src/cpu/sparc/vm/templateTable_sparc.cpp

Mon, 11 Oct 2010 04:18:58 -0700

author
twisti
date
Mon, 11 Oct 2010 04:18:58 -0700
changeset 2201
d55217dc206f
parent 2138
d5d065957597
child 2268
3b2dea75431e
permissions
-rw-r--r--

6829194: JSR 292 needs to support compressed oops
Reviewed-by: kvn, jrose

     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);
   344 }
   346 void TemplateTable::ldc2_w() {
   347   transition(vtos, vtos);
   348   Label retry, resolved, Long, exit;
   350   __ bind(retry);
   351   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   352   __ get_cpool_and_tags(O0, O2);
   354   const int base_offset = constantPoolOopDesc::header_size() * wordSize;
   355   const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
   356   // get type from tags
   357   __ add(O2, tags_offset, O2);
   358   __ ldub(O2, O1, O2);
   360   __ sll(O1, LogBytesPerWord, O1);
   361   __ add(O0, O1, G3_scratch);
   363   __ cmp(O2, JVM_CONSTANT_Double);
   364   __ brx(Assembler::notEqual, false, Assembler::pt, Long);
   365   __ delayed()->nop();
   366   // A double can be placed at word-aligned locations in the constant pool.
   367   // Check out Conversions.java for an example.
   368   // Also constantPoolOopDesc::header_size() is 20, which makes it very difficult
   369   // to double-align double on the constant pool.  SG, 11/7/97
   370 #ifdef _LP64
   371   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
   372 #else
   373   FloatRegister f = Ftos_d;
   374   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
   375   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
   376          f->successor());
   377 #endif
   378   __ push(dtos);
   379   __ ba(false, exit);
   380   __ delayed()->nop();
   382   __ bind(Long);
   383 #ifdef _LP64
   384   __ ldx(G3_scratch, base_offset, Otos_l);
   385 #else
   386   __ ld(G3_scratch, base_offset, Otos_l);
   387   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
   388 #endif
   389   __ push(ltos);
   391   __ bind(exit);
   392 }
   395 void TemplateTable::locals_index(Register reg, int offset) {
   396   __ ldub( at_bcp(offset), reg );
   397 }
   400 void TemplateTable::locals_index_wide(Register reg) {
   401   // offset is 2, not 1, because Lbcp points to wide prefix code
   402   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
   403 }
   405 void TemplateTable::iload() {
   406   transition(vtos, itos);
   407   // Rewrite iload,iload  pair into fast_iload2
   408   //         iload,caload pair into fast_icaload
   409   if (RewriteFrequentPairs) {
   410     Label rewrite, done;
   412     // get next byte
   413     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
   415     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
   416     // last two iloads in a pair.  Comparing against fast_iload means that
   417     // the next bytecode is neither an iload or a caload, and therefore
   418     // an iload pair.
   419     __ cmp(G3_scratch, (int)Bytecodes::_iload);
   420     __ br(Assembler::equal, false, Assembler::pn, done);
   421     __ delayed()->nop();
   423     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
   424     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   425     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
   427     __ cmp(G3_scratch, (int)Bytecodes::_caload);
   428     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   429     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
   431     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
   432     // rewrite
   433     // G4_scratch: fast bytecode
   434     __ bind(rewrite);
   435     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
   436     __ bind(done);
   437   }
   439   // Get the local value into tos
   440   locals_index(G3_scratch);
   441   __ access_local_int( G3_scratch, Otos_i );
   442 }
   444 void TemplateTable::fast_iload2() {
   445   transition(vtos, itos);
   446   locals_index(G3_scratch);
   447   __ access_local_int( G3_scratch, Otos_i );
   448   __ push_i();
   449   locals_index(G3_scratch, 3);  // get next bytecode's local index.
   450   __ access_local_int( G3_scratch, Otos_i );
   451 }
   453 void TemplateTable::fast_iload() {
   454   transition(vtos, itos);
   455   locals_index(G3_scratch);
   456   __ access_local_int( G3_scratch, Otos_i );
   457 }
   459 void TemplateTable::lload() {
   460   transition(vtos, ltos);
   461   locals_index(G3_scratch);
   462   __ access_local_long( G3_scratch, Otos_l );
   463 }
   466 void TemplateTable::fload() {
   467   transition(vtos, ftos);
   468   locals_index(G3_scratch);
   469   __ access_local_float( G3_scratch, Ftos_f );
   470 }
   473 void TemplateTable::dload() {
   474   transition(vtos, dtos);
   475   locals_index(G3_scratch);
   476   __ access_local_double( G3_scratch, Ftos_d );
   477 }
   480 void TemplateTable::aload() {
   481   transition(vtos, atos);
   482   locals_index(G3_scratch);
   483   __ access_local_ptr( G3_scratch, Otos_i);
   484 }
   487 void TemplateTable::wide_iload() {
   488   transition(vtos, itos);
   489   locals_index_wide(G3_scratch);
   490   __ access_local_int( G3_scratch, Otos_i );
   491 }
   494 void TemplateTable::wide_lload() {
   495   transition(vtos, ltos);
   496   locals_index_wide(G3_scratch);
   497   __ access_local_long( G3_scratch, Otos_l );
   498 }
   501 void TemplateTable::wide_fload() {
   502   transition(vtos, ftos);
   503   locals_index_wide(G3_scratch);
   504   __ access_local_float( G3_scratch, Ftos_f );
   505 }
   508 void TemplateTable::wide_dload() {
   509   transition(vtos, dtos);
   510   locals_index_wide(G3_scratch);
   511   __ access_local_double( G3_scratch, Ftos_d );
   512 }
   515 void TemplateTable::wide_aload() {
   516   transition(vtos, atos);
   517   locals_index_wide(G3_scratch);
   518   __ access_local_ptr( G3_scratch, Otos_i );
   519   __ verify_oop(Otos_i);
   520 }
   523 void TemplateTable::iaload() {
   524   transition(itos, itos);
   525   // Otos_i: index
   526   // tos: array
   527   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   528   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
   529 }
   532 void TemplateTable::laload() {
   533   transition(itos, ltos);
   534   // Otos_i: index
   535   // O2: array
   536   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   537   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
   538 }
   541 void TemplateTable::faload() {
   542   transition(itos, ftos);
   543   // Otos_i: index
   544   // O2: array
   545   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   546   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
   547 }
   550 void TemplateTable::daload() {
   551   transition(itos, dtos);
   552   // Otos_i: index
   553   // O2: array
   554   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   555   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
   556 }
   559 void TemplateTable::aaload() {
   560   transition(itos, atos);
   561   // Otos_i: index
   562   // tos: array
   563   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
   564   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   565   __ verify_oop(Otos_i);
   566 }
   569 void TemplateTable::baload() {
   570   transition(itos, itos);
   571   // Otos_i: index
   572   // tos: array
   573   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
   574   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
   575 }
   578 void TemplateTable::caload() {
   579   transition(itos, itos);
   580   // Otos_i: index
   581   // tos: array
   582   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   583   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   584 }
   586 void TemplateTable::fast_icaload() {
   587   transition(vtos, itos);
   588   // Otos_i: index
   589   // tos: array
   590   locals_index(G3_scratch);
   591   __ access_local_int( G3_scratch, Otos_i );
   592   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   593   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   594 }
   597 void TemplateTable::saload() {
   598   transition(itos, itos);
   599   // Otos_i: index
   600   // tos: array
   601   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   602   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
   603 }
   606 void TemplateTable::iload(int n) {
   607   transition(vtos, itos);
   608   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   609 }
   612 void TemplateTable::lload(int n) {
   613   transition(vtos, ltos);
   614   assert(n+1 < Argument::n_register_parameters, "would need more code");
   615   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
   616 }
   619 void TemplateTable::fload(int n) {
   620   transition(vtos, ftos);
   621   assert(n < Argument::n_register_parameters, "would need more code");
   622   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
   623 }
   626 void TemplateTable::dload(int n) {
   627   transition(vtos, dtos);
   628   FloatRegister dst = Ftos_d;
   629   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
   630 }
   633 void TemplateTable::aload(int n) {
   634   transition(vtos, atos);
   635   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   636 }
   639 void TemplateTable::aload_0() {
   640   transition(vtos, atos);
   642   // According to bytecode histograms, the pairs:
   643   //
   644   // _aload_0, _fast_igetfield (itos)
   645   // _aload_0, _fast_agetfield (atos)
   646   // _aload_0, _fast_fgetfield (ftos)
   647   //
   648   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
   649   // bytecode checks the next bytecode and then rewrites the current
   650   // bytecode into a pair bytecode; otherwise it rewrites the current
   651   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
   652   //
   653   if (RewriteFrequentPairs) {
   654     Label rewrite, done;
   656     // get next byte
   657     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
   659     // do actual aload_0
   660     aload(0);
   662     // if _getfield then wait with rewrite
   663     __ cmp(G3_scratch, (int)Bytecodes::_getfield);
   664     __ br(Assembler::equal, false, Assembler::pn, done);
   665     __ delayed()->nop();
   667     // if _igetfield then rewrite to _fast_iaccess_0
   668     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   669     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
   670     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   671     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
   673     // if _agetfield then rewrite to _fast_aaccess_0
   674     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   675     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
   676     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   677     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
   679     // if _fgetfield then rewrite to _fast_faccess_0
   680     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   681     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
   682     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   683     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
   685     // else rewrite to _fast_aload0
   686     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   687     __ set(Bytecodes::_fast_aload_0, G4_scratch);
   689     // rewrite
   690     // G4_scratch: fast bytecode
   691     __ bind(rewrite);
   692     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
   693     __ bind(done);
   694   } else {
   695     aload(0);
   696   }
   697 }
   700 void TemplateTable::istore() {
   701   transition(itos, vtos);
   702   locals_index(G3_scratch);
   703   __ store_local_int( G3_scratch, Otos_i );
   704 }
   707 void TemplateTable::lstore() {
   708   transition(ltos, vtos);
   709   locals_index(G3_scratch);
   710   __ store_local_long( G3_scratch, Otos_l );
   711 }
   714 void TemplateTable::fstore() {
   715   transition(ftos, vtos);
   716   locals_index(G3_scratch);
   717   __ store_local_float( G3_scratch, Ftos_f );
   718 }
   721 void TemplateTable::dstore() {
   722   transition(dtos, vtos);
   723   locals_index(G3_scratch);
   724   __ store_local_double( G3_scratch, Ftos_d );
   725 }
   728 void TemplateTable::astore() {
   729   transition(vtos, vtos);
   730   __ load_ptr(0, Otos_i);
   731   __ inc(Lesp, Interpreter::stackElementSize);
   732   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   733   locals_index(G3_scratch);
   734   __ store_local_ptr(G3_scratch, Otos_i);
   735 }
   738 void TemplateTable::wide_istore() {
   739   transition(vtos, vtos);
   740   __ pop_i();
   741   locals_index_wide(G3_scratch);
   742   __ store_local_int( G3_scratch, Otos_i );
   743 }
   746 void TemplateTable::wide_lstore() {
   747   transition(vtos, vtos);
   748   __ pop_l();
   749   locals_index_wide(G3_scratch);
   750   __ store_local_long( G3_scratch, Otos_l );
   751 }
   754 void TemplateTable::wide_fstore() {
   755   transition(vtos, vtos);
   756   __ pop_f();
   757   locals_index_wide(G3_scratch);
   758   __ store_local_float( G3_scratch, Ftos_f );
   759 }
   762 void TemplateTable::wide_dstore() {
   763   transition(vtos, vtos);
   764   __ pop_d();
   765   locals_index_wide(G3_scratch);
   766   __ store_local_double( G3_scratch, Ftos_d );
   767 }
   770 void TemplateTable::wide_astore() {
   771   transition(vtos, vtos);
   772   __ load_ptr(0, Otos_i);
   773   __ inc(Lesp, Interpreter::stackElementSize);
   774   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   775   locals_index_wide(G3_scratch);
   776   __ store_local_ptr(G3_scratch, Otos_i);
   777 }
   780 void TemplateTable::iastore() {
   781   transition(itos, vtos);
   782   __ pop_i(O2); // index
   783   // Otos_i: val
   784   // O3: array
   785   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   786   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
   787 }
   790 void TemplateTable::lastore() {
   791   transition(ltos, vtos);
   792   __ pop_i(O2); // index
   793   // Otos_l: val
   794   // O3: array
   795   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   796   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
   797 }
   800 void TemplateTable::fastore() {
   801   transition(ftos, vtos);
   802   __ pop_i(O2); // index
   803   // Ftos_f: val
   804   // O3: array
   805   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   806   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
   807 }
   810 void TemplateTable::dastore() {
   811   transition(dtos, vtos);
   812   __ pop_i(O2); // index
   813   // Fos_d: val
   814   // O3: array
   815   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   816   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
   817 }
   820 void TemplateTable::aastore() {
   821   Label store_ok, is_null, done;
   822   transition(vtos, vtos);
   823   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
   824   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
   825   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
   826   // Otos_i: val
   827   // O2: index
   828   // O3: array
   829   __ verify_oop(Otos_i);
   830   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
   832   // do array store check - check for NULL value first
   833   __ br_null( Otos_i, false, Assembler::pn, is_null );
   834   __ delayed()->nop();
   836   __ load_klass(O3, O4); // get array klass
   837   __ load_klass(Otos_i, O5); // get value klass
   839   // do fast instanceof cache test
   841   __ ld_ptr(O4,     sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes(),  O4);
   843   assert(Otos_i == O0, "just checking");
   845   // Otos_i:    value
   846   // O1:        addr - offset
   847   // O2:        index
   848   // O3:        array
   849   // O4:        array element klass
   850   // O5:        value klass
   852   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
   854   // Generate a fast subtype check.  Branch to store_ok if no
   855   // failure.  Throw if failure.
