
  2602   Register index  = G4_scratch;
  2603   Register Rclass = Rcache;
  2604   Register Roffset= G4_scratch;
  2605   Register Rflags = G1_scratch;
  2606   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2608   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2609   jvmti_post_field_mod(Rcache, index, is_static);
  2610   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2612   Assembler::Membar_mask_bits read_bits =
  2613     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2614   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2616   Label notVolatile, checkVolatile, exit;
  2617   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2618     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2619     __ and3(Rflags, Lscratch, Lscratch);
  2621     if (__ membar_has_effect(read_bits)) {
  2622       __ tst(Lscratch);
  2623       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2624       __ delayed()->nop();
  2625       volatile_barrier(read_bits);
  2626       __ bind(notVolatile);
  2630   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2631   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2632   ConstantPoolCacheEntry::verify_tosBits();
  2634   // compute field type
  2635   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
  2637   if (is_static) {
  2638     // putstatic with object type most likely, check that first
  2639     __ cmp(Rflags, atos );
  2640     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2641     __ delayed() ->cmp(Rflags, itos );
  2643     // atos
  2644     __ pop_ptr();
  2645     __ verify_oop(Otos_i);
  2647     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2649     __ ba(false, checkVolatile);
  2650     __ delayed()->tst(Lscratch);
  2652     __ bind(notObj);
  2654     // cmp(Rflags, itos );
  2655     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2656     __ delayed() ->cmp(Rflags, btos );
  2658     // itos
  2659     __ pop_i();
  2660     __ st(Otos_i, Rclass, Roffset);
  2661     __ ba(false, checkVolatile);
  2662     __ delayed()->tst(Lscratch);
  2664     __ bind(notInt);
  2666   } else {
  2667     // putfield with int type most likely, check that first
  2668     __ cmp(Rflags, itos );
  2669     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2670     __ delayed() ->cmp(Rflags, atos );
  2672     // itos
  2673     __ pop_i();
  2674     pop_and_check_object(Rclass);
  2675     __ st(Otos_i, Rclass, Roffset);
  2676     patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch);
  2677     __ ba(false, checkVolatile);
  2678     __ delayed()->tst(Lscratch);
  2680     __ bind(notInt);
  2681     // cmp(Rflags, atos );
  2682     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2683     __ delayed() ->cmp(Rflags, btos );
  2685     // atos
  2686     __ pop_ptr();
  2687     pop_and_check_object(Rclass);
  2688     __ verify_oop(Otos_i);
  2690     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2692     patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch);
  2693     __ ba(false, checkVolatile);
  2694     __ delayed()->tst(Lscratch);
  2696     __ bind(notObj);
  2699   // cmp(Rflags, btos );
  2700   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2701   __ delayed() ->cmp(Rflags, ltos );
  2703   // btos
  2704   __ pop_i();
  2705   if (!is_static) pop_and_check_object(Rclass);
  2706   __ stb(Otos_i, Rclass, Roffset);
  2707   if (!is_static) {
  2708     patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch);
  2710   __ ba(false, checkVolatile);
  2711   __ delayed()->tst(Lscratch);
  2713   __ bind(notByte);
  2715   // cmp(Rflags, ltos );
  2716   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2717   __ delayed() ->cmp(Rflags, ctos );
  2719   // ltos
  2720   __ pop_l();
  2721   if (!is_static) pop_and_check_object(Rclass);
  2722   __ st_long(Otos_l, Rclass, Roffset);
  2723   if (!is_static) {
  2724     patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch);
  2726   __ ba(false, checkVolatile);
  2727   __ delayed()->tst(Lscratch);
  2729   __ bind(notLong);
  2731   // cmp(Rflags, ctos );
  2732   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2733   __ delayed() ->cmp(Rflags, stos );
  2735   // ctos (char)
  2736   __ pop_i();
  2737   if (!is_static) pop_and_check_object(Rclass);
  2738   __ sth(Otos_i, Rclass, Roffset);
  2739   if (!is_static) {
  2740     patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch);
  2742   __ ba(false, checkVolatile);
  2743   __ delayed()->tst(Lscratch);
  2745   __ bind(notChar);
  2746   // cmp(Rflags, stos );
  2747   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2748   __ delayed() ->cmp(Rflags, ftos );
  2750   // stos (char)
  2751   __ pop_i();
  2752   if (!is_static) pop_and_check_object(Rclass);
  2753   __ sth(Otos_i, Rclass, Roffset);
  2754   if (!is_static) {
  2755     patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch);
  2757   __ ba(false, checkVolatile);
  2758   __ delayed()->tst(Lscratch);
  2760   __ bind(notShort);
  2761   // cmp(Rflags, ftos );
  2762   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
  2763   __ delayed()->nop();
  2765   // ftos
