src/cpu/sparc/vm/templateTable_sparc.cpp

Tue, 24 Dec 2013 11:48:39 -0800

author
mikael
date
Tue, 24 Dec 2013 11:48:39 -0800
changeset 6198
55fb97c4c58d
parent 6039
bd3237e0e18d
child 6223
add2caa66e7e
permissions
-rw-r--r--

8029233: Update copyright year to match last edit in jdk8 hotspot repository for 2013
Summary: Copyright year updated for files modified during 2013
Reviewed-by: twisti, iveresov

     1 /*
     2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "interpreter/interpreter.hpp"
    27 #include "interpreter/interpreterRuntime.hpp"
    28 #include "interpreter/templateTable.hpp"
    29 #include "memory/universe.inline.hpp"
    30 #include "oops/methodData.hpp"
    31 #include "oops/objArrayKlass.hpp"
    32 #include "oops/oop.inline.hpp"
    33 #include "prims/methodHandles.hpp"
    34 #include "runtime/sharedRuntime.hpp"
    35 #include "runtime/stubRoutines.hpp"
    36 #include "runtime/synchronizer.hpp"
    37 #include "utilities/macros.hpp"
    39 #ifndef CC_INTERP
    40 #define __ _masm->
    42 // Misc helpers
    44 // Do an oop store like *(base + index + offset) = val
    45 // index can be noreg,
    46 static void do_oop_store(InterpreterMacroAssembler* _masm,
    47                          Register base,
    48                          Register index,
    49                          int offset,
    50                          Register val,
    51                          Register tmp,
    52                          BarrierSet::Name barrier,
    53                          bool precise) {
    54   assert(tmp != val && tmp != base && tmp != index, "register collision");
    55   assert(index == noreg || offset == 0, "only one offset");
    56   switch (barrier) {
    57 #if INCLUDE_ALL_GCS
    58     case BarrierSet::G1SATBCT:
    59     case BarrierSet::G1SATBCTLogging:
    60       {
    61         // Load and record the previous value.
    62         __ g1_write_barrier_pre(base, index, offset,
    63                                 noreg /* pre_val */,
    64                                 tmp, true /*preserve_o_regs*/);
    66         // G1 barrier needs uncompressed oop for region cross check.
    67         Register new_val = val;
    68         if (UseCompressedOops && val != G0) {
    69           new_val = tmp;
    70           __ mov(val, new_val);
    71         }
    73         if (index == noreg ) {
    74           assert(Assembler::is_simm13(offset), "fix this code");
    75           __ store_heap_oop(val, base, offset);
    76         } else {
    77           __ store_heap_oop(val, base, index);
    78         }
    80         // No need for post barrier if storing NULL
    81         if (val != G0) {
    82           if (precise) {
    83             if (index == noreg) {
    84               __ add(base, offset, base);
    85             } else {
    86               __ add(base, index, base);
    87             }
    88           }
    89           __ g1_write_barrier_post(base, new_val, tmp);
    90         }
    91       }
    92       break;
    93 #endif // INCLUDE_ALL_GCS
    94     case BarrierSet::CardTableModRef:
    95     case BarrierSet::CardTableExtension:
    96       {
    97         if (index == noreg ) {
    98           assert(Assembler::is_simm13(offset), "fix this code");
    99           __ store_heap_oop(val, base, offset);
   100         } else {
   101           __ store_heap_oop(val, base, index);
   102         }
   103         // No need for post barrier if storing NULL
   104         if (val != G0) {
   105           if (precise) {
   106             if (index == noreg) {
   107               __ add(base, offset, base);
   108             } else {
   109               __ add(base, index, base);
   110             }
   111           }
   112           __ card_write_barrier_post(base, val, tmp);
   113         }
   114       }
   115       break;
   116     case BarrierSet::ModRef:
   117     case BarrierSet::Other:
   118       ShouldNotReachHere();
   119       break;
   120     default      :
   121       ShouldNotReachHere();
   123   }
   124 }
   127 //----------------------------------------------------------------------------------------------------
   128 // Platform-dependent initialization
   130 void TemplateTable::pd_initialize() {
   131   // (none)
   132 }
   135 //----------------------------------------------------------------------------------------------------
   136 // Condition conversion
   137 Assembler::Condition ccNot(TemplateTable::Condition cc) {
   138   switch (cc) {
   139     case TemplateTable::equal        : return Assembler::notEqual;
   140     case TemplateTable::not_equal    : return Assembler::equal;
   141     case TemplateTable::less         : return Assembler::greaterEqual;
   142     case TemplateTable::less_equal   : return Assembler::greater;
   143     case TemplateTable::greater      : return Assembler::lessEqual;
   144     case TemplateTable::greater_equal: return Assembler::less;
   145   }
   146   ShouldNotReachHere();
   147   return Assembler::zero;
   148 }
   150 //----------------------------------------------------------------------------------------------------
   151 // Miscelaneous helper routines
   154 Address TemplateTable::at_bcp(int offset) {
   155   assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
   156   return Address(Lbcp, offset);
   157 }
   160 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
   161                                    Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
   162                                    int byte_no) {
   163   // With sharing on, may need to test Method* flag.
   164   if (!RewriteBytecodes)  return;
   165   Label L_patch_done;
   167   switch (bc) {
   168   case Bytecodes::_fast_aputfield:
   169   case Bytecodes::_fast_bputfield:
   170   case Bytecodes::_fast_cputfield:
   171   case Bytecodes::_fast_dputfield:
   172   case Bytecodes::_fast_fputfield:
   173   case Bytecodes::_fast_iputfield:
   174   case Bytecodes::_fast_lputfield:
   175   case Bytecodes::_fast_sputfield:
   176     {
   177       // We skip bytecode quickening for putfield instructions when
   178       // the put_code written to the constant pool cache is zero.
   179       // This is required so that every execution of this instruction
   180       // calls out to InterpreterRuntime::resolve_get_put to do
   181       // additional, required work.
   182       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
   183       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
   184       __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
   185       __ set(bc, bc_reg);
   186       __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
   187     }
   188     break;
   189   default:
   190     assert(byte_no == -1, "sanity");
   191     if (load_bc_into_bc_reg) {
   192       __ set(bc, bc_reg);
   193     }
   194   }
   196   if (JvmtiExport::can_post_breakpoint()) {
   197     Label L_fast_patch;
   198     __ ldub(at_bcp(0), temp_reg);
   199     __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
   200     // perform the quickening, slowly, in the bowels of the breakpoint table
   201     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
   202     __ ba_short(L_patch_done);
   203     __ bind(L_fast_patch);
   204   }
   206 #ifdef ASSERT
   207   Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
   208   Label L_okay;
   209   __ ldub(at_bcp(0), temp_reg);
   210   __ cmp(temp_reg, orig_bytecode);
   211   __ br(Assembler::equal, false, Assembler::pt, L_okay);
   212   __ delayed()->cmp(temp_reg, bc_reg);
   213   __ br(Assembler::equal, false, Assembler::pt, L_okay);
   214   __ delayed()->nop();
   215   __ stop("patching the wrong bytecode");
   216   __ bind(L_okay);
   217 #endif
   219   // patch bytecode
   220   __ stb(bc_reg, at_bcp(0));
   221   __ bind(L_patch_done);
   222 }
   224 //----------------------------------------------------------------------------------------------------
   225 // Individual instructions
   227 void TemplateTable::nop() {
   228   transition(vtos, vtos);
   229   // nothing to do
   230 }
   232 void TemplateTable::shouldnotreachhere() {
   233   transition(vtos, vtos);
   234   __ stop("shouldnotreachhere bytecode");
   235 }
   237 void TemplateTable::aconst_null() {
   238   transition(vtos, atos);
   239   __ clr(Otos_i);
   240 }
   243 void TemplateTable::iconst(int value) {
   244   transition(vtos, itos);
   245   __ set(value, Otos_i);
   246 }
   249 void TemplateTable::lconst(int value) {
   250   transition(vtos, ltos);
   251   assert(value >= 0, "check this code");
   252 #ifdef _LP64
   253   __ set(value, Otos_l);
   254 #else
   255   __ set(value, Otos_l2);
   256   __ clr( Otos_l1);
   257 #endif
   258 }
   261 void TemplateTable::fconst(int value) {
   262   transition(vtos, ftos);
   263   static float zero = 0.0, one = 1.0, two = 2.0;
   264   float* p;
   265   switch( value ) {
   266    default: ShouldNotReachHere();
   267    case 0:  p = &zero;  break;
   268    case 1:  p = &one;   break;
   269    case 2:  p = &two;   break;
   270   }
   271   AddressLiteral a(p);
   272   __ sethi(a, G3_scratch);
   273   __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
   274 }
   277 void TemplateTable::dconst(int value) {
   278   transition(vtos, dtos);
   279   static double zero = 0.0, one = 1.0;
   280   double* p;
   281   switch( value ) {
   282    default: ShouldNotReachHere();
   283    case 0:  p = &zero;  break;
   284    case 1:  p = &one;   break;
   285   }
   286   AddressLiteral a(p);
   287   __ sethi(a, G3_scratch);
   288   __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
   289 }
   292 // %%%%% Should factore most snippet templates across platforms
   294 void TemplateTable::bipush() {
   295   transition(vtos, itos);
   296   __ ldsb( at_bcp(1), Otos_i );
   297 }
   299 void TemplateTable::sipush() {
   300   transition(vtos, itos);
   301   __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
   302 }
   304 void TemplateTable::ldc(bool wide) {
   305   transition(vtos, vtos);
   306   Label call_ldc, notInt, isString, notString, notClass, exit;
   308   if (wide) {
   309     __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   310   } else {
   311     __ ldub(Lbcp, 1, O1);
   312   }
   313   __ get_cpool_and_tags(O0, O2);
   315   const int base_offset = ConstantPool::header_size() * wordSize;
   316   const int tags_offset = Array<u1>::base_offset_in_bytes();
   318   // get type from tags
   319   __ add(O2, tags_offset, O2);
   320   __ ldub(O2, O1, O2);
   322   // unresolved class? If so, must resolve
   323   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClass, Assembler::equal, Assembler::pt, call_ldc);
   325   // unresolved class in error state
   326   __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClassInError, Assembler::equal, Assembler::pn, call_ldc);
   328   __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
   329   __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
   330   __ delayed()->add(O0, base_offset, O0);
   332   __ bind(call_ldc);
   333   __ set(wide, O1);
   334   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
   335   __ push(atos);
   336   __ ba_short(exit);
   338   __ bind(notClass);
   339  // __ add(O0, base_offset, O0);
   340   __ sll(O1, LogBytesPerWord, O1);
   341   __ cmp(O2, JVM_CONSTANT_Integer);
   342   __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
   343   __ delayed()->cmp(O2, JVM_CONSTANT_String);
   344   __ ld(O0, O1, Otos_i);
   345   __ push(itos);
   346   __ ba_short(exit);
   348   __ bind(notInt);
   349  // __ cmp(O2, JVM_CONSTANT_String);
   350   __ brx(Assembler::notEqual, true, Assembler::pt, notString);
   351   __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   352   __ bind(isString);
   353   __ stop("string should be rewritten to fast_aldc");
   354   __ ba_short(exit);
   356   __ bind(notString);
   357  // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
   358   __ push(ftos);
   360   __ bind(exit);
   361 }
   363 // Fast path for caching oop constants.
   364 // %%% We should use this to handle Class and String constants also.
   365 // %%% It will simplify the ldc/primitive path considerably.
   366 void TemplateTable::fast_aldc(bool wide) {
   367   transition(vtos, atos);
   369   int index_size = wide ? sizeof(u2) : sizeof(u1);
   370   Label resolved;
   372   // We are resolved if the resolved reference cache entry contains a
   373   // non-null object (CallSite, etc.)
   374   assert_different_registers(Otos_i, G3_scratch);
   375   __ get_cache_index_at_bcp(Otos_i, G3_scratch, 1, index_size);  // load index => G3_scratch
   376   __ load_resolved_reference_at_index(Otos_i, G3_scratch);
   377   __ tst(Otos_i);
   378   __ br(Assembler::notEqual, false, Assembler::pt, resolved);
   379   __ delayed()->set((int)bytecode(), O1);
   381   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
   383   // first time invocation - must resolve first
   384   __ call_VM(Otos_i, entry, O1);
   385   __ bind(resolved);
   386   __ verify_oop(Otos_i);
   387 }
   390 void TemplateTable::ldc2_w() {
   391   transition(vtos, vtos);
   392   Label Long, exit;
   394   __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
   395   __ get_cpool_and_tags(O0, O2);
   397   const int base_offset = ConstantPool::header_size() * wordSize;
   398   const int tags_offset = Array<u1>::base_offset_in_bytes();
   399   // get type from tags
   400   __ add(O2, tags_offset, O2);
   401   __ ldub(O2, O1, O2);
   403   __ sll(O1, LogBytesPerWord, O1);
   404   __ add(O0, O1, G3_scratch);
   406   __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, Long);
   407   // A double can be placed at word-aligned locations in the constant pool.
   408   // Check out Conversions.java for an example.
   409   // Also ConstantPool::header_size() is 20, which makes it very difficult
   410   // to double-align double on the constant pool.  SG, 11/7/97
   411 #ifdef _LP64
   412   __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
   413 #else
   414   FloatRegister f = Ftos_d;
   415   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
   416   __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
   417          f->successor());
   418 #endif
   419   __ push(dtos);
   420   __ ba_short(exit);
   422   __ bind(Long);
   423 #ifdef _LP64
   424   __ ldx(G3_scratch, base_offset, Otos_l);
   425 #else
   426   __ ld(G3_scratch, base_offset, Otos_l);
   427   __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
   428 #endif
   429   __ push(ltos);
   431   __ bind(exit);
   432 }
   435 void TemplateTable::locals_index(Register reg, int offset) {
   436   __ ldub( at_bcp(offset), reg );
   437 }
   440 void TemplateTable::locals_index_wide(Register reg) {
   441   // offset is 2, not 1, because Lbcp points to wide prefix code
   442   __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
   443 }
   445 void TemplateTable::iload() {
   446   transition(vtos, itos);
   447   // Rewrite iload,iload  pair into fast_iload2
   448   //         iload,caload pair into fast_icaload
   449   if (RewriteFrequentPairs) {
   450     Label rewrite, done;
   452     // get next byte
   453     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
   455     // if _iload, wait to rewrite to iload2.  We only want to rewrite the
   456     // last two iloads in a pair.  Comparing against fast_iload means that
   457     // the next bytecode is neither an iload or a caload, and therefore
   458     // an iload pair.