   856   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
   858   // Not a subtype; so must throw exception
   859   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
   861   // Store is OK.
   862   __ bind(store_ok);
   863   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
   865   __ ba(false,done);
   866   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
   868   __ bind(is_null);
   869   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
   871   __ profile_null_seen(G3_scratch);
   872   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
   873   __ bind(done);
   874 }
   877 void TemplateTable::bastore() {
   878   transition(itos, vtos);
   879   __ pop_i(O2); // index
   880   // Otos_i: val
   881   // O3: array
   882   __ index_check(O3, O2, 0, G3_scratch, O2);
   883   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
   884 }
   887 void TemplateTable::castore() {
   888   transition(itos, vtos);
   889   __ pop_i(O2); // index
   890   // Otos_i: val
   891   // O3: array
   892   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
   893   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
   894 }
   897 void TemplateTable::sastore() {
   898   // %%%%% Factor across platform
   899   castore();
   900 }
   903 void TemplateTable::istore(int n) {
   904   transition(itos, vtos);
   905   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
   906 }
   909 void TemplateTable::lstore(int n) {
   910   transition(ltos, vtos);
   911   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
   912   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
   914 }
   917 void TemplateTable::fstore(int n) {
   918   transition(ftos, vtos);
   919   assert(n < Argument::n_register_parameters, "only handle register cases");
   920   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
   921 }
   924 void TemplateTable::dstore(int n) {
   925   transition(dtos, vtos);
   926   FloatRegister src = Ftos_d;
   927   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
   928 }
   931 void TemplateTable::astore(int n) {
   932   transition(vtos, vtos);
   933   __ load_ptr(0, Otos_i);
   934   __ inc(Lesp, Interpreter::stackElementSize);
   935   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   936   __ store_local_ptr(n, Otos_i);
   937 }
   940 void TemplateTable::pop() {
   941   transition(vtos, vtos);
   942   __ inc(Lesp, Interpreter::stackElementSize);
   943 }
   946 void TemplateTable::pop2() {
   947   transition(vtos, vtos);
   948   __ inc(Lesp, 2 * Interpreter::stackElementSize);
   949 }
   952 void TemplateTable::dup() {
   953   transition(vtos, vtos);
   954   // stack: ..., a
   955   // load a and tag
   956   __ load_ptr(0, Otos_i);
   957   __ push_ptr(Otos_i);
   958   // stack: ..., a, a
   959 }
   962 void TemplateTable::dup_x1() {
   963   transition(vtos, vtos);
   964   // stack: ..., a, b
   965   __ load_ptr( 1, G3_scratch);  // get a
   966   __ load_ptr( 0, Otos_l1);     // get b
   967   __ store_ptr(1, Otos_l1);     // put b
   968   __ store_ptr(0, G3_scratch);  // put a - like swap
   969   __ push_ptr(Otos_l1);         // push b
   970   // stack: ..., b, a, b
   971 }
   974 void TemplateTable::dup_x2() {
   975   transition(vtos, vtos);
   976   // stack: ..., a, b, c
   977   // get c and push on stack, reuse registers
   978   __ load_ptr( 0, G3_scratch);  // get c
   979   __ push_ptr(G3_scratch);      // push c with tag
   980   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
   981   // (stack offsets n+1 now)
   982   __ load_ptr( 3, Otos_l1);     // get a
   983   __ store_ptr(3, G3_scratch);  // put c at 3
   984   // stack: ..., c, b, c, c  (a in reg)
   985   __ load_ptr( 2, G3_scratch);  // get b
   986   __ store_ptr(2, Otos_l1);     // put a at 2
   987   // stack: ..., c, a, c, c  (b in reg)
   988   __ store_ptr(1, G3_scratch);  // put b at 1
   989   // stack: ..., c, a, b, c
   990 }
   993 void TemplateTable::dup2() {
   994   transition(vtos, vtos);
   995   __ load_ptr(1, G3_scratch);  // get a
   996   __ load_ptr(0, Otos_l1);     // get b
   997   __ push_ptr(G3_scratch);     // push a
   998   __ push_ptr(Otos_l1);        // push b
   999   // stack: ..., a, b, a, b
  1003 void TemplateTable::dup2_x1() {
  1004   transition(vtos, vtos);
  1005   // stack: ..., a, b, c
  1006   __ load_ptr( 1, Lscratch);    // get b
  1007   __ load_ptr( 2, Otos_l1);     // get a
  1008   __ store_ptr(2, Lscratch);    // put b at a
  1009   // stack: ..., b, b, c
  1010   __ load_ptr( 0, G3_scratch);  // get c
  1011   __ store_ptr(1, G3_scratch);  // put c at b
  1012   // stack: ..., b, c, c
  1013   __ store_ptr(0, Otos_l1);     // put a at c
  1014   // stack: ..., b, c, a
  1015   __ push_ptr(Lscratch);        // push b
  1016   __ push_ptr(G3_scratch);      // push c
  1017   // stack: ..., b, c, a, b, c
  1021 // The spec says that these types can be a mixture of category 1 (1 word)
  1022 // types and/or category 2 types (long and doubles)
  1023 void TemplateTable::dup2_x2() {
  1024   transition(vtos, vtos);
  1025   // stack: ..., a, b, c, d
  1026   __ load_ptr( 1, Lscratch);    // get c
  1027   __ load_ptr( 3, Otos_l1);     // get a
  1028   __ store_ptr(3, Lscratch);    // put c at 3
  1029   __ store_ptr(1, Otos_l1);     // put a at 1
  1030   // stack: ..., c, b, a, d
  1031   __ load_ptr( 2, G3_scratch);  // get b
  1032   __ load_ptr( 0, Otos_l1);     // get d
  1033   __ store_ptr(0, G3_scratch);  // put b at 0
  1034   __ store_ptr(2, Otos_l1);     // put d at 2
  1035   // stack: ..., c, d, a, b
  1036   __ push_ptr(Lscratch);        // push c
  1037   __ push_ptr(Otos_l1);         // push d
  1038   // stack: ..., c, d, a, b, c, d
  1042 void TemplateTable::swap() {
  1043   transition(vtos, vtos);
  1044   // stack: ..., a, b
  1045   __ load_ptr( 1, G3_scratch);  // get a
  1046   __ load_ptr( 0, Otos_l1);     // get b
  1047   __ store_ptr(0, G3_scratch);  // put b
  1048   __ store_ptr(1, Otos_l1);     // put a
  1049   // stack: ..., b, a
  1053 void TemplateTable::iop2(Operation op) {
  1054   transition(itos, itos);
  1055   __ pop_i(O1);
  1056   switch (op) {
  1057    case  add:  __  add(O1, Otos_i, Otos_i);  break;
  1058    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
  1059      // %%%%% Mul may not exist: better to call .mul?
  1060    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
  1061    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
  1062    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
  1063    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
  1064    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
  1065    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
  1066    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
  1067    default: ShouldNotReachHere();
  1072 void TemplateTable::lop2(Operation op) {
  1073   transition(ltos, ltos);
  1074   __ pop_l(O2);
  1075   switch (op) {
  1076 #ifdef _LP64
  1077    case  add:  __  add(O2, Otos_l, Otos_l);  break;
  1078    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
  1079    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
  1080    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
  1081    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
  1082 #else
  1083    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
  1084    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
  1085    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
  1086    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
  1087    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
  1088 #endif
  1089    default: ShouldNotReachHere();
  1094 void TemplateTable::idiv() {
  1095   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
  1096   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
  1098   transition(itos, itos);
  1099   __ pop_i(O1); // get 1st op
  1101   // Y contains upper 32 bits of result, set it to 0 or all ones
  1102   __ wry(G0);
  1103   __ mov(~0, G3_scratch);
  1105   __ tst(O1);
  1106      Label neg;
  1107   __ br(Assembler::negative, true, Assembler::pn, neg);
  1108   __ delayed()->wry(G3_scratch);
  1109   __ bind(neg);
  1111      Label ok;
  1112   __ tst(Otos_i);
  1113   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
  1115   const int min_int = 0x80000000;
  1116   Label regular;
  1117   __ cmp(Otos_i, -1);
  1118   __ br(Assembler::notEqual, false, Assembler::pt, regular);
  1119 #ifdef _LP64
  1120   // Don't put set in delay slot
  1121   // Set will turn into multiple instructions in 64 bit mode
  1122   __ delayed()->nop();
  1123   __ set(min_int, G4_scratch);
  1124 #else
  1125   __ delayed()->set(min_int, G4_scratch);
  1126 #endif
  1127   Label done;
  1128   __ cmp(O1, G4_scratch);
  1129   __ br(Assembler::equal, true, Assembler::pt, done);
  1130   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
  1132   __ bind(regular);
  1133   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
  1134   __ bind(done);
  1138 void TemplateTable::irem() {
  1139   transition(itos, itos);
  1140   __ mov(Otos_i, O2); // save divisor
  1141   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
  1142   __ smul(Otos_i, O2, Otos_i);
  1143   __ sub(O1, Otos_i, Otos_i);
  1147 void TemplateTable::lmul() {
  1148   transition(ltos, ltos);
  1149   __ pop_l(O2);
  1150 #ifdef _LP64
  1151   __ mulx(Otos_l, O2, Otos_l);
  1152 #else
  1153   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
  1154 #endif
  1159 void TemplateTable::ldiv() {
  1160   transition(ltos, ltos);
  1162   // check for zero
  1163   __ pop_l(O2);
  1164 #ifdef _LP64
  1165   __ tst(Otos_l);
  1166   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1167   __ sdivx(O2, Otos_l, Otos_l);
  1168 #else
  1169   __ orcc(Otos_l1, Otos_l2, G0);
  1170   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1171   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
  1172 #endif
  1176 void TemplateTable::lrem() {
  1177   transition(ltos, ltos);
  1179   // check for zero
  1180   __ pop_l(O2);
  1181 #ifdef _LP64
  1182   __ tst(Otos_l);
  1183   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1184   __ sdivx(O2, Otos_l, Otos_l2);
  1185   __ mulx (Otos_l2, Otos_l, Otos_l2);
  1186   __ sub  (O2, Otos_l2, Otos_l);
  1187 #else
  1188   __ orcc(Otos_l1, Otos_l2, G0);
  1189   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1190   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
  1191 #endif
  1195 void TemplateTable::lshl() {
  1196   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
  1198   __ pop_l(O2);                          // shift value in O2, O3
  1199 #ifdef _LP64
  1200   __ sllx(O2, Otos_i, Otos_l);
  1201 #else
  1202   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1203 #endif
  1207 void TemplateTable::lshr() {
  1208   transition(itos, ltos); // %%%% see lshl comment
  1210   __ pop_l(O2);                          // shift value in O2, O3
  1211 #ifdef _LP64
  1212   __ srax(O2, Otos_i, Otos_l);
  1213 #else
  1214   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1215 #endif
  1220 void TemplateTable::lushr() {
  1221   transition(itos, ltos); // %%%% see lshl comment
  1223   __ pop_l(O2);                          // shift value in O2, O3
  1224 #ifdef _LP64
  1225   __ srlx(O2, Otos_i, Otos_l);
  1226 #else
  1227   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1228 #endif
  1232 void TemplateTable::fop2(Operation op) {
  1233   transition(ftos, ftos);
  1234   switch (op) {
  1235    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1236    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1237    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1238    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1239    case  rem:
  1240      assert(Ftos_f == F0, "just checking");
  1241 #ifdef _LP64
  1242      // LP64 calling conventions use F1, F3 for passing 2 floats
  1243      __ pop_f(F1);
  1244      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
  1245 #else
  1246      __ pop_i(O0);
  1247      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
  1248      __ ld( __ d_tmp, O1 );
  1249 #endif
  1250      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
  1251      assert( Ftos_f == F0, "fix this code" );
  1252      break;
  1254    default: ShouldNotReachHere();
  1259 void TemplateTable::dop2(Operation op) {
  1260   transition(dtos, dtos);
  1261   switch (op) {
  1262    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1263    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1264    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1265    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1266    case  rem:
  1267 #ifdef _LP64
  1268      // Pass arguments in D0, D2
  1269      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
  1270      __ pop_d( F0 );
  1271 #else
  1272      // Pass arguments in O0O1, O2O3
  1273      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1274      __ ldd( __ d_tmp, O2 );
  1275      __ pop_d(Ftos_f);
  1276      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1277      __ ldd( __ d_tmp, O0 );
  1278 #endif
  1279      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
  1280      assert( Ftos_d == F0, "fix this code" );
  1281      break;
  1283    default: ShouldNotReachHere();
  1288 void TemplateTable::ineg() {
  1289   transition(itos, itos);
  1290   __ neg(Otos_i);
  1294 void TemplateTable::lneg() {
  1295   transition(ltos, ltos);
  1296 #ifdef _LP64
  1297   __ sub(G0, Otos_l, Otos_l);
