1 /*
   2  * Copyright (c) 2014, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright 2013, 2015 SAP AG. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterRuntime.hpp"
  30 #include "interpreter/interp_masm.hpp"
  31 #include "interpreter/templateInterpreter.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "memory/universe.inline.hpp"
  34 #include "oops/objArrayKlass.hpp"
  35 #include "oops/oop.inline.hpp"
  36 #include "prims/methodHandles.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/stubRoutines.hpp"
  39 #include "runtime/synchronizer.hpp"
  40 #include "utilities/macros.hpp"
  41 
  42 #undef __
  43 #define __ _masm->
  44 
  45 // ============================================================================
  46 // Misc helpers
  47 
  48 // Do an oop store like *(base + index) = val OR *(base + offset) = val
  49 // (only one of both variants is possible at the same time).
  50 // Index can be noreg.
  51 // Kills:
  52 //   Rbase, Rtmp
  53 static void do_oop_store(InterpreterMacroAssembler* _masm,
  54                          Register           Rbase,
  55                          RegisterOrConstant offset,
  56                          Register           Rval,         // Noreg means always null.
  57                          Register           Rtmp1,
  58                          Register           Rtmp2,
  59                          Register           Rtmp3,
  60                          BarrierSet::Name   barrier,
  61                          bool               precise,
  62                          bool               check_null) {
  63   assert_different_registers(Rtmp1, Rtmp2, Rtmp3, Rval, Rbase);
  64 
  65   switch (barrier) {
  66 #if INCLUDE_ALL_GCS
  67     case BarrierSet::G1SATBCTLogging:
  68       {
  69         // Load and record the previous value.
  70         __ g1_write_barrier_pre(Rbase, offset,
  71                                 Rtmp3, /* holder of pre_val ? */
  72                                 Rtmp1, Rtmp2, false /* frame */);
  73 
  74         Label Lnull, Ldone;
  75         if (Rval != noreg) {
  76           if (check_null) {
  77             __ cmpdi(CCR0, Rval, 0);
  78             __ beq(CCR0, Lnull);
  79           }
  80           __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval must stay uncompressed.*/ Rtmp1);
  81           // Mark the card.
  82           if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
  83             __ add(Rbase, offset, Rbase);
  84           }
  85           __ g1_write_barrier_post(Rbase, Rval, Rtmp1, Rtmp2, Rtmp3, /*filtered (fast path)*/ &Ldone);
  86           if (check_null) { __ b(Ldone); }
  87         }
  88 
  89         if (Rval == noreg || check_null) { // Store null oop.
  90           Register Rnull = Rval;
  91           __ bind(Lnull);
  92           if (Rval == noreg) {
  93             Rnull = Rtmp1;
  94             __ li(Rnull, 0);
  95           }
  96           if (UseCompressedOops) {
  97             __ stw(Rnull, offset, Rbase);
  98           } else {
  99             __ std(Rnull, offset, Rbase);
 100           }
 101         }
 102         __ bind(Ldone);
 103       }
 104       break;
 105 #endif // INCLUDE_ALL_GCS
 106     case BarrierSet::CardTableForRS:
 107     case BarrierSet::CardTableExtension:
 108       {
 109         Label Lnull, Ldone;
 110         if (Rval != noreg) {
 111           if (check_null) {
 112             __ cmpdi(CCR0, Rval, 0);
 113             __ beq(CCR0, Lnull);
 114           }
 115           __ store_heap_oop_not_null(Rval, offset, Rbase, /*Rval should better stay uncompressed.*/ Rtmp1);
 116           // Mark the card.
 117           if (!(offset.is_constant() && offset.as_constant() == 0) && precise) {
 118             __ add(Rbase, offset, Rbase);
 119           }
 120           __ card_write_barrier_post(Rbase, Rval, Rtmp1);
 121           if (check_null) {
 122             __ b(Ldone);
 123           }
 124         }
 125 
 126         if (Rval == noreg || check_null) { // Store null oop.
 127           Register Rnull = Rval;
 128           __ bind(Lnull);
 129           if (Rval == noreg) {
 130             Rnull = Rtmp1;
 131             __ li(Rnull, 0);
 132           }
 133           if (UseCompressedOops) {
 134             __ stw(Rnull, offset, Rbase);
 135           } else {
 136             __ std(Rnull, offset, Rbase);
 137           }
 138         }
 139         __ bind(Ldone);
 140       }
 141       break;
 142     case BarrierSet::ModRef:
 143       ShouldNotReachHere();
 144       break;
 145     default:
 146       ShouldNotReachHere();
 147   }
 148 }
 149 
 150 // ============================================================================
 151 // Platform-dependent initialization
 152 
 153 void TemplateTable::pd_initialize() {
 154   // No ppc64 specific initialization.
 155 }
 156 
 157 Address TemplateTable::at_bcp(int offset) {
 158   // Not used on ppc.
 159   ShouldNotReachHere();
 160   return Address();
 161 }
 162 
 163 // Patches the current bytecode (ptr to it located in bcp)
 164 // in the bytecode stream with a new one.
 165 void TemplateTable::patch_bytecode(Bytecodes::Code new_bc, Register Rnew_bc, Register Rtemp, bool load_bc_into_bc_reg /*=true*/, int byte_no) {
 166   // With sharing on, may need to test method flag.
 167   if (!RewriteBytecodes) return;
 168   Label L_patch_done;
 169 
 170   switch (new_bc) {
 171     case Bytecodes::_fast_aputfield:
 172     case Bytecodes::_fast_bputfield:
 173     case Bytecodes::_fast_cputfield:
 174     case Bytecodes::_fast_dputfield:
 175     case Bytecodes::_fast_fputfield:
 176     case Bytecodes::_fast_iputfield:
 177     case Bytecodes::_fast_lputfield:
 178     case Bytecodes::_fast_sputfield:
 179     {
 180       // We skip bytecode quickening for putfield instructions when
 181       // the put_code written to the constant pool cache is zero.
 182       // This is required so that every execution of this instruction
 183       // calls out to InterpreterRuntime::resolve_get_put to do
 184       // additional, required work.
 185       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 186       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 187       __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
 188       // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
 189 #if defined(VM_LITTLE_ENDIAN)
 190       __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
 191 #else
 192       __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
 193 #endif
 194       __ cmpwi(CCR0, Rnew_bc, 0);
 195       __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
 196       __ beq(CCR0, L_patch_done);
 197       // __ isync(); // acquire not needed
 198       break;
 199     }
 200 
 201     default:
 202       assert(byte_no == -1, "sanity");
 203       if (load_bc_into_bc_reg) {
 204         __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
 205       }
 206   }
 207 
 208   if (JvmtiExport::can_post_breakpoint()) {
 209     Label L_fast_patch;
 210     __ lbz(Rtemp, 0, R14_bcp);
 211     __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
 212     __ bne(CCR0, L_fast_patch);
 213     // Perform the quickening, slowly, in the bowels of the breakpoint table.
 214     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
 215     __ b(L_patch_done);
 216     __ bind(L_fast_patch);
 217   }
 218 
 219   // Patch bytecode.
 220   __ stb(Rnew_bc, 0, R14_bcp);
 221 
 222   __ bind(L_patch_done);
 223 }
 224 
 225 // ============================================================================
 226 // Individual instructions
 227 
 228 void TemplateTable::nop() {
 229   transition(vtos, vtos);
 230   // Nothing to do.
 231 }
 232 
 233 void TemplateTable::shouldnotreachhere() {
 234   transition(vtos, vtos);
 235   __ stop("shouldnotreachhere bytecode");
 236 }
 237 
 238 void TemplateTable::aconst_null() {
 239   transition(vtos, atos);
 240   __ li(R17_tos, 0);
 241 }
 242 
 243 void TemplateTable::iconst(int value) {
 244   transition(vtos, itos);
 245   assert(value >= -1 && value <= 5, "");
 246   __ li(R17_tos, value);
 247 }
 248 
 249 void TemplateTable::lconst(int value) {
 250   transition(vtos, ltos);
 251   assert(value >= -1 && value <= 5, "");
 252   __ li(R17_tos, value);
 253 }
 254 
 255 void TemplateTable::fconst(int value) {
 256   transition(vtos, ftos);
 257   static float zero = 0.0;
 258   static float one  = 1.0;
 259   static float two  = 2.0;
 260   switch (value) {
 261     default: ShouldNotReachHere();
 262     case 0: {
 263       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
 264       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 265       break;
 266     }
 267     case 1: {
 268       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
 269       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 270       break;
 271     }
 272     case 2: {
 273       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
 274       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 275       break;
 276     }
 277   }
 278 }
 279 
 280 void TemplateTable::dconst(int value) {
 281   transition(vtos, dtos);
 282   static double zero = 0.0;
 283   static double one  = 1.0;
 284   switch (value) {
 285     case 0: {
 286       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
 287       __ lfd(F15_ftos, simm16_offset, R11_scratch1);
 288       break;
 289     }
 290     case 1: {
 291       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
 292       __ lfd(F15_ftos, simm16_offset, R11_scratch1);
 293       break;
 294     }
 295     default: ShouldNotReachHere();
 296   }
 297 }
 298 
 299 void TemplateTable::bipush() {
 300   transition(vtos, itos);
 301   __ lbz(R17_tos, 1, R14_bcp);
 302   __ extsb(R17_tos, R17_tos);
 303 }
 304 
 305 void TemplateTable::sipush() {
 306   transition(vtos, itos);
 307   __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
 308 }
 309 
 310 void TemplateTable::ldc(bool wide) {
 311   Register Rscratch1 = R11_scratch1,
 312            Rscratch2 = R12_scratch2,
 313            Rcpool    = R3_ARG1;
 314 
 315   transition(vtos, vtos);
 316   Label notInt, notClass, exit;
 317 
 318   __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
 319   if (wide) { // Read index.
 320     __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
 321   } else {
 322     __ lbz(Rscratch1, 1, R14_bcp);
 323   }
 324 
 325   const int base_offset = ConstantPool::header_size() * wordSize;
 326   const int tags_offset = Array<u1>::base_offset_in_bytes();
 327 
 328   // Get type from tags.
 329   __ addi(Rscratch2, Rscratch2, tags_offset);
 330   __ lbzx(Rscratch2, Rscratch2, Rscratch1);
 331 
 332   __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
 333   __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
 334   __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
 335 
 336   // Resolved class - need to call vm to get java mirror of the class.
 337   __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
 338   __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
 339   __ beq(CCR0, notClass);
 340 
 341   __ li(R4, wide ? 1 : 0);
 342   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
 343   __ push(atos);
 344   __ b(exit);
 345 
 346   __ align(32, 12);
 347   __ bind(notClass);
 348   __ addi(Rcpool, Rcpool, base_offset);
 349   __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
 350   __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
 351   __ bne(CCR0, notInt);
 352   __ lwax(R17_tos, Rcpool, Rscratch1);
 353   __ push(itos);
 354   __ b(exit);
 355 
 356   __ align(32, 12);
 357   __ bind(notInt);
 358 #ifdef ASSERT
 359   // String and Object are rewritten to fast_aldc
 360   __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
 361   __ asm_assert_eq("unexpected type", 0x8765);
 362 #endif
 363   __ lfsx(F15_ftos, Rcpool, Rscratch1);
 364   __ push(ftos);
 365 
 366   __ align(32, 12);
 367   __ bind(exit);
 368 }
 369 
 370 // Fast path for caching oop constants.
 371 void TemplateTable::fast_aldc(bool wide) {
 372   transition(vtos, atos);
 373 
 374   int index_size = wide ? sizeof(u2) : sizeof(u1);
 375   const Register Rscratch = R11_scratch1;
 376   Label is_null;
 377 
 378   // We are resolved if the resolved reference cache entry contains a
 379   // non-null object (CallSite, etc.)
 380   __ get_cache_index_at_bcp(Rscratch, 1, index_size);  // Load index.
 381   __ load_resolved_reference_at_index(R17_tos, Rscratch, &is_null);
 382   __ verify_oop(R17_tos);
 383   __ dispatch_epilog(atos, Bytecodes::length_for(bytecode()));
 384 
 385   __ bind(is_null);
 386   __ load_const_optimized(R3_ARG1, (int)bytecode());
 387 
 388   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 389 
 390   // First time invocation - must resolve first.
 391   __ call_VM(R17_tos, entry, R3_ARG1);
 392   __ verify_oop(R17_tos);
 393 }
 394 
 395 void TemplateTable::ldc2_w() {
 396   transition(vtos, vtos);
 397   Label Llong, Lexit;
 398 
 399   Register Rindex = R11_scratch1,
 400            Rcpool = R12_scratch2,
 401            Rtag   = R3_ARG1;
 402   __ get_cpool_and_tags(Rcpool, Rtag);
 403   __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
 404 
 405   const int base_offset = ConstantPool::header_size() * wordSize;
 406   const int tags_offset = Array<u1>::base_offset_in_bytes();
 407   // Get type from tags.
 408   __ addi(Rcpool, Rcpool, base_offset);
 409   __ addi(Rtag, Rtag, tags_offset);
 410 
 411   __ lbzx(Rtag, Rtag, Rindex);
 412 
 413   __ sldi(Rindex, Rindex, LogBytesPerWord);
 414   __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
 415   __ bne(CCR0, Llong);
 416   // A double can be placed at word-aligned locations in the constant pool.
 417   // Check out Conversions.java for an example.
 418   // Also ConstantPool::header_size() is 20, which makes it very difficult
 419   // to double-align double on the constant pool. SG, 11/7/97
 420   __ lfdx(F15_ftos, Rcpool, Rindex);
 421   __ push(dtos);
 422   __ b(Lexit);
 423 
 424   __ bind(Llong);
 425   __ ldx(R17_tos, Rcpool, Rindex);
 426   __ push(ltos);
 427 
 428   __ bind(Lexit);
 429 }
 430 
 431 // Get the locals index located in the bytecode stream at bcp + offset.
 432 void TemplateTable::locals_index(Register Rdst, int offset) {
 433   __ lbz(Rdst, offset, R14_bcp);
 434 }
 435 
 436 void TemplateTable::iload() {
 437   iload_internal();
 438 }
 439 
 440 void TemplateTable::nofast_iload() {
 441   iload_internal(may_not_rewrite);
 442 }
 443 
 444 void TemplateTable::iload_internal(RewriteControl rc) {
 445   transition(vtos, itos);
 446 
 447   // Get the local value into tos
 448   const Register Rindex = R22_tmp2;
 449   locals_index(Rindex);
 450 
 451   // Rewrite iload,iload  pair into fast_iload2
 452   //         iload,caload pair into fast_icaload
 453   if (RewriteFrequentPairs && rc == may_rewrite) {
 454     Label Lrewrite, Ldone;
 455     Register Rnext_byte  = R3_ARG1,
 456              Rrewrite_to = R6_ARG4,
 457              Rscratch    = R11_scratch1;
 458 
 459     // get next byte
 460     __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
 461 
 462     // if _iload, wait to rewrite to iload2. We only want to rewrite the
 463     // last two iloads in a pair. Comparing against fast_iload means that
 464     // the next bytecode is neither an iload or a caload, and therefore
 465     // an iload pair.