   459     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
   461     __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
   462     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   463     __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
   465     __ cmp(G3_scratch, (int)Bytecodes::_caload);
   466     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   467     __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
   469     __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
   470     // rewrite
   471     // G4_scratch: fast bytecode
   472     __ bind(rewrite);
   473     patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
   474     __ bind(done);
   475   }
   477   // Get the local value into tos
   478   locals_index(G3_scratch);
   479   __ access_local_int( G3_scratch, Otos_i );
   480 }
   482 void TemplateTable::fast_iload2() {
   483   transition(vtos, itos);
   484   locals_index(G3_scratch);
   485   __ access_local_int( G3_scratch, Otos_i );
   486   __ push_i();
   487   locals_index(G3_scratch, 3);  // get next bytecode's local index.
   488   __ access_local_int( G3_scratch, Otos_i );
   489 }
   491 void TemplateTable::fast_iload() {
   492   transition(vtos, itos);
   493   locals_index(G3_scratch);
   494   __ access_local_int( G3_scratch, Otos_i );
   495 }
   497 void TemplateTable::lload() {
   498   transition(vtos, ltos);
   499   locals_index(G3_scratch);
   500   __ access_local_long( G3_scratch, Otos_l );
   501 }
   504 void TemplateTable::fload() {
   505   transition(vtos, ftos);
   506   locals_index(G3_scratch);
   507   __ access_local_float( G3_scratch, Ftos_f );
   508 }
   511 void TemplateTable::dload() {
   512   transition(vtos, dtos);
   513   locals_index(G3_scratch);
   514   __ access_local_double( G3_scratch, Ftos_d );
   515 }
   518 void TemplateTable::aload() {
   519   transition(vtos, atos);
   520   locals_index(G3_scratch);
   521   __ access_local_ptr( G3_scratch, Otos_i);
   522 }
   525 void TemplateTable::wide_iload() {
   526   transition(vtos, itos);
   527   locals_index_wide(G3_scratch);
   528   __ access_local_int( G3_scratch, Otos_i );
   529 }
   532 void TemplateTable::wide_lload() {
   533   transition(vtos, ltos);
   534   locals_index_wide(G3_scratch);
   535   __ access_local_long( G3_scratch, Otos_l );
   536 }
   539 void TemplateTable::wide_fload() {
   540   transition(vtos, ftos);
   541   locals_index_wide(G3_scratch);
   542   __ access_local_float( G3_scratch, Ftos_f );
   543 }
   546 void TemplateTable::wide_dload() {
   547   transition(vtos, dtos);
   548   locals_index_wide(G3_scratch);
   549   __ access_local_double( G3_scratch, Ftos_d );
   550 }
   553 void TemplateTable::wide_aload() {
   554   transition(vtos, atos);
   555   locals_index_wide(G3_scratch);
   556   __ access_local_ptr( G3_scratch, Otos_i );
   557   __ verify_oop(Otos_i);
   558 }
   561 void TemplateTable::iaload() {
   562   transition(itos, itos);
   563   // Otos_i: index
   564   // tos: array
   565   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   566   __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
   567 }
   570 void TemplateTable::laload() {
   571   transition(itos, ltos);
   572   // Otos_i: index
   573   // O2: array
   574   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   575   __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
   576 }
   579 void TemplateTable::faload() {
   580   transition(itos, ftos);
   581   // Otos_i: index
   582   // O2: array
   583   __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
   584   __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
   585 }
   588 void TemplateTable::daload() {
   589   transition(itos, dtos);
   590   // Otos_i: index
   591   // O2: array
   592   __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
   593   __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
   594 }
   597 void TemplateTable::aaload() {
   598   transition(itos, atos);
   599   // Otos_i: index
   600   // tos: array
   601   __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
   602   __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
   603   __ verify_oop(Otos_i);
   604 }
   607 void TemplateTable::baload() {
   608   transition(itos, itos);
   609   // Otos_i: index
   610   // tos: array
   611   __ index_check(O2, Otos_i, 0, G3_scratch, O3);
   612   __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
   613 }
   616 void TemplateTable::caload() {
   617   transition(itos, itos);
   618   // Otos_i: index
   619   // tos: array
   620   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   621   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   622 }
   624 void TemplateTable::fast_icaload() {
   625   transition(vtos, itos);
   626   // Otos_i: index
   627   // tos: array
   628   locals_index(G3_scratch);
   629   __ access_local_int( G3_scratch, Otos_i );
   630   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   631   __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
   632 }
   635 void TemplateTable::saload() {
   636   transition(itos, itos);
   637   // Otos_i: index
   638   // tos: array
   639   __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
   640   __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
   641 }
   644 void TemplateTable::iload(int n) {
   645   transition(vtos, itos);
   646   __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   647 }
   650 void TemplateTable::lload(int n) {
   651   transition(vtos, ltos);
   652   assert(n+1 < Argument::n_register_parameters, "would need more code");
   653   __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
   654 }
   657 void TemplateTable::fload(int n) {
   658   transition(vtos, ftos);
   659   assert(n < Argument::n_register_parameters, "would need more code");
   660   __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
   661 }
   664 void TemplateTable::dload(int n) {
   665   transition(vtos, dtos);
   666   FloatRegister dst = Ftos_d;
   667   __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
   668 }
   671 void TemplateTable::aload(int n) {
   672   transition(vtos, atos);
   673   __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
   674 }
   677 void TemplateTable::aload_0() {
   678   transition(vtos, atos);
   680   // According to bytecode histograms, the pairs:
   681   //
   682   // _aload_0, _fast_igetfield (itos)
   683   // _aload_0, _fast_agetfield (atos)
   684   // _aload_0, _fast_fgetfield (ftos)
   685   //
   686   // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
   687   // bytecode checks the next bytecode and then rewrites the current
   688   // bytecode into a pair bytecode; otherwise it rewrites the current
   689   // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
   690   //
   691   if (RewriteFrequentPairs) {
   692     Label rewrite, done;
   694     // get next byte
   695     __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
   697     // do actual aload_0
   698     aload(0);
   700     // if _getfield then wait with rewrite
   701     __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
   703     // if _igetfield then rewrite to _fast_iaccess_0
   704     assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   705     __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
   706     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   707     __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
   709     // if _agetfield then rewrite to _fast_aaccess_0
   710     assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   711     __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
   712     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   713     __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
   715     // if _fgetfield then rewrite to _fast_faccess_0
   716     assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   717     __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
   718     __ br(Assembler::equal, false, Assembler::pn, rewrite);
   719     __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
   721     // else rewrite to _fast_aload0
   722     assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
   723     __ set(Bytecodes::_fast_aload_0, G4_scratch);
   725     // rewrite
   726     // G4_scratch: fast bytecode
   727     __ bind(rewrite);
   728     patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
   729     __ bind(done);
   730   } else {
   731     aload(0);
   732   }
   733 }
   736 void TemplateTable::istore() {
   737   transition(itos, vtos);
   738   locals_index(G3_scratch);
   739   __ store_local_int( G3_scratch, Otos_i );
   740 }
   743 void TemplateTable::lstore() {
   744   transition(ltos, vtos);
   745   locals_index(G3_scratch);
   746   __ store_local_long( G3_scratch, Otos_l );
   747 }
   750 void TemplateTable::fstore() {
   751   transition(ftos, vtos);
   752   locals_index(G3_scratch);
   753   __ store_local_float( G3_scratch, Ftos_f );
   754 }
   757 void TemplateTable::dstore() {
   758   transition(dtos, vtos);
   759   locals_index(G3_scratch);
   760   __ store_local_double( G3_scratch, Ftos_d );
   761 }
   764 void TemplateTable::astore() {
   765   transition(vtos, vtos);
   766   __ load_ptr(0, Otos_i);
   767   __ inc(Lesp, Interpreter::stackElementSize);
   768   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   769   locals_index(G3_scratch);
   770   __ store_local_ptr(G3_scratch, Otos_i);
   771 }
   774 void TemplateTable::wide_istore() {
   775   transition(vtos, vtos);
   776   __ pop_i();
   777   locals_index_wide(G3_scratch);
   778   __ store_local_int( G3_scratch, Otos_i );
   779 }
   782 void TemplateTable::wide_lstore() {
   783   transition(vtos, vtos);
   784   __ pop_l();
   785   locals_index_wide(G3_scratch);
   786   __ store_local_long( G3_scratch, Otos_l );
   787 }
   790 void TemplateTable::wide_fstore() {
   791   transition(vtos, vtos);
   792   __ pop_f();
   793   locals_index_wide(G3_scratch);
   794   __ store_local_float( G3_scratch, Ftos_f );
   795 }
   798 void TemplateTable::wide_dstore() {
   799   transition(vtos, vtos);
   800   __ pop_d();
   801   locals_index_wide(G3_scratch);
   802   __ store_local_double( G3_scratch, Ftos_d );
   803 }
   806 void TemplateTable::wide_astore() {
   807   transition(vtos, vtos);
   808   __ load_ptr(0, Otos_i);
   809   __ inc(Lesp, Interpreter::stackElementSize);
   810   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   811   locals_index_wide(G3_scratch);
   812   __ store_local_ptr(G3_scratch, Otos_i);
   813 }
   816 void TemplateTable::iastore() {
   817   transition(itos, vtos);
   818   __ pop_i(O2); // index
   819   // Otos_i: val
   820   // O3: array
   821   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   822   __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
   823 }
   826 void TemplateTable::lastore() {
   827   transition(ltos, vtos);
   828   __ pop_i(O2); // index
   829   // Otos_l: val
   830   // O3: array
   831   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   832   __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
   833 }
   836 void TemplateTable::fastore() {
   837   transition(ftos, vtos);
   838   __ pop_i(O2); // index
   839   // Ftos_f: val
   840   // O3: array
   841   __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
   842   __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
   843 }
   846 void TemplateTable::dastore() {
   847   transition(dtos, vtos);
   848   __ pop_i(O2); // index
   849   // Fos_d: val
   850   // O3: array
   851   __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
   852   __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
   853 }
   856 void TemplateTable::aastore() {
   857   Label store_ok, is_null, done;
   858   transition(vtos, vtos);
   859   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
   860   __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
   861   __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
   862   // Otos_i: val
   863   // O2: index
   864   // O3: array
   865   __ verify_oop(Otos_i);
   866   __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
   868   // do array store check - check for NULL value first
   869   __ br_null_short( Otos_i, Assembler::pn, is_null );
   871   __ load_klass(O3, O4); // get array klass
   872   __ load_klass(Otos_i, O5); // get value klass
   874   // do fast instanceof cache test
   876   __ ld_ptr(O4,     in_bytes(ObjArrayKlass::element_klass_offset()),  O4);
   878   assert(Otos_i == O0, "just checking");
   880   // Otos_i:    value
   881   // O1:        addr - offset
   882   // O2:        index
   883   // O3:        array
   884   // O4:        array element klass
   885   // O5:        value klass
   887   // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
   889   // Generate a fast subtype check.  Branch to store_ok if no
   890   // failure.  Throw if failure.
   891   __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
   893   // Not a subtype; so must throw exception
   894   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
   896   // Store is OK.
   897   __ bind(store_ok);
   898   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
   900   __ ba(done);
   901   __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
   903   __ bind(is_null);
   904   do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
   906   __ profile_null_seen(G3_scratch);
   907   __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
   908   __ bind(done);
   909 }
   912 void TemplateTable::bastore() {
   913   transition(itos, vtos);
   914   __ pop_i(O2); // index
   915   // Otos_i: val
   916   // O3: array
   917   __ index_check(O3, O2, 0, G3_scratch, O2);
   918   __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
   919 }
   922 void TemplateTable::castore() {
   923   transition(itos, vtos);
   924   __ pop_i(O2); // index
   925   // Otos_i: val
   926   // O3: array
   927   __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
   928   __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
   929 }
   932 void TemplateTable::sastore() {
   933   // %%%%% Factor across platform
   934   castore();
   935 }
   938 void TemplateTable::istore(int n) {
   939   transition(itos, vtos);
   940   __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
   941 }
   944 void TemplateTable::lstore(int n) {
   945   transition(ltos, vtos);
   946   assert(n+1 < Argument::n_register_parameters, "only handle register cases");
   947   __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
   949 }
   952 void TemplateTable::fstore(int n) {
   953   transition(ftos, vtos);
   954   assert(n < Argument::n_register_parameters, "only handle register cases");
   955   __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
   956 }
   959 void TemplateTable::dstore(int n) {
   960   transition(dtos, vtos);
   961   FloatRegister src = Ftos_d;
   962   __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
   963 }
   966 void TemplateTable::astore(int n) {
   967   transition(vtos, vtos);
   968   __ load_ptr(0, Otos_i);
   969   __ inc(Lesp, Interpreter::stackElementSize);
   970   __ verify_oop_or_return_address(Otos_i, G3_scratch);
   971   __ store_local_ptr(n, Otos_i);
   972 }
   975 void TemplateTable::pop() {
   976   transition(vtos, vtos);
   977   __ inc(Lesp, Interpreter::stackElementSize);
   978 }
   981 void TemplateTable::pop2() {
   982   transition(vtos, vtos);
   983   __ inc(Lesp, 2 * Interpreter::stackElementSize);
   984 }
   987 void TemplateTable::dup() {
   988   transition(vtos, vtos);
   989   // stack: ..., a
   990   // load a and tag
   991   __ load_ptr(0, Otos_i);
   992   __ push_ptr(Otos_i);
   993   // stack: ..., a, a
   994 }
   997 void TemplateTable::dup_x1() {
   998   transition(vtos, vtos);
   999   // stack: ..., a, b
  1000   __ load_ptr( 1, G3_scratch);  // get a
  1001   __ load_ptr( 0, Otos_l1);     // get b
  1002   __ store_ptr(1, Otos_l1);     // put b
  1003   __ store_ptr(0, G3_scratch);  // put a - like swap
  1004   __ push_ptr(Otos_l1);         // push b
  1005   // stack: ..., b, a, b
  1009 void TemplateTable::dup_x2() {
  1010   transition(vtos, vtos);
  1011   // stack: ..., a, b, c
  1012   // get c and push on stack, reuse registers
  1013   __ load_ptr( 0, G3_scratch);  // get c
  1014   __ push_ptr(G3_scratch);      // push c with tag
  1015   // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
  1016   // (stack offsets n+1 now)
  1017   __ load_ptr( 3, Otos_l1);     // get a
  1018   __ store_ptr(3, G3_scratch);  // put c at 3
  1019   // stack: ..., c, b, c, c  (a in reg)
  1020   __ load_ptr( 2, G3_scratch);  // get b
  1021   __ store_ptr(2, Otos_l1);     // put a at 2
  1022   // stack: ..., c, a, c, c  (b in reg)
  1023   __ store_ptr(1, G3_scratch);  // put b at 1
  1024   // stack: ..., c, a, b, c
  1028 void TemplateTable::dup2() {
  1029   transition(vtos, vtos);
  1030   __ load_ptr(1, G3_scratch);  // get a
  1031   __ load_ptr(0, Otos_l1);     // get b
  1032   __ push_ptr(G3_scratch);     // push a
  1033   __ push_ptr(Otos_l1);        // push b
  1034   // stack: ..., a, b, a, b
  1038 void TemplateTable::dup2_x1() {
  1039   transition(vtos, vtos);
  1040   // stack: ..., a, b, c
  1041   __ load_ptr( 1, Lscratch);    // get b
  1042   __ load_ptr( 2, Otos_l1);     // get a
  1043   __ store_ptr(2, Lscratch);    // put b at a
  1044   // stack: ..., b, b, c
  1045   __ load_ptr( 0, G3_scratch);  // get c
  1046   __ store_ptr(1, G3_scratch);  // put c at b
  1047   // stack: ..., b, c, c
  1048   __ store_ptr(0, Otos_l1);     // put a at c
  1049   // stack: ..., b, c, a
  1050   __ push_ptr(Lscratch);        // push b
  1051   __ push_ptr(G3_scratch);      // push c
  1052   // stack: ..., b, c, a, b, c
  1056 // The spec says that these types can be a mixture of category 1 (1 word)
  1057 // types and/or category 2 types (long and doubles)
  1058 void TemplateTable::dup2_x2() {
  1059   transition(vtos, vtos);
  1060   // stack: ..., a, b, c, d
  1061   __ load_ptr( 1, Lscratch);    // get c
  1062   __ load_ptr( 3, Otos_l1);     // get a
  1063   __ store_ptr(3, Lscratch);    // put c at 3
  1064   __ store_ptr(1, Otos_l1);     // put a at 1
  1065   // stack: ..., c, b, a, d
  1066   __ load_ptr( 2, G3_scratch);  // get b
  1067   __ load_ptr( 0, Otos_l1);     // get d
  1068   __ store_ptr(0, G3_scratch);  // put b at 0
  1069   __ store_ptr(2, Otos_l1);     // put d at 2
  1070   // stack: ..., c, d, a, b
  1071   __ push_ptr(Lscratch);        // push c
  1072   __ push_ptr(Otos_l1);         // push d
  1073   // stack: ..., c, d, a, b, c, d
  1077 void TemplateTable::swap() {
  1078   transition(vtos, vtos);
  1079   // stack: ..., a, b
  1080   __ load_ptr( 1, G3_scratch);  // get a
  1081   __ load_ptr( 0, Otos_l1);     // get b
  1082   __ store_ptr(0, G3_scratch);  // put b
  1083   __ store_ptr(1, Otos_l1);     // put a
  1084   // stack: ..., b, a
  1088 void TemplateTable::iop2(Operation op) {
  1089   transition(itos, itos);
  1090   __ pop_i(O1);
  1091   switch (op) {
  1092    case  add:  __  add(O1, Otos_i, Otos_i);  break;
  1093    case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
  1094      // %%%%% Mul may not exist: better to call .mul?