  2766   __ pop_f();
  2767   if (!is_static) pop_and_check_object(Rclass);
  2768   __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2769   if (!is_static) {
  2770     patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch);
  2772   __ ba(false, checkVolatile);
  2773   __ delayed()->tst(Lscratch);
  2775   __ bind(notFloat);
  2777   // dtos
  2778   __ pop_d();
  2779   if (!is_static) pop_and_check_object(Rclass);
  2780   __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2781   if (!is_static) {
  2782     patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch);
  2785   __ bind(checkVolatile);
  2786   __ tst(Lscratch);
  2788   if (__ membar_has_effect(write_bits)) {
  2789     // __ tst(Lscratch); in delay slot
  2790     __ br(Assembler::zero, false, Assembler::pt, exit);
  2791     __ delayed()->nop();
  2792     volatile_barrier(Assembler::StoreLoad);
  2793     __ bind(exit);
  2797 void TemplateTable::fast_storefield(TosState state) {
  2798   transition(state, vtos);
  2799   Register Rcache = G3_scratch;
  2800   Register Rclass = Rcache;
  2801   Register Roffset= G4_scratch;
  2802   Register Rflags = G1_scratch;
  2803   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2805   jvmti_post_fast_field_mod();
  2807   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
  2809   Assembler::Membar_mask_bits read_bits =
  2810     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2811   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2813   Label notVolatile, checkVolatile, exit;
  2814   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2815     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2816     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2817     __ and3(Rflags, Lscratch, Lscratch);
  2818     if (__ membar_has_effect(read_bits)) {
  2819       __ tst(Lscratch);
  2820       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2821       __ delayed()->nop();
  2822       volatile_barrier(read_bits);
  2823       __ bind(notVolatile);
  2827   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2828   pop_and_check_object(Rclass);
  2830   switch (bytecode()) {
  2831     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
  2832     case Bytecodes::_fast_cputfield: /* fall through */
  2833     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
  2834     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
  2835     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
  2836     case Bytecodes::_fast_fputfield:
  2837       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2838       break;
  2839     case Bytecodes::_fast_dputfield:
  2840       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2841       break;
  2842     case Bytecodes::_fast_aputfield:
  2843       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2844       break;
  2845     default:
  2846       ShouldNotReachHere();
  2849   if (__ membar_has_effect(write_bits)) {
  2850     __ tst(Lscratch);
  2851     __ br(Assembler::zero, false, Assembler::pt, exit);
  2852     __ delayed()->nop();
  2853     volatile_barrier(Assembler::StoreLoad);
  2854     __ bind(exit);
  2859 void TemplateTable::putfield(int byte_no) {
  2860   putfield_or_static(byte_no, false);
  2863 void TemplateTable::putstatic(int byte_no) {
  2864   putfield_or_static(byte_no, true);
  2868 void TemplateTable::fast_xaccess(TosState state) {
  2869   transition(vtos, state);
  2870   Register Rcache = G3_scratch;
  2871   Register Roffset = G4_scratch;
  2872   Register Rflags  = G4_scratch;
  2873   Register Rreceiver = Lscratch;
  2875   __ ld_ptr(Llocals, 0, Rreceiver);
  2877   // access constant pool cache  (is resolved)
  2878   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
  2879   __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2880   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
  2882   __ verify_oop(Rreceiver);
  2883   __ null_check(Rreceiver);
  2884   if (state == atos) {
  2885     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
  2886   } else if (state == itos) {
  2887     __ ld (Rreceiver, Roffset, Otos_i) ;
  2888   } else if (state == ftos) {
  2889     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
  2890   } else {
  2891     ShouldNotReachHere();
  2894   Assembler::Membar_mask_bits membar_bits =
  2895     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2896   if (__ membar_has_effect(membar_bits)) {
  2898     // Get is_volatile value in Rflags and check if membar is needed
  2899     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2901     // Test volatile
  2902     Label notVolatile;
  2903     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2904     __ btst(Rflags, Lscratch);
  2905     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2906     __ delayed()->nop();
  2907     volatile_barrier(membar_bits);
  2908     __ bind(notVolatile);
  2911   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
  2912   __ sub(Lbcp, 1, Lbcp);
  2915 //----------------------------------------------------------------------------------------------------
  2916 // Calls
  2918 void TemplateTable::count_calls(Register method, Register temp) {