 466     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
 467     __ beq(CCR0, Ldone);
 468 
 469     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
 470     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
 471     __ beq(CCR1, Lrewrite);
 472 
 473     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
 474     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
 475     __ beq(CCR0, Lrewrite);
 476 
 477     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
 478 
 479     __ bind(Lrewrite);
 480     patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
 481     __ bind(Ldone);
 482   }
 483 
 484   __ load_local_int(R17_tos, Rindex, Rindex);
 485 }
 486 
 487 // Load 2 integers in a row without dispatching
 488 void TemplateTable::fast_iload2() {
 489   transition(vtos, itos);
 490 
 491   __ lbz(R3_ARG1, 1, R14_bcp);
 492   __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
 493 
 494   __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
 495   __ load_local_int(R17_tos, R12_scratch2, R17_tos);
 496   __ push_i(R3_ARG1);
 497 }
 498 
 499 void TemplateTable::fast_iload() {
 500   transition(vtos, itos);
 501   // Get the local value into tos
 502 
 503   const Register Rindex = R11_scratch1;
 504   locals_index(Rindex);
 505   __ load_local_int(R17_tos, Rindex, Rindex);
 506 }
 507 
 508 // Load a local variable type long from locals area to TOS cache register.
 509 // Local index resides in bytecodestream.
 510 void TemplateTable::lload() {
 511   transition(vtos, ltos);
 512 
 513   const Register Rindex = R11_scratch1;
 514   locals_index(Rindex);
 515   __ load_local_long(R17_tos, Rindex, Rindex);
 516 }
 517 
 518 void TemplateTable::fload() {
 519   transition(vtos, ftos);
 520 
 521   const Register Rindex = R11_scratch1;
 522   locals_index(Rindex);
 523   __ load_local_float(F15_ftos, Rindex, Rindex);
 524 }
 525 
 526 void TemplateTable::dload() {
 527   transition(vtos, dtos);
 528 
 529   const Register Rindex = R11_scratch1;
 530   locals_index(Rindex);
 531   __ load_local_double(F15_ftos, Rindex, Rindex);
 532 }
 533 
 534 void TemplateTable::aload() {
 535   transition(vtos, atos);
 536 
 537   const Register Rindex = R11_scratch1;
 538   locals_index(Rindex);
 539   __ load_local_ptr(R17_tos, Rindex, Rindex);
 540 }
 541 
 542 void TemplateTable::locals_index_wide(Register Rdst) {
 543   // Offset is 2, not 1, because Lbcp points to wide prefix code.
 544   __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
 545 }
 546 
 547 void TemplateTable::wide_iload() {
 548   // Get the local value into tos.
 549 
 550   const Register Rindex = R11_scratch1;
 551   locals_index_wide(Rindex);
 552   __ load_local_int(R17_tos, Rindex, Rindex);
 553 }
 554 
 555 void TemplateTable::wide_lload() {
 556   transition(vtos, ltos);
 557 
 558   const Register Rindex = R11_scratch1;
 559   locals_index_wide(Rindex);
 560   __ load_local_long(R17_tos, Rindex, Rindex);
 561 }
 562 
 563 void TemplateTable::wide_fload() {
 564   transition(vtos, ftos);
 565 
 566   const Register Rindex = R11_scratch1;
 567   locals_index_wide(Rindex);
 568   __ load_local_float(F15_ftos, Rindex, Rindex);
 569 }
 570 
 571 void TemplateTable::wide_dload() {
 572   transition(vtos, dtos);
 573 
 574   const Register Rindex = R11_scratch1;
 575   locals_index_wide(Rindex);
 576   __ load_local_double(F15_ftos, Rindex, Rindex);
 577 }
 578 
 579 void TemplateTable::wide_aload() {
 580   transition(vtos, atos);
 581 
 582   const Register Rindex = R11_scratch1;
 583   locals_index_wide(Rindex);
 584   __ load_local_ptr(R17_tos, Rindex, Rindex);
 585 }
 586 
 587 void TemplateTable::iaload() {
 588   transition(itos, itos);
 589 
 590   const Register Rload_addr = R3_ARG1,
 591                  Rarray     = R4_ARG2,
 592                  Rtemp      = R5_ARG3;
 593   __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
 594   __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
 595 }
 596 
 597 void TemplateTable::laload() {
 598   transition(itos, ltos);
 599 
 600   const Register Rload_addr = R3_ARG1,
 601                  Rarray     = R4_ARG2,
 602                  Rtemp      = R5_ARG3;
 603   __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
 604   __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
 605 }
 606 
 607 void TemplateTable::faload() {
 608   transition(itos, ftos);
 609 
 610   const Register Rload_addr = R3_ARG1,
 611                  Rarray     = R4_ARG2,
 612                  Rtemp      = R5_ARG3;
 613   __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
 614   __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
 615 }
 616 
 617 void TemplateTable::daload() {
 618   transition(itos, dtos);
 619 
 620   const Register Rload_addr = R3_ARG1,
 621                  Rarray     = R4_ARG2,
 622                  Rtemp      = R5_ARG3;
 623   __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
 624   __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
 625 }
 626 
 627 void TemplateTable::aaload() {
 628   transition(itos, atos);
 629 
 630   // tos: index
 631   // result tos: array
 632   const Register Rload_addr = R3_ARG1,
 633                  Rarray     = R4_ARG2,
 634                  Rtemp      = R5_ARG3;
 635   __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
 636   __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
 637   __ verify_oop(R17_tos);
 638   //__ dcbt(R17_tos); // prefetch
 639 }
 640 
 641 void TemplateTable::baload() {
 642   transition(itos, itos);
 643 
 644   const Register Rload_addr = R3_ARG1,
 645                  Rarray     = R4_ARG2,
 646                  Rtemp      = R5_ARG3;
 647   __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
 648   __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
 649   __ extsb(R17_tos, R17_tos);
 650 }
 651 
 652 void TemplateTable::caload() {
 653   transition(itos, itos);
 654 
 655   const Register Rload_addr = R3_ARG1,
 656                  Rarray     = R4_ARG2,
 657                  Rtemp      = R5_ARG3;
 658   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 659   __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
 660 }
 661 
 662 // Iload followed by caload frequent pair.
 663 void TemplateTable::fast_icaload() {
 664   transition(vtos, itos);
 665 
 666   const Register Rload_addr = R3_ARG1,
 667                  Rarray     = R4_ARG2,
 668                  Rtemp      = R11_scratch1;
 669 
 670   locals_index(R17_tos);
 671   __ load_local_int(R17_tos, Rtemp, R17_tos);
 672   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 673   __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
 674 }
 675 
 676 void TemplateTable::saload() {
 677   transition(itos, itos);
 678 
 679   const Register Rload_addr = R11_scratch1,
 680                  Rarray     = R12_scratch2,
 681                  Rtemp      = R3_ARG1;
 682   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 683   __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
 684 }
 685 
 686 void TemplateTable::iload(int n) {
 687   transition(vtos, itos);
 688 
 689   __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
 690 }
 691 
 692 void TemplateTable::lload(int n) {
 693   transition(vtos, ltos);
 694 
 695   __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
 696 }
 697 
 698 void TemplateTable::fload(int n) {
 699   transition(vtos, ftos);
 700 
 701   __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
 702 }
 703 
 704 void TemplateTable::dload(int n) {
 705   transition(vtos, dtos);
 706 
 707   __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
 708 }
 709 
 710 void TemplateTable::aload(int n) {
 711   transition(vtos, atos);
 712 
 713   __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
 714 }
 715 
 716 void TemplateTable::aload_0() {
 717   aload_0_internal();
 718 }
 719 
 720 void TemplateTable::nofast_aload_0() {
 721   aload_0_internal(may_not_rewrite);
 722 }
 723 
 724 void TemplateTable::aload_0_internal(RewriteControl rc) {
 725   transition(vtos, atos);
 726   // According to bytecode histograms, the pairs:
 727   //
 728   // _aload_0, _fast_igetfield
 729   // _aload_0, _fast_agetfield
 730   // _aload_0, _fast_fgetfield
 731   //
 732   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 733   // _aload_0 bytecode checks if the next bytecode is either
 734   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 735   // rewrites the current bytecode into a pair bytecode; otherwise it
 736   // rewrites the current bytecode into _0 that doesn't do
 737   // the pair check anymore.
 738   //
 739   // Note: If the next bytecode is _getfield, the rewrite must be
 740   //       delayed, otherwise we may miss an opportunity for a pair.
 741   //
 742   // Also rewrite frequent pairs
 743   //   aload_0, aload_1
 744   //   aload_0, iload_1
 745   // These bytecodes with a small amount of code are most profitable
 746   // to rewrite.
 747 
 748   if (RewriteFrequentPairs && rc == may_rewrite) {
 749 
 750     Label Lrewrite, Ldont_rewrite;
 751     Register Rnext_byte  = R3_ARG1,
 752              Rrewrite_to = R6_ARG4,
 753              Rscratch    = R11_scratch1;
 754 
 755     // Get next byte.
 756     __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
 757 
 758     // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
 759     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
 760     __ beq(CCR0, Ldont_rewrite);
 761 
 762     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
 763     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
 764     __ beq(CCR1, Lrewrite);
 765 
 766     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
 767     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
 768     __ beq(CCR0, Lrewrite);
 769 
 770     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
 771     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
 772     __ beq(CCR1, Lrewrite);
 773 
 774     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
 775 
 776     __ bind(Lrewrite);
 777     patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
 778     __ bind(Ldont_rewrite);
 779   }
 780 
 781   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 782   aload(0);
 783 }
 784 
 785 void TemplateTable::istore() {
 786   transition(itos, vtos);
 787 
 788   const Register Rindex = R11_scratch1;
 789   locals_index(Rindex);
 790   __ store_local_int(R17_tos, Rindex);
 791 }
 792 
 793 void TemplateTable::lstore() {
 794   transition(ltos, vtos);
 795   const Register Rindex = R11_scratch1;
 796   locals_index(Rindex);
 797   __ store_local_long(R17_tos, Rindex);
 798 }
 799 
 800 void TemplateTable::fstore() {
 801   transition(ftos, vtos);
 802 
 803   const Register Rindex = R11_scratch1;
 804   locals_index(Rindex);
 805   __ store_local_float(F15_ftos, Rindex);
 806 }
 807 
 808 void TemplateTable::dstore() {
 809   transition(dtos, vtos);
 810 
 811   const Register Rindex = R11_scratch1;
 812   locals_index(Rindex);
 813   __ store_local_double(F15_ftos, Rindex);
 814 }
 815 
 816 void TemplateTable::astore() {
 817   transition(vtos, vtos);
 818 
 819   const Register Rindex = R11_scratch1;
 820   __ pop_ptr();
 821   __ verify_oop_or_return_address(R17_tos, Rindex);
 822   locals_index(Rindex);
 823   __ store_local_ptr(R17_tos, Rindex);
 824 }
 825 
 826 void TemplateTable::wide_istore() {
 827   transition(vtos, vtos);
 828 
 829   const Register Rindex = R11_scratch1;
 830   __ pop_i();
 831   locals_index_wide(Rindex);
 832   __ store_local_int(R17_tos, Rindex);
 833 }
 834 
 835 void TemplateTable::wide_lstore() {
 836   transition(vtos, vtos);
 837 
 838   const Register Rindex = R11_scratch1;
 839   __ pop_l();
 840   locals_index_wide(Rindex);
 841   __ store_local_long(R17_tos, Rindex);
 842 }
 843 
 844 void TemplateTable::wide_fstore() {
 845   transition(vtos, vtos);
 846 
 847   const Register Rindex = R11_scratch1;
 848   __ pop_f();
 849   locals_index_wide(Rindex);
 850   __ store_local_float(F15_ftos, Rindex);
 851 }
 852 
 853 void TemplateTable::wide_dstore() {
 854   transition(vtos, vtos);
 855 
 856   const Register Rindex = R11_scratch1;
 857   __ pop_d();
 858   locals_index_wide(Rindex);
 859   __ store_local_double(F15_ftos, Rindex);
 860 }
 861 
 862 void TemplateTable::wide_astore() {
 863   transition(vtos, vtos);
 864 
 865   const Register Rindex = R11_scratch1;
 866   __ pop_ptr();
 867   __ verify_oop_or_return_address(R17_tos, Rindex);
 868   locals_index_wide(Rindex);
 869   __ store_local_ptr(R17_tos, Rindex);
 870 }
 871 
 872 void TemplateTable::iastore() {
 873   transition(itos, vtos);
 874 
 875   const Register Rindex      = R3_ARG1,
 876                  Rstore_addr = R4_ARG2,
 877                  Rarray      = R5_ARG3,
 878                  Rtemp       = R6_ARG4;
 879   __ pop_i(Rindex);
 880   __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
 881   __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
 882   }
 883 
 884 void TemplateTable::lastore() {
 885   transition(ltos, vtos);
 886 
 887   const Register Rindex      = R3_ARG1,
 888                  Rstore_addr = R4_ARG2,
 889                  Rarray      = R5_ARG3,
 890                  Rtemp       = R6_ARG4;
 891   __ pop_i(Rindex);
 892   __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
 893   __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
 894   }
 895 
 896 void TemplateTable::fastore() {
 897   transition(ftos, vtos);
 898 
 899   const Register Rindex      = R3_ARG1,
 900                  Rstore_addr = R4_ARG2,
 901                  Rarray      = R5_ARG3,
 902                  Rtemp       = R6_ARG4;
 903   __ pop_i(Rindex);
 904   __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
 905   __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
 906   }
 907 
 908 void TemplateTable::dastore() {
 909   transition(dtos, vtos);
 910 
 911   const Register Rindex      = R3_ARG1,
 912                  Rstore_addr = R4_ARG2,
 913                  Rarray      = R5_ARG3,
 914                  Rtemp       = R6_ARG4;
 915   __ pop_i(Rindex);
 916   __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
 917   __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
 918   }
 919 
 920 // Pop 3 values from the stack and...
 921 void TemplateTable::aastore() {
 922   transition(vtos, vtos);
 923 
 924   Label Lstore_ok, Lis_null, Ldone;
 925   const Register Rindex    = R3_ARG1,
 926                  Rarray    = R4_ARG2,
 927                  Rscratch  = R11_scratch1,
 928                  Rscratch2 = R12_scratch2,
 929                  Rarray_klass = R5_ARG3,
 930                  Rarray_element_klass = Rarray_klass,
 931                  Rvalue_klass = R6_ARG4,
 932                  Rstore_addr = R31;    // Use register which survives VM call.
 933 
 934   __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
 935   __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
 936   __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp);  // Get array.