  1095    case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
  1096    case _and:  __ and3(O1, Otos_i, Otos_i);  break;
  1097    case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
  1098    case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
  1099    case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
  1100    case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
  1101    case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
  1102    default: ShouldNotReachHere();
  1107 void TemplateTable::lop2(Operation op) {
  1108   transition(ltos, ltos);
  1109   __ pop_l(O2);
  1110   switch (op) {
  1111 #ifdef _LP64
  1112    case  add:  __  add(O2, Otos_l, Otos_l);  break;
  1113    case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
  1114    case _and:  __ and3(O2, Otos_l, Otos_l);  break;
  1115    case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
  1116    case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
  1117 #else
  1118    case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
  1119    case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
  1120    case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
  1121    case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
  1122    case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
  1123 #endif
  1124    default: ShouldNotReachHere();
  1129 void TemplateTable::idiv() {
  1130   // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
  1131   // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
  1133   transition(itos, itos);
  1134   __ pop_i(O1); // get 1st op
  1136   // Y contains upper 32 bits of result, set it to 0 or all ones
  1137   __ wry(G0);
  1138   __ mov(~0, G3_scratch);
  1140   __ tst(O1);
  1141      Label neg;
  1142   __ br(Assembler::negative, true, Assembler::pn, neg);
  1143   __ delayed()->wry(G3_scratch);
  1144   __ bind(neg);
  1146      Label ok;
  1147   __ tst(Otos_i);
  1148   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
  1150   const int min_int = 0x80000000;
  1151   Label regular;
  1152   __ cmp(Otos_i, -1);
  1153   __ br(Assembler::notEqual, false, Assembler::pt, regular);
  1154 #ifdef _LP64
  1155   // Don't put set in delay slot
  1156   // Set will turn into multiple instructions in 64 bit mode
  1157   __ delayed()->nop();
  1158   __ set(min_int, G4_scratch);
  1159 #else
  1160   __ delayed()->set(min_int, G4_scratch);
  1161 #endif
  1162   Label done;
  1163   __ cmp(O1, G4_scratch);
  1164   __ br(Assembler::equal, true, Assembler::pt, done);
  1165   __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
  1167   __ bind(regular);
  1168   __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
  1169   __ bind(done);
  1173 void TemplateTable::irem() {
  1174   transition(itos, itos);
  1175   __ mov(Otos_i, O2); // save divisor
  1176   idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
  1177   __ smul(Otos_i, O2, Otos_i);
  1178   __ sub(O1, Otos_i, Otos_i);
  1182 void TemplateTable::lmul() {
  1183   transition(ltos, ltos);
  1184   __ pop_l(O2);
  1185 #ifdef _LP64
  1186   __ mulx(Otos_l, O2, Otos_l);
  1187 #else
  1188   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
  1189 #endif
  1194 void TemplateTable::ldiv() {
  1195   transition(ltos, ltos);
  1197   // check for zero
  1198   __ pop_l(O2);
  1199 #ifdef _LP64
  1200   __ tst(Otos_l);
  1201   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1202   __ sdivx(O2, Otos_l, Otos_l);
  1203 #else
  1204   __ orcc(Otos_l1, Otos_l2, G0);
  1205   __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1206   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
  1207 #endif
  1211 void TemplateTable::lrem() {
  1212   transition(ltos, ltos);
  1214   // check for zero
  1215   __ pop_l(O2);
  1216 #ifdef _LP64
  1217   __ tst(Otos_l);
  1218   __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1219   __ sdivx(O2, Otos_l, Otos_l2);
  1220   __ mulx (Otos_l2, Otos_l, Otos_l2);
  1221   __ sub  (O2, Otos_l2, Otos_l);
  1222 #else
  1223   __ orcc(Otos_l1, Otos_l2, G0);
  1224   __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
  1225   __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
  1226 #endif
  1230 void TemplateTable::lshl() {
  1231   transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
  1233   __ pop_l(O2);                          // shift value in O2, O3
  1234 #ifdef _LP64
  1235   __ sllx(O2, Otos_i, Otos_l);
  1236 #else
  1237   __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1238 #endif
  1242 void TemplateTable::lshr() {
  1243   transition(itos, ltos); // %%%% see lshl comment
  1245   __ pop_l(O2);                          // shift value in O2, O3
  1246 #ifdef _LP64
  1247   __ srax(O2, Otos_i, Otos_l);
  1248 #else
  1249   __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1250 #endif
  1255 void TemplateTable::lushr() {
  1256   transition(itos, ltos); // %%%% see lshl comment
  1258   __ pop_l(O2);                          // shift value in O2, O3
  1259 #ifdef _LP64
  1260   __ srlx(O2, Otos_i, Otos_l);
  1261 #else
  1262   __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
  1263 #endif
  1267 void TemplateTable::fop2(Operation op) {
  1268   transition(ftos, ftos);
  1269   switch (op) {
  1270    case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1271    case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1272    case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1273    case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
  1274    case  rem:
  1275      assert(Ftos_f == F0, "just checking");
  1276 #ifdef _LP64
  1277      // LP64 calling conventions use F1, F3 for passing 2 floats
  1278      __ pop_f(F1);
  1279      __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
  1280 #else
  1281      __ pop_i(O0);
  1282      __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
  1283      __ ld( __ d_tmp, O1 );
  1284 #endif
  1285      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
  1286      assert( Ftos_f == F0, "fix this code" );
  1287      break;
  1289    default: ShouldNotReachHere();
  1294 void TemplateTable::dop2(Operation op) {
  1295   transition(dtos, dtos);
  1296   switch (op) {
  1297    case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1298    case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1299    case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1300    case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
  1301    case  rem:
  1302 #ifdef _LP64
  1303      // Pass arguments in D0, D2
  1304      __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
  1305      __ pop_d( F0 );
  1306 #else
  1307      // Pass arguments in O0O1, O2O3
  1308      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1309      __ ldd( __ d_tmp, O2 );
  1310      __ pop_d(Ftos_f);
  1311      __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
  1312      __ ldd( __ d_tmp, O0 );
  1313 #endif
  1314      __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
  1315      assert( Ftos_d == F0, "fix this code" );
  1316      break;
  1318    default: ShouldNotReachHere();
  1323 void TemplateTable::ineg() {
  1324   transition(itos, itos);
  1325   __ neg(Otos_i);
  1329 void TemplateTable::lneg() {
  1330   transition(ltos, ltos);
  1331 #ifdef _LP64
  1332   __ sub(G0, Otos_l, Otos_l);
  1333 #else
  1334   __ lneg(Otos_l1, Otos_l2);
  1335 #endif
  1339 void TemplateTable::fneg() {
  1340   transition(ftos, ftos);
  1341   __ fneg(FloatRegisterImpl::S, Ftos_f, Ftos_f);
  1345 void TemplateTable::dneg() {
  1346   transition(dtos, dtos);
  1347   __ fneg(FloatRegisterImpl::D, Ftos_f, Ftos_f);
  1351 void TemplateTable::iinc() {
  1352   transition(vtos, vtos);
  1353   locals_index(G3_scratch);
  1354   __ ldsb(Lbcp, 2, O2);  // load constant
  1355   __ access_local_int(G3_scratch, Otos_i);
  1356   __ add(Otos_i, O2, Otos_i);
  1357   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1361 void TemplateTable::wide_iinc() {
  1362   transition(vtos, vtos);
  1363   locals_index_wide(G3_scratch);
  1364   __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
  1365   __ access_local_int(G3_scratch, Otos_i);
  1366   __ add(Otos_i, O3, Otos_i);
  1367   __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
  1371 void TemplateTable::convert() {
  1372 // %%%%% Factor this first part accross platforms
  1373   #ifdef ASSERT
  1374     TosState tos_in  = ilgl;
  1375     TosState tos_out = ilgl;
  1376     switch (bytecode()) {
  1377       case Bytecodes::_i2l: // fall through
  1378       case Bytecodes::_i2f: // fall through
  1379       case Bytecodes::_i2d: // fall through
  1380       case Bytecodes::_i2b: // fall through
  1381       case Bytecodes::_i2c: // fall through
  1382       case Bytecodes::_i2s: tos_in = itos; break;
  1383       case Bytecodes::_l2i: // fall through
  1384       case Bytecodes::_l2f: // fall through
  1385       case Bytecodes::_l2d: tos_in = ltos; break;
  1386       case Bytecodes::_f2i: // fall through
  1387       case Bytecodes::_f2l: // fall through
  1388       case Bytecodes::_f2d: tos_in = ftos; break;
  1389       case Bytecodes::_d2i: // fall through
  1390       case Bytecodes::_d2l: // fall through
  1391       case Bytecodes::_d2f: tos_in = dtos; break;
  1392       default             : ShouldNotReachHere();
  1394     switch (bytecode()) {
  1395       case Bytecodes::_l2i: // fall through
  1396       case Bytecodes::_f2i: // fall through
  1397       case Bytecodes::_d2i: // fall through
  1398       case Bytecodes::_i2b: // fall through
  1399       case Bytecodes::_i2c: // fall through
  1400       case Bytecodes::_i2s: tos_out = itos; break;
  1401       case Bytecodes::_i2l: // fall through
  1402       case Bytecodes::_f2l: // fall through
  1403       case Bytecodes::_d2l: tos_out = ltos; break;
  1404       case Bytecodes::_i2f: // fall through
  1405       case Bytecodes::_l2f: // fall through
  1406       case Bytecodes::_d2f: tos_out = ftos; break;
  1407       case Bytecodes::_i2d: // fall through
  1408       case Bytecodes::_l2d: // fall through
  1409       case Bytecodes::_f2d: tos_out = dtos; break;
  1410       default             : ShouldNotReachHere();
  1412     transition(tos_in, tos_out);
  1413   #endif
  1416   // Conversion
  1417   Label done;
  1418   switch (bytecode()) {
  1419    case Bytecodes::_i2l:
  1420 #ifdef _LP64
  1421     // Sign extend the 32 bits
  1422     __ sra ( Otos_i, 0, Otos_l );
  1423 #else
  1424     __ addcc(Otos_i, 0, Otos_l2);
  1425     __ br(Assembler::greaterEqual, true, Assembler::pt, done);
  1426     __ delayed()->clr(Otos_l1);
  1427     __ set(~0, Otos_l1);
  1428 #endif
  1429     break;
  1431    case Bytecodes::_i2f:
  1432     __ st(Otos_i, __ d_tmp );
  1433     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1434     __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
  1435     break;
  1437    case Bytecodes::_i2d:
  1438     __ st(Otos_i, __ d_tmp);
  1439     __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
  1440     __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
  1441     break;
  1443    case Bytecodes::_i2b:
  1444     __ sll(Otos_i, 24, Otos_i);
  1445     __ sra(Otos_i, 24, Otos_i);
  1446     break;
  1448    case Bytecodes::_i2c:
  1449     __ sll(Otos_i, 16, Otos_i);
  1450     __ srl(Otos_i, 16, Otos_i);
  1451     break;
  1453    case Bytecodes::_i2s:
  1454     __ sll(Otos_i, 16, Otos_i);
  1455     __ sra(Otos_i, 16, Otos_i);
  1456     break;
  1458    case Bytecodes::_l2i:
  1459 #ifndef _LP64
  1460     __ mov(Otos_l2, Otos_i);
  1461 #else
  1462     // Sign-extend into the high 32 bits
  1463     __ sra(Otos_l, 0, Otos_i);
  1464 #endif
  1465     break;
  1467    case Bytecodes::_l2f:
  1468    case Bytecodes::_l2d:
  1469     __ st_long(Otos_l, __ d_tmp);
  1470     __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
  1472     if (bytecode() == Bytecodes::_l2f) {
  1473       __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1474     } else {
  1475       __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
  1477     break;
  1479   case Bytecodes::_f2i:  {
  1480       Label isNaN;
  1481       // result must be 0 if value is NaN; test by comparing value to itself
  1482       __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
  1483       __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
  1484       __ delayed()->clr(Otos_i);                                     // NaN
  1485       __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
  1486       __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
  1487       __ ld(__ d_tmp, Otos_i);
  1488       __ bind(isNaN);
  1490     break;
  1492    case Bytecodes::_f2l:
  1493     // must uncache tos
  1494     __ push_f();
  1495 #ifdef _LP64
  1496     __ pop_f(F1);
  1497 #else
  1498     __ pop_i(O0);
  1499 #endif
  1500     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
  1501     break;
  1503    case Bytecodes::_f2d:
  1504     __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
  1505     break;
  1507    case Bytecodes::_d2i:
  1508    case Bytecodes::_d2l:
  1509     // must uncache tos
  1510     __ push_d();
  1511 #ifdef _LP64
  1512     // LP64 calling conventions pass first double arg in D0
  1513     __ pop_d( Ftos_d );
  1514 #else
  1515     __ pop_i( O0 );
  1516     __ pop_i( O1 );
  1517 #endif
  1518     __ call_VM_leaf(Lscratch,
  1519         bytecode() == Bytecodes::_d2i
  1520           ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
  1521           : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
  1522     break;
  1524     case Bytecodes::_d2f:
  1525       __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
  1526     break;
  1528     default: ShouldNotReachHere();
  1530   __ bind(done);
  1534 void TemplateTable::lcmp() {
  1535   transition(ltos, itos);
  1537 #ifdef _LP64
  1538   __ pop_l(O1); // pop off value 1, value 2 is in O0
  1539   __ lcmp( O1, Otos_l, Otos_i );
  1540 #else
  1541   __ pop_l(O2); // cmp O2,3 to O0,1
  1542   __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
  1543 #endif
  1547 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
  1549   if (is_float) __ pop_f(F2);
  1550   else          __ pop_d(F2);
  1552   assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
  1554   __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
  1557 void TemplateTable::branch(bool is_jsr, bool is_wide) {
  1558   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
  1559   __ verify_thread();
  1561   const Register O2_bumped_count = O2;
  1562   __ profile_taken_branch(G3_scratch, O2_bumped_count);
  1564   // get (wide) offset to O1_disp
  1565   const Register O1_disp = O1;
  1566   if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
  1567   else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
  1569   // Handle all the JSR stuff here, then exit.