  2919   // implemented elsewhere
  2920   ShouldNotReachHere();
  2923 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
  2924   Register Rtemp = G4_scratch;
  2925   Register Rcall = Rindex;
  2926   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  2928   // get target methodOop & entry point
  2929   const int base = instanceKlass::vtable_start_offset() * wordSize;
  2930   if (vtableEntry::size() % 3 == 0) {
  2931     // scale the vtable index by 12:
  2932     int one_third = vtableEntry::size() / 3;
  2933     __ sll(Rindex, exact_log2(one_third * 1 * wordSize), Rtemp);
  2934     __ sll(Rindex, exact_log2(one_third * 2 * wordSize), Rindex);
  2935     __ add(Rindex, Rtemp, Rindex);
  2936   } else {
  2937     // scale the vtable index by 8:
  2938     __ sll(Rindex, exact_log2(vtableEntry::size() * wordSize), Rindex);
  2941   __ add(Rrecv, Rindex, Rrecv);
  2942   __ ld_ptr(Rrecv, base + vtableEntry::method_offset_in_bytes(), G5_method);
  2944   __ call_from_interpreter(Rcall, Gargs, Rret);
  2947 void TemplateTable::invokevirtual(int byte_no) {
  2948   transition(vtos, vtos);
  2949   assert(byte_no == f2_byte, "use this argument");
  2951   Register Rscratch = G3_scratch;
  2952   Register Rtemp = G4_scratch;
  2953   Register Rret = Lscratch;
  2954   Register Rrecv = G5_method;
  2955   Label notFinal;
  2957   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
  2958   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  2960   // Check for vfinal
  2961   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), G4_scratch);
  2962   __ btst(Rret, G4_scratch);
  2963   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  2964   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
  2966   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
  2968   invokevfinal_helper(Rscratch, Rret);
  2970   __ bind(notFinal);
  2972   __ mov(G5_method, Rscratch);  // better scratch register
  2973   __ load_receiver(G4_scratch, O0);  // gets receiverOop
  2974   // receiver is in O0
  2975   __ verify_oop(O0);
  2977   // get return address
  2978   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  2979   __ set(table, Rtemp);
  2980   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  2981   // Make sure we don't need to mask Rret for tosBits after the above shift
  2982   ConstantPoolCacheEntry::verify_tosBits();
  2983   __ sll(Rret,  LogBytesPerWord, Rret);
  2984   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  2986   // get receiver klass
  2987   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  2988   __ load_klass(O0, Rrecv);
  2989   __ verify_oop(Rrecv);
  2991   __ profile_virtual_call(Rrecv, O4);
  2993   generate_vtable_call(Rrecv, Rscratch, Rret);
  2996 void TemplateTable::fast_invokevfinal(int byte_no) {
  2997   transition(vtos, vtos);
  2998   assert(byte_no == f2_byte, "use this argument");
  3000   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
  3001                              /*is_invokevfinal*/true, false);
  3002   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3003   invokevfinal_helper(G3_scratch, Lscratch);
  3006 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
  3007   Register Rtemp = G4_scratch;
  3009   __ verify_oop(G5_method);
  3011   // Load receiver from stack slot
  3012   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  3013   __ load_receiver(G4_scratch, O0);
  3015   // receiver NULL check
  3016   __ null_check(O0);
  3018   __ profile_final_call(O4);
  3020   // get return address
  3021   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3022   __ set(table, Rtemp);
  3023   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3024   // Make sure we don't need to mask Rret for tosBits after the above shift
  3025   ConstantPoolCacheEntry::verify_tosBits();
  3026   __ sll(Rret,  LogBytesPerWord, Rret);
  3027   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3030   // do the call
  3031   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3034 void TemplateTable::invokespecial(int byte_no) {
  3035   transition(vtos, vtos);
  3036   assert(byte_no == f1_byte, "use this argument");
  3038   Register Rscratch = G3_scratch;
  3039   Register Rtemp = G4_scratch;
  3040   Register Rret = Lscratch;
  3042   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3043   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3045   __ verify_oop(G5_method);
  3047   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  3048   __ load_receiver(G4_scratch, O0);
  3050   // receiver NULL check
  3051   __ null_check(O0);
  3053   __ profile_call(O4);
  3055   // get return address
  3056   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3057   __ set(table, Rtemp);
  3058   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3059   // Make sure we don't need to mask Rret for tosBits after the above shift
  3060   ConstantPoolCacheEntry::verify_tosBits();
  3061   __ sll(Rret,  LogBytesPerWord, Rret);
  3062   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3064   // do the call
  3065   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3068 void TemplateTable::invokestatic(int byte_no) {