 937 
 938   __ verify_oop(R17_tos);
 939   __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
 940   // Rindex is dead!
 941   Register Rscratch3 = Rindex;
 942 
 943   // Do array store check - check for NULL value first.
 944   __ cmpdi(CCR0, R17_tos, 0);
 945   __ beq(CCR0, Lis_null);
 946 
 947   __ load_klass(Rarray_klass, Rarray);
 948   __ load_klass(Rvalue_klass, R17_tos);
 949 
 950   // Do fast instanceof cache test.
 951   __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
 952 
 953   // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
 954   __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
 955 
 956   // Fell through: subtype check failed => throw an exception.
 957   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
 958   __ mtctr(R11_scratch1);
 959   __ bctr();
 960 
 961   __ bind(Lis_null);
 962   do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
 963                Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
 964   __ profile_null_seen(Rscratch, Rscratch2);
 965   __ b(Ldone);
 966 
 967   // Store is OK.
 968   __ bind(Lstore_ok);
 969   do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
 970                Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
 971 
 972   __ bind(Ldone);
 973   // Adjust sp (pops array, index and value).
 974   __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
 975 }
 976 
 977 void TemplateTable::bastore() {
 978   transition(itos, vtos);
 979 
 980   const Register Rindex   = R11_scratch1,
 981                  Rarray   = R12_scratch2,
 982                  Rscratch = R3_ARG1;
 983   __ pop_i(Rindex);
 984   // tos: val
 985   // Rarray: array ptr (popped by index_check)
 986   __ index_check(Rarray, Rindex, 0, Rscratch, Rarray);
 987   __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
 988 }
 989 
 990 void TemplateTable::castore() {
 991   transition(itos, vtos);
 992 
 993   const Register Rindex   = R11_scratch1,
 994                  Rarray   = R12_scratch2,
 995                  Rscratch = R3_ARG1;
 996   __ pop_i(Rindex);
 997   // tos: val
 998   // Rarray: array ptr (popped by index_check)
 999   __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1000   __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1001 }
1002 
1003 void TemplateTable::sastore() {
1004   castore();
1005 }
1006 
1007 void TemplateTable::istore(int n) {
1008   transition(itos, vtos);
1009   __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1010 }
1011 
1012 void TemplateTable::lstore(int n) {
1013   transition(ltos, vtos);
1014   __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1015 }
1016 
1017 void TemplateTable::fstore(int n) {
1018   transition(ftos, vtos);
1019   __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1020 }
1021 
1022 void TemplateTable::dstore(int n) {
1023   transition(dtos, vtos);
1024   __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1025 }
1026 
1027 void TemplateTable::astore(int n) {
1028   transition(vtos, vtos);
1029 
1030   __ pop_ptr();
1031   __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1032   __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1033 }
1034 
1035 void TemplateTable::pop() {
1036   transition(vtos, vtos);
1037 
1038   __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1039 }
1040 
1041 void TemplateTable::pop2() {
1042   transition(vtos, vtos);
1043 
1044   __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1045 }
1046 
1047 void TemplateTable::dup() {
1048   transition(vtos, vtos);
1049 
1050   __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1051   __ push_ptr(R11_scratch1);
1052 }
1053 
1054 void TemplateTable::dup_x1() {
1055   transition(vtos, vtos);
1056 
1057   Register Ra = R11_scratch1,
1058            Rb = R12_scratch2;
1059   // stack: ..., a, b
1060   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1061   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1062   __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1063   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1064   __ push_ptr(Rb);
1065   // stack: ..., b, a, b
1066 }
1067 
1068 void TemplateTable::dup_x2() {
1069   transition(vtos, vtos);
1070 
1071   Register Ra = R11_scratch1,
1072            Rb = R12_scratch2,
1073            Rc = R3_ARG1;
1074 
1075   // stack: ..., a, b, c
1076   __ ld(Rc, Interpreter::stackElementSize,     R15_esp);  // load c
1077   __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);  // load a
1078   __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1079   __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);  // load b
1080   // stack: ..., c, b, c
1081   __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1082   // stack: ..., c, a, c
1083   __ std(Rb, Interpreter::stackElementSize,     R15_esp); // store b in c
1084   __ push_ptr(Rc);                                        // push c
1085   // stack: ..., c, a, b, c
1086 }
1087 
1088 void TemplateTable::dup2() {
1089   transition(vtos, vtos);
1090 
1091   Register Ra = R11_scratch1,
1092            Rb = R12_scratch2;
1093   // stack: ..., a, b
1094   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1095   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1096   __ push_2ptrs(Ra, Rb);
1097   // stack: ..., a, b, a, b
1098 }
1099 
1100 void TemplateTable::dup2_x1() {
1101   transition(vtos, vtos);
1102 
1103   Register Ra = R11_scratch1,
1104            Rb = R12_scratch2,
1105            Rc = R3_ARG1;
1106   // stack: ..., a, b, c
1107   __ ld(Rc, Interpreter::stackElementSize,     R15_esp);
1108   __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1109   __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1110   __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1111   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1112   __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1113   // stack: ..., b, c, a
1114   __ push_2ptrs(Rb, Rc);
1115   // stack: ..., b, c, a, b, c
1116 }
1117 
1118 void TemplateTable::dup2_x2() {
1119   transition(vtos, vtos);
1120 
1121   Register Ra = R11_scratch1,
1122            Rb = R12_scratch2,
1123            Rc = R3_ARG1,
1124            Rd = R4_ARG2;
1125   // stack: ..., a, b, c, d
1126   __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1127   __ ld(Rd, Interpreter::stackElementSize,     R15_esp);
1128   __ std(Rb, Interpreter::stackElementSize,     R15_esp);  // store b in d
1129   __ std(Rd, Interpreter::stackElementSize * 3, R15_esp);  // store d in b
1130   __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1131   __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1132   __ std(Ra, Interpreter::stackElementSize * 2, R15_esp);  // store a in c
1133   __ std(Rc, Interpreter::stackElementSize * 4, R15_esp);  // store c in a
1134   // stack: ..., c, d, a, b
1135   __ push_2ptrs(Rc, Rd);
1136   // stack: ..., c, d, a, b, c, d
1137 }
1138 
1139 void TemplateTable::swap() {
1140   transition(vtos, vtos);
1141   // stack: ..., a, b
1142 
1143   Register Ra = R11_scratch1,
1144            Rb = R12_scratch2;
1145   // stack: ..., a, b
1146   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1147   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1148   __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1149   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1150   // stack: ..., b, a
1151 }
1152 
1153 void TemplateTable::iop2(Operation op) {
1154   transition(itos, itos);
1155 
1156   Register Rscratch = R11_scratch1;
1157 
1158   __ pop_i(Rscratch);
1159   // tos  = number of bits to shift
1160   // Rscratch = value to shift
1161   switch (op) {
1162     case  add:   __ add(R17_tos, Rscratch, R17_tos); break;
1163     case  sub:   __ sub(R17_tos, Rscratch, R17_tos); break;
1164     case  mul:   __ mullw(R17_tos, Rscratch, R17_tos); break;
1165     case  _and:  __ andr(R17_tos, Rscratch, R17_tos); break;
1166     case  _or:   __ orr(R17_tos, Rscratch, R17_tos); break;
1167     case  _xor:  __ xorr(R17_tos, Rscratch, R17_tos); break;
1168     case  shl:   __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1169     case  shr:   __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1170     case  ushr:  __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1171     default:     ShouldNotReachHere();
1172   }
1173 }
1174 
1175 void TemplateTable::lop2(Operation op) {
1176   transition(ltos, ltos);
1177 
1178   Register Rscratch = R11_scratch1;
1179   __ pop_l(Rscratch);
1180   switch (op) {
1181     case  add:   __ add(R17_tos, Rscratch, R17_tos); break;
1182     case  sub:   __ sub(R17_tos, Rscratch, R17_tos); break;
1183     case  _and:  __ andr(R17_tos, Rscratch, R17_tos); break;
1184     case  _or:   __ orr(R17_tos, Rscratch, R17_tos); break;
1185     case  _xor:  __ xorr(R17_tos, Rscratch, R17_tos); break;
1186     default:     ShouldNotReachHere();
1187   }
1188 }
1189 
1190 void TemplateTable::idiv() {
1191   transition(itos, itos);
1192 
1193   Label Lnormal, Lexception, Ldone;
1194   Register Rdividend = R11_scratch1; // Used by irem.
1195 
1196   __ addi(R0, R17_tos, 1);
1197   __ cmplwi(CCR0, R0, 2);
1198   __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1199 
1200   __ cmpwi(CCR1, R17_tos, 0);
1201   __ beq(CCR1, Lexception); // divisor == 0
1202 
1203   __ pop_i(Rdividend);
1204   __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1205   __ b(Ldone);
1206 
1207   __ bind(Lexception);
1208   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1209   __ mtctr(R11_scratch1);
1210   __ bctr();
1211 
1212   __ align(32, 12);
1213   __ bind(Lnormal);
1214   __ pop_i(Rdividend);
1215   __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1216   __ bind(Ldone);
1217 }
1218 
1219 void TemplateTable::irem() {
1220   transition(itos, itos);
1221 
1222   __ mr(R12_scratch2, R17_tos);
1223   idiv();
1224   __ mullw(R17_tos, R17_tos, R12_scratch2);
1225   __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1226 }
1227 
1228 void TemplateTable::lmul() {
1229   transition(ltos, ltos);
1230 
1231   __ pop_l(R11_scratch1);
1232   __ mulld(R17_tos, R11_scratch1, R17_tos);
1233 }
1234 
1235 void TemplateTable::ldiv() {
1236   transition(ltos, ltos);
1237 
1238   Label Lnormal, Lexception, Ldone;
1239   Register Rdividend = R11_scratch1; // Used by lrem.
1240 
1241   __ addi(R0, R17_tos, 1);
1242   __ cmpldi(CCR0, R0, 2);
1243   __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1244 
1245   __ cmpdi(CCR1, R17_tos, 0);
1246   __ beq(CCR1, Lexception); // divisor == 0
1247 
1248   __ pop_l(Rdividend);
1249   __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1250   __ b(Ldone);
1251 
1252   __ bind(Lexception);
1253   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1254   __ mtctr(R11_scratch1);
1255   __ bctr();
1256 
1257   __ align(32, 12);
1258   __ bind(Lnormal);
1259   __ pop_l(Rdividend);
1260   __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1261   __ bind(Ldone);
1262 }
1263 
1264 void TemplateTable::lrem() {
1265   transition(ltos, ltos);
1266 
1267   __ mr(R12_scratch2, R17_tos);
1268   ldiv();
1269   __ mulld(R17_tos, R17_tos, R12_scratch2);
1270   __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1271 }
1272 
1273 void TemplateTable::lshl() {
1274   transition(itos, ltos);
1275 
1276   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1277   __ pop_l(R11_scratch1);
1278   __ sld(R17_tos, R11_scratch1, R17_tos);
1279 }
1280 
1281 void TemplateTable::lshr() {
1282   transition(itos, ltos);
1283 
1284   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1285   __ pop_l(R11_scratch1);
1286   __ srad(R17_tos, R11_scratch1, R17_tos);
1287 }
1288 
1289 void TemplateTable::lushr() {
1290   transition(itos, ltos);
1291 
1292   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1293   __ pop_l(R11_scratch1);
1294   __ srd(R17_tos, R11_scratch1, R17_tos);
1295 }
1296 
1297 void TemplateTable::fop2(Operation op) {
1298   transition(ftos, ftos);
1299 
1300   switch (op) {
1301     case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1302     case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1303     case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1304     case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1305     case rem:
1306       __ pop_f(F1_ARG1);
1307       __ fmr(F2_ARG2, F15_ftos);
1308       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1309       __ fmr(F15_ftos, F1_RET);
1310       break;
1311 
1312     default: ShouldNotReachHere();
1313   }
1314 }
1315 
1316 void TemplateTable::dop2(Operation op) {
1317   transition(dtos, dtos);
1318 
1319   switch (op) {
1320     case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1321     case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1322     case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1323     case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1324     case rem:
1325       __ pop_d(F1_ARG1);
1326       __ fmr(F2_ARG2, F15_ftos);
1327       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1328       __ fmr(F15_ftos, F1_RET);
1329       break;
1330 
1331     default: ShouldNotReachHere();
1332   }
1333 }
1334 
1335 // Negate the value in the TOS cache.
1336 void TemplateTable::ineg() {
1337   transition(itos, itos);
1338 
1339   __ neg(R17_tos, R17_tos);
1340 }
1341 
1342 // Negate the value in the TOS cache.
1343 void TemplateTable::lneg() {
1344   transition(ltos, ltos);
1345 
1346   __ neg(R17_tos, R17_tos);
1347 }
1348 
1349 void TemplateTable::fneg() {
1350   transition(ftos, ftos);
1351 
1352   __ fneg(F15_ftos, F15_ftos);
1353 }
1354 
1355 void TemplateTable::dneg() {
1356   transition(dtos, dtos);
1357 
1358   __ fneg(F15_ftos, F15_ftos);
1359 }
1360 
1361 // Increments a local variable in place.
1362 void TemplateTable::iinc() {
1363   transition(vtos, vtos);
1364 
1365   const Register Rindex     = R11_scratch1,
1366                  Rincrement = R0,
1367                  Rvalue     = R12_scratch2;
1368 
1369   locals_index(Rindex);              // Load locals index from bytecode stream.
1370   __ lbz(Rincrement, 2, R14_bcp);    // Load increment from the bytecode stream.