  1298 #else
  1299   __ lneg(Otos_l1, Otos_l2);
  1300 #endif
  1304 void TemplateTable::fneg() {
  1305   transition(ftos, ftos);
  1306   __ fneg(FloatRegisterImpl::S, Ftos_f);
  1310 void TemplateTable::dneg() {
  1311   transition(dtos, dtos);
  1312   // v8 has fnegd if source and dest are the same
  1313   __ fneg(FloatRegisterImpl::D, Ftos_f);
  1317 void TemplateTable::iinc() {
  1318   transition(vtos, vtos);
  1319   locals_index(G3_scratch);
  1320   __ ldsb(Lbcp, 2, O2);  // load constant
  1321   __ access_local_int(G3_scratch, Otos_i);
  1322   __ add(Otos_i, O2, Otos_i);
  1323   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1327 void TemplateTable::wide_iinc() {
  1328   transition(vtos, vtos);
  1329   locals_index_wide(G3_scratch);
  1330   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
  1331   __ access_local_int(G3_scratch, Otos_i);
  1332   __ add(Otos_i, O3, Otos_i);
  1333   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1337 void TemplateTable::convert() {
  1338 // %%%%% Factor this first part accross platforms
  1339   #ifdef ASSERT
  1340     TosState tos_in  = ilgl;
  1341     TosState tos_out = ilgl;
  1342     switch (bytecode()) {
  1343       case Bytecodes::_i2l: // fall through
  1344       case Bytecodes::_i2f: // fall through
  1345       case Bytecodes::_i2d: // fall through
  1346       case Bytecodes::_i2b: // fall through
  1347       case Bytecodes::_i2c: // fall through
  1348       case Bytecodes::_i2s: tos_in = itos; break;
  1349       case Bytecodes::_l2i: // fall through
  1350       case Bytecodes::_l2f: // fall through
  1351       case Bytecodes::_l2d: tos_in = ltos; break;
  1352       case Bytecodes::_f2i: // fall through
  1353       case Bytecodes::_f2l: // fall through
  1354       case Bytecodes::_f2d: tos_in = ftos; break;
  1355       case Bytecodes::_d2i: // fall through
  1356       case Bytecodes::_d2l: // fall through
  1357       case Bytecodes::_d2f: tos_in = dtos; break;
  1358       default             : ShouldNotReachHere();
  1360     switch (bytecode()) {
  1361       case Bytecodes::_l2i: // fall through
  1362       case Bytecodes::_f2i: // fall through
  1363       case Bytecodes::_d2i: // fall through
  1364       case Bytecodes::_i2b: // fall through
  1365       case Bytecodes::_i2c: // fall through
  1366       case Bytecodes::_i2s: tos_out = itos; break;
  1367       case Bytecodes::_i2l: // fall through
  1368       case Bytecodes::_f2l: // fall through
  1369       case Bytecodes::_d2l: tos_out = ltos; break;
  1370       case Bytecodes::_i2f: // fall through
  1371       case Bytecodes::_l2f: // fall through
  1372       case Bytecodes::_d2f: tos_out = ftos; break;
  1373       case Bytecodes::_i2d: // fall through
  1374       case Bytecodes::_l2d: // fall through
  1375       case Bytecodes::_f2d: tos_out = dtos; break;
  1376       default             : ShouldNotReachHere();
  1378     transition(tos_in, tos_out);
  1379   #endif
  1382   // Conversion
  1383   Label done;
  1384   switch (bytecode()) {
  1385    case Bytecodes::_i2l:
  1386 #ifdef _LP64
  1387     // Sign extend the 32 bits
  1388     __ sra ( Otos_i, 0, Otos_l );
  1389 #else
  1390     __ addcc(Otos_i, 0, Otos_l2);
  1391     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
  1392     __ delayed()->clr(Otos_l1);
  1393     __ set(~0, Otos_l1);
  1394 #endif
  1395     break;
  1397    case Bytecodes::_i2f:
  1398     __ st(Otos_i, __ d_tmp );
  1399     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1400     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
  1401     break;
  1403    case Bytecodes::_i2d:
  1404     __ st(Otos_i, __ d_tmp);
  1405     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1406     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
  1407     break;
  1409    case Bytecodes::_i2b:
  1410     __ sll(Otos_i, 24, Otos_i);
  1411     __ sra(Otos_i, 24, Otos_i);
  1412     break;
  1414    case Bytecodes::_i2c:
  1415     __ sll(Otos_i, 16, Otos_i);
  1416     __ srl(Otos_i, 16, Otos_i);
  1417     break;
  1419    case Bytecodes::_i2s:
  1420     __ sll(Otos_i, 16, Otos_i);
  1421     __ sra(Otos_i, 16, Otos_i);
  1422     break;
  1424    case Bytecodes::_l2i:
  1425 #ifndef _LP64
  1426     __ mov(Otos_l2, Otos_i);
  1427 #else
  1428     // Sign-extend into the high 32 bits
  1429     __ sra(Otos_l, 0, Otos_i);
  1430 #endif
  1431     break;
  1433    case Bytecodes::_l2f:
  1434    case Bytecodes::_l2d:
  1435     __ st_long(Otos_l, __ d_tmp);
  1436     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
  1438     if (VM_Version::v9_instructions_work()) {
  1439       if (bytecode() == Bytecodes::_l2f) {
  1440         __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1441       } else {
  1442         __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
  1444     } else {
  1445       __ call_VM_leaf(
  1446         Lscratch,
  1447         bytecode() == Bytecodes::_l2f
  1448           ? CAST_FROM_FN_PTR(address, SharedRuntime::l2f)
  1449           : CAST_FROM_FN_PTR(address, SharedRuntime::l2d)
  1450       );
  1452     break;
  1454   case Bytecodes::_f2i:  {
  1455       Label isNaN;
  1456       // result must be 0 if value is NaN; test by comparing value to itself
  1457       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
  1458       // According to the v8 manual, you have to have a non-fp instruction
  1459       // between fcmp and fb.
  1460       if (!VM_Version::v9_instructions_work()) {
  1461         __ nop();
  1463       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
  1464       __ delayed()->clr(Otos_i);                                     // NaN
  1465       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
  1466       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
  1467       __ ld(__ d_tmp, Otos_i);
  1468       __ bind(isNaN);
  1470     break;
  1472    case Bytecodes::_f2l:
  1473     // must uncache tos
  1474     __ push_f();
  1475 #ifdef _LP64
  1476     __ pop_f(F1);
  1477 #else
  1478     __ pop_i(O0);
  1479 #endif
  1480     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
  1481     break;
  1483    case Bytecodes::_f2d:
  1484     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
  1485     break;
  1487    case Bytecodes::_d2i:
  1488    case Bytecodes::_d2l:
  1489     // must uncache tos
  1490     __ push_d();
  1491 #ifdef _LP64
  1492     // LP64 calling conventions pass first double arg in D0
  1493     __ pop_d( Ftos_d );
  1494 #else
  1495     __ pop_i( O0 );
  1496     __ pop_i( O1 );
  1497 #endif
  1498     __ call_VM_leaf(Lscratch,
  1499         bytecode() == Bytecodes::_d2i
  1500           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
  1501           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
  1502     break;
  1504     case Bytecodes::_d2f:
  1505     if (VM_Version::v9_instructions_work()) {
  1506       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1508     else {
  1509       // must uncache tos
  1510       __ push_d();
  1511       __ pop_i(O0);
  1512       __ pop_i(O1);
  1513       __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::d2f));
  1515     break;
  1517     default: ShouldNotReachHere();
  1519   __ bind(done);
  1523 void TemplateTable::lcmp() {
  1524   transition(ltos, itos);
  1526 #ifdef _LP64
  1527   __ pop_l(O1); // pop off value 1, value 2 is in O0
  1528   __ lcmp( O1, Otos_l, Otos_i );
  1529 #else
  1530   __ pop_l(O2); // cmp O2,3 to O0,1
  1531   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
  1532 #endif
  1536 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
  1538   if (is_float) __ pop_f(F2);
  1539   else          __ pop_d(F2);
  1541   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
  1543   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
  1546 void TemplateTable::branch(bool is_jsr, bool is_wide) {
  1547   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
  1548   __ verify_oop(Lmethod);
  1549   __ verify_thread();
  1551   const Register O2_bumped_count = O2;
  1552   __ profile_taken_branch(G3_scratch, O2_bumped_count);
  1554   // get (wide) offset to O1_disp
  1555   const Register O1_disp = O1;
  1556   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
  1557   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
  1559   // Handle all the JSR stuff here, then exit.
  1560   // It's much shorter and cleaner than intermingling with the
  1561   // non-JSR normal-branch stuff occurring below.
  1562   if( is_jsr ) {
  1563     // compute return address as bci in Otos_i
  1564     __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1565     __ sub(Lbcp, G3_scratch, G3_scratch);
  1566     __ sub(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
  1568     // Bump Lbcp to target of JSR
  1569     __ add(Lbcp, O1_disp, Lbcp);
  1570     // Push returnAddress for "ret" on stack
  1571     __ push_ptr(Otos_i);
  1572     // And away we go!
  1573     __ dispatch_next(vtos);
  1574     return;
  1577   // Normal (non-jsr) branch handling
  1579   // Save the current Lbcp
  1580   const Register O0_cur_bcp = O0;
  1581   __ mov( Lbcp, O0_cur_bcp );
  1584   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
  1585   if ( increment_invocation_counter_for_backward_branches ) {
  1586     Label Lforward;
  1587     // check branch direction
  1588     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
  1589     // Bump bytecode pointer by displacement (take the branch)
  1590     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
  1592     if (TieredCompilation) {
  1593       Label Lno_mdo, Loverflow;
  1594       int increment = InvocationCounter::count_increment;
  1595       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
  1596       if (ProfileInterpreter) {
  1597         // If no method data exists, go to profile_continue.
  1598         __ ld_ptr(Lmethod, methodOopDesc::method_data_offset(), G4_scratch);
  1599         __ br_null(G4_scratch, false, Assembler::pn, Lno_mdo);
  1600         __ delayed()->nop();
  1602         // Increment backedge counter in the MDO
  1603         Address mdo_backedge_counter(G4_scratch, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
  1604                                                  in_bytes(InvocationCounter::counter_offset()));
  1605         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, Lscratch,
  1606                                    Assembler::notZero, &Lforward);
  1607         __ ba(false, Loverflow);
  1608         __ delayed()->nop();
  1611       // If there's no MDO, increment counter in methodOop
  1612       __ bind(Lno_mdo);
  1613       Address backedge_counter(Lmethod, in_bytes(methodOopDesc::backedge_counter_offset()) +
  1614                                         in_bytes(InvocationCounter::counter_offset()));
  1615       __ increment_mask_and_jump(backedge_counter, increment, mask, G3_scratch, Lscratch,
  1616                                  Assembler::notZero, &Lforward);
  1617       __ bind(Loverflow);
  1619       // notify point for loop, pass branch bytecode
  1620       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O0_cur_bcp);
  1622       // Was an OSR adapter generated?
  1623       // O0 = osr nmethod
  1624       __ br_null(O0, false, Assembler::pn, Lforward);
  1625       __ delayed()->nop();
  1627       // Has the nmethod been invalidated already?
  1628       __ ld(O0, nmethod::entry_bci_offset(), O2);
  1629       __ cmp(O2, InvalidOSREntryBci);
  1630       __ br(Assembler::equal, false, Assembler::pn, Lforward);
  1631       __ delayed()->nop();
  1633       // migrate the interpreter frame off of the stack
  1635       __ mov(G2_thread, L7);
  1636       // save nmethod
  1637       __ mov(O0, L6);
  1638       __ set_last_Java_frame(SP, noreg);
  1639       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
  1640       __ reset_last_Java_frame();
  1641       __ mov(L7, G2_thread);
  1643       // move OSR nmethod to I1
  1644       __ mov(L6, I1);
  1646       // OSR buffer to I0
  1647       __ mov(O0, I0);
  1649       // remove the interpreter frame
  1650       __ restore(I5_savedSP, 0, SP);
  1652       // Jump to the osr code.
  1653       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
  1654       __ jmp(O2, G0);
  1655       __ delayed()->nop();
  1657     } else {
  1658       // Update Backedge branch separately from invocations
  1659       const Register G4_invoke_ctr = G4;
  1660       __ increment_backedge_counter(G4_invoke_ctr, G1_scratch);
  1661       if (ProfileInterpreter) {
  1662         __ test_invocation_counter_for_mdp(G4_invoke_ctr, Lbcp, G3_scratch, Lforward);
  1663         if (UseOnStackReplacement) {
  1664           __ test_backedge_count_for_osr(O2_bumped_count, O0_cur_bcp, G3_scratch);
  1666       } else {
  1667         if (UseOnStackReplacement) {
  1668           __ test_backedge_count_for_osr(G4_invoke_ctr, O0_cur_bcp, G3_scratch);
  1673     __ bind(Lforward);
  1674   } else
  1675     // Bump bytecode pointer by displacement (take the branch)
  1676     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
  1678   // continue with bytecode @ target
  1679   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
  1680   // %%%%% and changing dispatch_next to dispatch_only
  1681   __ dispatch_next(vtos);
  1685 // Note Condition in argument is TemplateTable::Condition
  1686 // arg scope is within class scope
  1688 void TemplateTable::if_0cmp(Condition cc) {
  1689   // no pointers, integer only!