  3069   transition(vtos, vtos);
  3070   assert(byte_no == f1_byte, "use this argument");
  3072   Register Rscratch = G3_scratch;
  3073   Register Rtemp = G4_scratch;
  3074   Register Rret = Lscratch;
  3076   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3077   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3079   __ verify_oop(G5_method);
  3081   __ profile_call(O4);
  3083   // get return address
  3084   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3085   __ set(table, Rtemp);
  3086   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3087   // Make sure we don't need to mask Rret for tosBits after the above shift
  3088   ConstantPoolCacheEntry::verify_tosBits();
  3089   __ sll(Rret,  LogBytesPerWord, Rret);
  3090   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3092   // do the call
  3093   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3097 void TemplateTable::invokeinterface_object_method(Register RklassOop,
  3098                                                   Register Rcall,
  3099                                                   Register Rret,
  3100                                                   Register Rflags) {
  3101   Register Rscratch = G4_scratch;
  3102   Register Rindex = Lscratch;
  3104   assert_different_registers(Rscratch, Rindex, Rret);
  3106   Label notFinal;
  3108   // Check for vfinal
  3109   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), Rscratch);
  3110   __ btst(Rflags, Rscratch);
  3111   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  3112   __ delayed()->nop();
  3114   __ profile_final_call(O4);
  3116   // do the call - the index (f2) contains the methodOop
  3117   assert_different_registers(G5_method, Gargs, Rcall);
  3118   __ mov(Rindex, G5_method);
  3119   __ call_from_interpreter(Rcall, Gargs, Rret);
  3120   __ bind(notFinal);
  3122   __ profile_virtual_call(RklassOop, O4);
  3123   generate_vtable_call(RklassOop, Rindex, Rret);
  3127 void TemplateTable::invokeinterface(int byte_no) {
  3128   transition(vtos, vtos);
  3129   assert(byte_no == f1_byte, "use this argument");
  3131   Register Rscratch = G4_scratch;
  3132   Register Rret = G3_scratch;
  3133   Register Rindex = Lscratch;
  3134   Register Rinterface = G1_scratch;
  3135   Register RklassOop = G5_method;
  3136   Register Rflags = O1;
  3137   assert_different_registers(Rscratch, G5_method);
  3139   load_invoke_cp_cache_entry(byte_no, Rinterface, Rindex, Rflags, /*virtual*/ false, false, false);
  3140   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3142   // get receiver
  3143   __ and3(Rflags, 0xFF, Rscratch);       // gets number of parameters
  3144   __ load_receiver(Rscratch, O0);
  3145   __ verify_oop(O0);
  3147   __ mov(Rflags, Rret);
  3149   // get return address
  3150   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3151   __ set(table, Rscratch);
  3152   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3153   // Make sure we don't need to mask Rret for tosBits after the above shift
  3154   ConstantPoolCacheEntry::verify_tosBits();
  3155   __ sll(Rret,  LogBytesPerWord, Rret);
  3156   __ ld_ptr(Rscratch, Rret, Rret);      // get return address
  3158   // get receiver klass
  3159   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  3160   __ load_klass(O0, RklassOop);
  3161   __ verify_oop(RklassOop);
  3163   // Special case of invokeinterface called for virtual method of
  3164   // java.lang.Object.  See cpCacheOop.cpp for details.
  3165   // This code isn't produced by javac, but could be produced by
  3166   // another compliant java compiler.
  3167   Label notMethod;
  3168   __ set((1 << ConstantPoolCacheEntry::methodInterface), Rscratch);
  3169   __ btst(Rflags, Rscratch);
  3170   __ br(Assembler::zero, false, Assembler::pt, notMethod);
  3171   __ delayed()->nop();
  3173   invokeinterface_object_method(RklassOop, Rinterface, Rret, Rflags);
  3175   __ bind(notMethod);
  3177   __ profile_virtual_call(RklassOop, O4);
  3179   //
  3180   // find entry point to call
  3181   //
  3183   // compute start of first itableOffsetEntry (which is at end of vtable)
  3184   const int base = instanceKlass::vtable_start_offset() * wordSize;
  3185   Label search;
  3186   Register Rtemp = Rflags;
  3188   __ ld(RklassOop, instanceKlass::vtable_length_offset() * wordSize, Rtemp);
  3189   if (align_object_offset(1) > 1) {
  3190     __ round_to(Rtemp, align_object_offset(1));
  3192   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
  3193   if (Assembler::is_simm13(base)) {
  3194     __ add(Rtemp, base, Rtemp);
  3195   } else {
  3196     __ set(base, Rscratch);
  3197     __ add(Rscratch, Rtemp, Rtemp);
  3199   __ add(RklassOop, Rtemp, Rscratch);
  3201   __ bind(search);
  3203   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
  3205     Label ok;
  3207     // Check that entry is non-null.  Null entries are probably a bytecode
  3208     // problem.  If the interface isn't implemented by the receiver class,
  3209     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
  3210     // this too but that's only if the entry isn't already resolved, so we
  3211     // need to check again.