1371   __ extsb(Rincrement, Rincrement);
1372 
1373   __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1374 
1375   __ add(Rvalue, Rincrement, Rvalue);
1376   __ stw(Rvalue, 0, Rindex);
1377 }
1378 
1379 void TemplateTable::wide_iinc() {
1380   transition(vtos, vtos);
1381 
1382   Register Rindex       = R11_scratch1,
1383            Rlocals_addr = Rindex,
1384            Rincr        = R12_scratch2;
1385   locals_index_wide(Rindex);
1386   __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1387   __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1388   __ add(R17_tos, Rincr, R17_tos);
1389   __ stw(R17_tos, 0, Rlocals_addr);
1390 }
1391 
1392 void TemplateTable::convert() {
1393   // %%%%% Factor this first part accross platforms
1394 #ifdef ASSERT
1395   TosState tos_in  = ilgl;
1396   TosState tos_out = ilgl;
1397   switch (bytecode()) {
1398     case Bytecodes::_i2l: // fall through
1399     case Bytecodes::_i2f: // fall through
1400     case Bytecodes::_i2d: // fall through
1401     case Bytecodes::_i2b: // fall through
1402     case Bytecodes::_i2c: // fall through
1403     case Bytecodes::_i2s: tos_in = itos; break;
1404     case Bytecodes::_l2i: // fall through
1405     case Bytecodes::_l2f: // fall through
1406     case Bytecodes::_l2d: tos_in = ltos; break;
1407     case Bytecodes::_f2i: // fall through
1408     case Bytecodes::_f2l: // fall through
1409     case Bytecodes::_f2d: tos_in = ftos; break;
1410     case Bytecodes::_d2i: // fall through
1411     case Bytecodes::_d2l: // fall through
1412     case Bytecodes::_d2f: tos_in = dtos; break;
1413     default             : ShouldNotReachHere();
1414   }
1415   switch (bytecode()) {
1416     case Bytecodes::_l2i: // fall through
1417     case Bytecodes::_f2i: // fall through
1418     case Bytecodes::_d2i: // fall through
1419     case Bytecodes::_i2b: // fall through
1420     case Bytecodes::_i2c: // fall through
1421     case Bytecodes::_i2s: tos_out = itos; break;
1422     case Bytecodes::_i2l: // fall through
1423     case Bytecodes::_f2l: // fall through
1424     case Bytecodes::_d2l: tos_out = ltos; break;
1425     case Bytecodes::_i2f: // fall through
1426     case Bytecodes::_l2f: // fall through
1427     case Bytecodes::_d2f: tos_out = ftos; break;
1428     case Bytecodes::_i2d: // fall through
1429     case Bytecodes::_l2d: // fall through
1430     case Bytecodes::_f2d: tos_out = dtos; break;
1431     default             : ShouldNotReachHere();
1432   }
1433   transition(tos_in, tos_out);
1434 #endif
1435 
1436   // Conversion
1437   Label done;
1438   switch (bytecode()) {
1439     case Bytecodes::_i2l:
1440       __ extsw(R17_tos, R17_tos);
1441       break;
1442 
1443     case Bytecodes::_l2i:
1444       // Nothing to do, we'll continue to work with the lower bits.
1445       break;
1446 
1447     case Bytecodes::_i2b:
1448       __ extsb(R17_tos, R17_tos);
1449       break;
1450 
1451     case Bytecodes::_i2c:
1452       __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1453       break;
1454 
1455     case Bytecodes::_i2s:
1456       __ extsh(R17_tos, R17_tos);
1457       break;
1458 
1459     case Bytecodes::_i2d:
1460       __ extsw(R17_tos, R17_tos);
1461     case Bytecodes::_l2d:
1462       __ push_l_pop_d();
1463       __ fcfid(F15_ftos, F15_ftos);
1464       break;
1465 
1466     case Bytecodes::_i2f:
1467       __ extsw(R17_tos, R17_tos);
1468       __ push_l_pop_d();
1469       if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1470         // Comment: alternatively, load with sign extend could be done by lfiwax.
1471         __ fcfids(F15_ftos, F15_ftos);
1472       } else {
1473         __ fcfid(F15_ftos, F15_ftos);
1474         __ frsp(F15_ftos, F15_ftos);
1475       }
1476       break;
1477 
1478     case Bytecodes::_l2f:
1479       if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1480         __ push_l_pop_d();
1481         __ fcfids(F15_ftos, F15_ftos);
1482       } else {
1483         // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1484         __ mr(R3_ARG1, R17_tos);
1485         __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1486         __ fmr(F15_ftos, F1_RET);
1487       }
1488       break;
1489 
1490     case Bytecodes::_f2d:
1491       // empty
1492       break;
1493 
1494     case Bytecodes::_d2f:
1495       __ frsp(F15_ftos, F15_ftos);
1496       break;
1497 
1498     case Bytecodes::_d2i:
1499     case Bytecodes::_f2i:
1500       __ fcmpu(CCR0, F15_ftos, F15_ftos);
1501       __ li(R17_tos, 0); // 0 in case of NAN
1502       __ bso(CCR0, done);
1503       __ fctiwz(F15_ftos, F15_ftos);
1504       __ push_d_pop_l();
1505       break;
1506 
1507     case Bytecodes::_d2l:
1508     case Bytecodes::_f2l:
1509       __ fcmpu(CCR0, F15_ftos, F15_ftos);
1510       __ li(R17_tos, 0); // 0 in case of NAN
1511       __ bso(CCR0, done);
1512       __ fctidz(F15_ftos, F15_ftos);
1513       __ push_d_pop_l();
1514       break;
1515 
1516     default: ShouldNotReachHere();
1517   }
1518   __ bind(done);
1519 }
1520 
1521 // Long compare
1522 void TemplateTable::lcmp() {
1523   transition(ltos, itos);
1524 
1525   const Register Rscratch = R11_scratch1;
1526   __ pop_l(Rscratch); // first operand, deeper in stack
1527 
1528   __ cmpd(CCR0, Rscratch, R17_tos); // compare
1529   __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1530   __ srwi(Rscratch, R17_tos, 30);
1531   __ srawi(R17_tos, R17_tos, 31);
1532   __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1533 }
1534 
1535 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1536 // unordered_result == -1 => fcmpl or dcmpl
1537 // unordered_result ==  1 => fcmpg or dcmpg
1538 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1539   const FloatRegister Rfirst  = F0_SCRATCH,
1540                       Rsecond = F15_ftos;
1541   const Register Rscratch = R11_scratch1;
1542 
1543   if (is_float) {
1544     __ pop_f(Rfirst);
1545   } else {
1546     __ pop_d(Rfirst);
1547   }
1548 
1549   Label Lunordered, Ldone;
1550   __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1551   if (unordered_result) {
1552     __ bso(CCR0, Lunordered);
1553   }
1554   __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1555   __ srwi(Rscratch, R17_tos, 30);
1556   __ srawi(R17_tos, R17_tos, 31);
1557   __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1558   if (unordered_result) {
1559     __ b(Ldone);
1560     __ bind(Lunordered);
1561     __ load_const_optimized(R17_tos, unordered_result);
1562   }
1563   __ bind(Ldone);
1564 }
1565 
1566 // Branch_conditional which takes TemplateTable::Condition.
1567 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1568   bool positive = false;
1569   Assembler::Condition cond = Assembler::equal;
1570   switch (cc) {
1571     case TemplateTable::equal:         positive = true ; cond = Assembler::equal  ; break;
1572     case TemplateTable::not_equal:     positive = false; cond = Assembler::equal  ; break;
1573     case TemplateTable::less:          positive = true ; cond = Assembler::less   ; break;
1574     case TemplateTable::less_equal:    positive = false; cond = Assembler::greater; break;
1575     case TemplateTable::greater:       positive = true ; cond = Assembler::greater; break;
1576     case TemplateTable::greater_equal: positive = false; cond = Assembler::less   ; break;
1577     default: ShouldNotReachHere();
1578   }
1579   int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1580   int bi = Assembler::bi0(crx, cond);
1581   __ bc(bo, bi, L);
1582 }
1583 
1584 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1585 
1586   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1587   __ verify_thread();
1588 
1589   const Register Rscratch1    = R11_scratch1,
1590                  Rscratch2    = R12_scratch2,
1591                  Rscratch3    = R3_ARG1,
1592                  R4_counters  = R4_ARG2,
1593                  bumped_count = R31,
1594                  Rdisp        = R22_tmp2;
1595 
1596   __ profile_taken_branch(Rscratch1, bumped_count);
1597 
1598   // Get (wide) offset.
1599   if (is_wide) {
1600     __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1601   } else {
1602     __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1603   }
1604 
1605   // --------------------------------------------------------------------------
1606   // Handle all the JSR stuff here, then exit.
1607   // It's much shorter and cleaner than intermingling with the
1608   // non-JSR normal-branch stuff occurring below.
1609   if (is_jsr) {
1610     // Compute return address as bci in Otos_i.
1611     __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1612     __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1613     __ subf(R17_tos, Rscratch1, Rscratch2);
1614 
1615     // Bump bcp to target of JSR.
1616     __ add(R14_bcp, Rdisp, R14_bcp);
1617     // Push returnAddress for "ret" on stack.
1618     __ push_ptr(R17_tos);
1619     // And away we go!
1620     __ dispatch_next(vtos);
1621     return;
1622   }
1623 
1624   // --------------------------------------------------------------------------
1625   // Normal (non-jsr) branch handling
1626 
1627   const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1628   if (increment_invocation_counter_for_backward_branches) {
1629     //__ unimplemented("branch invocation counter");
1630 
1631     Label Lforward;
1632     __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1633 
1634     // Check branch direction.
1635     __ cmpdi(CCR0, Rdisp, 0);
1636     __ bgt(CCR0, Lforward);
1637 
1638     __ get_method_counters(R19_method, R4_counters, Lforward);
1639 
1640     if (TieredCompilation) {
1641       Label Lno_mdo, Loverflow;
1642       const int increment = InvocationCounter::count_increment;
1643       const int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1644       if (ProfileInterpreter) {
1645         Register Rmdo = Rscratch1;
1646 
1647         // If no method data exists, go to profile_continue.
1648         __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1649         __ cmpdi(CCR0, Rmdo, 0);
1650         __ beq(CCR0, Lno_mdo);
1651 
1652         // Increment backedge counter in the MDO.
1653         const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1654         __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1655         __ load_const_optimized(Rscratch3, mask, R0);
1656         __ addi(Rscratch2, Rscratch2, increment);
1657         __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1658         __ and_(Rscratch3, Rscratch2, Rscratch3);
1659         __ bne(CCR0, Lforward);
1660         __ b(Loverflow);
1661       }
1662 
1663       // If there's no MDO, increment counter in method.
1664       const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1665       __ bind(Lno_mdo);
1666       __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1667       __ load_const_optimized(Rscratch3, mask, R0);
1668       __ addi(Rscratch2, Rscratch2, increment);
1669       __ stw(Rscratch2, mo_bc_offs, R19_method);
1670       __ and_(Rscratch3, Rscratch2, Rscratch3);
1671       __ bne(CCR0, Lforward);
1672 
1673       __ bind(Loverflow);
1674 
1675       // Notify point for loop, pass branch bytecode.
1676       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R14_bcp, true);
1677 
1678       // Was an OSR adapter generated?
1679       // O0 = osr nmethod
1680       __ cmpdi(CCR0, R3_RET, 0);
1681       __ beq(CCR0, Lforward);
1682 
1683       // Has the nmethod been invalidated already?
1684       __ lbz(R0, nmethod::state_offset(), R3_RET);
1685       __ cmpwi(CCR0, R0, nmethod::in_use);
1686       __ bne(CCR0, Lforward);
1687 
1688       // Migrate the interpreter frame off of the stack.
1689       // We can use all registers because we will not return to interpreter from this point.
1690 
1691       // Save nmethod.
1692       const Register osr_nmethod = R31;
1693       __ mr(osr_nmethod, R3_RET);
1694       __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1695       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1696       __ reset_last_Java_frame();
1697       // OSR buffer is in ARG1.
1698 
1699       // Remove the interpreter frame.
1700       __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1701 
1702       // Jump to the osr code.
1703       __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1704       __ mtlr(R0);
1705       __ mtctr(R11_scratch1);
1706       __ bctr();
1707 
1708     } else {
1709 
1710       const Register invoke_ctr = Rscratch1;
1711       // Update Backedge branch separately from invocations.
1712       __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1713 
1714       if (ProfileInterpreter) {
1715         __ test_invocation_counter_for_mdp(invoke_ctr, Rscratch2, Lforward);
1716         if (UseOnStackReplacement) {
1717           __ test_backedge_count_for_osr(bumped_count, R14_bcp, Rscratch2);
1718         }
1719       } else {
1720         if (UseOnStackReplacement) {
1721           __ test_backedge_count_for_osr(invoke_ctr, R14_bcp, Rscratch2);
1722         }
1723       }
1724     }
1725 
1726     __ bind(Lforward);
1727 
1728   } else {
1729     // Bump bytecode pointer by displacement (take the branch).
1730     __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1731   }
1732   // Continue with bytecode @ target.
1733   // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1734   // %%%%% and changing dispatch_next to dispatch_only.
1735   __ dispatch_next(vtos);
1736 }
1737 
1738 // Helper function for if_cmp* methods below.
1739 // Factored out common compare and branch code.
1740 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1741   Label Lnot_taken;
1742   // Note: The condition code we get is the condition under which we
1743   // *fall through*! So we have to inverse the CC here.
1744 
1745   if (is_jint) {
1746     if (cmp0) {
1747       __ cmpwi(CCR0, Rfirst, 0);
1748     } else {
1749       __ cmpw(CCR0, Rfirst, Rsecond);
1750     }
1751   } else {
1752     if (cmp0) {
1753       __ cmpdi(CCR0, Rfirst, 0);
1754     } else {
1755       __ cmpd(CCR0, Rfirst, Rsecond);
1756     }
1757   }
1758   branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1759 
1760   // Conition is false => Jump!
1761   branch(false, false);
1762 
1763   // Condition is not true => Continue.
1764   __ align(32, 12);
1765   __ bind(Lnot_taken);
1766   __ profile_not_taken_branch(Rscratch1, Rscratch2);
1767 }
1768 
1769 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1770 void TemplateTable::if_0cmp(Condition cc) {
1771   transition(itos, vtos);
1772 
1773   if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1774 }
1775 
1776 // Compare integer values and fall through if CC holds, branch away otherwise.
1777 //
1778 // Interface:
1779 //  - Rfirst: First operand  (older stack value)
1780 //  - tos:    Second operand (younger stack value)
1781 void TemplateTable::if_icmp(Condition cc) {
1782   transition(itos, vtos);
1783 
1784   const Register Rfirst  = R0,
1785                  Rsecond = R17_tos;
1786 
1787   __ pop_i(Rfirst);
1788   if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1789 }
1790 
1791 void TemplateTable::if_nullcmp(Condition cc) {
1792   transition(atos, vtos);
1793 
1794   if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1795 }
1796 
1797 void TemplateTable::if_acmp(Condition cc) {
1798   transition(atos, vtos);
1799 
1800   const Register Rfirst  = R0,
1801                  Rsecond = R17_tos;
1802 
1803   __ pop_ptr(Rfirst);
1804   if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1805 }
1806 
1807 void TemplateTable::ret() {
1808   locals_index(R11_scratch1);
1809   __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1810 
1811   __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1812 
1813   __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1814   __ add(R11_scratch1, R17_tos, R11_scratch1);
1815   __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1816   __ dispatch_next(vtos);
1817 }
1818 
1819 void TemplateTable::wide_ret() {
1820   transition(vtos, vtos);
1821 
1822   const Register Rindex = R3_ARG1,
1823                  Rscratch1 = R11_scratch1,
1824                  Rscratch2 = R12_scratch2;
1825 
1826   locals_index_wide(Rindex);
1827   __ load_local_ptr(R17_tos, R17_tos, Rindex);
1828   __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1829   // Tos now contains the bci, compute the bcp from that.