  1570   // It's much shorter and cleaner than intermingling with the
  1571   // non-JSR normal-branch stuff occurring below.
  1572   if( is_jsr ) {
  1573     // compute return address as bci in Otos_i
  1574     __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
  1575     __ sub(Lbcp, G3_scratch, G3_scratch);
  1576     __ sub(G3_scratch, in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
  1578     // Bump Lbcp to target of JSR
  1579     __ add(Lbcp, O1_disp, Lbcp);
  1580     // Push returnAddress for "ret" on stack
  1581     __ push_ptr(Otos_i);
  1582     // And away we go!
  1583     __ dispatch_next(vtos);
  1584     return;
  1587   // Normal (non-jsr) branch handling
  1589   // Save the current Lbcp
  1590   const Register l_cur_bcp = Lscratch;
  1591   __ mov( Lbcp, l_cur_bcp );
  1593   bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
  1594   if ( increment_invocation_counter_for_backward_branches ) {
  1595     Label Lforward;
  1596     // check branch direction
  1597     __ br( Assembler::positive, false,  Assembler::pn, Lforward );
  1598     // Bump bytecode pointer by displacement (take the branch)
  1599     __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
  1601     const Register Rcounters = G3_scratch;
  1602     __ get_method_counters(Lmethod, Rcounters, Lforward);
  1604     if (TieredCompilation) {
  1605       Label Lno_mdo, Loverflow;
  1606       int increment = InvocationCounter::count_increment;
  1607       int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
  1608       if (ProfileInterpreter) {
  1609         // If no method data exists, go to profile_continue.
  1610         __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
  1611         __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
  1613         // Increment backedge counter in the MDO
  1614         Address mdo_backedge_counter(G4_scratch, in_bytes(MethodData::backedge_counter_offset()) +
  1615                                                  in_bytes(InvocationCounter::counter_offset()));
  1616         __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, O0,
  1617                                    Assembler::notZero, &Lforward);
  1618         __ ba_short(Loverflow);
  1621       // If there's no MDO, increment counter in MethodCounters*
  1622       __ bind(Lno_mdo);
  1623       Address backedge_counter(Rcounters,
  1624               in_bytes(MethodCounters::backedge_counter_offset()) +
  1625               in_bytes(InvocationCounter::counter_offset()));
  1626       __ increment_mask_and_jump(backedge_counter, increment, mask, G4_scratch, O0,
  1627                                  Assembler::notZero, &Lforward);
  1628       __ bind(Loverflow);
  1630       // notify point for loop, pass branch bytecode
  1631       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), l_cur_bcp);
  1633       // Was an OSR adapter generated?
  1634       // O0 = osr nmethod
  1635       __ br_null_short(O0, Assembler::pn, Lforward);
  1637       // Has the nmethod been invalidated already?
  1638       __ ld(O0, nmethod::entry_bci_offset(), O2);
  1639       __ cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, Lforward);
  1641       // migrate the interpreter frame off of the stack
  1643       __ mov(G2_thread, L7);
  1644       // save nmethod
  1645       __ mov(O0, L6);
  1646       __ set_last_Java_frame(SP, noreg);
  1647       __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
  1648       __ reset_last_Java_frame();
  1649       __ mov(L7, G2_thread);
  1651       // move OSR nmethod to I1
  1652       __ mov(L6, I1);
  1654       // OSR buffer to I0
  1655       __ mov(O0, I0);
  1657       // remove the interpreter frame
  1658       __ restore(I5_savedSP, 0, SP);
  1660       // Jump to the osr code.
  1661       __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
  1662       __ jmp(O2, G0);
  1663       __ delayed()->nop();
  1665     } else {
  1666       // Update Backedge branch separately from invocations
  1667       const Register G4_invoke_ctr = G4;
  1668       __ increment_backedge_counter(Rcounters, G4_invoke_ctr, G1_scratch);
  1669       if (ProfileInterpreter) {
  1670         __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
  1671         if (UseOnStackReplacement) {
  1672           __ test_backedge_count_for_osr(O2_bumped_count, l_cur_bcp, G3_scratch);
  1674       } else {
  1675         if (UseOnStackReplacement) {
  1676           __ test_backedge_count_for_osr(G4_invoke_ctr, l_cur_bcp, G3_scratch);
  1681     __ bind(Lforward);
  1682   } else
  1683     // Bump bytecode pointer by displacement (take the branch)
  1684     __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
  1686   // continue with bytecode @ target
  1687   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
  1688   // %%%%% and changing dispatch_next to dispatch_only
  1689   __ dispatch_next(vtos);
  1693 // Note Condition in argument is TemplateTable::Condition
  1694 // arg scope is within class scope
  1696 void TemplateTable::if_0cmp(Condition cc) {
  1697   // no pointers, integer only!
  1698   transition(itos, vtos);
  1699   // assume branch is more often taken than not (loops use backward branches)
  1700   __ cmp( Otos_i, 0);
  1701   __ if_cmp(ccNot(cc), false);
  1705 void TemplateTable::if_icmp(Condition cc) {
  1706   transition(itos, vtos);
  1707   __ pop_i(O1);
  1708   __ cmp(O1, Otos_i);
  1709   __ if_cmp(ccNot(cc), false);
  1713 void TemplateTable::if_nullcmp(Condition cc) {
  1714   transition(atos, vtos);
  1715   __ tst(Otos_i);
  1716   __ if_cmp(ccNot(cc), true);
  1720 void TemplateTable::if_acmp(Condition cc) {
  1721   transition(atos, vtos);
  1722   __ pop_ptr(O1);
  1723   __ verify_oop(O1);
  1724   __ verify_oop(Otos_i);
  1725   __ cmp(O1, Otos_i);
  1726   __ if_cmp(ccNot(cc), true);
  1731 void TemplateTable::ret() {
  1732   transition(vtos, vtos);
  1733   locals_index(G3_scratch);
  1734   __ access_local_returnAddress(G3_scratch, Otos_i);
  1735   // Otos_i contains the bci, compute the bcp from that
  1737 #ifdef _LP64
  1738 #ifdef ASSERT
  1739   // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
  1740   // the result.  The return address (really a BCI) was stored with an
  1741   // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
  1742   // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
  1743   // loaded value.
  1744   { Label zzz ;
  1745      __ set (65536, G3_scratch) ;
  1746      __ cmp (Otos_i, G3_scratch) ;
  1747      __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
  1748      __ delayed()->nop();
  1749      __ stop("BCI is in the wrong register half?");
  1750      __ bind (zzz) ;
  1752 #endif
  1753 #endif
  1755   __ profile_ret(vtos, Otos_i, G4_scratch);
  1757   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
  1758   __ add(G3_scratch, Otos_i, G3_scratch);
  1759   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
  1760   __ dispatch_next(vtos);
  1764 void TemplateTable::wide_ret() {
  1765   transition(vtos, vtos);
  1766   locals_index_wide(G3_scratch);
  1767   __ access_local_returnAddress(G3_scratch, Otos_i);
  1768   // Otos_i contains the bci, compute the bcp from that
  1770   __ profile_ret(vtos, Otos_i, G4_scratch);
  1772   __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
  1773   __ add(G3_scratch, Otos_i, G3_scratch);
  1774   __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
  1775   __ dispatch_next(vtos);
  1779 void TemplateTable::tableswitch() {
  1780   transition(itos, vtos);
  1781   Label default_case, continue_execution;
  1783   // align bcp
  1784   __ add(Lbcp, BytesPerInt, O1);
  1785   __ and3(O1, -BytesPerInt, O1);
  1786   // load lo, hi
  1787   __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
  1788   __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
  1789 #ifdef _LP64
  1790   // Sign extend the 32 bits
  1791   __ sra ( Otos_i, 0, Otos_i );
  1792 #endif /* _LP64 */
  1794   // check against lo & hi
  1795   __ cmp( Otos_i, O2);
  1796   __ br( Assembler::less, false, Assembler::pn, default_case);
  1797   __ delayed()->cmp( Otos_i, O3 );
  1798   __ br( Assembler::greater, false, Assembler::pn, default_case);
  1799   // lookup dispatch offset
  1800   __ delayed()->sub(Otos_i, O2, O2);
  1801   __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
  1802   __ sll(O2, LogBytesPerInt, O2);
  1803   __ add(O2, 3 * BytesPerInt, O2);
  1804   __ ba(continue_execution);
  1805   __ delayed()->ld(O1, O2, O2);
  1806   // handle default
  1807   __ bind(default_case);
  1808   __ profile_switch_default(O3);
  1809   __ ld(O1, 0, O2); // get default offset
  1810   // continue execution
  1811   __ bind(continue_execution);
  1812   __ add(Lbcp, O2, Lbcp);
  1813   __ dispatch_next(vtos);
  1817 void TemplateTable::lookupswitch() {
  1818   transition(itos, itos);
  1819   __ stop("lookupswitch bytecode should have been rewritten");
  1822 void TemplateTable::fast_linearswitch() {
  1823   transition(itos, vtos);
  1824     Label loop_entry, loop, found, continue_execution;
  1825   // align bcp
  1826   __ add(Lbcp, BytesPerInt, O1);
  1827   __ and3(O1, -BytesPerInt, O1);
  1828  // set counter
  1829   __ ld(O1, BytesPerInt, O2);
  1830   __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
  1831   __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
  1832   __ ba(loop_entry);
  1833   __ delayed()->add(O3, O2, O2); // counter now points past last pair
  1835   // table search
  1836   __ bind(loop);
  1837   __ cmp(O4, Otos_i);
  1838   __ br(Assembler::equal, true, Assembler::pn, found);
  1839   __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
  1840   __ inc(O3, 2 * BytesPerInt);
  1842   __ bind(loop_entry);
  1843   __ cmp(O2, O3);
  1844   __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
  1845   __ delayed()->ld(O3, 0, O4);
  1847   // default case
  1848   __ ld(O1, 0, O4); // get default offset
  1849   if (ProfileInterpreter) {
  1850     __ profile_switch_default(O3);
  1851     __ ba_short(continue_execution);
  1854   // entry found -> get offset
  1855   __ bind(found);
  1856   if (ProfileInterpreter) {
  1857     __ sub(O3, O1, O3);
  1858     __ sub(O3, 2*BytesPerInt, O3);
  1859     __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
  1860     __ profile_switch_case(O3, O1, O2, G3_scratch);
  1862     __ bind(continue_execution);
  1864   __ add(Lbcp, O4, Lbcp);
  1865   __ dispatch_next(vtos);
  1869 void TemplateTable::fast_binaryswitch() {
  1870   transition(itos, vtos);
  1871   // Implementation using the following core algorithm: (copied from Intel)
  1872   //
  1873   // int binary_search(int key, LookupswitchPair* array, int n) {
  1874   //   // Binary search according to "Methodik des Programmierens" by
  1875   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
  1876   //   int i = 0;
  1877   //   int j = n;
  1878   //   while (i+1 < j) {
  1879   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
  1880   //     // with      Q: for all i: 0 <= i < n: key < a[i]
  1881   //     // where a stands for the array and assuming that the (inexisting)
  1882   //     // element a[n] is infinitely big.
  1883   //     int h = (i + j) >> 1;
  1884   //     // i < h < j
  1885   //     if (key < array[h].fast_match()) {
  1886   //       j = h;
  1887   //     } else {
  1888   //       i = h;
  1889   //     }
  1890   //   }
  1891   //   // R: a[i] <= key < a[i+1] or Q
  1892   //   // (i.e., if key is within array, i is the correct index)
  1893   //   return i;
  1894   // }
  1896   // register allocation
  1897   assert(Otos_i == O0, "alias checking");
  1898   const Register Rkey     = Otos_i;                    // already set (tosca)
  1899   const Register Rarray   = O1;
  1900   const Register Ri       = O2;
  1901   const Register Rj       = O3;
  1902   const Register Rh       = O4;
  1903   const Register Rscratch = O5;
  1905   const int log_entry_size = 3;
  1906   const int entry_size = 1 << log_entry_size;
  1908   Label found;
  1909   // Find Array start
  1910   __ add(Lbcp, 3 * BytesPerInt, Rarray);
  1911   __ and3(Rarray, -BytesPerInt, Rarray);
  1912   // initialize i & j (in delay slot)
  1913   __ clr( Ri );
  1915   // and start
  1916   Label entry;
  1917   __ ba(entry);
  1918   __ delayed()->ld( Rarray, -BytesPerInt, Rj);
  1919   // (Rj is already in the native byte-ordering.)
  1921   // binary search loop
  1922   { Label loop;
  1923     __ bind( loop );
  1924     // int h = (i + j) >> 1;
  1925     __ sra( Rh, 1, Rh );
  1926     // if (key < array[h].fast_match()) {
  1927     //   j = h;
  1928     // } else {
  1929     //   i = h;
  1930     // }
  1931     __ sll( Rh, log_entry_size, Rscratch );
  1932     __ ld( Rarray, Rscratch, Rscratch );
  1933     // (Rscratch is already in the native byte-ordering.)