  1690   transition(itos, vtos);
  1691   // assume branch is more often taken than not (loops use backward branches)
  1692   __ cmp( Otos_i, 0);
  1693   __ if_cmp(ccNot(cc), false);
  1697 void TemplateTable::if_icmp(Condition cc) {
  1698   transition(itos, vtos);
  1699   __ pop_i(O1);
  1700   __ cmp(O1, Otos_i);
  1701   __ if_cmp(ccNot(cc), false);
  1705 void TemplateTable::if_nullcmp(Condition cc) {
  1706   transition(atos, vtos);
  1707   __ tst(Otos_i);
  1708   __ if_cmp(ccNot(cc), true);
  1712 void TemplateTable::if_acmp(Condition cc) {
  1713   transition(atos, vtos);
  1714   __ pop_ptr(O1);
  1715   __ verify_oop(O1);
  1716   __ verify_oop(Otos_i);
  1717   __ cmp(O1, Otos_i);
  1718   __ if_cmp(ccNot(cc), true);
  1723 void TemplateTable::ret() {
  1724   transition(vtos, vtos);
  1725   locals_index(G3_scratch);
  1726   __ access_local_returnAddress(G3_scratch, Otos_i);
  1727   // Otos_i contains the bci, compute the bcp from that
  1729 #ifdef _LP64
  1730 #ifdef ASSERT
  1731   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
  1732   // the result.  The return address (really a BCI) was stored with an
  1733   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
  1734   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
  1735   // loaded value.
  1736   { Label zzz ;
  1737      __ set (65536, G3_scratch) ;
  1738      __ cmp (Otos_i, G3_scratch) ;
  1739      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
  1740      __ delayed()->nop();
  1741      __ stop("BCI is in the wrong register half?");
  1742      __ bind (zzz) ;
  1744 #endif
  1745 #endif
  1747   __ profile_ret(vtos, Otos_i, G4_scratch);
  1749   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1750   __ add(G3_scratch, Otos_i, G3_scratch);
  1751   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1752   __ dispatch_next(vtos);
  1756 void TemplateTable::wide_ret() {
  1757   transition(vtos, vtos);
  1758   locals_index_wide(G3_scratch);
  1759   __ access_local_returnAddress(G3_scratch, Otos_i);
  1760   // Otos_i contains the bci, compute the bcp from that
  1762   __ profile_ret(vtos, Otos_i, G4_scratch);
  1764   __ ld_ptr(Lmethod, methodOopDesc::const_offset(), G3_scratch);
  1765   __ add(G3_scratch, Otos_i, G3_scratch);
  1766   __ add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), Lbcp);
  1767   __ dispatch_next(vtos);
  1771 void TemplateTable::tableswitch() {
  1772   transition(itos, vtos);
  1773   Label default_case, continue_execution;
  1775   // align bcp
  1776   __ add(Lbcp, BytesPerInt, O1);
  1777   __ and3(O1, -BytesPerInt, O1);
  1778   // load lo, hi
  1779   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
  1780   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
  1781 #ifdef _LP64
  1782   // Sign extend the 32 bits
  1783   __ sra ( Otos_i, 0, Otos_i );
  1784 #endif /* _LP64 */
  1786   // check against lo & hi
  1787   __ cmp( Otos_i, O2);
  1788   __ br( Assembler::less, false, Assembler::pn, default_case);
  1789   __ delayed()->cmp( Otos_i, O3 );
  1790   __ br( Assembler::greater, false, Assembler::pn, default_case);
  1791   // lookup dispatch offset
  1792   __ delayed()->sub(Otos_i, O2, O2);
  1793   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
  1794   __ sll(O2, LogBytesPerInt, O2);
  1795   __ add(O2, 3 * BytesPerInt, O2);
  1796   __ ba(false, continue_execution);
  1797   __ delayed()->ld(O1, O2, O2);
  1798   // handle default
  1799   __ bind(default_case);
  1800   __ profile_switch_default(O3);
  1801   __ ld(O1, 0, O2); // get default offset
  1802   // continue execution
  1803   __ bind(continue_execution);
  1804   __ add(Lbcp, O2, Lbcp);
  1805   __ dispatch_next(vtos);
  1809 void TemplateTable::lookupswitch() {
  1810   transition(itos, itos);
  1811   __ stop("lookupswitch bytecode should have been rewritten");
  1814 void TemplateTable::fast_linearswitch() {
  1815   transition(itos, vtos);
  1816     Label loop_entry, loop, found, continue_execution;
  1817   // align bcp
  1818   __ add(Lbcp, BytesPerInt, O1);
  1819   __ and3(O1, -BytesPerInt, O1);
  1820  // set counter
  1821   __ ld(O1, BytesPerInt, O2);
  1822   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
  1823   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
  1824   __ ba(false, loop_entry);
  1825   __ delayed()->add(O3, O2, O2); // counter now points past last pair
  1827   // table search
  1828   __ bind(loop);
  1829   __ cmp(O4, Otos_i);
  1830   __ br(Assembler::equal, true, Assembler::pn, found);
  1831   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
  1832   __ inc(O3, 2 * BytesPerInt);
  1834   __ bind(loop_entry);
  1835   __ cmp(O2, O3);
  1836   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
  1837   __ delayed()->ld(O3, 0, O4);
  1839   // default case
  1840   __ ld(O1, 0, O4); // get default offset
  1841   if (ProfileInterpreter) {
  1842     __ profile_switch_default(O3);
  1843     __ ba(false, continue_execution);
  1844     __ delayed()->nop();
  1847   // entry found -> get offset
  1848   __ bind(found);
  1849   if (ProfileInterpreter) {
  1850     __ sub(O3, O1, O3);
  1851     __ sub(O3, 2*BytesPerInt, O3);
  1852     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
  1853     __ profile_switch_case(O3, O1, O2, G3_scratch);
  1855     __ bind(continue_execution);
  1857   __ add(Lbcp, O4, Lbcp);
  1858   __ dispatch_next(vtos);
  1862 void TemplateTable::fast_binaryswitch() {
  1863   transition(itos, vtos);
  1864   // Implementation using the following core algorithm: (copied from Intel)
  1865   //
  1866   // int binary_search(int key, LookupswitchPair* array, int n) {
  1867   //   // Binary search according to "Methodik des Programmierens" by
  1868   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
  1869   //   int i = 0;
  1870   //   int j = n;
  1871   //   while (i+1 < j) {
  1872   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
  1873   //     // with      Q: for all i: 0 <= i < n: key < a[i]
  1874   //     // where a stands for the array and assuming that the (inexisting)
  1875   //     // element a[n] is infinitely big.
  1876   //     int h = (i + j) >> 1;
  1877   //     // i < h < j
  1878   //     if (key < array[h].fast_match()) {
  1879   //       j = h;
  1880   //     } else {
  1881   //       i = h;
  1882   //     }
  1883   //   }
  1884   //   // R: a[i] <= key < a[i+1] or Q
  1885   //   // (i.e., if key is within array, i is the correct index)
  1886   //   return i;
  1887   // }
  1889   // register allocation
  1890   assert(Otos_i == O0, "alias checking");
  1891   const Register Rkey     = Otos_i;                    // already set (tosca)
  1892   const Register Rarray   = O1;
  1893   const Register Ri       = O2;
  1894   const Register Rj       = O3;
  1895   const Register Rh       = O4;
  1896   const Register Rscratch = O5;
  1898   const int log_entry_size = 3;
  1899   const int entry_size = 1 << log_entry_size;
  1901   Label found;
  1902   // Find Array start
  1903   __ add(Lbcp, 3 * BytesPerInt, Rarray);
  1904   __ and3(Rarray, -BytesPerInt, Rarray);
  1905   // initialize i & j (in delay slot)
  1906   __ clr( Ri );
  1908   // and start
  1909   Label entry;
  1910   __ ba(false, entry);
  1911   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
  1912   // (Rj is already in the native byte-ordering.)
  1914   // binary search loop
  1915   { Label loop;
  1916     __ bind( loop );
  1917     // int h = (i + j) >> 1;
  1918     __ sra( Rh, 1, Rh );
  1919     // if (key < array[h].fast_match()) {
  1920     //   j = h;
  1921     // } else {
  1922     //   i = h;
  1923     // }
  1924     __ sll( Rh, log_entry_size, Rscratch );
  1925     __ ld( Rarray, Rscratch, Rscratch );
  1926     // (Rscratch is already in the native byte-ordering.)
  1927     __ cmp( Rkey, Rscratch );
  1928     if ( VM_Version::v9_instructions_work() ) {
  1929       __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
  1930       __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
  1932     else {
  1933       Label end_of_if;
  1934       __ br( Assembler::less, true, Assembler::pt, end_of_if );
  1935       __ delayed()->mov( Rh, Rj ); // if (<) Rj = Rh
  1936       __ mov( Rh, Ri );            // else i = h
  1937       __ bind(end_of_if);          // }
  1940     // while (i+1 < j)
  1941     __ bind( entry );
  1942     __ add( Ri, 1, Rscratch );
  1943     __ cmp(Rscratch, Rj);
  1944     __ br( Assembler::less, true, Assembler::pt, loop );
  1945     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
  1948   // end of binary search, result index is i (must check again!)
  1949   Label default_case;
  1950   Label continue_execution;
  1951   if (ProfileInterpreter) {
  1952     __ mov( Ri, Rh );              // Save index in i for profiling
  1954   __ sll( Ri, log_entry_size, Ri );
  1955   __ ld( Rarray, Ri, Rscratch );
  1956   // (Rscratch is already in the native byte-ordering.)
  1957   __ cmp( Rkey, Rscratch );
  1958   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
  1959   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
  1961   // entry found -> j = offset
  1962   __ inc( Ri, BytesPerInt );
  1963   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
  1964   __ ld( Rarray, Ri, Rj );
  1965   // (Rj is already in the native byte-ordering.)
  1967   if (ProfileInterpreter) {
  1968     __ ba(false, continue_execution);
  1969     __ delayed()->nop();
  1972   __ bind(default_case); // fall through (if not profiling)
  1973   __ profile_switch_default(Ri);
  1975   __ bind(continue_execution);
  1976   __ add( Lbcp, Rj, Lbcp );
  1977   __ dispatch_next( vtos );
  1981 void TemplateTable::_return(TosState state) {
  1982   transition(state, state);
  1983   assert(_desc->calls_vm(), "inconsistent calls_vm information");
  1985   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
  1986     assert(state == vtos, "only valid state");
  1987     __ mov(G0, G3_scratch);
  1988     __ access_local_ptr(G3_scratch, Otos_i);
  1989     __ load_klass(Otos_i, O2);
  1990     __ set(JVM_ACC_HAS_FINALIZER, G3);
  1991     __ ld(O2, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), O2);
  1992     __ andcc(G3, O2, G0);
  1993     Label skip_register_finalizer;
  1994     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
  1995     __ delayed()->nop();
  1997     // Call out to do finalizer registration
  1998     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
  2000     __ bind(skip_register_finalizer);
  2003   __ remove_activation(state, /* throw_monitor_exception */ true);
  2005   // The caller's SP was adjusted upon method entry to accomodate
  2006   // the callee's non-argument locals. Undo that adjustment.
  2007   __ ret();                             // return to caller
  2008   __ delayed()->restore(I5_savedSP, G0, SP);
  2012 // ----------------------------------------------------------------------------
  2013 // Volatile variables demand their effects be made known to all CPU's in
  2014 // order.  Store buffers on most chips allow reads & writes to reorder; the
  2015 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
  2016 // memory barrier (i.e., it's not sufficient that the interpreter does not
  2017 // reorder volatile references, the hardware also must not reorder them).
  2018 //
  2019 // According to the new Java Memory Model (JMM):
  2020 // (1) All volatiles are serialized wrt to each other.
  2021 // ALSO reads & writes act as aquire & release, so:
  2022 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
  2023 // the read float up to before the read.  It's OK for non-volatile memory refs
  2024 // that happen before the volatile read to float down below it.
  2025 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
  2026 // that happen BEFORE the write float down to after the write.  It's OK for
  2027 // non-volatile memory refs that happen after the volatile write to float up
  2028 // before it.
  2029 //
  2030 // We only put in barriers around volatile refs (they are expensive), not
  2031 // _between_ memory refs (that would require us to track the flavor of the
  2032 // previous memory refs).  Requirements (2) and (3) require some barriers
  2033 // before volatile stores and after volatile loads.  These nearly cover
  2034 // requirement (1) but miss the volatile-store-volatile-load case.  This final
  2035 // case is placed after volatile-stores although it could just as well go
  2036 // before volatile-loads.
  2037 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
  2038   // Helper function to insert a is-volatile test and memory barrier
  2039   // All current sparc implementations run in TSO, needing only StoreLoad
  2040   if ((order_constraint & Assembler::StoreLoad) == 0) return;
  2041   __ membar( order_constraint );
  2044 // ----------------------------------------------------------------------------
  2045 void TemplateTable::resolve_cache_and_index(int byte_no,
  2046                                             Register result,
  2047                                             Register Rcache,
  2048                                             Register index,
  2049                                             size_t index_size) {
  2050   // Depends on cpCacheOop layout!
  2051   Label resolved;
  2053   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2054   if (byte_no == f1_oop) {
  2055     // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
  2056     // This kind of CP cache entry does not need to match the flags byte, because
  2057     // there is a 1-1 relation between bytecode type and CP entry type.