1830   __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1831   __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1832   __ add(R14_bcp, Rscratch1, Rscratch2);
1833   __ dispatch_next(vtos);
1834 }
1835 
1836 void TemplateTable::tableswitch() {
1837   transition(itos, vtos);
1838 
1839   Label Ldispatch, Ldefault_case;
1840   Register Rlow_byte         = R3_ARG1,
1841            Rindex            = Rlow_byte,
1842            Rhigh_byte        = R4_ARG2,
1843            Rdef_offset_addr  = R5_ARG3, // is going to contain address of default offset
1844            Rscratch1         = R11_scratch1,
1845            Rscratch2         = R12_scratch2,
1846            Roffset           = R6_ARG4;
1847 
1848   // Align bcp.
1849   __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1850   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1851 
1852   // Load lo & hi.
1853   __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1854   __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1855 
1856   // Check for default case (=index outside [low,high]).
1857   __ cmpw(CCR0, R17_tos, Rlow_byte);
1858   __ cmpw(CCR1, R17_tos, Rhigh_byte);
1859   __ blt(CCR0, Ldefault_case);
1860   __ bgt(CCR1, Ldefault_case);
1861 
1862   // Lookup dispatch offset.
1863   __ sub(Rindex, R17_tos, Rlow_byte);
1864   __ extsw(Rindex, Rindex);
1865   __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1866   __ sldi(Rindex, Rindex, LogBytesPerInt);
1867   __ addi(Rindex, Rindex, 3 * BytesPerInt);
1868 #if defined(VM_LITTLE_ENDIAN)
1869   __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1870   __ extsw(Roffset, Roffset);
1871 #else
1872   __ lwax(Roffset, Rdef_offset_addr, Rindex);
1873 #endif
1874   __ b(Ldispatch);
1875 
1876   __ bind(Ldefault_case);
1877   __ profile_switch_default(Rhigh_byte, Rscratch1);
1878   __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1879 
1880   __ bind(Ldispatch);
1881 
1882   __ add(R14_bcp, Roffset, R14_bcp);
1883   __ dispatch_next(vtos);
1884 }
1885 
1886 void TemplateTable::lookupswitch() {
1887   transition(itos, itos);
1888   __ stop("lookupswitch bytecode should have been rewritten");
1889 }
1890 
1891 // Table switch using linear search through cases.
1892 // Bytecode stream format:
1893 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1894 // Note: Everything is big-endian format here.
1895 void TemplateTable::fast_linearswitch() {
1896   transition(itos, vtos);
1897 
1898   Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1899   Register Rcount           = R3_ARG1,
1900            Rcurrent_pair    = R4_ARG2,
1901            Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1902            Roffset          = R31,     // Might need to survive C call.
1903            Rvalue           = R12_scratch2,
1904            Rscratch         = R11_scratch1,
1905            Rcmp_value       = R17_tos;
1906 
1907   // Align bcp.
1908   __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1909   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1910 
1911   // Setup loop counter and limit.
1912   __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1913   __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1914 
1915   __ mtctr(Rcount);
1916   __ cmpwi(CCR0, Rcount, 0);
1917   __ bne(CCR0, Lloop_entry);
1918 
1919   // Default case
1920   __ bind(Ldefault_case);
1921   __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1922   if (ProfileInterpreter) {
1923     __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1924   }
1925   __ b(Lcontinue_execution);
1926 
1927   // Next iteration
1928   __ bind(Lsearch_loop);
1929   __ bdz(Ldefault_case);
1930   __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1931   __ bind(Lloop_entry);
1932   __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1933   __ cmpw(CCR0, Rvalue, Rcmp_value);
1934   __ bne(CCR0, Lsearch_loop);
1935 
1936   // Found, load offset.
1937   __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1938   // Calculate case index and profile
1939   __ mfctr(Rcurrent_pair);
1940   if (ProfileInterpreter) {
1941     __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1942     __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1943   }
1944 
1945   __ bind(Lcontinue_execution);
1946   __ add(R14_bcp, Roffset, R14_bcp);
1947   __ dispatch_next(vtos);
1948 }
1949 
1950 // Table switch using binary search (value/offset pairs are ordered).
1951 // Bytecode stream format:
1952 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1953 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1954 void TemplateTable::fast_binaryswitch() {
1955 
1956   transition(itos, vtos);
1957   // Implementation using the following core algorithm: (copied from Intel)
1958   //
1959   // int binary_search(int key, LookupswitchPair* array, int n) {
1960   //   // Binary search according to "Methodik des Programmierens" by
1961   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1962   //   int i = 0;
1963   //   int j = n;
1964   //   while (i+1 < j) {
1965   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1966   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1967   //     // where a stands for the array and assuming that the (inexisting)
1968   //     // element a[n] is infinitely big.
1969   //     int h = (i + j) >> 1;
1970   //     // i < h < j
1971   //     if (key < array[h].fast_match()) {
1972   //       j = h;
1973   //     } else {
1974   //       i = h;
1975   //     }
1976   //   }
1977   //   // R: a[i] <= key < a[i+1] or Q
1978   //   // (i.e., if key is within array, i is the correct index)
1979   //   return i;
1980   // }
1981 
1982   // register allocation
1983   const Register Rkey     = R17_tos;          // already set (tosca)
1984   const Register Rarray   = R3_ARG1;
1985   const Register Ri       = R4_ARG2;
1986   const Register Rj       = R5_ARG3;
1987   const Register Rh       = R6_ARG4;
1988   const Register Rscratch = R11_scratch1;
1989 
1990   const int log_entry_size = 3;
1991   const int entry_size = 1 << log_entry_size;
1992 
1993   Label found;
1994 
1995   // Find Array start,
1996   __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
1997   __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
1998 
1999   // initialize i & j
2000   __ li(Ri,0);
2001   __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2002 
2003   // and start.
2004   Label entry;
2005   __ b(entry);
2006 
2007   // binary search loop
2008   { Label loop;
2009     __ bind(loop);
2010     // int h = (i + j) >> 1;
2011     __ srdi(Rh, Rh, 1);
2012     // if (key < array[h].fast_match()) {
2013     //   j = h;
2014     // } else {
2015     //   i = h;
2016     // }
2017     __ sldi(Rscratch, Rh, log_entry_size);
2018 #if defined(VM_LITTLE_ENDIAN)
2019     __ lwbrx(Rscratch, Rscratch, Rarray);
2020 #else
2021     __ lwzx(Rscratch, Rscratch, Rarray);
2022 #endif
2023 
2024     // if (key < current value)
2025     //   Rh = Rj
2026     // else
2027     //   Rh = Ri
2028     Label Lgreater;
2029     __ cmpw(CCR0, Rkey, Rscratch);
2030     __ bge(CCR0, Lgreater);
2031     __ mr(Rj, Rh);
2032     __ b(entry);
2033     __ bind(Lgreater);
2034     __ mr(Ri, Rh);
2035 
2036     // while (i+1 < j)
2037     __ bind(entry);
2038     __ addi(Rscratch, Ri, 1);
2039     __ cmpw(CCR0, Rscratch, Rj);
2040     __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2041 
2042     __ blt(CCR0, loop);
2043   }
2044 
2045   // End of binary search, result index is i (must check again!).
2046   Label default_case;
2047   Label continue_execution;
2048   if (ProfileInterpreter) {
2049     __ mr(Rh, Ri);              // Save index in i for profiling.
2050   }
2051   // Ri = value offset
2052   __ sldi(Ri, Ri, log_entry_size);
2053   __ add(Ri, Ri, Rarray);
2054   __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2055 
2056   Label not_found;
2057   // Ri = offset offset
2058   __ cmpw(CCR0, Rkey, Rscratch);
2059   __ beq(CCR0, not_found);
2060   // entry not found -> j = default offset
2061   __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2062   __ b(default_case);
2063 
2064   __ bind(not_found);
2065   // entry found -> j = offset
2066   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2067   __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2068 
2069   if (ProfileInterpreter) {
2070     __ b(continue_execution);
2071   }
2072 
2073   __ bind(default_case); // fall through (if not profiling)
2074   __ profile_switch_default(Ri, Rscratch);
2075 
2076   __ bind(continue_execution);
2077 
2078   __ extsw(Rj, Rj);
2079   __ add(R14_bcp, Rj, R14_bcp);
2080   __ dispatch_next(vtos);
2081 }
2082 
2083 void TemplateTable::_return(TosState state) {
2084   transition(state, state);
2085   assert(_desc->calls_vm(),
2086          "inconsistent calls_vm information"); // call in remove_activation
2087 
2088   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2089 
2090     Register Rscratch     = R11_scratch1,
2091              Rklass       = R12_scratch2,
2092              Rklass_flags = Rklass;
2093     Label Lskip_register_finalizer;
2094 
2095     // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2096     assert(state == vtos, "only valid state");
2097     __ ld(R17_tos, 0, R18_locals);
2098 
2099     // Load klass of this obj.
2100     __ load_klass(Rklass, R17_tos);
2101     __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2102     __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2103     __ bfalse(CCR0, Lskip_register_finalizer);
2104 
2105     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2106 
2107     __ align(32, 12);
2108     __ bind(Lskip_register_finalizer);
2109   }
2110 
2111   // Move the result value into the correct register and remove memory stack frame.
2112   __ remove_activation(state, /* throw_monitor_exception */ true);
2113   // Restoration of lr done by remove_activation.
2114   switch (state) {
2115     case ltos:
2116     case btos:
2117     case ctos:
2118     case stos:
2119     case atos:
2120     case itos: __ mr(R3_RET, R17_tos); break;
2121     case ftos:
2122     case dtos: __ fmr(F1_RET, F15_ftos); break;
2123     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2124                // to get visible before the reference to the object gets stored anywhere.
2125                __ membar(Assembler::StoreStore); break;
2126     default  : ShouldNotReachHere();
2127   }
2128   __ blr();
2129 }
2130 
2131 // ============================================================================
2132 // Constant pool cache access
2133 //
2134 // Memory ordering:
2135 //
2136 // Like done in C++ interpreter, we load the fields
2137 //   - _indices
2138 //   - _f12_oop
2139 // acquired, because these are asked if the cache is already resolved. We don't
2140 // want to float loads above this check.
2141 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2142 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2143 
2144 // Call into the VM if call site is not yet resolved
2145 //
2146 // Input regs:
2147 //   - None, all passed regs are outputs.
2148 //
2149 // Returns:
2150 //   - Rcache:  The const pool cache entry that contains the resolved result.
2151 //   - Rresult: Either noreg or output for f1/f2.
2152 //
2153 // Kills:
2154 //   - Rscratch
2155 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2156 
2157   __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2158   Label Lresolved, Ldone;
2159 
2160   Bytecodes::Code code = bytecode();
2161   switch (code) {
2162   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2163   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2164   }
2165 
2166   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2167   // We are resolved if the indices offset contains the current bytecode.
2168 #if defined(VM_LITTLE_ENDIAN)
2169   __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2170 #else
2171   __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2172 #endif
2173   // Acquire by cmp-br-isync (see below).
2174   __ cmpdi(CCR0, Rscratch, (int)code);
2175   __ beq(CCR0, Lresolved);
2176 
2177   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2178   __ li(R4_ARG2, code);
2179   __ call_VM(noreg, entry, R4_ARG2, true);
2180 
2181   // Update registers with resolved info.
2182   __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2183   __ b(Ldone);
2184 
2185   __ bind(Lresolved);
2186   __ isync(); // Order load wrt. succeeding loads.
2187   __ bind(Ldone);
2188 }
2189 
2190 // Load the constant pool cache entry at field accesses into registers.
2191 // The Rcache and Rindex registers must be set before call.
2192 // Input:
2193 //   - Rcache, Rindex
2194 // Output:
2195 //   - Robj, Roffset, Rflags
2196 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2197                                               Register Rcache,
2198                                               Register Rindex /* unused on PPC64 */,
2199                                               Register Roffset,
2200                                               Register Rflags,
2201                                               bool is_static = false) {
2202   assert_different_registers(Rcache, Rflags, Roffset);
2203   // assert(Rindex == noreg, "parameter not used on PPC64");
2204 
2205   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2206   __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2207   __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2208   if (is_static) {
2209     __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2210     __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2211     // Acquire not needed here. Following access has an address dependency on this value.
2212   }
2213 }
2214 
2215 // Load the constant pool cache entry at invokes into registers.
2216 // Resolve if necessary.
2217 
2218 // Input Registers:
2219 //   - None, bcp is used, though
2220 //
2221 // Return registers:
2222 //   - Rmethod       (f1 field or f2 if invokevirtual)
2223 //   - Ritable_index (f2 field)
2224 //   - Rflags        (flags field)
2225 //
2226 // Kills:
2227 //   - R21
2228 //
2229 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2230                                                Register Rmethod,
2231                                                Register Ritable_index,
2232                                                Register Rflags,
2233                                                bool is_invokevirtual,
2234                                                bool is_invokevfinal,
2235                                                bool is_invokedynamic) {
2236 
2237   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2238   // Determine constant pool cache field offsets.
2239   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2240   const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2241   const int flags_offset  = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2242   // Access constant pool cache fields.
2243   const int index_offset  = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2244 
2245   Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2246 
2247   if (is_invokevfinal) {
2248     assert(Ritable_index == noreg, "register not used");
2249     // Already resolved.
2250     __ get_cache_and_index_at_bcp(Rcache, 1);
2251   } else {
2252     resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2253   }
2254 
2255   __ ld(Rmethod, method_offset, Rcache);
2256   __ ld(Rflags, flags_offset, Rcache);
2257 
2258   if (Ritable_index != noreg) {
2259     __ ld(Ritable_index, index_offset, Rcache);
2260   }
2261 }
2262 
2263 // ============================================================================
2264 // Field access
2265 
2266 // Volatile variables demand their effects be made known to all CPU's
2267 // in order. Store buffers on most chips allow reads & writes to
2268 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2269 // without some kind of memory barrier (i.e., it's not sufficient that
2270 // the interpreter does not reorder volatile references, the hardware
2271 // also must not reorder them).
2272 //
2273 // According to the new Java Memory Model (JMM):
2274 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2275 //     writes act as aquire & release, so:
2276 // (2) A read cannot let unrelated NON-volatile memory refs that
2277 //     happen after the read float up to before the read. It's OK for
2278 //     non-volatile memory refs that happen before the volatile read to
2279 //     float down below it.
2280 // (3) Similar a volatile write cannot let unrelated NON-volatile
2281 //     memory refs that happen BEFORE the write float down to after the
2282 //     write. It's OK for non-volatile memory refs that happen after the
2283 //     volatile write to float up before it.