  3212     __ br_notnull( Rtemp, false, Assembler::pt, ok);
  3213     __ delayed()->nop();
  3214     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
  3215     __ should_not_reach_here();
  3216     __ bind(ok);
  3217     __ verify_oop(Rtemp);
  3220   __ verify_oop(Rinterface);
  3222   __ cmp(Rinterface, Rtemp);
  3223   __ brx(Assembler::notEqual, true, Assembler::pn, search);
  3224   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
  3226   // entry found and Rscratch points to it
  3227   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
  3229   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
  3230   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
  3231   __ add(Rscratch, Rindex, Rscratch);
  3232   __ ld_ptr(RklassOop, Rscratch, G5_method);
  3234   // Check for abstract method error.
  3236     Label ok;
  3237     __ tst(G5_method);
  3238     __ brx(Assembler::notZero, false, Assembler::pt, ok);
  3239     __ delayed()->nop();
  3240     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
  3241     __ should_not_reach_here();
  3242     __ bind(ok);
  3245   Register Rcall = Rinterface;
  3246   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  3248   __ verify_oop(G5_method);
  3249   __ call_from_interpreter(Rcall, Gargs, Rret);
  3254 void TemplateTable::invokedynamic(int byte_no) {
  3255   transition(vtos, vtos);
  3256   assert(byte_no == f1_oop, "use this argument");
  3258   if (!EnableInvokeDynamic) {
  3259     // We should not encounter this bytecode if !EnableInvokeDynamic.
  3260     // The verifier will stop it.  However, if we get past the verifier,
  3261     // this will stop the thread in a reasonable way, without crashing the JVM.
  3262     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
  3263                      InterpreterRuntime::throw_IncompatibleClassChangeError));
  3264     // the call_VM checks for exception, so we should never return here.
  3265     __ should_not_reach_here();
  3266     return;
  3269   // G5: CallSite object (f1)
  3270   // XX: unused (f2)
  3271   // XX: flags (unused)
  3273   Register G5_callsite = G5_method;
  3274   Register Rscratch    = G3_scratch;
  3275   Register Rtemp       = G1_scratch;
  3276   Register Rret        = Lscratch;
  3278   load_invoke_cp_cache_entry(byte_no, G5_callsite, noreg, Rret,
  3279                              /*virtual*/ false, /*vfinal*/ false, /*indy*/ true);
  3280   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
  3282   __ verify_oop(G5_callsite);
  3284   // profile this call
  3285   __ profile_call(O4);
  3287   // get return address
  3288   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3289   __ set(table, Rtemp);
  3290   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);  // get return type
  3291   // Make sure we don't need to mask Rret for tosBits after the above shift
  3292   ConstantPoolCacheEntry::verify_tosBits();
  3293   __ sll(Rret, LogBytesPerWord, Rret);
  3294   __ ld_ptr(Rtemp, Rret, Rret);  // get return address
  3296   __ load_heap_oop(G5_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle);
  3297   __ null_check(G3_method_handle);
  3299   // Adjust Rret first so Llast_SP can be same as Rret
  3300   __ add(Rret, -frame::pc_return_offset, O7);
  3301   __ add(Lesp, BytesPerWord, Gargs);  // setup parameter pointer
  3302   __ jump_to_method_handle_entry(G3_method_handle, Rtemp, /* emit_delayed_nop */ false);
  3303   // Record SP so we can remove any stack space allocated by adapter transition
  3304   __ delayed()->mov(SP, Llast_SP);
  3308 //----------------------------------------------------------------------------------------------------
  3309 // Allocation
  3311 void TemplateTable::_new() {
  3312   transition(vtos, atos);
  3314   Label slow_case;
  3315   Label done;
  3316   Label initialize_header;
  3317   Label initialize_object;  // including clearing the fields
  3319   Register RallocatedObject = Otos_i;
  3320   Register RinstanceKlass = O1;
  3321   Register Roffset = O3;
  3322   Register Rscratch = O4;
  3324   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3325   __ get_cpool_and_tags(Rscratch, G3_scratch);
  3326   // make sure the class we're about to instantiate has been resolved
  3327   // This is done before loading instanceKlass to be consistent with the order
  3328   // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
  3329   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3330   __ ldub(G3_scratch, Roffset, G3_scratch);
  3331   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3332   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3333   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3334   // get instanceKlass
  3335   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
  3336   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3337   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
  3339   // make sure klass is fully initialized:
  3340   __ ld(RinstanceKlass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_scratch);
  3341   __ cmp(G3_scratch, instanceKlass::fully_initialized);
  3342   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3343   __ delayed()->ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3345   // get instance_size in instanceKlass (already aligned)
  3346   //__ ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3348   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
  3349   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
  3350   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
  3351   __ delayed()->nop();
  3353   // allocate the instance
  3354   // 1) Try to allocate in the TLAB
  3355   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
  3356   // 3) if the above fails (or is not applicable), go to a slow case
  3357   // (creates a new TLAB, etc.)