  2058     assert_different_registers(result, Rcache);
  2059     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2060               ConstantPoolCacheEntry::f1_offset(), result);
  2061     __ tst(result);
  2062     __ br(Assembler::notEqual, false, Assembler::pt, resolved);
  2063     __ delayed()->set((int)bytecode(), O1);
  2064   } else {
  2065     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
  2066     assert(result == noreg, "");  //else change code for setting result
  2067     const int shift_count = (1 + byte_no)*BitsPerByte;
  2069     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2070               ConstantPoolCacheEntry::indices_offset(), Lbyte_code);
  2072     __ srl(  Lbyte_code, shift_count, Lbyte_code );
  2073     __ and3( Lbyte_code,        0xFF, Lbyte_code );
  2074     __ cmp(  Lbyte_code, (int)bytecode());
  2075     __ br(   Assembler::equal, false, Assembler::pt, resolved);
  2076     __ delayed()->set((int)bytecode(), O1);
  2079   address entry;
  2080   switch (bytecode()) {
  2081     case Bytecodes::_getstatic      : // fall through
  2082     case Bytecodes::_putstatic      : // fall through
  2083     case Bytecodes::_getfield       : // fall through
  2084     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
  2085     case Bytecodes::_invokevirtual  : // fall through
  2086     case Bytecodes::_invokespecial  : // fall through
  2087     case Bytecodes::_invokestatic   : // fall through
  2088     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
  2089     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
  2090     case Bytecodes::_fast_aldc      : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2091     case Bytecodes::_fast_aldc_w    : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);     break;
  2092     default                         : ShouldNotReachHere();                                 break;
  2094   // first time invocation - must resolve first
  2095   __ call_VM(noreg, entry, O1);
  2096   // Update registers with resolved info
  2097   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2098   if (result != noreg)
  2099     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() +
  2100               ConstantPoolCacheEntry::f1_offset(), result);
  2101   __ bind(resolved);
  2104 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
  2105                                                Register Rmethod,
  2106                                                Register Ritable_index,
  2107                                                Register Rflags,
  2108                                                bool is_invokevirtual,
  2109                                                bool is_invokevfinal,
  2110                                                bool is_invokedynamic) {
  2111   // Uses both G3_scratch and G4_scratch
  2112   Register Rcache = G3_scratch;
  2113   Register Rscratch = G4_scratch;
  2114   assert_different_registers(Rcache, Rmethod, Ritable_index);
  2116   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2118   // determine constant pool cache field offsets
  2119   const int method_offset = in_bytes(
  2120     cp_base_offset +
  2121       (is_invokevirtual
  2122        ? ConstantPoolCacheEntry::f2_offset()
  2123        : ConstantPoolCacheEntry::f1_offset()
  2125     );
  2126   const int flags_offset = in_bytes(cp_base_offset +
  2127                                     ConstantPoolCacheEntry::flags_offset());
  2128   // access constant pool cache fields
  2129   const int index_offset = in_bytes(cp_base_offset +
  2130                                     ConstantPoolCacheEntry::f2_offset());
  2132   if (is_invokevfinal) {
  2133     __ get_cache_and_index_at_bcp(Rcache, Rscratch, 1);
  2134     __ ld_ptr(Rcache, method_offset, Rmethod);
  2135   } else if (byte_no == f1_oop) {
  2136     // Resolved f1_oop goes directly into 'method' register.
  2137     resolve_cache_and_index(byte_no, Rmethod, Rcache, Rscratch, sizeof(u4));
  2138   } else {
  2139     resolve_cache_and_index(byte_no, noreg, Rcache, Rscratch, sizeof(u2));
  2140     __ ld_ptr(Rcache, method_offset, Rmethod);
  2143   if (Ritable_index != noreg) {
  2144     __ ld_ptr(Rcache, index_offset, Ritable_index);
  2146   __ ld_ptr(Rcache, flags_offset, Rflags);
  2149 // The Rcache register must be set before call
  2150 void TemplateTable::load_field_cp_cache_entry(Register Robj,
  2151                                               Register Rcache,
  2152                                               Register index,
  2153                                               Register Roffset,
  2154                                               Register Rflags,
  2155                                               bool is_static) {
  2156   assert_different_registers(Rcache, Rflags, Roffset);
  2158   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2160   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2161   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2162   if (is_static) {
  2163     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
  2167 // The registers Rcache and index expected to be set before call.
  2168 // Correct values of the Rcache and index registers are preserved.
  2169 void TemplateTable::jvmti_post_field_access(Register Rcache,
  2170                                             Register index,
  2171                                             bool is_static,
  2172                                             bool has_tos) {
  2173   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2175   if (JvmtiExport::can_post_field_access()) {
  2176     // Check to see if a field access watch has been set before we take
  2177     // the time to call into the VM.
  2178     Label Label1;
  2179     assert_different_registers(Rcache, index, G1_scratch);
  2180     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
  2181     __ load_contents(get_field_access_count_addr, G1_scratch);
  2182     __ tst(G1_scratch);
  2183     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2184     __ delayed()->nop();
  2186     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
  2188     if (is_static) {
  2189       __ clr(Otos_i);
  2190     } else {
  2191       if (has_tos) {
  2192       // save object pointer before call_VM() clobbers it
  2193         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
  2194       } else {
  2195         // Load top of stack (do not pop the value off the stack);
  2196         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
  2198       __ verify_oop(Otos_i);
  2200     // Otos_i: object pointer or NULL if static
  2201     // Rcache: cache entry pointer
  2202     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
  2203                Otos_i, Rcache);
  2204     if (!is_static && has_tos) {
  2205       __ pop_ptr(Otos_i);  // restore object pointer
  2206       __ verify_oop(Otos_i);
  2208     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2209     __ bind(Label1);
  2213 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
  2214   transition(vtos, vtos);
  2216   Register Rcache = G3_scratch;
  2217   Register index  = G4_scratch;
  2218   Register Rclass = Rcache;
  2219   Register Roffset= G4_scratch;
  2220   Register Rflags = G1_scratch;
  2221   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2223   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2224   jvmti_post_field_access(Rcache, index, is_static, false);
  2225   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2227   if (!is_static) {
  2228     pop_and_check_object(Rclass);
  2229   } else {
  2230     __ verify_oop(Rclass);
  2233   Label exit;
  2235   Assembler::Membar_mask_bits membar_bits =
  2236     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2238   if (__ membar_has_effect(membar_bits)) {
  2239     // Get volatile flag
  2240     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2241     __ and3(Rflags, Lscratch, Lscratch);
  2244   Label checkVolatile;
  2246   // compute field type
  2247   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
  2248   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2249   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2250   ConstantPoolCacheEntry::verify_tosBits();
  2252   // Check atos before itos for getstatic, more likely (in Queens at least)
  2253   __ cmp(Rflags, atos);
  2254   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2255   __ delayed() ->cmp(Rflags, itos);
  2257   // atos
  2258   __ load_heap_oop(Rclass, Roffset, Otos_i);
  2259   __ verify_oop(Otos_i);
  2260   __ push(atos);
  2261   if (!is_static) {
  2262     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
  2264   __ ba(false, checkVolatile);
  2265   __ delayed()->tst(Lscratch);
  2267   __ bind(notObj);
  2269   // cmp(Rflags, itos);
  2270   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2271   __ delayed() ->cmp(Rflags, ltos);
  2273   // itos
  2274   __ ld(Rclass, Roffset, Otos_i);
  2275   __ push(itos);
  2276   if (!is_static) {
  2277     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
  2279   __ ba(false, checkVolatile);
  2280   __ delayed()->tst(Lscratch);
  2282   __ bind(notInt);
  2284   // cmp(Rflags, ltos);
  2285   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2286   __ delayed() ->cmp(Rflags, btos);
  2288   // ltos
  2289   // load must be atomic
  2290   __ ld_long(Rclass, Roffset, Otos_l);
  2291   __ push(ltos);
  2292   if (!is_static) {
  2293     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
  2295   __ ba(false, checkVolatile);
  2296   __ delayed()->tst(Lscratch);
  2298   __ bind(notLong);
  2300   // cmp(Rflags, btos);
  2301   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2302   __ delayed() ->cmp(Rflags, ctos);
  2304   // btos
  2305   __ ldsb(Rclass, Roffset, Otos_i);
  2306   __ push(itos);
  2307   if (!is_static) {
  2308     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
  2310   __ ba(false, checkVolatile);
  2311   __ delayed()->tst(Lscratch);
  2313   __ bind(notByte);
  2315   // cmp(Rflags, ctos);
  2316   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2317   __ delayed() ->cmp(Rflags, stos);
  2319   // ctos
  2320   __ lduh(Rclass, Roffset, Otos_i);
  2321   __ push(itos);
  2322   if (!is_static) {
  2323     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
  2325   __ ba(false, checkVolatile);
  2326   __ delayed()->tst(Lscratch);
  2328   __ bind(notChar);
  2330   // cmp(Rflags, stos);
  2331   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2332   __ delayed() ->cmp(Rflags, ftos);
  2334   // stos
  2335   __ ldsh(Rclass, Roffset, Otos_i);
  2336   __ push(itos);
  2337   if (!is_static) {
  2338     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
  2340   __ ba(false, checkVolatile);
  2341   __ delayed()->tst(Lscratch);
  2343   __ bind(notShort);
  2346   // cmp(Rflags, ftos);
  2347   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
  2348   __ delayed() ->tst(Lscratch);
  2350   // ftos
  2351   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
  2352   __ push(ftos);
  2353   if (!is_static) {
  2354     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
  2356   __ ba(false, checkVolatile);
  2357   __ delayed()->tst(Lscratch);
  2359   __ bind(notFloat);
  2362   // dtos
  2363   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
  2364   __ push(dtos);
  2365   if (!is_static) {
  2366     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
  2369   __ bind(checkVolatile);
  2370   if (__ membar_has_effect(membar_bits)) {
  2371     // __ tst(Lscratch); executed in delay slot
  2372     __ br(Assembler::zero, false, Assembler::pt, exit);
  2373     __ delayed()->nop();
  2374     volatile_barrier(membar_bits);
  2377   __ bind(exit);
  2381 void TemplateTable::getfield(int byte_no) {
  2382   getfield_or_static(byte_no, false);
  2385 void TemplateTable::getstatic(int byte_no) {
  2386   getfield_or_static(byte_no, true);
  2390 void TemplateTable::fast_accessfield(TosState state) {
  2391   transition(atos, state);
  2392   Register Rcache  = G3_scratch;
  2393   Register index   = G4_scratch;
  2394   Register Roffset = G4_scratch;
  2395   Register Rflags  = Rcache;
  2396   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2398   __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2399   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
  2401   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2403   __ null_check(Otos_i);
  2404   __ verify_oop(Otos_i);
  2406   Label exit;
  2408   Assembler::Membar_mask_bits membar_bits =
  2409     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2410   if (__ membar_has_effect(membar_bits)) {
  2411     // Get volatile flag
  2412     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
  2413     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2416   switch (bytecode()) {
  2417     case Bytecodes::_fast_bgetfield:
  2418       __ ldsb(Otos_i, Roffset, Otos_i);
  2419       break;
  2420     case Bytecodes::_fast_cgetfield:
  2421       __ lduh(Otos_i, Roffset, Otos_i);
  2422       break;
  2423     case Bytecodes::_fast_sgetfield:
  2424       __ ldsh(Otos_i, Roffset, Otos_i);
  2425       break;
  2426     case Bytecodes::_fast_igetfield:
  2427       __ ld(Otos_i, Roffset, Otos_i);
  2428       break;
  2429     case Bytecodes::_fast_lgetfield:
  2430       __ ld_long(Otos_i, Roffset, Otos_l);
  2431       break;
  2432     case Bytecodes::_fast_fgetfield:
  2433       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
  2434       break;
  2435     case Bytecodes::_fast_dgetfield:
  2436       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
  2437       break;
  2438     case Bytecodes::_fast_agetfield:
  2439       __ load_heap_oop(Otos_i, Roffset, Otos_i);
  2440       break;
  2441     default:
  2442       ShouldNotReachHere();
  2445   if (__ membar_has_effect(membar_bits)) {
  2446     __ btst(Lscratch, Rflags);
  2447     __ br(Assembler::zero, false, Assembler::pt, exit);
  2448     __ delayed()->nop();
  2449     volatile_barrier(membar_bits);
  2450     __ bind(exit);
  2453   if (state == atos) {
  2454     __ verify_oop(Otos_i);    // does not blow flags!
  2458 void TemplateTable::jvmti_post_fast_field_mod() {
  2459   if (JvmtiExport::can_post_field_modification()) {
  2460     // Check to see if a field modification watch has been set before we take
  2461     // the time to call into the VM.
  2462     Label done;
  2463     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2464     __ load_contents(get_field_modification_count_addr, G4_scratch);
  2465     __ tst(G4_scratch);
  2466     __ br(Assembler::zero, false, Assembler::pt, done);
  2467     __ delayed()->nop();
  2468     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
  2469     __ verify_oop(G4_scratch);
  2470     __ push_ptr(G4_scratch);    // put the object pointer back on tos
  2471     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
  2472     // Save tos values before call_VM() clobbers them. Since we have
  2473     // to do it for every data type, we use the saved values as the
  2474     // jvalue object.
  2475     switch (bytecode()) {  // save tos values before call_VM() clobbers them
  2476     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
  2477     case Bytecodes::_fast_bputfield: // fall through
  2478     case Bytecodes::_fast_sputfield: // fall through
  2479     case Bytecodes::_fast_cputfield: // fall through
  2480     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
  2481     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
  2482     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
  2483     // get words in right order for use as jvalue object
  2484     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
  2486     // setup pointer to jvalue object
  2487     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
  2488     // G4_scratch:  object pointer
  2489     // G1_scratch: cache entry pointer
  2490     // G3_scratch: jvalue object on the stack
  2491     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
  2492     switch (bytecode()) {             // restore tos values
  2493     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
  2494     case Bytecodes::_fast_bputfield: // fall through
  2495     case Bytecodes::_fast_sputfield: // fall through
  2496     case Bytecodes::_fast_cputfield: // fall through
  2497     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
  2498     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
  2499     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
  2500     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
  2502     __ bind(done);
  2506 // The registers Rcache and index expected to be set before call.
  2507 // The function may destroy various registers, just not the Rcache and index registers.
  2508 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
  2509   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2511   if (JvmtiExport::can_post_field_modification()) {
  2512     // Check to see if a field modification watch has been set before we take
  2513     // the time to call into the VM.