2284 //
2285 // We only put in barriers around volatile refs (they are expensive),
2286 // not _between_ memory refs (that would require us to track the
2287 // flavor of the previous memory refs). Requirements (2) and (3)
2288 // require some barriers before volatile stores and after volatile
2289 // loads. These nearly cover requirement (1) but miss the
2290 // volatile-store-volatile-load case.  This final case is placed after
2291 // volatile-stores although it could just as well go before
2292 // volatile-loads.
2293 
2294 // The registers cache and index expected to be set before call.
2295 // Correct values of the cache and index registers are preserved.
2296 // Kills:
2297 //   Rcache (if has_tos)
2298 //   Rscratch
2299 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2300 
2301   assert_different_registers(Rcache, Rscratch);
2302 
2303   if (JvmtiExport::can_post_field_access()) {
2304     ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2305     Label Lno_field_access_post;
2306 
2307     // Check if post field access in enabled.
2308     int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2309     __ lwz(Rscratch, offs, Rscratch);
2310 
2311     __ cmpwi(CCR0, Rscratch, 0);
2312     __ beq(CCR0, Lno_field_access_post);
2313 
2314     // Post access enabled - do it!
2315     __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2316     if (is_static) {
2317       __ li(R17_tos, 0);
2318     } else {
2319       if (has_tos) {
2320         // The fast bytecode versions have obj ptr in register.
2321         // Thus, save object pointer before call_VM() clobbers it
2322         // put object on tos where GC wants it.
2323         __ push_ptr(R17_tos);
2324       } else {
2325         // Load top of stack (do not pop the value off the stack).
2326         __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2327       }
2328       __ verify_oop(R17_tos);
2329     }
2330     // tos:   object pointer or NULL if static
2331     // cache: cache entry pointer
2332     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2333     if (!is_static && has_tos) {
2334       // Restore object pointer.
2335       __ pop_ptr(R17_tos);
2336       __ verify_oop(R17_tos);
2337     } else {
2338       // Cache is still needed to get class or obj.
2339       __ get_cache_and_index_at_bcp(Rcache, 1);
2340     }
2341 
2342     __ align(32, 12);
2343     __ bind(Lno_field_access_post);
2344   }
2345 }
2346 
2347 // kills R11_scratch1
2348 void TemplateTable::pop_and_check_object(Register Roop) {
2349   Register Rtmp = R11_scratch1;
2350 
2351   assert_different_registers(Rtmp, Roop);
2352   __ pop_ptr(Roop);
2353   // For field access must check obj.
2354   __ null_check_throw(Roop, -1, Rtmp);
2355   __ verify_oop(Roop);
2356 }
2357 
2358 // PPC64: implement volatile loads as fence-store-acquire.
2359 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2360   transition(vtos, vtos);
2361 
2362   Label Lacquire, Lisync;
2363 
2364   const Register Rcache        = R3_ARG1,
2365                  Rclass_or_obj = R22_tmp2,
2366                  Roffset       = R23_tmp3,
2367                  Rflags        = R31,
2368                  Rbtable       = R5_ARG3,
2369                  Rbc           = R6_ARG4,
2370                  Rscratch      = R12_scratch2;
2371 
2372   static address field_branch_table[number_of_states],
2373                  static_branch_table[number_of_states];
2374 
2375   address* branch_table = (is_static || rc == may_not_rewrite) ? static_branch_table : field_branch_table;
2376 
2377   // Get field offset.
2378   resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2379 
2380   // JVMTI support
2381   jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2382 
2383   // Load after possible GC.
2384   load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2385 
2386   // Load pointer to branch table.
2387   __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2388 
2389   // Get volatile flag.
2390   __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2391   // Note: sync is needed before volatile load on PPC64.
2392 
2393   // Check field type.
2394   __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2395 
2396 #ifdef ASSERT
2397   Label LFlagInvalid;
2398   __ cmpldi(CCR0, Rflags, number_of_states);
2399   __ bge(CCR0, LFlagInvalid);
2400 #endif
2401 
2402   // Load from branch table and dispatch (volatile case: one instruction ahead).
2403   __ sldi(Rflags, Rflags, LogBytesPerWord);
2404   __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2405   if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2406     __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2407   }
2408   __ ldx(Rbtable, Rbtable, Rflags);
2409 
2410   // Get the obj from stack.
2411   if (!is_static) {
2412     pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2413   } else {
2414     __ verify_oop(Rclass_or_obj);
2415   }
2416 
2417   if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2418     __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2419   }
2420   __ mtctr(Rbtable);
2421   __ bctr();
2422 
2423 #ifdef ASSERT
2424   __ bind(LFlagInvalid);
2425   __ stop("got invalid flag", 0x654);
2426 #endif
2427 
2428   if (!is_static && rc == may_not_rewrite) {
2429     // We reuse the code from is_static.  It's jumped to via the table above.
2430     return;
2431   }
2432 
2433 #ifdef ASSERT
2434   // __ bind(Lvtos);
2435   address pc_before_fence = __ pc();
2436   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2437   assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2438   assert(branch_table[vtos] == 0, "can't compute twice");
2439   branch_table[vtos] = __ pc(); // non-volatile_entry point
2440   __ stop("vtos unexpected", 0x655);
2441 #endif
2442 
2443   __ align(32, 28, 28); // Align load.
2444   // __ bind(Ldtos);
2445   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2446   assert(branch_table[dtos] == 0, "can't compute twice");
2447   branch_table[dtos] = __ pc(); // non-volatile_entry point
2448   __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2449   __ push(dtos);
2450   if (!is_static && rc == may_rewrite) {
2451     patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2452   }
2453   {
2454     Label acquire_double;
2455     __ beq(CCR6, acquire_double); // Volatile?
2456     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2457 
2458     __ bind(acquire_double);
2459     __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2460     __ beq_predict_taken(CCR0, Lisync);
2461     __ b(Lisync); // In case of NAN.
2462   }
2463 
2464   __ align(32, 28, 28); // Align load.
2465   // __ bind(Lftos);
2466   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2467   assert(branch_table[ftos] == 0, "can't compute twice");
2468   branch_table[ftos] = __ pc(); // non-volatile_entry point
2469   __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2470   __ push(ftos);
2471   if (!is_static && rc == may_rewrite) {
2472     patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch);
2473   }
2474   {
2475     Label acquire_float;
2476     __ beq(CCR6, acquire_float); // Volatile?
2477     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2478 
2479     __ bind(acquire_float);
2480     __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2481     __ beq_predict_taken(CCR0, Lisync);
2482     __ b(Lisync); // In case of NAN.
2483   }
2484 
2485   __ align(32, 28, 28); // Align load.
2486   // __ bind(Litos);
2487   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2488   assert(branch_table[itos] == 0, "can't compute twice");
2489   branch_table[itos] = __ pc(); // non-volatile_entry point
2490   __ lwax(R17_tos, Rclass_or_obj, Roffset);
2491   __ push(itos);
2492   if (!is_static && rc == may_rewrite) {
2493     patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2494   }
2495   __ beq(CCR6, Lacquire); // Volatile?
2496   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2497 
2498   __ align(32, 28, 28); // Align load.
2499   // __ bind(Lltos);
2500   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2501   assert(branch_table[ltos] == 0, "can't compute twice");
2502   branch_table[ltos] = __ pc(); // non-volatile_entry point
2503   __ ldx(R17_tos, Rclass_or_obj, Roffset);
2504   __ push(ltos);
2505   if (!is_static && rc == may_rewrite) {
2506     patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2507   }
2508   __ beq(CCR6, Lacquire); // Volatile?
2509   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2510 
2511   __ align(32, 28, 28); // Align load.
2512   // __ bind(Lbtos);
2513   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2514   assert(branch_table[btos] == 0, "can't compute twice");
2515   branch_table[btos] = __ pc(); // non-volatile_entry point
2516   __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2517   __ extsb(R17_tos, R17_tos);
2518   __ push(btos);
2519   if (!is_static && rc == may_rewrite) {
2520     patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2521   }
2522   __ beq(CCR6, Lacquire); // Volatile?
2523   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2524 
2525   __ align(32, 28, 28); // Align load.
2526   // __ bind(Lctos);
2527   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2528   assert(branch_table[ctos] == 0, "can't compute twice");
2529   branch_table[ctos] = __ pc(); // non-volatile_entry point
2530   __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2531   __ push(ctos);
2532   if (!is_static && rc == may_rewrite) {
2533     patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2534   }
2535   __ beq(CCR6, Lacquire); // Volatile?
2536   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2537 
2538   __ align(32, 28, 28); // Align load.
2539   // __ bind(Lstos);
2540   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2541   assert(branch_table[stos] == 0, "can't compute twice");
2542   branch_table[stos] = __ pc(); // non-volatile_entry point
2543   __ lhax(R17_tos, Rclass_or_obj, Roffset);
2544   __ push(stos);
2545   if (!is_static && rc == may_rewrite) {
2546     patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2547   }
2548   __ beq(CCR6, Lacquire); // Volatile?
2549   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2550 
2551   __ align(32, 28, 28); // Align load.
2552   // __ bind(Latos);
2553   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2554   assert(branch_table[atos] == 0, "can't compute twice");
2555   branch_table[atos] = __ pc(); // non-volatile_entry point
2556   __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2557   __ verify_oop(R17_tos);
2558   __ push(atos);
2559   //__ dcbt(R17_tos); // prefetch
2560   if (!is_static && rc == may_rewrite) {
2561     patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2562   }
2563   __ beq(CCR6, Lacquire); // Volatile?
2564   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2565 
2566   __ align(32, 12);
2567   __ bind(Lacquire);
2568   __ twi_0(R17_tos);
2569   __ bind(Lisync);
2570   __ isync(); // acquire
2571 
2572 #ifdef ASSERT
2573   for (int i = 0; i<number_of_states; ++i) {
2574     assert(branch_table[i], "get initialization");
2575     //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2576     //              is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2577   }
2578 #endif
2579 }
2580 
2581 void TemplateTable::getfield(int byte_no) {
2582   getfield_or_static(byte_no, false);
2583 }
2584 
2585 void TemplateTable::nofast_getfield(int byte_no) {
2586   getfield_or_static(byte_no, false, may_not_rewrite);
2587 }
2588 
2589 void TemplateTable::getstatic(int byte_no) {
2590   getfield_or_static(byte_no, true);
2591 }
2592 
2593 // The registers cache and index expected to be set before call.
2594 // The function may destroy various registers, just not the cache and index registers.
2595 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2596 
2597   assert_different_registers(Rcache, Rscratch, R6_ARG4);
2598 
2599   if (JvmtiExport::can_post_field_modification()) {
2600     Label Lno_field_mod_post;
2601 
2602     // Check if post field access in enabled.
2603     int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2604     __ lwz(Rscratch, offs, Rscratch);
2605 
2606     __ cmpwi(CCR0, Rscratch, 0);
2607     __ beq(CCR0, Lno_field_mod_post);
2608 
2609     // Do the post
2610     ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2611     const Register Robj = Rscratch;
2612 
2613     __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2614     if (is_static) {
2615       // Life is simple. Null out the object pointer.
2616       __ li(Robj, 0);
2617     } else {
2618       // In case of the fast versions, value lives in registers => put it back on tos.
2619       int offs = Interpreter::expr_offset_in_bytes(0);
2620       Register base = R15_esp;
2621       switch(bytecode()) {
2622         case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2623         case Bytecodes::_fast_iputfield: // Fall through
2624         case Bytecodes::_fast_bputfield: // Fall through
2625         case Bytecodes::_fast_cputfield: // Fall through
2626         case Bytecodes::_fast_sputfield: __ push_i(); offs+=  Interpreter::stackElementSize; break;
2627         case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2628         case Bytecodes::_fast_fputfield: __ push_f(); offs+=  Interpreter::stackElementSize; break;
2629         case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2630         default: {
2631           offs = 0;
2632           base = Robj;
2633           const Register Rflags = Robj;
2634           Label is_one_slot;
2635           // Life is harder. The stack holds the value on top, followed by the
2636           // object. We don't know the size of the value, though; it could be
2637           // one or two words depending on its type. As a result, we must find
2638           // the type to determine where the object is.
2639           __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2640           __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2641 
2642           __ cmpwi(CCR0, Rflags, ltos);
2643           __ cmpwi(CCR1, Rflags, dtos);
2644           __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2645           __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2646           __ beq(CCR0, is_one_slot);
2647           __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2648           __ bind(is_one_slot);
2649           break;
2650         }
2651       }
2652       __ ld(Robj, offs, base);
2653       __ verify_oop(Robj);
2654     }
2655 
2656     __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2657     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2658     __ get_cache_and_index_at_bcp(Rcache, 1);
2659 
2660     // In case of the fast versions, value lives in registers => put it back on tos.
2661     switch(bytecode()) {
2662       case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2663       case Bytecodes::_fast_iputfield: // Fall through
2664       case Bytecodes::_fast_bputfield: // Fall through
2665       case Bytecodes::_fast_cputfield: // Fall through
2666       case Bytecodes::_fast_sputfield: __ pop_i(); break;
2667       case Bytecodes::_fast_lputfield: __ pop_l(); break;
2668       case Bytecodes::_fast_fputfield: __ pop_f(); break;
2669       case Bytecodes::_fast_dputfield: __ pop_d(); break;
2670       default: break; // Nothin' to do.
2671     }
2672 
2673     __ align(32, 12);
2674     __ bind(Lno_field_mod_post);
2675   }
2676 }
2677 
2678 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2679 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2680   Label Lvolatile;
2681 
2682   const Register Rcache        = R5_ARG3,  // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2683                  Rclass_or_obj = R31,      // Needs to survive C call.
2684                  Roffset       = R22_tmp2, // Needs to survive C call.
2685                  Rflags        = R3_ARG1,
2686                  Rbtable       = R4_ARG2,
2687                  Rscratch      = R11_scratch1,
2688                  Rscratch2     = R12_scratch2,
2689                  Rscratch3     = R6_ARG4,
2690                  Rbc           = Rscratch3;
2691   const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2692 
2693   static address field_rw_branch_table[number_of_states],
2694                  field_norw_branch_table[number_of_states],
2695                  static_branch_table[number_of_states];
2696 
2697   address* branch_table = is_static ? static_branch_table :
2698     (rc == may_rewrite ? field_rw_branch_table : field_norw_branch_table);
2699 
2700   // Stack (grows up):
2701   //  value
2702   //  obj
2703 
2704   // Load the field offset.
2705   resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2706   jvmti_post_field_mod(Rcache, Rscratch, is_static);
2707   load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2708 
2709   // Load pointer to branch table.
2710   __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2711 
2712   // Get volatile flag.
2713   __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2714 
2715   // Check the field type.
2716   __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2717 
2718 #ifdef ASSERT
2719   Label LFlagInvalid;
2720   __ cmpldi(CCR0, Rflags, number_of_states);
2721   __ bge(CCR0, LFlagInvalid);
2722 #endif
2723 
2724   // Load from branch table and dispatch (volatile case: one instruction ahead).