  1934     __ cmp( Rkey, Rscratch );
  1935     __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
  1936     __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
  1938     // while (i+1 < j)
  1939     __ bind( entry );
  1940     __ add( Ri, 1, Rscratch );
  1941     __ cmp(Rscratch, Rj);
  1942     __ br( Assembler::less, true, Assembler::pt, loop );
  1943     __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
  1946   // end of binary search, result index is i (must check again!)
  1947   Label default_case;
  1948   Label continue_execution;
  1949   if (ProfileInterpreter) {
  1950     __ mov( Ri, Rh );              // Save index in i for profiling
  1952   __ sll( Ri, log_entry_size, Ri );
  1953   __ ld( Rarray, Ri, Rscratch );
  1954   // (Rscratch is already in the native byte-ordering.)
  1955   __ cmp( Rkey, Rscratch );
  1956   __ br( Assembler::notEqual, true, Assembler::pn, default_case );
  1957   __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
  1959   // entry found -> j = offset
  1960   __ inc( Ri, BytesPerInt );
  1961   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
  1962   __ ld( Rarray, Ri, Rj );
  1963   // (Rj is already in the native byte-ordering.)
  1965   if (ProfileInterpreter) {
  1966     __ ba_short(continue_execution);
  1969   __ bind(default_case); // fall through (if not profiling)
  1970   __ profile_switch_default(Ri);
  1972   __ bind(continue_execution);
  1973   __ add( Lbcp, Rj, Lbcp );
  1974   __ dispatch_next( vtos );
  1978 void TemplateTable::_return(TosState state) {
  1979   transition(state, state);
  1980   assert(_desc->calls_vm(), "inconsistent calls_vm information");
  1982   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
  1983     assert(state == vtos, "only valid state");
  1984     __ mov(G0, G3_scratch);
  1985     __ access_local_ptr(G3_scratch, Otos_i);
  1986     __ load_klass(Otos_i, O2);
  1987     __ set(JVM_ACC_HAS_FINALIZER, G3);
  1988     __ ld(O2, in_bytes(Klass::access_flags_offset()), O2);
  1989     __ andcc(G3, O2, G0);
  1990     Label skip_register_finalizer;
  1991     __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
  1992     __ delayed()->nop();
  1994     // Call out to do finalizer registration
  1995     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
  1997     __ bind(skip_register_finalizer);
  2000   __ remove_activation(state, /* throw_monitor_exception */ true);
  2002   // The caller's SP was adjusted upon method entry to accomodate
  2003   // the callee's non-argument locals. Undo that adjustment.
  2004   __ ret();                             // return to caller
  2005   __ delayed()->restore(I5_savedSP, G0, SP);
  2009 // ----------------------------------------------------------------------------
  2010 // Volatile variables demand their effects be made known to all CPU's in
  2011 // order.  Store buffers on most chips allow reads & writes to reorder; the
  2012 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
  2013 // memory barrier (i.e., it's not sufficient that the interpreter does not
  2014 // reorder volatile references, the hardware also must not reorder them).
  2015 //
  2016 // According to the new Java Memory Model (JMM):
  2017 // (1) All volatiles are serialized wrt to each other.
  2018 // ALSO reads & writes act as aquire & release, so:
  2019 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
  2020 // the read float up to before the read.  It's OK for non-volatile memory refs
  2021 // that happen before the volatile read to float down below it.
  2022 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
  2023 // that happen BEFORE the write float down to after the write.  It's OK for
  2024 // non-volatile memory refs that happen after the volatile write to float up
  2025 // before it.
  2026 //
  2027 // We only put in barriers around volatile refs (they are expensive), not
  2028 // _between_ memory refs (that would require us to track the flavor of the
  2029 // previous memory refs).  Requirements (2) and (3) require some barriers
  2030 // before volatile stores and after volatile loads.  These nearly cover
  2031 // requirement (1) but miss the volatile-store-volatile-load case.  This final
  2032 // case is placed after volatile-stores although it could just as well go
  2033 // before volatile-loads.
  2034 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
  2035   // Helper function to insert a is-volatile test and memory barrier
  2036   // All current sparc implementations run in TSO, needing only StoreLoad
  2037   if ((order_constraint & Assembler::StoreLoad) == 0) return;
  2038   __ membar( order_constraint );
  2041 // ----------------------------------------------------------------------------
  2042 void TemplateTable::resolve_cache_and_index(int byte_no,
  2043                                             Register Rcache,
  2044                                             Register index,
  2045                                             size_t index_size) {
  2046   // Depends on cpCacheOop layout!
  2047   Label resolved;
  2049     assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
  2050     __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
  2051     __ cmp(Lbyte_code, (int) bytecode());  // have we resolved this bytecode?
  2052     __ br(Assembler::equal, false, Assembler::pt, resolved);
  2053     __ delayed()->set((int)bytecode(), O1);
  2055   address entry;
  2056   switch (bytecode()) {
  2057     case Bytecodes::_getstatic      : // fall through
  2058     case Bytecodes::_putstatic      : // fall through
  2059     case Bytecodes::_getfield       : // fall through
  2060     case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
  2061     case Bytecodes::_invokevirtual  : // fall through
  2062     case Bytecodes::_invokespecial  : // fall through
  2063     case Bytecodes::_invokestatic   : // fall through
  2064     case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
  2065     case Bytecodes::_invokehandle   : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);  break;
  2066     case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
  2067     default:
  2068       fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
  2069       break;
  2071   // first time invocation - must resolve first
  2072   __ call_VM(noreg, entry, O1);
  2073   // Update registers with resolved info
  2074   __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
  2075   __ bind(resolved);
  2078 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
  2079                                                Register method,
  2080                                                Register itable_index,
  2081                                                Register flags,
  2082                                                bool is_invokevirtual,
  2083                                                bool is_invokevfinal,
  2084                                                bool is_invokedynamic) {
  2085   // Uses both G3_scratch and G4_scratch
  2086   Register cache = G3_scratch;
  2087   Register index = G4_scratch;
  2088   assert_different_registers(cache, method, itable_index);
  2090   // determine constant pool cache field offsets
  2091   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
  2092   const int method_offset = in_bytes(
  2093       ConstantPoolCache::base_offset() +
  2094       ((byte_no == f2_byte)
  2095        ? ConstantPoolCacheEntry::f2_offset()
  2096        : ConstantPoolCacheEntry::f1_offset()
  2098     );
  2099   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
  2100                                     ConstantPoolCacheEntry::flags_offset());
  2101   // access constant pool cache fields
  2102   const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
  2103                                     ConstantPoolCacheEntry::f2_offset());
  2105   if (is_invokevfinal) {
  2106     __ get_cache_and_index_at_bcp(cache, index, 1);
  2107     __ ld_ptr(Address(cache, method_offset), method);
  2108   } else {
  2109     size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
  2110     resolve_cache_and_index(byte_no, cache, index, index_size);
  2111     __ ld_ptr(Address(cache, method_offset), method);
  2114   if (itable_index != noreg) {
  2115     // pick up itable or appendix index from f2 also:
  2116     __ ld_ptr(Address(cache, index_offset), itable_index);
  2118   __ ld_ptr(Address(cache, flags_offset), flags);
  2121 // The Rcache register must be set before call
  2122 void TemplateTable::load_field_cp_cache_entry(Register Robj,
  2123                                               Register Rcache,
  2124                                               Register index,
  2125                                               Register Roffset,
  2126                                               Register Rflags,
  2127                                               bool is_static) {
  2128   assert_different_registers(Rcache, Rflags, Roffset);
  2130   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2132   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2133   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2134   if (is_static) {
  2135     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
  2136     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
  2137     __ ld_ptr( Robj, mirror_offset, Robj);
  2141 // The registers Rcache and index expected to be set before call.
  2142 // Correct values of the Rcache and index registers are preserved.
  2143 void TemplateTable::jvmti_post_field_access(Register Rcache,
  2144                                             Register index,
  2145                                             bool is_static,
  2146                                             bool has_tos) {
  2147   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2149   if (JvmtiExport::can_post_field_access()) {
  2150     // Check to see if a field access watch has been set before we take
  2151     // the time to call into the VM.
  2152     Label Label1;
  2153     assert_different_registers(Rcache, index, G1_scratch);
  2154     AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
  2155     __ load_contents(get_field_access_count_addr, G1_scratch);
  2156     __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
  2158     __ add(Rcache, in_bytes(cp_base_offset), Rcache);
  2160     if (is_static) {
  2161       __ clr(Otos_i);
  2162     } else {
  2163       if (has_tos) {
  2164       // save object pointer before call_VM() clobbers it
  2165         __ push_ptr(Otos_i);  // put object on tos where GC wants it.
  2166       } else {
  2167         // Load top of stack (do not pop the value off the stack);
  2168         __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
  2170       __ verify_oop(Otos_i);
  2172     // Otos_i: object pointer or NULL if static
  2173     // Rcache: cache entry pointer
  2174     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
  2175                Otos_i, Rcache);
  2176     if (!is_static && has_tos) {
  2177       __ pop_ptr(Otos_i);  // restore object pointer
  2178       __ verify_oop(Otos_i);
  2180     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2181     __ bind(Label1);
  2185 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
  2186   transition(vtos, vtos);
  2188   Register Rcache = G3_scratch;
  2189   Register index  = G4_scratch;
  2190   Register Rclass = Rcache;
  2191   Register Roffset= G4_scratch;
  2192   Register Rflags = G1_scratch;
  2193   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2195   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
  2196   jvmti_post_field_access(Rcache, index, is_static, false);
  2197   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2199   if (!is_static) {
  2200     pop_and_check_object(Rclass);
  2201   } else {
  2202     __ verify_oop(Rclass);
  2205   Label exit;
  2207   Assembler::Membar_mask_bits membar_bits =
  2208     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2210   if (__ membar_has_effect(membar_bits)) {
  2211     // Get volatile flag
  2212     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
  2213     __ and3(Rflags, Lscratch, Lscratch);
  2216   Label checkVolatile;
  2218   // compute field type
  2219   Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
  2220   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
  2221   // Make sure we don't need to mask Rflags after the above shift
  2222   ConstantPoolCacheEntry::verify_tos_state_shift();
  2224   // Check atos before itos for getstatic, more likely (in Queens at least)
  2225   __ cmp(Rflags, atos);
  2226   __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2227   __ delayed() ->cmp(Rflags, itos);
  2229   // atos
  2230   __ load_heap_oop(Rclass, Roffset, Otos_i);
  2231   __ verify_oop(Otos_i);
  2232   __ push(atos);
  2233   if (!is_static) {
  2234     patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
  2236   __ ba(checkVolatile);
  2237   __ delayed()->tst(Lscratch);
  2239   __ bind(notObj);
  2241   // cmp(Rflags, itos);
  2242   __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2243   __ delayed() ->cmp(Rflags, ltos);
  2245   // itos
  2246   __ ld(Rclass, Roffset, Otos_i);
  2247   __ push(itos);
  2248   if (!is_static) {
  2249     patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
  2251   __ ba(checkVolatile);
  2252   __ delayed()->tst(Lscratch);
  2254   __ bind(notInt);
  2256   // cmp(Rflags, ltos);
  2257   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2258   __ delayed() ->cmp(Rflags, btos);
  2260   // ltos
  2261   // load must be atomic
  2262   __ ld_long(Rclass, Roffset, Otos_l);
  2263   __ push(ltos);
  2264   if (!is_static) {
  2265     patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
  2267   __ ba(checkVolatile);
  2268   __ delayed()->tst(Lscratch);
  2270   __ bind(notLong);
  2272   // cmp(Rflags, btos);
  2273   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2274   __ delayed() ->cmp(Rflags, ctos);
  2276   // btos
  2277   __ ldsb(Rclass, Roffset, Otos_i);
  2278   __ push(itos);
  2279   if (!is_static) {
  2280     patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
  2282   __ ba(checkVolatile);
  2283   __ delayed()->tst(Lscratch);
  2285   __ bind(notByte);
  2287   // cmp(Rflags, ctos);
  2288   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2289   __ delayed() ->cmp(Rflags, stos);
  2291   // ctos
  2292   __ lduh(Rclass, Roffset, Otos_i);
  2293   __ push(itos);
  2294   if (!is_static) {
  2295     patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
  2297   __ ba(checkVolatile);
  2298   __ delayed()->tst(Lscratch);
  2300   __ bind(notChar);
  2302   // cmp(Rflags, stos);
  2303   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2304   __ delayed() ->cmp(Rflags, ftos);
  2306   // stos
  2307   __ ldsh(Rclass, Roffset, Otos_i);
  2308   __ push(itos);
  2309   if (!is_static) {
  2310     patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
  2312   __ ba(checkVolatile);
  2313   __ delayed()->tst(Lscratch);
  2315   __ bind(notShort);
  2318   // cmp(Rflags, ftos);
  2319   __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
  2320   __ delayed() ->tst(Lscratch);
  2322   // ftos
  2323   __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
  2324   __ push(ftos);
  2325   if (!is_static) {
  2326     patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
  2328   __ ba(checkVolatile);
  2329   __ delayed()->tst(Lscratch);
  2331   __ bind(notFloat);
  2334   // dtos
  2335   __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
  2336   __ push(dtos);
  2337   if (!is_static) {
  2338     patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
  2341   __ bind(checkVolatile);
  2342   if (__ membar_has_effect(membar_bits)) {
  2343     // __ tst(Lscratch); executed in delay slot
  2344     __ br(Assembler::zero, false, Assembler::pt, exit);
  2345     __ delayed()->nop();
  2346     volatile_barrier(membar_bits);
  2349   __ bind(exit);
  2353 void TemplateTable::getfield(int byte_no) {
  2354   getfield_or_static(byte_no, false);
  2357 void TemplateTable::getstatic(int byte_no) {
  2358   getfield_or_static(byte_no, true);
  2362 void TemplateTable::fast_accessfield(TosState state) {
  2363   transition(atos, state);
  2364   Register Rcache  = G3_scratch;
  2365   Register index   = G4_scratch;
  2366   Register Roffset = G4_scratch;
  2367   Register Rflags  = Rcache;
  2368   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2370   __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2371   jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
  2373   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2375   __ null_check(Otos_i);
  2376   __ verify_oop(Otos_i);
  2378   Label exit;
  2380   Assembler::Membar_mask_bits membar_bits =
  2381     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2382   if (__ membar_has_effect(membar_bits)) {
  2383     // Get volatile flag
  2384     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
  2385     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
  2388   switch (bytecode()) {
  2389     case Bytecodes::_fast_bgetfield:
  2390       __ ldsb(Otos_i, Roffset, Otos_i);
  2391       break;
  2392     case Bytecodes::_fast_cgetfield:
  2393       __ lduh(Otos_i, Roffset, Otos_i);
  2394       break;
  2395     case Bytecodes::_fast_sgetfield:
  2396       __ ldsh(Otos_i, Roffset, Otos_i);
  2397       break;
  2398     case Bytecodes::_fast_igetfield:
  2399       __ ld(Otos_i, Roffset, Otos_i);
  2400       break;
  2401     case Bytecodes::_fast_lgetfield:
  2402       __ ld_long(Otos_i, Roffset, Otos_l);
  2403       break;
  2404     case Bytecodes::_fast_fgetfield:
  2405       __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
  2406       break;
  2407     case Bytecodes::_fast_dgetfield:
  2408       __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
  2409       break;
  2410     case Bytecodes::_fast_agetfield:
  2411       __ load_heap_oop(Otos_i, Roffset, Otos_i);
  2412       break;
  2413     default:
  2414       ShouldNotReachHere();
  2417   if (__ membar_has_effect(membar_bits)) {
  2418     __ btst(Lscratch, Rflags);
  2419     __ br(Assembler::zero, false, Assembler::pt, exit);
  2420     __ delayed()->nop();
  2421     volatile_barrier(membar_bits);
  2422     __ bind(exit);
  2425   if (state == atos) {
  2426     __ verify_oop(Otos_i);    // does not blow flags!