  2514     Label Label1;
  2515     assert_different_registers(Rcache, index, G1_scratch);
  2516     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2517     __ load_contents(get_field_modification_count_addr, G1_scratch);
  2518     __ tst(G1_scratch);
  2519     __ br(Assembler::zero, false, Assembler::pt, Label1);
  2520     __ delayed()->nop();
  2522     // The Rcache and index registers have been already set.
  2523     // This allows to eliminate this call but the Rcache and index
  2524     // registers must be correspondingly used after this line.
  2525     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
  2527     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
  2528     if (is_static) {
  2529       // Life is simple.  Null out the object pointer.
  2530       __ clr(G4_scratch);
  2531     } else {
  2532       Register Rflags = G1_scratch;
  2533       // Life is harder. The stack holds the value on top, followed by the
  2534       // object.  We don't know the size of the value, though; it could be
  2535       // one or two words depending on its type. As a result, we must find
  2536       // the type to determine where the object is.
  2538       Label two_word, valsizeknown;
  2539       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2540       __ mov(Lesp, G4_scratch);
  2541       __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2542       // Make sure we don't need to mask Rflags for tosBits after the above shift
  2543       ConstantPoolCacheEntry::verify_tosBits();
  2544       __ cmp(Rflags, ltos);
  2545       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2546       __ delayed()->cmp(Rflags, dtos);
  2547       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2548       __ delayed()->nop();
  2549       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
  2550       __ br(Assembler::always, false, Assembler::pt, valsizeknown);
  2551       __ delayed()->nop();
  2552       __ bind(two_word);
  2554       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
  2556       __ bind(valsizeknown);
  2557       // setup object pointer
  2558       __ ld_ptr(G4_scratch, 0, G4_scratch);
  2559       __ verify_oop(G4_scratch);
  2561     // setup pointer to jvalue object
  2562     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
  2563     // G4_scratch:  object pointer or NULL if static
  2564     // G3_scratch: cache entry pointer
  2565     // G1_scratch: jvalue object on the stack
  2566     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
  2567                G4_scratch, G3_scratch, G1_scratch);
  2568     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2569     __ bind(Label1);
  2573 void TemplateTable::pop_and_check_object(Register r) {
  2574   __ pop_ptr(r);
  2575   __ null_check(r);  // for field access must check obj.
  2576   __ verify_oop(r);
  2579 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
  2580   transition(vtos, vtos);
  2581   Register Rcache = G3_scratch;
  2582   Register index  = G4_scratch;
  2583   Register Rclass = Rcache;
  2584   Register Roffset= G4_scratch;
  2585   Register Rflags = G1_scratch;
  2586   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2588   resolve_cache_and_index(byte_no, noreg, Rcache, index, sizeof(u2));
  2589   jvmti_post_field_mod(Rcache, index, is_static);
  2590   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2592   Assembler::Membar_mask_bits read_bits =
  2593     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2594   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2596   Label notVolatile, checkVolatile, exit;
  2597   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2598     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2599     __ and3(Rflags, Lscratch, Lscratch);
  2601     if (__ membar_has_effect(read_bits)) {
  2602       __ tst(Lscratch);
  2603       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2604       __ delayed()->nop();
  2605       volatile_barrier(read_bits);
  2606       __ bind(notVolatile);
  2610   __ srl(Rflags, ConstantPoolCacheEntry::tosBits, Rflags);
  2611   // Make sure we don't need to mask Rflags for tosBits after the above shift
  2612   ConstantPoolCacheEntry::verify_tosBits();
  2614   // compute field type
  2615   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
  2617   if (is_static) {
  2618     // putstatic with object type most likely, check that first
  2619     __ cmp(Rflags, atos );
  2620     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2621     __ delayed() ->cmp(Rflags, itos );
  2623     // atos
  2624     __ pop_ptr();
  2625     __ verify_oop(Otos_i);
  2627     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2629     __ ba(false, checkVolatile);
  2630     __ delayed()->tst(Lscratch);
  2632     __ bind(notObj);
  2634     // cmp(Rflags, itos );
  2635     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2636     __ delayed() ->cmp(Rflags, btos );
  2638     // itos
  2639     __ pop_i();
  2640     __ st(Otos_i, Rclass, Roffset);
  2641     __ ba(false, checkVolatile);
  2642     __ delayed()->tst(Lscratch);
  2644     __ bind(notInt);
  2646   } else {
  2647     // putfield with int type most likely, check that first
  2648     __ cmp(Rflags, itos );
  2649     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2650     __ delayed() ->cmp(Rflags, atos );
  2652     // itos
  2653     __ pop_i();
  2654     pop_and_check_object(Rclass);
  2655     __ st(Otos_i, Rclass, Roffset);
  2656     patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch);
  2657     __ ba(false, checkVolatile);
  2658     __ delayed()->tst(Lscratch);
  2660     __ bind(notInt);
  2661     // cmp(Rflags, atos );
  2662     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2663     __ delayed() ->cmp(Rflags, btos );
  2665     // atos
  2666     __ pop_ptr();
  2667     pop_and_check_object(Rclass);
  2668     __ verify_oop(Otos_i);
  2670     do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2672     patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch);
  2673     __ ba(false, checkVolatile);
  2674     __ delayed()->tst(Lscratch);
  2676     __ bind(notObj);
  2679   // cmp(Rflags, btos );
  2680   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2681   __ delayed() ->cmp(Rflags, ltos );
  2683   // btos
  2684   __ pop_i();
  2685   if (!is_static) pop_and_check_object(Rclass);
  2686   __ stb(Otos_i, Rclass, Roffset);
  2687   if (!is_static) {
  2688     patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch);
  2690   __ ba(false, checkVolatile);
  2691   __ delayed()->tst(Lscratch);
  2693   __ bind(notByte);
  2695   // cmp(Rflags, ltos );
  2696   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2697   __ delayed() ->cmp(Rflags, ctos );
  2699   // ltos
  2700   __ pop_l();
  2701   if (!is_static) pop_and_check_object(Rclass);
  2702   __ st_long(Otos_l, Rclass, Roffset);
  2703   if (!is_static) {
  2704     patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch);
  2706   __ ba(false, checkVolatile);
  2707   __ delayed()->tst(Lscratch);
  2709   __ bind(notLong);
  2711   // cmp(Rflags, ctos );
  2712   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2713   __ delayed() ->cmp(Rflags, stos );
  2715   // ctos (char)
  2716   __ pop_i();
  2717   if (!is_static) pop_and_check_object(Rclass);
  2718   __ sth(Otos_i, Rclass, Roffset);
  2719   if (!is_static) {
  2720     patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch);
  2722   __ ba(false, checkVolatile);
  2723   __ delayed()->tst(Lscratch);
  2725   __ bind(notChar);
  2726   // cmp(Rflags, stos );
  2727   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2728   __ delayed() ->cmp(Rflags, ftos );
  2730   // stos (char)
  2731   __ pop_i();
  2732   if (!is_static) pop_and_check_object(Rclass);
  2733   __ sth(Otos_i, Rclass, Roffset);
  2734   if (!is_static) {
  2735     patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch);
  2737   __ ba(false, checkVolatile);
  2738   __ delayed()->tst(Lscratch);
  2740   __ bind(notShort);
  2741   // cmp(Rflags, ftos );
  2742   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
  2743   __ delayed()->nop();
  2745   // ftos
  2746   __ pop_f();
  2747   if (!is_static) pop_and_check_object(Rclass);
  2748   __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2749   if (!is_static) {
  2750     patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch);
  2752   __ ba(false, checkVolatile);
  2753   __ delayed()->tst(Lscratch);
  2755   __ bind(notFloat);
  2757   // dtos
  2758   __ pop_d();
  2759   if (!is_static) pop_and_check_object(Rclass);
  2760   __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2761   if (!is_static) {
  2762     patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch);
  2765   __ bind(checkVolatile);
  2766   __ tst(Lscratch);
  2768   if (__ membar_has_effect(write_bits)) {
  2769     // __ tst(Lscratch); in delay slot
  2770     __ br(Assembler::zero, false, Assembler::pt, exit);
  2771     __ delayed()->nop();
  2772     volatile_barrier(Assembler::StoreLoad);
  2773     __ bind(exit);
  2777 void TemplateTable::fast_storefield(TosState state) {
  2778   transition(state, vtos);
  2779   Register Rcache = G3_scratch;
  2780   Register Rclass = Rcache;
  2781   Register Roffset= G4_scratch;
  2782   Register Rflags = G1_scratch;
  2783   ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
  2785   jvmti_post_fast_field_mod();
  2787   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
  2789   Assembler::Membar_mask_bits read_bits =
  2790     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2791   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2793   Label notVolatile, checkVolatile, exit;
  2794   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2795     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2796     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2797     __ and3(Rflags, Lscratch, Lscratch);
  2798     if (__ membar_has_effect(read_bits)) {
  2799       __ tst(Lscratch);
  2800       __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2801       __ delayed()->nop();
  2802       volatile_barrier(read_bits);
  2803       __ bind(notVolatile);
  2807   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2808   pop_and_check_object(Rclass);
  2810   switch (bytecode()) {
  2811     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
  2812     case Bytecodes::_fast_cputfield: /* fall through */
  2813     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
  2814     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
  2815     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
  2816     case Bytecodes::_fast_fputfield:
  2817       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2818       break;
  2819     case Bytecodes::_fast_dputfield:
  2820       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2821       break;
  2822     case Bytecodes::_fast_aputfield:
  2823       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2824       break;
  2825     default:
  2826       ShouldNotReachHere();
  2829   if (__ membar_has_effect(write_bits)) {
  2830     __ tst(Lscratch);
  2831     __ br(Assembler::zero, false, Assembler::pt, exit);
  2832     __ delayed()->nop();
  2833     volatile_barrier(Assembler::StoreLoad);
  2834     __ bind(exit);
  2839 void TemplateTable::putfield(int byte_no) {
  2840   putfield_or_static(byte_no, false);
  2843 void TemplateTable::putstatic(int byte_no) {
  2844   putfield_or_static(byte_no, true);
  2848 void TemplateTable::fast_xaccess(TosState state) {
  2849   transition(vtos, state);
  2850   Register Rcache = G3_scratch;
  2851   Register Roffset = G4_scratch;
  2852   Register Rflags  = G4_scratch;
  2853   Register Rreceiver = Lscratch;
  2855   __ ld_ptr(Llocals, 0, Rreceiver);
  2857   // access constant pool cache  (is resolved)
  2858   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
  2859   __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2860   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
  2862   __ verify_oop(Rreceiver);
  2863   __ null_check(Rreceiver);
  2864   if (state == atos) {
  2865     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
  2866   } else if (state == itos) {
  2867     __ ld (Rreceiver, Roffset, Otos_i) ;
  2868   } else if (state == ftos) {
  2869     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
  2870   } else {
  2871     ShouldNotReachHere();
  2874   Assembler::Membar_mask_bits membar_bits =
  2875     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2876   if (__ membar_has_effect(membar_bits)) {
  2878     // Get is_volatile value in Rflags and check if membar is needed
  2879     __ ld_ptr(Rcache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2881     // Test volatile
  2882     Label notVolatile;
  2883     __ set((1 << ConstantPoolCacheEntry::volatileField), Lscratch);
  2884     __ btst(Rflags, Lscratch);
  2885     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2886     __ delayed()->nop();
  2887     volatile_barrier(membar_bits);
  2888     __ bind(notVolatile);
  2891   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
  2892   __ sub(Lbcp, 1, Lbcp);
  2895 //----------------------------------------------------------------------------------------------------
  2896 // Calls
  2898 void TemplateTable::count_calls(Register method, Register temp) {
  2899   // implemented elsewhere
  2900   ShouldNotReachHere();
  2903 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
  2904   Register Rtemp = G4_scratch;
  2905   Register Rcall = Rindex;
  2906   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  2908   // get target methodOop & entry point
  2909   const int base = instanceKlass::vtable_start_offset() * wordSize;
  2910   if (vtableEntry::size() % 3 == 0) {
  2911     // scale the vtable index by 12:
  2912     int one_third = vtableEntry::size() / 3;
  2913     __ sll(Rindex, exact_log2(one_third * 1 * wordSize), Rtemp);
  2914     __ sll(Rindex, exact_log2(one_third * 2 * wordSize), Rindex);
  2915     __ add(Rindex, Rtemp, Rindex);
  2916   } else {
  2917     // scale the vtable index by 8:
  2918     __ sll(Rindex, exact_log2(vtableEntry::size() * wordSize), Rindex);
  2921   __ add(Rrecv, Rindex, Rrecv);
  2922   __ ld_ptr(Rrecv, base + vtableEntry::method_offset_in_bytes(), G5_method);
  2924   __ call_from_interpreter(Rcall, Gargs, Rret);
  2927 void TemplateTable::invokevirtual(int byte_no) {
  2928   transition(vtos, vtos);
  2929   assert(byte_no == f2_byte, "use this argument");
  2931   Register Rscratch = G3_scratch;
  2932   Register Rtemp = G4_scratch;
  2933   Register Rret = Lscratch;
  2934   Register Rrecv = G5_method;
  2935   Label notFinal;
  2937   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
  2938   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  2940   // Check for vfinal