2725   __ sldi(Rflags, Rflags, LogBytesPerWord);
2726   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2727     __ cmpwi(CR_is_vol, Rscratch, 1);  // Volatile?
2728   }
2729   __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2730   __ ldx(Rbtable, Rbtable, Rflags);
2731 
2732   __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2733   __ mtctr(Rbtable);
2734   __ bctr();
2735 
2736 #ifdef ASSERT
2737   __ bind(LFlagInvalid);
2738   __ stop("got invalid flag", 0x656);
2739 
2740   // __ bind(Lvtos);
2741   address pc_before_release = __ pc();
2742   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2743   assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2744   assert(branch_table[vtos] == 0, "can't compute twice");
2745   branch_table[vtos] = __ pc(); // non-volatile_entry point
2746   __ stop("vtos unexpected", 0x657);
2747 #endif
2748 
2749   __ align(32, 28, 28); // Align pop.
2750   // __ bind(Ldtos);
2751   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2752   assert(branch_table[dtos] == 0, "can't compute twice");
2753   branch_table[dtos] = __ pc(); // non-volatile_entry point
2754   __ pop(dtos);
2755   if (!is_static) {
2756     pop_and_check_object(Rclass_or_obj);  // Kills R11_scratch1.
2757   }
2758   __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2759   if (!is_static && rc == may_rewrite) {
2760     patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no);
2761   }
2762   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2763     __ beq(CR_is_vol, Lvolatile); // Volatile?
2764   }
2765   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2766 
2767   __ align(32, 28, 28); // Align pop.
2768   // __ bind(Lftos);
2769   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2770   assert(branch_table[ftos] == 0, "can't compute twice");
2771   branch_table[ftos] = __ pc(); // non-volatile_entry point
2772   __ pop(ftos);
2773   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2774   __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2775   if (!is_static && rc == may_rewrite) {
2776     patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no);
2777   }
2778   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2779     __ beq(CR_is_vol, Lvolatile); // Volatile?
2780   }
2781   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2782 
2783   __ align(32, 28, 28); // Align pop.
2784   // __ bind(Litos);
2785   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2786   assert(branch_table[itos] == 0, "can't compute twice");
2787   branch_table[itos] = __ pc(); // non-volatile_entry point
2788   __ pop(itos);
2789   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2790   __ stwx(R17_tos, Rclass_or_obj, Roffset);
2791   if (!is_static && rc == may_rewrite) {
2792     patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no);
2793   }
2794   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2795     __ beq(CR_is_vol, Lvolatile); // Volatile?
2796   }
2797   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2798 
2799   __ align(32, 28, 28); // Align pop.
2800   // __ bind(Lltos);
2801   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2802   assert(branch_table[ltos] == 0, "can't compute twice");
2803   branch_table[ltos] = __ pc(); // non-volatile_entry point
2804   __ pop(ltos);
2805   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2806   __ stdx(R17_tos, Rclass_or_obj, Roffset);
2807   if (!is_static && rc == may_rewrite) {
2808     patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no);
2809   }
2810   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2811     __ beq(CR_is_vol, Lvolatile); // Volatile?
2812   }
2813   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2814 
2815   __ align(32, 28, 28); // Align pop.
2816   // __ bind(Lbtos);
2817   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2818   assert(branch_table[btos] == 0, "can't compute twice");
2819   branch_table[btos] = __ pc(); // non-volatile_entry point
2820   __ pop(btos);
2821   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2822   __ stbx(R17_tos, Rclass_or_obj, Roffset);
2823   if (!is_static && rc == may_rewrite) {
2824     patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no);
2825   }
2826   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2827     __ beq(CR_is_vol, Lvolatile); // Volatile?
2828   }
2829   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2830 
2831   __ align(32, 28, 28); // Align pop.
2832   // __ bind(Lctos);
2833   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2834   assert(branch_table[ctos] == 0, "can't compute twice");
2835   branch_table[ctos] = __ pc(); // non-volatile_entry point
2836   __ pop(ctos);
2837   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2838   __ sthx(R17_tos, Rclass_or_obj, Roffset);
2839   if (!is_static && rc == may_rewrite) {
2840     patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no);
2841   }
2842   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2843     __ beq(CR_is_vol, Lvolatile); // Volatile?
2844   }
2845   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2846 
2847   __ align(32, 28, 28); // Align pop.
2848   // __ bind(Lstos);
2849   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2850   assert(branch_table[stos] == 0, "can't compute twice");
2851   branch_table[stos] = __ pc(); // non-volatile_entry point
2852   __ pop(stos);
2853   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2854   __ sthx(R17_tos, Rclass_or_obj, Roffset);
2855   if (!is_static && rc == may_rewrite) {
2856     patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no);
2857   }
2858   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2859     __ beq(CR_is_vol, Lvolatile); // Volatile?
2860   }
2861   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2862 
2863   __ align(32, 28, 28); // Align pop.
2864   // __ bind(Latos);
2865   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2866   assert(branch_table[atos] == 0, "can't compute twice");
2867   branch_table[atos] = __ pc(); // non-volatile_entry point
2868   __ pop(atos);
2869   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2870   do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2871   if (!is_static && rc == may_rewrite) {
2872     patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no);
2873   }
2874   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2875     __ beq(CR_is_vol, Lvolatile); // Volatile?
2876     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2877 
2878     __ align(32, 12);
2879     __ bind(Lvolatile);
2880     __ fence();
2881   }
2882   // fallthru: __ b(Lexit);
2883 
2884 #ifdef ASSERT
2885   for (int i = 0; i<number_of_states; ++i) {
2886     assert(branch_table[i], "put initialization");
2887     //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2888     //              is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2889   }
2890 #endif
2891 }
2892 
2893 void TemplateTable::putfield(int byte_no) {
2894   putfield_or_static(byte_no, false);
2895 }
2896 
2897 void TemplateTable::nofast_putfield(int byte_no) {
2898   putfield_or_static(byte_no, false, may_not_rewrite);
2899 }
2900 
2901 void TemplateTable::putstatic(int byte_no) {
2902   putfield_or_static(byte_no, true);
2903 }
2904 
2905 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2906 void TemplateTable::jvmti_post_fast_field_mod() {
2907   __ should_not_reach_here();
2908 }
2909 
2910 void TemplateTable::fast_storefield(TosState state) {
2911   transition(state, vtos);
2912 
2913   const Register Rcache        = R5_ARG3,  // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2914                  Rclass_or_obj = R31,      // Needs to survive C call.
2915                  Roffset       = R22_tmp2, // Needs to survive C call.
2916                  Rflags        = R3_ARG1,
2917                  Rscratch      = R11_scratch1,
2918                  Rscratch2     = R12_scratch2,
2919                  Rscratch3     = R4_ARG2;
2920   const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2921 
2922   // Constant pool already resolved => Load flags and offset of field.
2923   __ get_cache_and_index_at_bcp(Rcache, 1);
2924   jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2925   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2926 
2927   // Get the obj and the final store addr.
2928   pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2929 
2930   // Get volatile flag.
2931   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2932   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2933   {
2934     Label LnotVolatile;
2935     __ beq(CCR0, LnotVolatile);
2936     __ release();
2937     __ align(32, 12);
2938     __ bind(LnotVolatile);
2939   }
2940 
2941   // Do the store and fencing.
2942   switch(bytecode()) {
2943     case Bytecodes::_fast_aputfield:
2944       // Store into the field.
2945       do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2946       break;
2947 
2948     case Bytecodes::_fast_iputfield:
2949       __ stwx(R17_tos, Rclass_or_obj, Roffset);
2950       break;
2951 
2952     case Bytecodes::_fast_lputfield:
2953       __ stdx(R17_tos, Rclass_or_obj, Roffset);
2954       break;
2955 
2956     case Bytecodes::_fast_bputfield:
2957       __ stbx(R17_tos, Rclass_or_obj, Roffset);
2958       break;
2959 
2960     case Bytecodes::_fast_cputfield:
2961     case Bytecodes::_fast_sputfield:
2962       __ sthx(R17_tos, Rclass_or_obj, Roffset);
2963       break;
2964 
2965     case Bytecodes::_fast_fputfield:
2966       __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2967       break;
2968 
2969     case Bytecodes::_fast_dputfield:
2970       __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2971       break;
2972 
2973     default: ShouldNotReachHere();
2974   }
2975 
2976   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2977     Label LVolatile;
2978     __ beq(CR_is_vol, LVolatile);
2979     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2980 
2981     __ align(32, 12);
2982     __ bind(LVolatile);
2983     __ fence();
2984   }
2985 }
2986 
2987 void TemplateTable::fast_accessfield(TosState state) {
2988   transition(atos, state);
2989 
2990   Label LisVolatile;
2991   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2992 
2993   const Register Rcache        = R3_ARG1,
2994                  Rclass_or_obj = R17_tos,
2995                  Roffset       = R22_tmp2,
2996                  Rflags        = R23_tmp3,
2997                  Rscratch      = R12_scratch2;
2998 
2999   // Constant pool already resolved. Get the field offset.
3000   __ get_cache_and_index_at_bcp(Rcache, 1);
3001   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3002 
3003   // JVMTI support
3004   jvmti_post_field_access(Rcache, Rscratch, false, true);
3005 
3006   // Get the load address.
3007   __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3008 
3009   // Get volatile flag.
3010   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3011   __ bne(CCR0, LisVolatile);
3012 
3013   switch(bytecode()) {
3014     case Bytecodes::_fast_agetfield:
3015     {
3016       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3017       __ verify_oop(R17_tos);
3018       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3019 
3020       __ bind(LisVolatile);
3021       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3022       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3023       __ verify_oop(R17_tos);
3024       __ twi_0(R17_tos);
3025       __ isync();
3026       break;
3027     }
3028     case Bytecodes::_fast_igetfield:
3029     {
3030       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3031       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3032 
3033       __ bind(LisVolatile);
3034       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3035       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3036       __ twi_0(R17_tos);
3037       __ isync();
3038       break;
3039     }
3040     case Bytecodes::_fast_lgetfield:
3041     {
3042       __ ldx(R17_tos, Rclass_or_obj, Roffset);
3043       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3044 
3045       __ bind(LisVolatile);
3046       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3047       __ ldx(R17_tos, Rclass_or_obj, Roffset);
3048       __ twi_0(R17_tos);
3049       __ isync();
3050       break;
3051     }
3052     case Bytecodes::_fast_bgetfield:
3053     {
3054       __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3055       __ extsb(R17_tos, R17_tos);
3056       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3057 
3058       __ bind(LisVolatile);
3059       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3060       __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3061       __ twi_0(R17_tos);
3062       __ extsb(R17_tos, R17_tos);
3063       __ isync();
3064       break;
3065     }
3066     case Bytecodes::_fast_cgetfield:
3067     {
3068       __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3069       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3070 
3071       __ bind(LisVolatile);
3072       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3073       __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3074       __ twi_0(R17_tos);
3075       __ isync();
3076       break;
3077     }
3078     case Bytecodes::_fast_sgetfield:
3079     {
3080       __ lhax(R17_tos, Rclass_or_obj, Roffset);
3081       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3082 
3083       __ bind(LisVolatile);
3084       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3085       __ lhax(R17_tos, Rclass_or_obj, Roffset);
3086       __ twi_0(R17_tos);
3087       __ isync();
3088       break;
3089     }
3090     case Bytecodes::_fast_fgetfield:
3091     {
3092       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3093       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3094 
3095       __ bind(LisVolatile);
3096       Label Ldummy;
3097       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3098       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3099       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3100       __ bne_predict_not_taken(CCR0, Ldummy);
3101       __ bind(Ldummy);
3102       __ isync();
3103       break;
3104     }
3105     case Bytecodes::_fast_dgetfield:
3106     {
3107       __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3108       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3109 
3110       __ bind(LisVolatile);
3111       Label Ldummy;
3112       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3113       __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3114       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3115       __ bne_predict_not_taken(CCR0, Ldummy);
3116       __ bind(Ldummy);
3117       __ isync();
3118       break;
3119     }
3120     default: ShouldNotReachHere();
3121   }
3122 }
3123 
3124 void TemplateTable::fast_xaccess(TosState state) {
3125   transition(vtos, state);
3126 
3127   Label LisVolatile;
3128   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3129   const Register Rcache        = R3_ARG1,
3130                  Rclass_or_obj = R17_tos,
3131                  Roffset       = R22_tmp2,
3132                  Rflags        = R23_tmp3,
3133                  Rscratch      = R12_scratch2;
3134 
3135   __ ld(Rclass_or_obj, 0, R18_locals);
3136 
3137   // Constant pool already resolved. Get the field offset.
3138   __ get_cache_and_index_at_bcp(Rcache, 2);
3139   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3140 
3141   // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3142 
3143   // Needed to report exception at the correct bcp.
3144   __ addi(R14_bcp, R14_bcp, 1);
3145 
3146   // Get the load address.
3147   __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3148 
3149   // Get volatile flag.
3150   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3151   __ bne(CCR0, LisVolatile);
3152 
3153   switch(state) {
3154   case atos:
3155     {
3156       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3157       __ verify_oop(R17_tos);
3158       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3159 
3160       __ bind(LisVolatile);
3161       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3162       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3163       __ verify_oop(R17_tos);
3164       __ twi_0(R17_tos);
3165       __ isync();
3166       break;
3167     }
3168   case itos:
3169     {
3170       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3171       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3172 
3173       __ bind(LisVolatile);
3174       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3175       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3176       __ twi_0(R17_tos);
3177       __ isync();
3178       break;
3179     }
3180   case ftos:
3181     {
3182       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3183       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3184 
3185       __ bind(LisVolatile);
3186       Label Ldummy;
3187       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3188       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3189       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3190       __ bne_predict_not_taken(CCR0, Ldummy);
3191       __ bind(Ldummy);
3192       __ isync();
3193       break;
3194     }
3195   default: ShouldNotReachHere();
3196   }
3197   __ addi(R14_bcp, R14_bcp, -1);
3198 }
3199 
3200 // ============================================================================
3201 // Calls
3202 
3203 // Common code for invoke
3204 //
3205 // Input:
3206 //   - byte_no
3207 //
3208 // Output:
3209 //   - Rmethod:        The method to invoke next.
3210 //   - Rret_addr:      The return address to return to.
3211 //   - Rindex:         MethodType (invokehandle) or CallSite obj (invokedynamic)
3212 //   - Rrecv:          Cache for "this" pointer, might be noreg if static call.
3213 //   - Rflags:         Method flags from const pool cache.