  2430 void TemplateTable::jvmti_post_fast_field_mod() {
  2431   if (JvmtiExport::can_post_field_modification()) {
  2432     // Check to see if a field modification watch has been set before we take
  2433     // the time to call into the VM.
  2434     Label done;
  2435     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2436     __ load_contents(get_field_modification_count_addr, G4_scratch);
  2437     __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
  2438     __ pop_ptr(G4_scratch);     // copy the object pointer from tos
  2439     __ verify_oop(G4_scratch);
  2440     __ push_ptr(G4_scratch);    // put the object pointer back on tos
  2441     __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
  2442     // Save tos values before call_VM() clobbers them. Since we have
  2443     // to do it for every data type, we use the saved values as the
  2444     // jvalue object.
  2445     switch (bytecode()) {  // save tos values before call_VM() clobbers them
  2446     case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
  2447     case Bytecodes::_fast_bputfield: // fall through
  2448     case Bytecodes::_fast_sputfield: // fall through
  2449     case Bytecodes::_fast_cputfield: // fall through
  2450     case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
  2451     case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
  2452     case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
  2453     // get words in right order for use as jvalue object
  2454     case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
  2456     // setup pointer to jvalue object
  2457     __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
  2458     // G4_scratch:  object pointer
  2459     // G1_scratch: cache entry pointer
  2460     // G3_scratch: jvalue object on the stack
  2461     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
  2462     switch (bytecode()) {             // restore tos values
  2463     case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
  2464     case Bytecodes::_fast_bputfield: // fall through
  2465     case Bytecodes::_fast_sputfield: // fall through
  2466     case Bytecodes::_fast_cputfield: // fall through
  2467     case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
  2468     case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
  2469     case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
  2470     case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
  2472     __ bind(done);
  2476 // The registers Rcache and index expected to be set before call.
  2477 // The function may destroy various registers, just not the Rcache and index registers.
  2478 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
  2479   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2481   if (JvmtiExport::can_post_field_modification()) {
  2482     // Check to see if a field modification watch has been set before we take
  2483     // the time to call into the VM.
  2484     Label Label1;
  2485     assert_different_registers(Rcache, index, G1_scratch);
  2486     AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
  2487     __ load_contents(get_field_modification_count_addr, G1_scratch);
  2488     __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
  2490     // The Rcache and index registers have been already set.
  2491     // This allows to eliminate this call but the Rcache and index
  2492     // registers must be correspondingly used after this line.
  2493     __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
  2495     __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
  2496     if (is_static) {
  2497       // Life is simple.  Null out the object pointer.
  2498       __ clr(G4_scratch);
  2499     } else {
  2500       Register Rflags = G1_scratch;
  2501       // Life is harder. The stack holds the value on top, followed by the
  2502       // object.  We don't know the size of the value, though; it could be
  2503       // one or two words depending on its type. As a result, we must find
  2504       // the type to determine where the object is.
  2506       Label two_word, valsizeknown;
  2507       __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2508       __ mov(Lesp, G4_scratch);
  2509       __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
  2510       // Make sure we don't need to mask Rflags after the above shift
  2511       ConstantPoolCacheEntry::verify_tos_state_shift();
  2512       __ cmp(Rflags, ltos);
  2513       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2514       __ delayed()->cmp(Rflags, dtos);
  2515       __ br(Assembler::equal, false, Assembler::pt, two_word);
  2516       __ delayed()->nop();
  2517       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
  2518       __ ba_short(valsizeknown);
  2519       __ bind(two_word);
  2521       __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
  2523       __ bind(valsizeknown);
  2524       // setup object pointer
  2525       __ ld_ptr(G4_scratch, 0, G4_scratch);
  2526       __ verify_oop(G4_scratch);
  2528     // setup pointer to jvalue object
  2529     __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
  2530     // G4_scratch:  object pointer or NULL if static
  2531     // G3_scratch: cache entry pointer
  2532     // G1_scratch: jvalue object on the stack
  2533     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
  2534                G4_scratch, G3_scratch, G1_scratch);
  2535     __ get_cache_and_index_at_bcp(Rcache, index, 1);
  2536     __ bind(Label1);
  2540 void TemplateTable::pop_and_check_object(Register r) {
  2541   __ pop_ptr(r);
  2542   __ null_check(r);  // for field access must check obj.
  2543   __ verify_oop(r);
  2546 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
  2547   transition(vtos, vtos);
  2548   Register Rcache = G3_scratch;
  2549   Register index  = G4_scratch;
  2550   Register Rclass = Rcache;
  2551   Register Roffset= G4_scratch;
  2552   Register Rflags = G1_scratch;
  2553   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2555   resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
  2556   jvmti_post_field_mod(Rcache, index, is_static);
  2557   load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
  2559   Assembler::Membar_mask_bits read_bits =
  2560     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2561   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2563   Label notVolatile, checkVolatile, exit;
  2564   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2565     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
  2566     __ and3(Rflags, Lscratch, Lscratch);
  2568     if (__ membar_has_effect(read_bits)) {
  2569       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
  2570       volatile_barrier(read_bits);
  2571       __ bind(notVolatile);
  2575   __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
  2576   // Make sure we don't need to mask Rflags after the above shift
  2577   ConstantPoolCacheEntry::verify_tos_state_shift();
  2579   // compute field type
  2580   Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
  2582   if (is_static) {
  2583     // putstatic with object type most likely, check that first
  2584     __ cmp(Rflags, atos);
  2585     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2586     __ delayed()->cmp(Rflags, itos);
  2588     // atos
  2590       __ pop_ptr();
  2591       __ verify_oop(Otos_i);
  2592       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2593       __ ba(checkVolatile);
  2594       __ delayed()->tst(Lscratch);
  2597     __ bind(notObj);
  2598     // cmp(Rflags, itos);
  2599     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2600     __ delayed()->cmp(Rflags, btos);
  2602     // itos
  2604       __ pop_i();
  2605       __ st(Otos_i, Rclass, Roffset);
  2606       __ ba(checkVolatile);
  2607       __ delayed()->tst(Lscratch);
  2610     __ bind(notInt);
  2611   } else {
  2612     // putfield with int type most likely, check that first
  2613     __ cmp(Rflags, itos);
  2614     __ br(Assembler::notEqual, false, Assembler::pt, notInt);
  2615     __ delayed()->cmp(Rflags, atos);
  2617     // itos
  2619       __ pop_i();
  2620       pop_and_check_object(Rclass);
  2621       __ st(Otos_i, Rclass, Roffset);
  2622       patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
  2623       __ ba(checkVolatile);
  2624       __ delayed()->tst(Lscratch);
  2627     __ bind(notInt);
  2628     // cmp(Rflags, atos);
  2629     __ br(Assembler::notEqual, false, Assembler::pt, notObj);
  2630     __ delayed()->cmp(Rflags, btos);
  2632     // atos
  2634       __ pop_ptr();
  2635       pop_and_check_object(Rclass);
  2636       __ verify_oop(Otos_i);
  2637       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2638       patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
  2639       __ ba(checkVolatile);
  2640       __ delayed()->tst(Lscratch);
  2643     __ bind(notObj);
  2646   // cmp(Rflags, btos);
  2647   __ br(Assembler::notEqual, false, Assembler::pt, notByte);
  2648   __ delayed()->cmp(Rflags, ltos);
  2650   // btos
  2652     __ pop_i();
  2653     if (!is_static) pop_and_check_object(Rclass);
  2654     __ stb(Otos_i, Rclass, Roffset);
  2655     if (!is_static) {
  2656       patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
  2658     __ ba(checkVolatile);
  2659     __ delayed()->tst(Lscratch);
  2662   __ bind(notByte);
  2663   // cmp(Rflags, ltos);
  2664   __ br(Assembler::notEqual, false, Assembler::pt, notLong);
  2665   __ delayed()->cmp(Rflags, ctos);
  2667   // ltos
  2669     __ pop_l();
  2670     if (!is_static) pop_and_check_object(Rclass);
  2671     __ st_long(Otos_l, Rclass, Roffset);
  2672     if (!is_static) {
  2673       patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
  2675     __ ba(checkVolatile);
  2676     __ delayed()->tst(Lscratch);
  2679   __ bind(notLong);
  2680   // cmp(Rflags, ctos);
  2681   __ br(Assembler::notEqual, false, Assembler::pt, notChar);
  2682   __ delayed()->cmp(Rflags, stos);
  2684   // ctos (char)
  2686     __ pop_i();
  2687     if (!is_static) pop_and_check_object(Rclass);
  2688     __ sth(Otos_i, Rclass, Roffset);
  2689     if (!is_static) {
  2690       patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
  2692     __ ba(checkVolatile);
  2693     __ delayed()->tst(Lscratch);
  2696   __ bind(notChar);
  2697   // cmp(Rflags, stos);
  2698   __ br(Assembler::notEqual, false, Assembler::pt, notShort);
  2699   __ delayed()->cmp(Rflags, ftos);
  2701   // stos (short)
  2703     __ pop_i();
  2704     if (!is_static) pop_and_check_object(Rclass);
  2705     __ sth(Otos_i, Rclass, Roffset);
  2706     if (!is_static) {
  2707       patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
  2709     __ ba(checkVolatile);
  2710     __ delayed()->tst(Lscratch);
  2713   __ bind(notShort);
  2714   // cmp(Rflags, ftos);
  2715   __ br(Assembler::notZero, false, Assembler::pt, notFloat);
  2716   __ delayed()->nop();
  2718   // ftos
  2720     __ pop_f();
  2721     if (!is_static) pop_and_check_object(Rclass);
  2722     __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2723     if (!is_static) {
  2724       patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
  2726     __ ba(checkVolatile);
  2727     __ delayed()->tst(Lscratch);
  2730   __ bind(notFloat);
  2732   // dtos
  2734     __ pop_d();
  2735     if (!is_static) pop_and_check_object(Rclass);
  2736     __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2737     if (!is_static) {
  2738       patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
  2742   __ bind(checkVolatile);
  2743   __ tst(Lscratch);
  2745   if (__ membar_has_effect(write_bits)) {
  2746     // __ tst(Lscratch); in delay slot
  2747     __ br(Assembler::zero, false, Assembler::pt, exit);
  2748     __ delayed()->nop();
  2749     volatile_barrier(Assembler::StoreLoad);
  2750     __ bind(exit);
  2754 void TemplateTable::fast_storefield(TosState state) {
  2755   transition(state, vtos);
  2756   Register Rcache = G3_scratch;
  2757   Register Rclass = Rcache;
  2758   Register Roffset= G4_scratch;
  2759   Register Rflags = G1_scratch;
  2760   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
  2762   jvmti_post_fast_field_mod();
  2764   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
  2766   Assembler::Membar_mask_bits read_bits =
  2767     Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
  2768   Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
  2770   Label notVolatile, checkVolatile, exit;
  2771   if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
  2772     __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2773     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
  2774     __ and3(Rflags, Lscratch, Lscratch);
  2775     if (__ membar_has_effect(read_bits)) {
  2776       __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
  2777       volatile_barrier(read_bits);
  2778       __ bind(notVolatile);
  2782   __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2783   pop_and_check_object(Rclass);
  2785   switch (bytecode()) {
  2786     case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
  2787     case Bytecodes::_fast_cputfield: /* fall through */
  2788     case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
  2789     case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
  2790     case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
  2791     case Bytecodes::_fast_fputfield:
  2792       __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
  2793       break;
  2794     case Bytecodes::_fast_dputfield:
  2795       __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
  2796       break;
  2797     case Bytecodes::_fast_aputfield:
  2798       do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
  2799       break;
  2800     default:
  2801       ShouldNotReachHere();
  2804   if (__ membar_has_effect(write_bits)) {
  2805     __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
  2806     volatile_barrier(Assembler::StoreLoad);
  2807     __ bind(exit);
  2812 void TemplateTable::putfield(int byte_no) {
  2813   putfield_or_static(byte_no, false);
  2816 void TemplateTable::putstatic(int byte_no) {
  2817   putfield_or_static(byte_no, true);
  2821 void TemplateTable::fast_xaccess(TosState state) {
  2822   transition(vtos, state);
  2823   Register Rcache = G3_scratch;
  2824   Register Roffset = G4_scratch;
  2825   Register Rflags  = G4_scratch;
  2826   Register Rreceiver = Lscratch;
  2828   __ ld_ptr(Llocals, 0, Rreceiver);
  2830   // access constant pool cache  (is resolved)
  2831   __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
  2832   __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
  2833   __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
  2835   __ verify_oop(Rreceiver);
  2836   __ null_check(Rreceiver);
  2837   if (state == atos) {
  2838     __ load_heap_oop(Rreceiver, Roffset, Otos_i);
  2839   } else if (state == itos) {
  2840     __ ld (Rreceiver, Roffset, Otos_i) ;
  2841   } else if (state == ftos) {
  2842     __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
  2843   } else {
  2844     ShouldNotReachHere();
  2847   Assembler::Membar_mask_bits membar_bits =
  2848     Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
  2849   if (__ membar_has_effect(membar_bits)) {
  2851     // Get is_volatile value in Rflags and check if membar is needed
  2852     __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
  2854     // Test volatile
  2855     Label notVolatile;
  2856     __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
  2857     __ btst(Rflags, Lscratch);
  2858     __ br(Assembler::zero, false, Assembler::pt, notVolatile);
  2859     __ delayed()->nop();
  2860     volatile_barrier(membar_bits);
  2861     __ bind(notVolatile);
  2864   __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
  2865   __ sub(Lbcp, 1, Lbcp);
  2868 //----------------------------------------------------------------------------------------------------
  2869 // Calls
  2871 void TemplateTable::count_calls(Register method, Register temp) {
  2872   // implemented elsewhere
  2873   ShouldNotReachHere();
  2876 void TemplateTable::prepare_invoke(int byte_no,
  2877                                    Register method,  // linked method (or i-klass)
  2878                                    Register ra,      // return address
  2879                                    Register index,   // itable index, MethodType, etc.
  2880                                    Register recv,    // if caller wants to see it
  2881                                    Register flags    // if caller wants to test it
  2882                                    ) {
  2883   // determine flags
  2884   const Bytecodes::Code code = bytecode();
  2885   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
  2886   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
  2887   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
  2888   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
  2889   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
  2890   const bool load_receiver       = (recv != noreg);
  2891   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
  2892   assert(recv  == noreg || recv  == O0, "");
  2893   assert(flags == noreg || flags == O1, "");
  2895   // setup registers & access constant pool cache
  2896   if (recv  == noreg)  recv  = O0;
  2897   if (flags == noreg)  flags = O1;
  2898   const Register temp = O2;
  2899   assert_different_registers(method, ra, index, recv, flags, temp);
  2901   load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
  2903   __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
  2905   // maybe push appendix to arguments
  2906   if (is_invokedynamic || is_invokehandle) {
  2907     Label L_no_push;
  2908     __ set((1 << ConstantPoolCacheEntry::has_appendix_shift), temp);
  2909     __ btst(flags, temp);
  2910     __ br(Assembler::zero, false, Assembler::pt, L_no_push);
  2911     __ delayed()->nop();
  2912     // Push the appendix as a trailing parameter.