  3359   const bool allow_shared_alloc =
  3360     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
  3362   if(UseTLAB) {
  3363     Register RoldTopValue = RallocatedObject;
  3364     Register RtopAddr = G3_scratch, RtlabWasteLimitValue = G3_scratch;
  3365     Register RnewTopValue = G1_scratch;
  3366     Register RendValue = Rscratch;
  3367     Register RfreeValue = RnewTopValue;
  3369     // check if we can allocate in the TLAB
  3370     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
  3371     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
  3372     __ add(RoldTopValue, Roffset, RnewTopValue);
  3374     // if there is enough space, we do not CAS and do not clear
  3375     __ cmp(RnewTopValue, RendValue);
  3376     if(ZeroTLAB) {
  3377       // the fields have already been cleared
  3378       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
  3379     } else {
  3380       // initialize both the header and fields
  3381       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
  3383     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
  3385     if (allow_shared_alloc) {
  3386     // Check if tlab should be discarded (refill_waste_limit >= free)
  3387     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
  3388     __ sub(RendValue, RoldTopValue, RfreeValue);
  3389 #ifdef _LP64
  3390     __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
  3391 #else
  3392     __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
  3393 #endif
  3394     __ cmp(RtlabWasteLimitValue, RfreeValue);
  3395     __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, slow_case); // tlab waste is small
  3396     __ delayed()->nop();
  3398     // increment waste limit to prevent getting stuck on this slow path
  3399     __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
  3400     __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
  3401     } else {
  3402       // No allocation in the shared eden.
  3403       __ br(Assembler::always, false, Assembler::pt, slow_case);
  3404       __ delayed()->nop();
  3408   // Allocation in the shared Eden
  3409   if (allow_shared_alloc) {
  3410     Register RoldTopValue = G1_scratch;
  3411     Register RtopAddr = G3_scratch;
  3412     Register RnewTopValue = RallocatedObject;
  3413     Register RendValue = Rscratch;
  3415     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
  3417     Label retry;
  3418     __ bind(retry);
  3419     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
  3420     __ ld_ptr(RendValue, 0, RendValue);
  3421     __ ld_ptr(RtopAddr, 0, RoldTopValue);
  3422     __ add(RoldTopValue, Roffset, RnewTopValue);
  3424     // RnewTopValue contains the top address after the new object
  3425     // has been allocated.
  3426     __ cmp(RnewTopValue, RendValue);
  3427     __ brx(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);
  3428     __ delayed()->nop();
  3430     __ casx_under_lock(RtopAddr, RoldTopValue, RnewTopValue,
  3431       VM_Version::v9_instructions_work() ? NULL :
  3432       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
  3434     // if someone beat us on the allocation, try again, otherwise continue
  3435     __ cmp(RoldTopValue, RnewTopValue);
  3436     __ brx(Assembler::notEqual, false, Assembler::pn, retry);
  3437     __ delayed()->nop();
  3440   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
  3441     // clear object fields
  3442     __ bind(initialize_object);
  3443     __ deccc(Roffset, sizeof(oopDesc));
  3444     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
  3445     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
  3447     // initialize remaining object fields
  3448     { Label loop;
  3449       __ subcc(Roffset, wordSize, Roffset);
  3450       __ bind(loop);
  3451       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
  3452       __ st_ptr(G0, G3_scratch, Roffset);
  3453       __ br(Assembler::notEqual, false, Assembler::pt, loop);
  3454       __ delayed()->subcc(Roffset, wordSize, Roffset);
  3456     __ br(Assembler::always, false, Assembler::pt, initialize_header);
  3457     __ delayed()->nop();
  3460   // slow case
  3461   __ bind(slow_case);
  3462   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3463   __ get_constant_pool(O1);
  3465   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
  3467   __ ba(false, done);
  3468   __ delayed()->nop();
  3470   // Initialize the header: mark, klass
  3471   __ bind(initialize_header);
  3473   if (UseBiasedLocking) {
  3474     __ ld_ptr(RinstanceKlass, Klass::prototype_header_offset_in_bytes() + sizeof(oopDesc), G4_scratch);
  3475   } else {
  3476     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
  3478   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
  3479   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
  3480   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
  3483     SkipIfEqual skip_if(
  3484       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
  3485     // Trigger dtrace event
  3486     __ push(atos);
  3487     __ call_VM_leaf(noreg,
  3488        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
  3489     __ pop(atos);
  3492   // continue
  3493   __ bind(done);
  3498 void TemplateTable::newarray() {
  3499   transition(itos, atos);
  3500   __ ldub(Lbcp, 1, O1);
  3501      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
  3505 void TemplateTable::anewarray() {
  3506   transition(itos, atos);
  3507   __ get_constant_pool(O1);
  3508   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3509      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
  3513 void TemplateTable::arraylength() {