  2941   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), G4_scratch);
  2942   __ btst(Rret, G4_scratch);
  2943   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  2944   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
  2946   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
  2948   invokevfinal_helper(Rscratch, Rret);
  2950   __ bind(notFinal);
  2952   __ mov(G5_method, Rscratch);  // better scratch register
  2953   __ load_receiver(G4_scratch, O0);  // gets receiverOop
  2954   // receiver is in O0
  2955   __ verify_oop(O0);
  2957   // get return address
  2958   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  2959   __ set(table, Rtemp);
  2960   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  2961   // Make sure we don't need to mask Rret for tosBits after the above shift
  2962   ConstantPoolCacheEntry::verify_tosBits();
  2963   __ sll(Rret,  LogBytesPerWord, Rret);
  2964   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  2966   // get receiver klass
  2967   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  2968   __ load_klass(O0, Rrecv);
  2969   __ verify_oop(Rrecv);
  2971   __ profile_virtual_call(Rrecv, O4);
  2973   generate_vtable_call(Rrecv, Rscratch, Rret);
  2976 void TemplateTable::fast_invokevfinal(int byte_no) {
  2977   transition(vtos, vtos);
  2978   assert(byte_no == f2_byte, "use this argument");
  2980   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
  2981                              /*is_invokevfinal*/true, false);
  2982   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  2983   invokevfinal_helper(G3_scratch, Lscratch);
  2986 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
  2987   Register Rtemp = G4_scratch;
  2989   __ verify_oop(G5_method);
  2991   // Load receiver from stack slot
  2992   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  2993   __ load_receiver(G4_scratch, O0);
  2995   // receiver NULL check
  2996   __ null_check(O0);
  2998   __ profile_final_call(O4);
  3000   // get return address
  3001   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3002   __ set(table, Rtemp);
  3003   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3004   // Make sure we don't need to mask Rret for tosBits after the above shift
  3005   ConstantPoolCacheEntry::verify_tosBits();
  3006   __ sll(Rret,  LogBytesPerWord, Rret);
  3007   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3010   // do the call
  3011   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3014 void TemplateTable::invokespecial(int byte_no) {
  3015   transition(vtos, vtos);
  3016   assert(byte_no == f1_byte, "use this argument");
  3018   Register Rscratch = G3_scratch;
  3019   Register Rtemp = G4_scratch;
  3020   Register Rret = Lscratch;
  3022   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3023   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3025   __ verify_oop(G5_method);
  3027   __ lduh(G5_method, in_bytes(methodOopDesc::size_of_parameters_offset()), G4_scratch);
  3028   __ load_receiver(G4_scratch, O0);
  3030   // receiver NULL check
  3031   __ null_check(O0);
  3033   __ profile_call(O4);
  3035   // get return address
  3036   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3037   __ set(table, Rtemp);
  3038   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3039   // Make sure we don't need to mask Rret for tosBits after the above shift
  3040   ConstantPoolCacheEntry::verify_tosBits();
  3041   __ sll(Rret,  LogBytesPerWord, Rret);
  3042   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3044   // do the call
  3045   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3048 void TemplateTable::invokestatic(int byte_no) {
  3049   transition(vtos, vtos);
  3050   assert(byte_no == f1_byte, "use this argument");
  3052   Register Rscratch = G3_scratch;
  3053   Register Rtemp = G4_scratch;
  3054   Register Rret = Lscratch;
  3056   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, /*virtual*/ false, false, false);
  3057   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3059   __ verify_oop(G5_method);
  3061   __ profile_call(O4);
  3063   // get return address
  3064   AddressLiteral table(Interpreter::return_3_addrs_by_index_table());
  3065   __ set(table, Rtemp);
  3066   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3067   // Make sure we don't need to mask Rret for tosBits after the above shift
  3068   ConstantPoolCacheEntry::verify_tosBits();
  3069   __ sll(Rret,  LogBytesPerWord, Rret);
  3070   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3072   // do the call
  3073   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3077 void TemplateTable::invokeinterface_object_method(Register RklassOop,
  3078                                                   Register Rcall,
  3079                                                   Register Rret,
  3080                                                   Register Rflags) {
  3081   Register Rscratch = G4_scratch;
  3082   Register Rindex = Lscratch;
  3084   assert_different_registers(Rscratch, Rindex, Rret);
  3086   Label notFinal;
  3088   // Check for vfinal
  3089   __ set((1 << ConstantPoolCacheEntry::vfinalMethod), Rscratch);
  3090   __ btst(Rflags, Rscratch);
  3091   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  3092   __ delayed()->nop();
  3094   __ profile_final_call(O4);
  3096   // do the call - the index (f2) contains the methodOop
  3097   assert_different_registers(G5_method, Gargs, Rcall);
  3098   __ mov(Rindex, G5_method);
  3099   __ call_from_interpreter(Rcall, Gargs, Rret);
  3100   __ bind(notFinal);
  3102   __ profile_virtual_call(RklassOop, O4);
  3103   generate_vtable_call(RklassOop, Rindex, Rret);
  3107 void TemplateTable::invokeinterface(int byte_no) {
  3108   transition(vtos, vtos);
  3109   assert(byte_no == f1_byte, "use this argument");
  3111   Register Rscratch = G4_scratch;
  3112   Register Rret = G3_scratch;
  3113   Register Rindex = Lscratch;
  3114   Register Rinterface = G1_scratch;
  3115   Register RklassOop = G5_method;
  3116   Register Rflags = O1;
  3117   assert_different_registers(Rscratch, G5_method);
  3119   load_invoke_cp_cache_entry(byte_no, Rinterface, Rindex, Rflags, /*virtual*/ false, false, false);
  3120   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3122   // get receiver
  3123   __ and3(Rflags, 0xFF, Rscratch);       // gets number of parameters
  3124   __ load_receiver(Rscratch, O0);
  3125   __ verify_oop(O0);
  3127   __ mov(Rflags, Rret);
  3129   // get return address
  3130   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3131   __ set(table, Rscratch);
  3132   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);          // get return type
  3133   // Make sure we don't need to mask Rret for tosBits after the above shift
  3134   ConstantPoolCacheEntry::verify_tosBits();
  3135   __ sll(Rret,  LogBytesPerWord, Rret);
  3136   __ ld_ptr(Rscratch, Rret, Rret);      // get return address
  3138   // get receiver klass
  3139   __ null_check(O0, oopDesc::klass_offset_in_bytes());
  3140   __ load_klass(O0, RklassOop);
  3141   __ verify_oop(RklassOop);
  3143   // Special case of invokeinterface called for virtual method of
  3144   // java.lang.Object.  See cpCacheOop.cpp for details.
  3145   // This code isn't produced by javac, but could be produced by
  3146   // another compliant java compiler.
  3147   Label notMethod;
  3148   __ set((1 << ConstantPoolCacheEntry::methodInterface), Rscratch);
  3149   __ btst(Rflags, Rscratch);
  3150   __ br(Assembler::zero, false, Assembler::pt, notMethod);
  3151   __ delayed()->nop();
  3153   invokeinterface_object_method(RklassOop, Rinterface, Rret, Rflags);
  3155   __ bind(notMethod);
  3157   __ profile_virtual_call(RklassOop, O4);
  3159   //
  3160   // find entry point to call
  3161   //
  3163   // compute start of first itableOffsetEntry (which is at end of vtable)
  3164   const int base = instanceKlass::vtable_start_offset() * wordSize;
  3165   Label search;
  3166   Register Rtemp = Rflags;
  3168   __ ld(RklassOop, instanceKlass::vtable_length_offset() * wordSize, Rtemp);
  3169   if (align_object_offset(1) > 1) {
  3170     __ round_to(Rtemp, align_object_offset(1));
  3172   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
  3173   if (Assembler::is_simm13(base)) {
  3174     __ add(Rtemp, base, Rtemp);
  3175   } else {
  3176     __ set(base, Rscratch);
  3177     __ add(Rscratch, Rtemp, Rtemp);
  3179   __ add(RklassOop, Rtemp, Rscratch);
  3181   __ bind(search);
  3183   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
  3185     Label ok;
  3187     // Check that entry is non-null.  Null entries are probably a bytecode
  3188     // problem.  If the interface isn't implemented by the receiver class,
  3189     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
  3190     // this too but that's only if the entry isn't already resolved, so we
  3191     // need to check again.
  3192     __ br_notnull( Rtemp, false, Assembler::pt, ok);
  3193     __ delayed()->nop();
  3194     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
  3195     __ should_not_reach_here();
  3196     __ bind(ok);
  3197     __ verify_oop(Rtemp);
  3200   __ verify_oop(Rinterface);
  3202   __ cmp(Rinterface, Rtemp);
  3203   __ brx(Assembler::notEqual, true, Assembler::pn, search);
  3204   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
  3206   // entry found and Rscratch points to it
  3207   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
  3209   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
  3210   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
  3211   __ add(Rscratch, Rindex, Rscratch);
  3212   __ ld_ptr(RklassOop, Rscratch, G5_method);
  3214   // Check for abstract method error.
  3216     Label ok;
  3217     __ tst(G5_method);
  3218     __ brx(Assembler::notZero, false, Assembler::pt, ok);
  3219     __ delayed()->nop();
  3220     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
  3221     __ should_not_reach_here();
  3222     __ bind(ok);
  3225   Register Rcall = Rinterface;
  3226   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  3228   __ verify_oop(G5_method);
  3229   __ call_from_interpreter(Rcall, Gargs, Rret);
  3234 void TemplateTable::invokedynamic(int byte_no) {
  3235   transition(vtos, vtos);
  3236   assert(byte_no == f1_oop, "use this argument");
  3238   if (!EnableInvokeDynamic) {
  3239     // We should not encounter this bytecode if !EnableInvokeDynamic.
  3240     // The verifier will stop it.  However, if we get past the verifier,
  3241     // this will stop the thread in a reasonable way, without crashing the JVM.
  3242     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
  3243                      InterpreterRuntime::throw_IncompatibleClassChangeError));
  3244     // the call_VM checks for exception, so we should never return here.
  3245     __ should_not_reach_here();
  3246     return;
  3249   // G5: CallSite object (f1)
  3250   // XX: unused (f2)
  3251   // XX: flags (unused)
  3253   Register G5_callsite = G5_method;
  3254   Register Rscratch    = G3_scratch;
  3255   Register Rtemp       = G1_scratch;
  3256   Register Rret        = Lscratch;
  3258   load_invoke_cp_cache_entry(byte_no, G5_callsite, noreg, Rret,
  3259                              /*virtual*/ false, /*vfinal*/ false, /*indy*/ true);
  3260   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
  3262   __ verify_oop(G5_callsite);
  3264   // profile this call
  3265   __ profile_call(O4);
  3267   // get return address
  3268   AddressLiteral table(Interpreter::return_5_addrs_by_index_table());
  3269   __ set(table, Rtemp);
  3270   __ srl(Rret, ConstantPoolCacheEntry::tosBits, Rret);  // get return type
  3271   // Make sure we don't need to mask Rret for tosBits after the above shift
  3272   ConstantPoolCacheEntry::verify_tosBits();
  3273   __ sll(Rret, LogBytesPerWord, Rret);
  3274   __ ld_ptr(Rtemp, Rret, Rret);  // get return address
  3276   __ load_heap_oop(G5_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle);
  3277   __ null_check(G3_method_handle);
  3279   // Adjust Rret first so Llast_SP can be same as Rret
  3280   __ add(Rret, -frame::pc_return_offset, O7);
  3281   __ add(Lesp, BytesPerWord, Gargs);  // setup parameter pointer
  3282   __ jump_to_method_handle_entry(G3_method_handle, Rtemp, /* emit_delayed_nop */ false);
  3283   // Record SP so we can remove any stack space allocated by adapter transition
  3284   __ delayed()->mov(SP, Llast_SP);
  3288 //----------------------------------------------------------------------------------------------------
  3289 // Allocation
  3291 void TemplateTable::_new() {
  3292   transition(vtos, atos);
  3294   Label slow_case;
  3295   Label done;
  3296   Label initialize_header;
  3297   Label initialize_object;  // including clearing the fields
  3299   Register RallocatedObject = Otos_i;
  3300   Register RinstanceKlass = O1;
  3301   Register Roffset = O3;
  3302   Register Rscratch = O4;
  3304   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3305   __ get_cpool_and_tags(Rscratch, G3_scratch);
  3306   // make sure the class we're about to instantiate has been resolved
  3307   // This is done before loading instanceKlass to be consistent with the order
  3308   // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
  3309   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3310   __ ldub(G3_scratch, Roffset, G3_scratch);
  3311   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3312   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3313   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3314   // get instanceKlass
  3315   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
  3316   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3317   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
  3319   // make sure klass is fully initialized:
  3320   __ ld(RinstanceKlass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_scratch);
  3321   __ cmp(G3_scratch, instanceKlass::fully_initialized);
  3322   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3323   __ delayed()->ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3325   // get instance_size in instanceKlass (already aligned)
  3326   //__ ld(RinstanceKlass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), Roffset);
  3328   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
  3329   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
  3330   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
  3331   __ delayed()->nop();
  3333   // allocate the instance
  3334   // 1) Try to allocate in the TLAB
  3335   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
  3336   // 3) if the above fails (or is not applicable), go to a slow case
  3337   // (creates a new TLAB, etc.)