  2913     // This must be done before we get the receiver,
  2914     // since the parameter_size includes it.
  2915     assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
  2916     __ load_resolved_reference_at_index(temp, index);
  2917     __ verify_oop(temp);
  2918     __ push_ptr(temp);  // push appendix (MethodType, CallSite, etc.)
  2919     __ bind(L_no_push);
  2922   // load receiver if needed (after appendix is pushed so parameter size is correct)
  2923   if (load_receiver) {
  2924     __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, temp);  // get parameter size
  2925     __ load_receiver(temp, recv);  //  __ argument_address uses Gargs but we need Lesp
  2926     __ verify_oop(recv);
  2929   // compute return type
  2930   __ srl(flags, ConstantPoolCacheEntry::tos_state_shift, ra);
  2931   // Make sure we don't need to mask flags after the above shift
  2932   ConstantPoolCacheEntry::verify_tos_state_shift();
  2933   // load return address
  2935     const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
  2936     AddressLiteral table(table_addr);
  2937     __ set(table, temp);
  2938     __ sll(ra, LogBytesPerWord, ra);
  2939     __ ld_ptr(Address(temp, ra), ra);
  2944 void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
  2945   Register Rtemp = G4_scratch;
  2946   Register Rcall = Rindex;
  2947   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  2949   // get target Method* & entry point
  2950   __ lookup_virtual_method(Rrecv, Rindex, G5_method);
  2951   __ call_from_interpreter(Rcall, Gargs, Rret);
  2954 void TemplateTable::invokevirtual(int byte_no) {
  2955   transition(vtos, vtos);
  2956   assert(byte_no == f2_byte, "use this argument");
  2958   Register Rscratch = G3_scratch;
  2959   Register Rtemp    = G4_scratch;
  2960   Register Rret     = Lscratch;
  2961   Register O0_recv  = O0;
  2962   Label notFinal;
  2964   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
  2965   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  2967   // Check for vfinal
  2968   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), G4_scratch);
  2969   __ btst(Rret, G4_scratch);
  2970   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  2971   __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
  2973   patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
  2975   invokevfinal_helper(Rscratch, Rret);
  2977   __ bind(notFinal);
  2979   __ mov(G5_method, Rscratch);  // better scratch register
  2980   __ load_receiver(G4_scratch, O0_recv);  // gets receiverOop
  2981   // receiver is in O0_recv
  2982   __ verify_oop(O0_recv);
  2984   // get return address
  2985   AddressLiteral table(Interpreter::invoke_return_entry_table());
  2986   __ set(table, Rtemp);
  2987   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
  2988   // Make sure we don't need to mask Rret after the above shift
  2989   ConstantPoolCacheEntry::verify_tos_state_shift();
  2990   __ sll(Rret,  LogBytesPerWord, Rret);
  2991   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  2993   // get receiver klass
  2994   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
  2995   __ load_klass(O0_recv, O0_recv);
  2996   __ verify_klass_ptr(O0_recv);
  2998   __ profile_virtual_call(O0_recv, O4);
  3000   generate_vtable_call(O0_recv, Rscratch, Rret);
  3003 void TemplateTable::fast_invokevfinal(int byte_no) {
  3004   transition(vtos, vtos);
  3005   assert(byte_no == f2_byte, "use this argument");
  3007   load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
  3008                              /*is_invokevfinal*/true, false);
  3009   __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
  3010   invokevfinal_helper(G3_scratch, Lscratch);
  3013 void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
  3014   Register Rtemp = G4_scratch;
  3016   // Load receiver from stack slot
  3017   __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G4_scratch);
  3018   __ lduh(G4_scratch, in_bytes(ConstMethod::size_of_parameters_offset()), G4_scratch);
  3019   __ load_receiver(G4_scratch, O0);
  3021   // receiver NULL check
  3022   __ null_check(O0);
  3024   __ profile_final_call(O4);
  3026   // get return address
  3027   AddressLiteral table(Interpreter::invoke_return_entry_table());
  3028   __ set(table, Rtemp);
  3029   __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
  3030   // Make sure we don't need to mask Rret after the above shift
  3031   ConstantPoolCacheEntry::verify_tos_state_shift();
  3032   __ sll(Rret,  LogBytesPerWord, Rret);
  3033   __ ld_ptr(Rtemp, Rret, Rret);         // get return address
  3036   // do the call
  3037   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3041 void TemplateTable::invokespecial(int byte_no) {
  3042   transition(vtos, vtos);
  3043   assert(byte_no == f1_byte, "use this argument");
  3045   const Register Rret     = Lscratch;
  3046   const Register O0_recv  = O0;
  3047   const Register Rscratch = G3_scratch;
  3049   prepare_invoke(byte_no, G5_method, Rret, noreg, O0_recv);  // get receiver also for null check
  3050   __ null_check(O0_recv);
  3052   // do the call
  3053   __ profile_call(O4);
  3054   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3058 void TemplateTable::invokestatic(int byte_no) {
  3059   transition(vtos, vtos);
  3060   assert(byte_no == f1_byte, "use this argument");
  3062   const Register Rret     = Lscratch;
  3063   const Register Rscratch = G3_scratch;
  3065   prepare_invoke(byte_no, G5_method, Rret);  // get f1 Method*
  3067   // do the call
  3068   __ profile_call(O4);
  3069   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3072 void TemplateTable::invokeinterface_object_method(Register RKlass,
  3073                                                   Register Rcall,
  3074                                                   Register Rret,
  3075                                                   Register Rflags) {
  3076   Register Rscratch = G4_scratch;
  3077   Register Rindex = Lscratch;
  3079   assert_different_registers(Rscratch, Rindex, Rret);
  3081   Label notFinal;
  3083   // Check for vfinal
  3084   __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), Rscratch);
  3085   __ btst(Rflags, Rscratch);
  3086   __ br(Assembler::zero, false, Assembler::pt, notFinal);
  3087   __ delayed()->nop();
  3089   __ profile_final_call(O4);
  3091   // do the call - the index (f2) contains the Method*
  3092   assert_different_registers(G5_method, Gargs, Rcall);
  3093   __ mov(Rindex, G5_method);
  3094   __ call_from_interpreter(Rcall, Gargs, Rret);
  3095   __ bind(notFinal);
  3097   __ profile_virtual_call(RKlass, O4);
  3098   generate_vtable_call(RKlass, Rindex, Rret);
  3102 void TemplateTable::invokeinterface(int byte_no) {
  3103   transition(vtos, vtos);
  3104   assert(byte_no == f1_byte, "use this argument");
  3106   const Register Rinterface  = G1_scratch;
  3107   const Register Rret        = G3_scratch;
  3108   const Register Rindex      = Lscratch;
  3109   const Register O0_recv     = O0;
  3110   const Register O1_flags    = O1;
  3111   const Register O2_Klass    = O2;
  3112   const Register Rscratch    = G4_scratch;
  3113   assert_different_registers(Rscratch, G5_method);
  3115   prepare_invoke(byte_no, Rinterface, Rret, Rindex, O0_recv, O1_flags);
  3117   // get receiver klass
  3118   __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
  3119   __ load_klass(O0_recv, O2_Klass);
  3121   // Special case of invokeinterface called for virtual method of
  3122   // java.lang.Object.  See cpCacheOop.cpp for details.
  3123   // This code isn't produced by javac, but could be produced by
  3124   // another compliant java compiler.
  3125   Label notMethod;
  3126   __ set((1 << ConstantPoolCacheEntry::is_forced_virtual_shift), Rscratch);
  3127   __ btst(O1_flags, Rscratch);
  3128   __ br(Assembler::zero, false, Assembler::pt, notMethod);
  3129   __ delayed()->nop();
  3131   invokeinterface_object_method(O2_Klass, Rinterface, Rret, O1_flags);
  3133   __ bind(notMethod);
  3135   __ profile_virtual_call(O2_Klass, O4);
  3137   //
  3138   // find entry point to call
  3139   //
  3141   // compute start of first itableOffsetEntry (which is at end of vtable)
  3142   const int base = InstanceKlass::vtable_start_offset() * wordSize;
  3143   Label search;
  3144   Register Rtemp = O1_flags;
  3146   __ ld(O2_Klass, InstanceKlass::vtable_length_offset() * wordSize, Rtemp);
  3147   if (align_object_offset(1) > 1) {
  3148     __ round_to(Rtemp, align_object_offset(1));
  3150   __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
  3151   if (Assembler::is_simm13(base)) {
  3152     __ add(Rtemp, base, Rtemp);
  3153   } else {
  3154     __ set(base, Rscratch);
  3155     __ add(Rscratch, Rtemp, Rtemp);
  3157   __ add(O2_Klass, Rtemp, Rscratch);
  3159   __ bind(search);
  3161   __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
  3163     Label ok;
  3165     // Check that entry is non-null.  Null entries are probably a bytecode
  3166     // problem.  If the interface isn't implemented by the receiver class,
  3167     // the VM should throw IncompatibleClassChangeError.  linkResolver checks
  3168     // this too but that's only if the entry isn't already resolved, so we
  3169     // need to check again.
  3170     __ br_notnull_short( Rtemp, Assembler::pt, ok);
  3171     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
  3172     __ should_not_reach_here();
  3173     __ bind(ok);
  3176   __ cmp(Rinterface, Rtemp);
  3177   __ brx(Assembler::notEqual, true, Assembler::pn, search);
  3178   __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
  3180   // entry found and Rscratch points to it
  3181   __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
  3183   assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
  3184   __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
  3185   __ add(Rscratch, Rindex, Rscratch);
  3186   __ ld_ptr(O2_Klass, Rscratch, G5_method);
  3188   // Check for abstract method error.
  3190     Label ok;
  3191     __ br_notnull_short(G5_method, Assembler::pt, ok);
  3192     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
  3193     __ should_not_reach_here();
  3194     __ bind(ok);
  3197   Register Rcall = Rinterface;
  3198   assert_different_registers(Rcall, G5_method, Gargs, Rret);
  3200   __ call_from_interpreter(Rcall, Gargs, Rret);
  3203 void TemplateTable::invokehandle(int byte_no) {
  3204   transition(vtos, vtos);
  3205   assert(byte_no == f1_byte, "use this argument");
  3207   if (!EnableInvokeDynamic) {
  3208     // rewriter does not generate this bytecode
  3209     __ should_not_reach_here();
  3210     return;
  3213   const Register Rret       = Lscratch;
  3214   const Register G4_mtype   = G4_scratch;
  3215   const Register O0_recv    = O0;
  3216   const Register Rscratch   = G3_scratch;
  3218   prepare_invoke(byte_no, G5_method, Rret, G4_mtype, O0_recv);
  3219   __ null_check(O0_recv);
  3221   // G4: MethodType object (from cpool->resolved_references[f1], if necessary)
  3222   // G5: MH.invokeExact_MT method (from f2)
  3224   // Note:  G4_mtype is already pushed (if necessary) by prepare_invoke
  3226   // do the call
  3227   __ verify_oop(G4_mtype);
  3228   __ profile_final_call(O4);  // FIXME: profile the LambdaForm also
  3229   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3233 void TemplateTable::invokedynamic(int byte_no) {
  3234   transition(vtos, vtos);
  3235   assert(byte_no == f1_byte, "use this argument");
  3237   if (!EnableInvokeDynamic) {
  3238     // We should not encounter this bytecode if !EnableInvokeDynamic.
  3239     // The verifier will stop it.  However, if we get past the verifier,
  3240     // this will stop the thread in a reasonable way, without crashing the JVM.
  3241     __ call_VM(noreg, CAST_FROM_FN_PTR(address,
  3242                      InterpreterRuntime::throw_IncompatibleClassChangeError));
  3243     // the call_VM checks for exception, so we should never return here.
  3244     __ should_not_reach_here();
  3245     return;
  3248   const Register Rret        = Lscratch;
  3249   const Register G4_callsite = G4_scratch;
  3250   const Register Rscratch    = G3_scratch;
  3252   prepare_invoke(byte_no, G5_method, Rret, G4_callsite);
  3254   // G4: CallSite object (from cpool->resolved_references[f1])
  3255   // G5: MH.linkToCallSite method (from f2)
  3257   // Note:  G4_callsite is already pushed by prepare_invoke
  3259   // %%% should make a type profile for any invokedynamic that takes a ref argument
  3260   // profile this call
  3261   __ profile_call(O4);
  3263   // do the call
  3264   __ verify_oop(G4_callsite);
  3265   __ call_from_interpreter(Rscratch, Gargs, Rret);
  3269 //----------------------------------------------------------------------------------------------------
  3270 // Allocation
  3272 void TemplateTable::_new() {
  3273   transition(vtos, atos);
  3275   Label slow_case;
  3276   Label done;
  3277   Label initialize_header;
  3278   Label initialize_object;  // including clearing the fields
  3280   Register RallocatedObject = Otos_i;
  3281   Register RinstanceKlass = O1;
  3282   Register Roffset = O3;
  3283   Register Rscratch = O4;
  3285   __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3286   __ get_cpool_and_tags(Rscratch, G3_scratch);
  3287   // make sure the class we're about to instantiate has been resolved
  3288   // This is done before loading InstanceKlass to be consistent with the order
  3289   // how Constant Pool is updated (see ConstantPool::klass_at_put)
  3290   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
  3291   __ ldub(G3_scratch, Roffset, G3_scratch);
  3292   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3293   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3294   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3295   // get InstanceKlass
  3296   //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
  3297   __ add(Roffset, sizeof(ConstantPool), Roffset);
  3298   __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
  3300   // make sure klass is fully initialized:
  3301   __ ldub(RinstanceKlass, in_bytes(InstanceKlass::init_state_offset()), G3_scratch);
  3302   __ cmp(G3_scratch, InstanceKlass::fully_initialized);
  3303   __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
  3304   __ delayed()->ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
  3306   // get instance_size in InstanceKlass (already aligned)
  3307   //__ ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
  3309   // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
  3310   __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
  3311   __ br(Assembler::notZero, false, Assembler::pn, slow_case);
  3312   __ delayed()->nop();
  3314   // allocate the instance
  3315   // 1) Try to allocate in the TLAB
  3316   // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
  3317   // 3) if the above fails (or is not applicable), go to a slow case
  3318   // (creates a new TLAB, etc.)