  3514   transition(atos, itos);
  3515   Label ok;
  3516   __ verify_oop(Otos_i);
  3517   __ tst(Otos_i);
  3518   __ throw_if_not_1_x( Assembler::notZero, ok );
  3519   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
  3520   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3524 void TemplateTable::checkcast() {
  3525   transition(atos, atos);
  3526   Label done, is_null, quicked, cast_ok, resolved;
  3527   Register Roffset = G1_scratch;
  3528   Register RobjKlass = O5;
  3529   Register RspecifiedKlass = O4;
  3531   // Check for casting a NULL
  3532   __ br_null(Otos_i, false, Assembler::pn, is_null);
  3533   __ delayed()->nop();
  3535   // Get value klass in RobjKlass
  3536   __ load_klass(Otos_i, RobjKlass); // get value klass
  3538   // Get constant pool tag
  3539   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3541   // See if the checkcast has been quickened
  3542   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3543   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3544   __ ldub(G3_scratch, Roffset, G3_scratch);
  3545   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3546   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3547   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3549   __ push_ptr(); // save receiver for result, and for GC
  3550   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3551   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3553   __ br(Assembler::always, false, Assembler::pt, resolved);
  3554   __ delayed()->nop();
  3556   // Extract target class from constant pool
  3557   __ bind(quicked);
  3558   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3559   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3560   __ bind(resolved);
  3561   __ load_klass(Otos_i, RobjKlass); // get value klass
  3563   // Generate a fast subtype check.  Branch to cast_ok if no
  3564   // failure.  Throw exception if failure.
  3565   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
  3567   // Not a subtype; so must throw exception
  3568   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
  3570   __ bind(cast_ok);
  3572   if (ProfileInterpreter) {
  3573     __ ba(false, done);
  3574     __ delayed()->nop();
  3576   __ bind(is_null);
  3577   __ profile_null_seen(G3_scratch);
  3578   __ bind(done);
  3582 void TemplateTable::instanceof() {
  3583   Label done, is_null, quicked, resolved;
  3584   transition(atos, itos);
  3585   Register Roffset = G1_scratch;
  3586   Register RobjKlass = O5;
  3587   Register RspecifiedKlass = O4;
  3589   // Check for casting a NULL
  3590   __ br_null(Otos_i, false, Assembler::pt, is_null);
  3591   __ delayed()->nop();
  3593   // Get value klass in RobjKlass
  3594   __ load_klass(Otos_i, RobjKlass); // get value klass
  3596   // Get constant pool tag
  3597   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3599   // See if the checkcast has been quickened
  3600   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3601   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3602   __ ldub(G3_scratch, Roffset, G3_scratch);
  3603   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3604   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3605   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3607   __ push_ptr(); // save receiver for result, and for GC
  3608   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3609   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3611   __ br(Assembler::always, false, Assembler::pt, resolved);
  3612   __ delayed()->nop();
  3615   // Extract target class from constant pool
  3616   __ bind(quicked);
  3617   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3618   __ get_constant_pool(Lscratch);
  3619   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3620   __ bind(resolved);
  3621   __ load_klass(Otos_i, RobjKlass); // get value klass
  3623   // Generate a fast subtype check.  Branch to cast_ok if no
  3624   // failure.  Return 0 if failure.
  3625   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
  3626   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
  3627   // Not a subtype; return 0;
  3628   __ clr( Otos_i );
  3630   if (ProfileInterpreter) {
  3631     __ ba(false, done);
  3632     __ delayed()->nop();
  3634   __ bind(is_null);
  3635   __ profile_null_seen(G3_scratch);
  3636   __ bind(done);
  3639 void TemplateTable::_breakpoint() {
  3641    // Note: We get here even if we are single stepping..
  3642    // jbug inists on setting breakpoints at every bytecode
  3643    // even if we are in single step mode.
  3645    transition(vtos, vtos);
  3646    // get the unpatched byte code
  3647    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
  3648    __ mov(O0, Lbyte_code);
  3650    // post the breakpoint event
  3651    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
  3653    // complete the execution of original bytecode
  3654    __ dispatch_normal(vtos);
  3658 //----------------------------------------------------------------------------------------------------
  3659 // Exceptions
  3661 void TemplateTable::athrow() {
  3662   transition(atos, vtos);
  3664   // This works because exception is cached in Otos_i which is same as O0,
  3665   // which is same as what throw_exception_entry_expects
  3666   assert(Otos_i == Oexception, "see explanation above");
  3668   __ verify_oop(Otos_i);
  3669   __ null_check(Otos_i);
  3670   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
  3674 //----------------------------------------------------------------------------------------------------
  3675 // Synchronization
  3678 // See frame_sparc.hpp for monitor block layout.