  3339   const bool allow_shared_alloc =
  3340     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
  3342   if(UseTLAB) {
  3343     Register RoldTopValue = RallocatedObject;
  3344     Register RtopAddr = G3_scratch, RtlabWasteLimitValue = G3_scratch;
  3345     Register RnewTopValue = G1_scratch;
  3346     Register RendValue = Rscratch;
  3347     Register RfreeValue = RnewTopValue;
  3349     // check if we can allocate in the TLAB
  3350     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
  3351     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
  3352     __ add(RoldTopValue, Roffset, RnewTopValue);
  3354     // if there is enough space, we do not CAS and do not clear
  3355     __ cmp(RnewTopValue, RendValue);
  3356     if(ZeroTLAB) {
  3357       // the fields have already been cleared
  3358       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
  3359     } else {
  3360       // initialize both the header and fields
  3361       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
  3363     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
  3365     if (allow_shared_alloc) {
  3366     // Check if tlab should be discarded (refill_waste_limit >= free)
  3367     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
  3368     __ sub(RendValue, RoldTopValue, RfreeValue);
  3369 #ifdef _LP64
  3370     __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
  3371 #else
  3372     __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
  3373 #endif
  3374     __ cmp(RtlabWasteLimitValue, RfreeValue);
  3375     __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, slow_case); // tlab waste is small
  3376     __ delayed()->nop();
  3378     // increment waste limit to prevent getting stuck on this slow path
  3379     __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
  3380     __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
  3381     } else {
  3382       // No allocation in the shared eden.
  3383       __ br(Assembler::always, false, Assembler::pt, slow_case);
  3384       __ delayed()->nop();
  3388   // Allocation in the shared Eden
  3389   if (allow_shared_alloc) {
  3390     Register RoldTopValue = G1_scratch;
  3391     Register RtopAddr = G3_scratch;
  3392     Register RnewTopValue = RallocatedObject;
  3393     Register RendValue = Rscratch;
  3395     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
  3397     Label retry;
  3398     __ bind(retry);
  3399     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
  3400     __ ld_ptr(RendValue, 0, RendValue);
  3401     __ ld_ptr(RtopAddr, 0, RoldTopValue);
  3402     __ add(RoldTopValue, Roffset, RnewTopValue);
  3404     // RnewTopValue contains the top address after the new object
  3405     // has been allocated.
  3406     __ cmp(RnewTopValue, RendValue);
  3407     __ brx(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);
  3408     __ delayed()->nop();
  3410     __ casx_under_lock(RtopAddr, RoldTopValue, RnewTopValue,
  3411       VM_Version::v9_instructions_work() ? NULL :
  3412       (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
  3414     // if someone beat us on the allocation, try again, otherwise continue
  3415     __ cmp(RoldTopValue, RnewTopValue);
  3416     __ brx(Assembler::notEqual, false, Assembler::pn, retry);
  3417     __ delayed()->nop();
  3420   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
  3421     // clear object fields
  3422     __ bind(initialize_object);
  3423     __ deccc(Roffset, sizeof(oopDesc));
  3424     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
  3425     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
  3427     // initialize remaining object fields
  3428     { Label loop;
  3429       __ subcc(Roffset, wordSize, Roffset);
  3430       __ bind(loop);
  3431       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
  3432       __ st_ptr(G0, G3_scratch, Roffset);
  3433       __ br(Assembler::notEqual, false, Assembler::pt, loop);
  3434       __ delayed()->subcc(Roffset, wordSize, Roffset);
  3436     __ br(Assembler::always, false, Assembler::pt, initialize_header);
  3437     __ delayed()->nop();
  3440   // slow case
  3441   __ bind(slow_case);
  3442   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3443   __ get_constant_pool(O1);
  3445   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
  3447   __ ba(false, done);
  3448   __ delayed()->nop();
  3450   // Initialize the header: mark, klass
  3451   __ bind(initialize_header);
  3453   if (UseBiasedLocking) {
  3454     __ ld_ptr(RinstanceKlass, Klass::prototype_header_offset_in_bytes() + sizeof(oopDesc), G4_scratch);
  3455   } else {
  3456     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
  3458   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
  3459   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
  3460   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
  3463     SkipIfEqual skip_if(
  3464       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
  3465     // Trigger dtrace event
  3466     __ push(atos);
  3467     __ call_VM_leaf(noreg,
  3468        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
  3469     __ pop(atos);
  3472   // continue
  3473   __ bind(done);
  3478 void TemplateTable::newarray() {
  3479   transition(itos, atos);
  3480   __ ldub(Lbcp, 1, O1);
  3481      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
  3485 void TemplateTable::anewarray() {
  3486   transition(itos, atos);
  3487   __ get_constant_pool(O1);
  3488   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3489      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
  3493 void TemplateTable::arraylength() {
  3494   transition(atos, itos);
  3495   Label ok;
  3496   __ verify_oop(Otos_i);
  3497   __ tst(Otos_i);
  3498   __ throw_if_not_1_x( Assembler::notZero, ok );
  3499   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
  3500   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3504 void TemplateTable::checkcast() {
  3505   transition(atos, atos);
  3506   Label done, is_null, quicked, cast_ok, resolved;
  3507   Register Roffset = G1_scratch;
  3508   Register RobjKlass = O5;
  3509   Register RspecifiedKlass = O4;
  3511   // Check for casting a NULL
  3512   __ br_null(Otos_i, false, Assembler::pn, is_null);
  3513   __ delayed()->nop();
  3515   // Get value klass in RobjKlass
  3516   __ load_klass(Otos_i, RobjKlass); // get value klass
  3518   // Get constant pool tag
  3519   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3521   // See if the checkcast has been quickened
  3522   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3523   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3524   __ ldub(G3_scratch, Roffset, G3_scratch);
  3525   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3526   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3527   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3529   __ push_ptr(); // save receiver for result, and for GC
  3530   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3531   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3533   __ br(Assembler::always, false, Assembler::pt, resolved);
  3534   __ delayed()->nop();
  3536   // Extract target class from constant pool
  3537   __ bind(quicked);
  3538   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3539   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3540   __ bind(resolved);
  3541   __ load_klass(Otos_i, RobjKlass); // get value klass
  3543   // Generate a fast subtype check.  Branch to cast_ok if no
  3544   // failure.  Throw exception if failure.
  3545   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
  3547   // Not a subtype; so must throw exception
  3548   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
  3550   __ bind(cast_ok);
  3552   if (ProfileInterpreter) {
  3553     __ ba(false, done);
  3554     __ delayed()->nop();
  3556   __ bind(is_null);
  3557   __ profile_null_seen(G3_scratch);
  3558   __ bind(done);
  3562 void TemplateTable::instanceof() {
  3563   Label done, is_null, quicked, resolved;
  3564   transition(atos, itos);
  3565   Register Roffset = G1_scratch;
  3566   Register RobjKlass = O5;
  3567   Register RspecifiedKlass = O4;
  3569   // Check for casting a NULL
  3570   __ br_null(Otos_i, false, Assembler::pt, is_null);
  3571   __ delayed()->nop();
  3573   // Get value klass in RobjKlass
  3574   __ load_klass(Otos_i, RobjKlass); // get value klass
  3576   // Get constant pool tag
  3577   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3579   // See if the checkcast has been quickened
  3580   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3581   __ add(G3_scratch, typeArrayOopDesc::header_size(T_BYTE) * wordSize, G3_scratch);
  3582   __ ldub(G3_scratch, Roffset, G3_scratch);
  3583   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3584   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3585   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3587   __ push_ptr(); // save receiver for result, and for GC
  3588   call_VM(RspecifiedKlass, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3589   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3591   __ br(Assembler::always, false, Assembler::pt, resolved);
  3592   __ delayed()->nop();
  3595   // Extract target class from constant pool
  3596   __ bind(quicked);
  3597   __ add(Roffset, sizeof(constantPoolOopDesc), Roffset);
  3598   __ get_constant_pool(Lscratch);
  3599   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3600   __ bind(resolved);
  3601   __ load_klass(Otos_i, RobjKlass); // get value klass
  3603   // Generate a fast subtype check.  Branch to cast_ok if no
  3604   // failure.  Return 0 if failure.
  3605   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
  3606   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
  3607   // Not a subtype; return 0;
  3608   __ clr( Otos_i );
  3610   if (ProfileInterpreter) {
  3611     __ ba(false, done);
  3612     __ delayed()->nop();
  3614   __ bind(is_null);
  3615   __ profile_null_seen(G3_scratch);
  3616   __ bind(done);
  3619 void TemplateTable::_breakpoint() {
  3621    // Note: We get here even if we are single stepping..
  3622    // jbug inists on setting breakpoints at every bytecode
  3623    // even if we are in single step mode.
  3625    transition(vtos, vtos);
  3626    // get the unpatched byte code
  3627    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
  3628    __ mov(O0, Lbyte_code);
  3630    // post the breakpoint event
  3631    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
  3633    // complete the execution of original bytecode
  3634    __ dispatch_normal(vtos);
  3638 //----------------------------------------------------------------------------------------------------
  3639 // Exceptions
  3641 void TemplateTable::athrow() {
  3642   transition(atos, vtos);
  3644   // This works because exception is cached in Otos_i which is same as O0,
  3645   // which is same as what throw_exception_entry_expects
  3646   assert(Otos_i == Oexception, "see explanation above");
  3648   __ verify_oop(Otos_i);
  3649   __ null_check(Otos_i);
  3650   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
  3654 //----------------------------------------------------------------------------------------------------
  3655 // Synchronization
  3658 // See frame_sparc.hpp for monitor block layout.
  3659 // Monitor elements are dynamically allocated by growing stack as needed.
  3661 void TemplateTable::monitorenter() {
  3662   transition(atos, vtos);
  3663   __ verify_oop(Otos_i);
  3664   // Try to acquire a lock on the object
  3665   // Repeat until succeeded (i.e., until
  3666   // monitorenter returns true).
  3668   {   Label ok;
  3669     __ tst(Otos_i);
  3670     __ throw_if_not_1_x( Assembler::notZero,  ok);
  3671     __ delayed()->mov(Otos_i, Lscratch); // save obj
  3672     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3675   assert(O0 == Otos_i, "Be sure where the object to lock is");
  3677   // find a free slot in the monitor block
  3680   // initialize entry pointer
  3681   __ clr(O1); // points to free slot or NULL
  3684     Label entry, loop, exit;
  3685     __ add( __ top_most_monitor(), O2 ); // last one to check
  3686     __ ba( false, entry );
  3687     __ delayed()->mov( Lmonitors, O3 ); // first one to check
  3690     __ bind( loop );
  3692     __ verify_oop(O4);          // verify each monitor's oop
  3693     __ tst(O4); // is this entry unused?
  3694     if (VM_Version::v9_instructions_work())
  3695       __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
  3696     else {
  3697       Label L;
  3698       __ br( Assembler::zero, true, Assembler::pn, L );
  3699       __ delayed()->mov(O3, O1); // rememeber this one if match
  3700       __ bind(L);
  3703     __ cmp(O4, O0); // check if current entry is for same object
  3704     __ brx( Assembler::equal, false, Assembler::pn, exit );
  3705     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
  3707     __ bind( entry );
  3709     __ cmp( O3, O2 );
  3710     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3711     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
  3713     __ bind( exit );
  3716   { Label allocated;
  3718     // found free slot?
  3719     __ br_notnull(O1, false, Assembler::pn, allocated);
  3720     __ delayed()->nop();
  3722     __ add_monitor_to_stack( false, O2, O3 );
  3723     __ mov(Lmonitors, O1);
  3725     __ bind(allocated);
  3728   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
  3729   // The object has already been poped from the stack, so the expression stack looks correct.
  3730   __ inc(Lbcp);
  3732   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
  3733   __ lock_object(O1, O0);
  3735   // check if there's enough space on the stack for the monitors after locking
  3736   __ generate_stack_overflow_check(0);
  3738   // The bcp has already been incremented. Just need to dispatch to next instruction.
  3739   __ dispatch_next(vtos);
  3743 void TemplateTable::monitorexit() {
  3744   transition(atos, vtos);
  3745   __ verify_oop(Otos_i);
  3746   __ tst(Otos_i);
  3747   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
  3749   assert(O0 == Otos_i, "just checking");
  3751   { Label entry, loop, found;
  3752     __ add( __ top_most_monitor(), O2 ); // last one to check
  3753     __ ba(false, entry );
  3754     // use Lscratch to hold monitor elem to check, start with most recent monitor,
  3755     // By using a local it survives the call to the C routine.
  3756     __ delayed()->mov( Lmonitors, Lscratch );
  3758     __ bind( loop );
  3760     __ verify_oop(O4);          // verify each monitor's oop
  3761     __ cmp(O4, O0); // check if current entry is for desired object
  3762     __ brx( Assembler::equal, true, Assembler::pt, found );
  3763     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
  3765     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
  3767     __ bind( entry );
  3769     __ cmp( Lscratch, O2 );
  3770     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3771     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
  3773     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
  3774     __ should_not_reach_here();
  3776     __ bind(found);
  3778   __ unlock_object(O1);
  3782 //----------------------------------------------------------------------------------------------------
  3783 // Wide instructions
  3785 void TemplateTable::wide() {
  3786   transition(vtos, vtos);
  3787   __ ldub(Lbcp, 1, G3_scratch);// get next bc
  3788   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
  3789   AddressLiteral ep(Interpreter::_wentry_point);
  3790   __ set(ep, G4_scratch);
  3791   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
  3792   __ jmp(G3_scratch, G0);
  3793   __ delayed()->nop();
  3794   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
  3798 //----------------------------------------------------------------------------------------------------
  3799 // Multi arrays
  3801 void TemplateTable::multianewarray() {
  3802   transition(vtos, atos);
  3803      // put ndims * wordSize into Lscratch
  3804   __ ldub( Lbcp,     3,               Lscratch);
  3805   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
  3806      // Lesp points past last_dim, so set to O1 to first_dim address
  3807   __ add(  Lesp,     Lscratch,        O1);
  3808      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
  3809   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
  3811 #endif /* !CC_INTERP */

mercurial