  3320   const bool allow_shared_alloc =
  3321     Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
  3323   if(UseTLAB) {
  3324     Register RoldTopValue = RallocatedObject;
  3325     Register RtlabWasteLimitValue = G3_scratch;
  3326     Register RnewTopValue = G1_scratch;
  3327     Register RendValue = Rscratch;
  3328     Register RfreeValue = RnewTopValue;
  3330     // check if we can allocate in the TLAB
  3331     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
  3332     __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
  3333     __ add(RoldTopValue, Roffset, RnewTopValue);
  3335     // if there is enough space, we do not CAS and do not clear
  3336     __ cmp(RnewTopValue, RendValue);
  3337     if(ZeroTLAB) {
  3338       // the fields have already been cleared
  3339       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
  3340     } else {
  3341       // initialize both the header and fields
  3342       __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
  3344     __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
  3346     if (allow_shared_alloc) {
  3347       // Check if tlab should be discarded (refill_waste_limit >= free)
  3348       __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
  3349       __ sub(RendValue, RoldTopValue, RfreeValue);
  3350 #ifdef _LP64
  3351       __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
  3352 #else
  3353       __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
  3354 #endif
  3355       __ cmp_and_brx_short(RtlabWasteLimitValue, RfreeValue, Assembler::greaterEqualUnsigned, Assembler::pt, slow_case); // tlab waste is small
  3357       // increment waste limit to prevent getting stuck on this slow path
  3358       __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
  3359       __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
  3360     } else {
  3361       // No allocation in the shared eden.
  3362       __ ba_short(slow_case);
  3366   // Allocation in the shared Eden
  3367   if (allow_shared_alloc) {
  3368     Register RoldTopValue = G1_scratch;
  3369     Register RtopAddr = G3_scratch;
  3370     Register RnewTopValue = RallocatedObject;
  3371     Register RendValue = Rscratch;
  3373     __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
  3375     Label retry;
  3376     __ bind(retry);
  3377     __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
  3378     __ ld_ptr(RendValue, 0, RendValue);
  3379     __ ld_ptr(RtopAddr, 0, RoldTopValue);
  3380     __ add(RoldTopValue, Roffset, RnewTopValue);
  3382     // RnewTopValue contains the top address after the new object
  3383     // has been allocated.
  3384     __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
  3386     __ cas_ptr(RtopAddr, RoldTopValue, RnewTopValue);
  3388     // if someone beat us on the allocation, try again, otherwise continue
  3389     __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
  3391     // bump total bytes allocated by this thread
  3392     // RoldTopValue and RtopAddr are dead, so can use G1 and G3
  3393     __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
  3396   if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
  3397     // clear object fields
  3398     __ bind(initialize_object);
  3399     __ deccc(Roffset, sizeof(oopDesc));
  3400     __ br(Assembler::zero, false, Assembler::pt, initialize_header);
  3401     __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
  3403     // initialize remaining object fields
  3404     if (UseBlockZeroing) {
  3405       // Use BIS for zeroing
  3406       __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
  3407     } else {
  3408       Label loop;
  3409       __ subcc(Roffset, wordSize, Roffset);
  3410       __ bind(loop);
  3411       //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
  3412       __ st_ptr(G0, G3_scratch, Roffset);
  3413       __ br(Assembler::notEqual, false, Assembler::pt, loop);
  3414       __ delayed()->subcc(Roffset, wordSize, Roffset);
  3416     __ ba_short(initialize_header);
  3419   // slow case
  3420   __ bind(slow_case);
  3421   __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3422   __ get_constant_pool(O1);
  3424   call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
  3426   __ ba_short(done);
  3428   // Initialize the header: mark, klass
  3429   __ bind(initialize_header);
  3431   if (UseBiasedLocking) {
  3432     __ ld_ptr(RinstanceKlass, in_bytes(Klass::prototype_header_offset()), G4_scratch);
  3433   } else {
  3434     __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
  3436   __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
  3437   __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
  3438   __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
  3441     SkipIfEqual skip_if(
  3442       _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
  3443     // Trigger dtrace event
  3444     __ push(atos);
  3445     __ call_VM_leaf(noreg,
  3446        CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
  3447     __ pop(atos);
  3450   // continue
  3451   __ bind(done);
  3456 void TemplateTable::newarray() {
  3457   transition(itos, atos);
  3458   __ ldub(Lbcp, 1, O1);
  3459      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
  3463 void TemplateTable::anewarray() {
  3464   transition(itos, atos);
  3465   __ get_constant_pool(O1);
  3466   __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
  3467      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
  3471 void TemplateTable::arraylength() {
  3472   transition(atos, itos);
  3473   Label ok;
  3474   __ verify_oop(Otos_i);
  3475   __ tst(Otos_i);
  3476   __ throw_if_not_1_x( Assembler::notZero, ok );
  3477   __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
  3478   __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3482 void TemplateTable::checkcast() {
  3483   transition(atos, atos);
  3484   Label done, is_null, quicked, cast_ok, resolved;
  3485   Register Roffset = G1_scratch;
  3486   Register RobjKlass = O5;
  3487   Register RspecifiedKlass = O4;
  3489   // Check for casting a NULL
  3490   __ br_null_short(Otos_i, Assembler::pn, is_null);
  3492   // Get value klass in RobjKlass
  3493   __ load_klass(Otos_i, RobjKlass); // get value klass
  3495   // Get constant pool tag
  3496   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3498   // See if the checkcast has been quickened
  3499   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3500   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
  3501   __ ldub(G3_scratch, Roffset, G3_scratch);
  3502   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3503   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3504   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3506   __ push_ptr(); // save receiver for result, and for GC
  3507   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3508   __ get_vm_result_2(RspecifiedKlass);
  3509   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3511   __ ba_short(resolved);
  3513   // Extract target class from constant pool
  3514   __ bind(quicked);
  3515   __ add(Roffset, sizeof(ConstantPool), Roffset);
  3516   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3517   __ bind(resolved);
  3518   __ load_klass(Otos_i, RobjKlass); // get value klass
  3520   // Generate a fast subtype check.  Branch to cast_ok if no
  3521   // failure.  Throw exception if failure.
  3522   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
  3524   // Not a subtype; so must throw exception
  3525   __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
  3527   __ bind(cast_ok);
  3529   if (ProfileInterpreter) {
  3530     __ ba_short(done);
  3532   __ bind(is_null);
  3533   __ profile_null_seen(G3_scratch);
  3534   __ bind(done);
  3538 void TemplateTable::instanceof() {
  3539   Label done, is_null, quicked, resolved;
  3540   transition(atos, itos);
  3541   Register Roffset = G1_scratch;
  3542   Register RobjKlass = O5;
  3543   Register RspecifiedKlass = O4;
  3545   // Check for casting a NULL
  3546   __ br_null_short(Otos_i, Assembler::pt, is_null);
  3548   // Get value klass in RobjKlass
  3549   __ load_klass(Otos_i, RobjKlass); // get value klass
  3551   // Get constant pool tag
  3552   __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
  3554   // See if the checkcast has been quickened
  3555   __ get_cpool_and_tags(Lscratch, G3_scratch);
  3556   __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
  3557   __ ldub(G3_scratch, Roffset, G3_scratch);
  3558   __ cmp(G3_scratch, JVM_CONSTANT_Class);
  3559   __ br(Assembler::equal, true, Assembler::pt, quicked);
  3560   __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
  3562   __ push_ptr(); // save receiver for result, and for GC
  3563   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
  3564   __ get_vm_result_2(RspecifiedKlass);
  3565   __ pop_ptr(Otos_i, G3_scratch); // restore receiver
  3567   __ ba_short(resolved);
  3569   // Extract target class from constant pool
  3570   __ bind(quicked);
  3571   __ add(Roffset, sizeof(ConstantPool), Roffset);
  3572   __ get_constant_pool(Lscratch);
  3573   __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
  3574   __ bind(resolved);
  3575   __ load_klass(Otos_i, RobjKlass); // get value klass
  3577   // Generate a fast subtype check.  Branch to cast_ok if no
  3578   // failure.  Return 0 if failure.
  3579   __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
  3580   __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
  3581   // Not a subtype; return 0;
  3582   __ clr( Otos_i );
  3584   if (ProfileInterpreter) {
  3585     __ ba_short(done);
  3587   __ bind(is_null);
  3588   __ profile_null_seen(G3_scratch);
  3589   __ bind(done);
  3592 void TemplateTable::_breakpoint() {
  3594    // Note: We get here even if we are single stepping..
  3595    // jbug inists on setting breakpoints at every bytecode
  3596    // even if we are in single step mode.
  3598    transition(vtos, vtos);
  3599    // get the unpatched byte code
  3600    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
  3601    __ mov(O0, Lbyte_code);
  3603    // post the breakpoint event
  3604    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
  3606    // complete the execution of original bytecode
  3607    __ dispatch_normal(vtos);
  3611 //----------------------------------------------------------------------------------------------------
  3612 // Exceptions
  3614 void TemplateTable::athrow() {
  3615   transition(atos, vtos);
  3617   // This works because exception is cached in Otos_i which is same as O0,
  3618   // which is same as what throw_exception_entry_expects
  3619   assert(Otos_i == Oexception, "see explanation above");
  3621   __ verify_oop(Otos_i);
  3622   __ null_check(Otos_i);
  3623   __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
  3627 //----------------------------------------------------------------------------------------------------
  3628 // Synchronization
  3631 // See frame_sparc.hpp for monitor block layout.
  3632 // Monitor elements are dynamically allocated by growing stack as needed.
  3634 void TemplateTable::monitorenter() {
  3635   transition(atos, vtos);
  3636   __ verify_oop(Otos_i);
  3637   // Try to acquire a lock on the object
  3638   // Repeat until succeeded (i.e., until
  3639   // monitorenter returns true).
  3641   {   Label ok;
  3642     __ tst(Otos_i);
  3643     __ throw_if_not_1_x( Assembler::notZero,  ok);
  3644     __ delayed()->mov(Otos_i, Lscratch); // save obj
  3645     __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
  3648   assert(O0 == Otos_i, "Be sure where the object to lock is");
  3650   // find a free slot in the monitor block
  3653   // initialize entry pointer
  3654   __ clr(O1); // points to free slot or NULL
  3657     Label entry, loop, exit;
  3658     __ add( __ top_most_monitor(), O2 ); // last one to check
  3659     __ ba( entry );
  3660     __ delayed()->mov( Lmonitors, O3 ); // first one to check
  3663     __ bind( loop );
  3665     __ verify_oop(O4);          // verify each monitor's oop
  3666     __ tst(O4); // is this entry unused?
  3667     __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
  3669     __ cmp(O4, O0); // check if current entry is for same object
  3670     __ brx( Assembler::equal, false, Assembler::pn, exit );
  3671     __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
  3673     __ bind( entry );
  3675     __ cmp( O3, O2 );
  3676     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3677     __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
  3679     __ bind( exit );
  3682   { Label allocated;
  3684     // found free slot?
  3685     __ br_notnull_short(O1, Assembler::pn, allocated);
  3687     __ add_monitor_to_stack( false, O2, O3 );
  3688     __ mov(Lmonitors, O1);
  3690     __ bind(allocated);
  3693   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
  3694   // The object has already been poped from the stack, so the expression stack looks correct.
  3695   __ inc(Lbcp);
  3697   __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
  3698   __ lock_object(O1, O0);
  3700   // check if there's enough space on the stack for the monitors after locking
  3701   __ generate_stack_overflow_check(0);
  3703   // The bcp has already been incremented. Just need to dispatch to next instruction.
  3704   __ dispatch_next(vtos);
  3708 void TemplateTable::monitorexit() {
  3709   transition(atos, vtos);
  3710   __ verify_oop(Otos_i);
  3711   __ tst(Otos_i);
  3712   __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
  3714   assert(O0 == Otos_i, "just checking");
  3716   { Label entry, loop, found;
  3717     __ add( __ top_most_monitor(), O2 ); // last one to check
  3718     __ ba(entry);
  3719     // use Lscratch to hold monitor elem to check, start with most recent monitor,
  3720     // By using a local it survives the call to the C routine.
  3721     __ delayed()->mov( Lmonitors, Lscratch );
  3723     __ bind( loop );
  3725     __ verify_oop(O4);          // verify each monitor's oop
  3726     __ cmp(O4, O0); // check if current entry is for desired object
  3727     __ brx( Assembler::equal, true, Assembler::pt, found );
  3728     __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
  3730     __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
  3732     __ bind( entry );
  3734     __ cmp( Lscratch, O2 );
  3735     __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
  3736     __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
  3738     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
  3739     __ should_not_reach_here();
  3741     __ bind(found);
  3743   __ unlock_object(O1);
  3747 //----------------------------------------------------------------------------------------------------
  3748 // Wide instructions
  3750 void TemplateTable::wide() {
  3751   transition(vtos, vtos);
  3752   __ ldub(Lbcp, 1, G3_scratch);// get next bc
  3753   __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
  3754   AddressLiteral ep(Interpreter::_wentry_point);
  3755   __ set(ep, G4_scratch);
  3756   __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
  3757   __ jmp(G3_scratch, G0);
  3758   __ delayed()->nop();
  3759   // Note: the Lbcp increment step is part of the individual wide bytecode implementations
  3763 //----------------------------------------------------------------------------------------------------
  3764 // Multi arrays
  3766 void TemplateTable::multianewarray() {
  3767   transition(vtos, atos);
  3768      // put ndims * wordSize into Lscratch
  3769   __ ldub( Lbcp,     3,               Lscratch);
  3770   __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
  3771      // Lesp points past last_dim, so set to O1 to first_dim address
  3772   __ add(  Lesp,     Lscratch,        O1);
  3773      call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
  3774   __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
  3776 #endif /* !CC_INTERP */

mercurial