  3679 // Monitor elements are dynamically allocated by growing stack as needed.
  3681 void TemplateTable::monitorenter() {
  3682   transition(atos, vtos);
  3683   __ verify_oop(Otos_i);
  3684   // Try to acquire a lock on the object
  3685   // Repeat until succeeded (i.e., until
  3686   // monitorenter returns true).
  3688   {   Label ok;
  3689     __ tst(Otos_i);
  3690     __ throw_if_not_1_x( Assembler::notZero,  ok);
  3691     __ delayed()->mov(Otos_i, Lscratch); // save obj
  3692     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3695   assert(O0 == Otos_i, "Be sure where the object to lock is");
  3697   // find a free slot in the monitor block
  3700   // initialize entry pointer
  3701   __ clr(O1); // points to free slot or NULL
  3704     Label entry, loop, exit;
  3705     __ add( __ top_most_monitor(), O2 ); // last one to check
  3706     __ ba( false, entry );
  3707     __ delayed()->mov( Lmonitors, O3 ); // first one to check
  3710     __ bind( loop );
  3712     __ verify_oop(O4);          // verify each monitor's oop
  3713     __ tst(O4); // is this entry unused?
  3714     if (VM_Version::v9_instructions_work())
  3715       __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
  3716     else {
  3717       Label L;
  3718       __ br( Assembler::zero, true, Assembler::pn, L );
  3719       __ delayed()->mov(O3, O1); // rememeber this one if match
  3720       __ bind(L);
  3723     __ cmp(O4, O0); // check if current entry is for same object
  3724     __ brx( Assembler::equal, false, Assembler::pn, exit );
  3725     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
  3727     __ bind( entry );
  3729     __ cmp( O3, O2 );
  3730     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3731     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
  3733     __ bind( exit );
  3736   { Label allocated;
  3738     // found free slot?
  3739     __ br_notnull(O1, false, Assembler::pn, allocated);
  3740     __ delayed()->nop();
  3742     __ add_monitor_to_stack( false, O2, O3 );
  3743     __ mov(Lmonitors, O1);
  3745     __ bind(allocated);
  3748   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
  3749   // The object has already been poped from the stack, so the expression stack looks correct.
  3750   __ inc(Lbcp);
  3752   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
  3753   __ lock_object(O1, O0);
  3755   // check if there's enough space on the stack for the monitors after locking
  3756   __ generate_stack_overflow_check(0);
  3758   // The bcp has already been incremented. Just need to dispatch to next instruction.
  3759   __ dispatch_next(vtos);
  3763 void TemplateTable::monitorexit() {
  3764   transition(atos, vtos);
  3765   __ verify_oop(Otos_i);
  3766   __ tst(Otos_i);
  3767   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
  3769   assert(O0 == Otos_i, "just checking");
  3771   { Label entry, loop, found;
  3772     __ add( __ top_most_monitor(), O2 ); // last one to check
  3773     __ ba(false, entry );
  3774     // use Lscratch to hold monitor elem to check, start with most recent monitor,
  3775     // By using a local it survives the call to the C routine.
  3776     __ delayed()->mov( Lmonitors, Lscratch );
  3778     __ bind( loop );
  3780     __ verify_oop(O4);          // verify each monitor's oop
  3781     __ cmp(O4, O0); // check if current entry is for desired object
  3782     __ brx( Assembler::equal, true, Assembler::pt, found );
  3783     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
  3785     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
  3787     __ bind( entry );
  3789     __ cmp( Lscratch, O2 );
  3790     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3791     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
  3793     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
  3794     __ should_not_reach_here();
  3796     __ bind(found);
  3798   __ unlock_object(O1);
  3802 //----------------------------------------------------------------------------------------------------
  3803 // Wide instructions
  3805 void TemplateTable::wide() {
  3806   transition(vtos, vtos);
  3807   __ ldub(Lbcp, 1, G3_scratch);// get next bc
  3808   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
  3809   AddressLiteral ep(Interpreter::_wentry_point);
  3810   __ set(ep, G4_scratch);
  3811   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
  3812   __ jmp(G3_scratch, G0);
  3813   __ delayed()->nop();
  3814   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
  3818 //----------------------------------------------------------------------------------------------------
  3819 // Multi arrays
  3821 void TemplateTable::multianewarray() {
  3822   transition(vtos, atos);
  3823      // put ndims * wordSize into Lscratch
  3824   __ ldub( Lbcp,     3,               Lscratch);
  3825   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
  3826      // Lesp points past last_dim, so set to O1 to first_dim address
  3827   __ add(  Lesp,     Lscratch,        O1);
  3828      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
  3829   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
  3831 #endif /* !CC_INTERP */

mercurial