3214 //
3215 //  Kills:
3216 //   - Rscratch1
3217 //
3218 void TemplateTable::prepare_invoke(int byte_no,
3219                                    Register Rmethod,  // linked method (or i-klass)
3220                                    Register Rret_addr,// return address
3221                                    Register Rindex,   // itable index, MethodType, etc.
3222                                    Register Rrecv,    // If caller wants to see it.
3223                                    Register Rflags,   // If caller wants to test it.
3224                                    Register Rscratch
3225                                    ) {
3226   // Determine flags.
3227   const Bytecodes::Code code = bytecode();
3228   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3229   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3230   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3231   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3232   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3233   const bool load_receiver       = (Rrecv != noreg);
3234   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3235 
3236   assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3237   assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3238   assert_different_registers(Rret_addr, Rscratch);
3239 
3240   load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3241 
3242   // Saving of SP done in call_from_interpreter.
3243 
3244   // Maybe push "appendix" to arguments.
3245   if (is_invokedynamic || is_invokehandle) {
3246     Label Ldone;
3247     __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3248     __ beq(CCR0, Ldone);
3249     // Push "appendix" (MethodType, CallSite, etc.).
3250     // This must be done before we get the receiver,
3251     // since the parameter_size includes it.
3252     __ load_resolved_reference_at_index(Rscratch, Rindex);
3253     __ verify_oop(Rscratch);
3254     __ push_ptr(Rscratch);
3255     __ bind(Ldone);
3256   }
3257 
3258   // Load receiver if needed (after appendix is pushed so parameter size is correct).
3259   if (load_receiver) {
3260     const Register Rparam_count = Rscratch;
3261     __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3262     __ load_receiver(Rparam_count, Rrecv);
3263     __ verify_oop(Rrecv);
3264   }
3265 
3266   // Get return address.
3267   {
3268     Register Rtable_addr = Rscratch;
3269     Register Rret_type = Rret_addr;
3270     address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3271 
3272     // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3273     __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3274     __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3275     __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3276     // Get return address.
3277     __ ldx(Rret_addr, Rtable_addr, Rret_type);
3278   }
3279 }
3280 
3281 // Helper for virtual calls. Load target out of vtable and jump off!
3282 // Kills all passed registers.
3283 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3284 
3285   assert_different_registers(Rrecv_klass, Rtemp, Rret);
3286   const Register Rtarget_method = Rindex;
3287 
3288   // Get target method & entry point.
3289   const int base = InstanceKlass::vtable_start_offset() * wordSize;
3290   // Calc vtable addr scale the vtable index by 8.
3291   __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size() * wordSize));
3292   // Load target.
3293   __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3294   __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3295   // Argument and return type profiling.
3296   __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3297   __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3298 }
3299 
3300 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3301 void TemplateTable::invokevirtual(int byte_no) {
3302   transition(vtos, vtos);
3303 
3304   Register Rtable_addr = R11_scratch1,
3305            Rret_type = R12_scratch2,
3306            Rret_addr = R5_ARG3,
3307            Rflags = R22_tmp2, // Should survive C call.
3308            Rrecv = R3_ARG1,
3309            Rrecv_klass = Rrecv,
3310            Rvtableindex_or_method = R31, // Should survive C call.
3311            Rnum_params = R4_ARG2,
3312            Rnew_bc = R6_ARG4;
3313 
3314   Label LnotFinal;
3315 
3316   load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3317 
3318   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3319   __ bfalse(CCR0, LnotFinal);
3320 
3321   if (RewriteBytecodes && !UseSharedSpaces) {
3322     patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3323   }
3324   invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3325 
3326   __ align(32, 12);
3327   __ bind(LnotFinal);
3328   // Load "this" pointer (receiver).
3329   __ rldicl(Rnum_params, Rflags, 64, 48);
3330   __ load_receiver(Rnum_params, Rrecv);
3331   __ verify_oop(Rrecv);
3332 
3333   // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3334   __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3335   __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3336   __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3337   __ ldx(Rret_addr, Rret_type, Rtable_addr);
3338   __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3339   __ load_klass(Rrecv_klass, Rrecv);
3340   __ verify_klass_ptr(Rrecv_klass);
3341   __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3342 
3343   generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3344 }
3345 
3346 void TemplateTable::fast_invokevfinal(int byte_no) {
3347   transition(vtos, vtos);
3348 
3349   assert(byte_no == f2_byte, "use this argument");
3350   Register Rflags  = R22_tmp2,
3351            Rmethod = R31;
3352   load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3353   invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3354 }
3355 
3356 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3357 
3358   assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3359 
3360   // Load receiver from stack slot.
3361   Register Rrecv = Rscratch2;
3362   Register Rnum_params = Rrecv;
3363 
3364   __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3365   __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3366 
3367   // Get return address.
3368   Register Rtable_addr = Rscratch1,
3369            Rret_addr   = Rflags,
3370            Rret_type   = Rret_addr;
3371   // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3372   __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3373   __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3374   __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3375   __ ldx(Rret_addr, Rret_type, Rtable_addr);
3376 
3377   // Load receiver and receiver NULL check.
3378   __ load_receiver(Rnum_params, Rrecv);
3379   __ null_check_throw(Rrecv, -1, Rscratch1);
3380 
3381   __ profile_final_call(Rrecv, Rscratch1);
3382   // Argument and return type profiling.
3383   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3384 
3385   // Do the call.
3386   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3387 }
3388 
3389 void TemplateTable::invokespecial(int byte_no) {
3390   assert(byte_no == f1_byte, "use this argument");
3391   transition(vtos, vtos);
3392 
3393   Register Rtable_addr = R3_ARG1,
3394            Rret_addr   = R4_ARG2,
3395            Rflags      = R5_ARG3,
3396            Rreceiver   = R6_ARG4,
3397            Rmethod     = R31;
3398 
3399   prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3400 
3401   // Receiver NULL check.
3402   __ null_check_throw(Rreceiver, -1, R11_scratch1);
3403 
3404   __ profile_call(R11_scratch1, R12_scratch2);
3405   // Argument and return type profiling.
3406   __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3407   __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3408 }
3409 
3410 void TemplateTable::invokestatic(int byte_no) {
3411   assert(byte_no == f1_byte, "use this argument");
3412   transition(vtos, vtos);
3413 
3414   Register Rtable_addr = R3_ARG1,
3415            Rret_addr   = R4_ARG2,
3416            Rflags      = R5_ARG3;
3417 
3418   prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3419 
3420   __ profile_call(R11_scratch1, R12_scratch2);
3421   // Argument and return type profiling.
3422   __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3423   __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3424 }
3425 
3426 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3427                                                   Register Rret,
3428                                                   Register Rflags,
3429                                                   Register Rindex,
3430                                                   Register Rtemp1,
3431                                                   Register Rtemp2) {
3432 
3433   assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3434   Label LnotFinal;
3435 
3436   // Check for vfinal.
3437   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3438   __ bfalse(CCR0, LnotFinal);
3439 
3440   Register Rscratch = Rflags; // Rflags is dead now.
3441 
3442   // Final call case.
3443   __ profile_final_call(Rtemp1, Rscratch);
3444   // Argument and return type profiling.
3445   __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
3446   // Do the final call - the index (f2) contains the method.
3447   __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3448 
3449   // Non-final callc case.
3450   __ bind(LnotFinal);
3451   __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3452   generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3453 }
3454 
3455 void TemplateTable::invokeinterface(int byte_no) {
3456   assert(byte_no == f1_byte, "use this argument");
3457   transition(vtos, vtos);
3458 
3459   const Register Rscratch1        = R11_scratch1,
3460                  Rscratch2        = R12_scratch2,
3461                  Rscratch3        = R9_ARG7,
3462                  Rscratch4        = R10_ARG8,
3463                  Rtable_addr      = Rscratch2,
3464                  Rinterface_klass = R5_ARG3,
3465                  Rret_type        = R8_ARG6,
3466                  Rret_addr        = Rret_type,
3467                  Rindex           = R6_ARG4,
3468                  Rreceiver        = R4_ARG2,
3469                  Rrecv_klass      = Rreceiver,
3470                  Rflags           = R7_ARG5;
3471 
3472   prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3473 
3474   // Get receiver klass.
3475   __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3476   __ load_klass(Rrecv_klass, Rreceiver);
3477 
3478   // Check corner case object method.
3479   Label LobjectMethod;
3480 
3481   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3482   __ btrue(CCR0, LobjectMethod);
3483 
3484   // Fallthrough: The normal invokeinterface case.
3485   __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3486 
3487   // Find entry point to call.
3488   Label Lthrow_icc, Lthrow_ame;
3489   // Result will be returned in Rindex.
3490   __ mr(Rscratch4, Rrecv_klass);
3491   __ mr(Rscratch3, Rindex);
3492   __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3493 
3494   __ cmpdi(CCR0, Rindex, 0);
3495   __ beq(CCR0, Lthrow_ame);
3496   // Found entry. Jump off!
3497   // Argument and return type profiling.
3498   __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
3499   __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3500 
3501   // Vtable entry was NULL => Throw abstract method error.
3502   __ bind(Lthrow_ame);
3503   __ mr(Rrecv_klass, Rscratch4);
3504   __ mr(Rindex, Rscratch3);
3505   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3506 
3507   // Interface was not found => Throw incompatible class change error.
3508   __ bind(Lthrow_icc);
3509   __ mr(Rrecv_klass, Rscratch4);
3510   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3511 
3512   __ should_not_reach_here();
3513 
3514   // Special case of invokeinterface called for virtual method of
3515   // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3516   // The invokeinterface was rewritten to a invokevirtual, hence we have
3517   // to handle this corner case. This code isn't produced by javac, but could
3518   // be produced by another compliant java compiler.
3519   __ bind(LobjectMethod);
3520   invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3521 }
3522 
3523 void TemplateTable::invokedynamic(int byte_no) {
3524   transition(vtos, vtos);
3525 
3526   const Register Rret_addr = R3_ARG1,
3527                  Rflags    = R4_ARG2,
3528                  Rmethod   = R22_tmp2,
3529                  Rscratch1 = R11_scratch1,
3530                  Rscratch2 = R12_scratch2;
3531 
3532   prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3533 
3534   // Profile this call.
3535   __ profile_call(Rscratch1, Rscratch2);
3536 
3537   // Off we go. With the new method handles, we don't jump to a method handle
3538   // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3539   // to be the callsite object the bootstrap method returned. This is passed to a
3540   // "link" method which does the dispatch (Most likely just grabs the MH stored
3541   // inside the callsite and does an invokehandle).
3542   // Argument and return type profiling.
3543   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3544   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3545 }
3546 
3547 void TemplateTable::invokehandle(int byte_no) {
3548   transition(vtos, vtos);
3549 
3550   const Register Rret_addr = R3_ARG1,
3551                  Rflags    = R4_ARG2,
3552                  Rrecv     = R5_ARG3,
3553                  Rmethod   = R22_tmp2,
3554                  Rscratch1 = R11_scratch1,
3555                  Rscratch2 = R12_scratch2;
3556 
3557   prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3558   __ verify_method_ptr(Rmethod);
3559   __ null_check_throw(Rrecv, -1, Rscratch2);
3560 
3561   __ profile_final_call(Rrecv, Rscratch1);
3562 
3563   // Still no call from handle => We call the method handle interpreter here.
3564   // Argument and return type profiling.
3565   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3566   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3567 }
3568 
3569 // =============================================================================
3570 // Allocation
3571 
3572 // Puts allocated obj ref onto the expression stack.
3573 void TemplateTable::_new() {
3574   transition(vtos, atos);
3575 
3576   Label Lslow_case,
3577         Ldone,
3578         Linitialize_header,
3579         Lallocate_shared,
3580         Linitialize_object;  // Including clearing the fields.
3581 
3582   const Register RallocatedObject = R17_tos,
3583                  RinstanceKlass   = R9_ARG7,
3584                  Rscratch         = R11_scratch1,
3585                  Roffset          = R8_ARG6,
3586                  Rinstance_size   = Roffset,
3587                  Rcpool           = R4_ARG2,
3588                  Rtags            = R3_ARG1,
3589                  Rindex           = R5_ARG3;
3590 
3591   const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc();
3592 
3593   // --------------------------------------------------------------------------
3594   // Check if fast case is possible.
3595 
3596   // Load pointers to const pool and const pool's tags array.
3597   __ get_cpool_and_tags(Rcpool, Rtags);
3598   // Load index of constant pool entry.
3599   __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3600 
3601   if (UseTLAB) {
3602     // Make sure the class we're about to instantiate has been resolved
3603     // This is done before loading instanceKlass to be consistent with the order
3604     // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3605     __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3606     __ lbzx(Rtags, Rindex, Rtags);
3607 
3608     __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3609     __ bne(CCR0, Lslow_case);
3610 
3611     // Get instanceKlass (load from Rcpool + sizeof(ConstantPool) + Rindex*BytesPerWord).
3612     __ sldi(Roffset, Rindex, LogBytesPerWord);
3613     __ addi(Rscratch, Rcpool, sizeof(ConstantPool));
3614     __ isync(); // Order load of instance Klass wrt. tags.
3615     __ ldx(RinstanceKlass, Roffset, Rscratch);
3616 
3617     // Make sure klass is fully initialized and get instance_size.
3618     __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3619     __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3620 
3621     __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3622     // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3623     __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3624 
3625     __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3626     __ beq(CCR0, Lslow_case);
3627 
3628     // --------------------------------------------------------------------------
3629     // Fast case:
3630     // Allocate the instance.
3631     // 1) Try to allocate in the TLAB.
3632     // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3633     // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3634 
3635     Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3636     Register RnewTopValue = R6_ARG4;
3637     Register RendValue    = R7_ARG5;
3638 
3639     // Check if we can allocate in the TLAB.
3640     __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3641     __ ld(RendValue,    in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3642 
3643     __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3644 
3645     // If there is enough space, we do not CAS and do not clear.
3646     __ cmpld(CCR0, RnewTopValue, RendValue);
3647     __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3648 
3649     __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3650 
3651     if (ZeroTLAB) {
3652       // The fields have already been cleared.
3653       __ b(Linitialize_header);
3654     } else {
3655       // Initialize both the header and fields.
3656       __ b(Linitialize_object);
3657     }
3658 
3659     // Fall through: TLAB was too small.
3660     if (allow_shared_alloc) {
3661       Register RtlabWasteLimitValue = R10_ARG8;
3662       Register RfreeValue = RnewTopValue;
3663 
3664       __ bind(Lallocate_shared);
3665       // Check if tlab should be discarded (refill_waste_limit >= free).
3666       __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3667       __ subf(RfreeValue, RoldTopValue, RendValue);
3668       __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3669       __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3670       __ bge(CCR0, Lslow_case);
3671 
3672       // Increment waste limit to prevent getting stuck on this slow path.
3673       __ addi(RtlabWasteLimitValue, RtlabWasteLimitValue, (int)ThreadLocalAllocBuffer::refill_waste_limit_increment());
3674       __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3675     }
3676     // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3677   }
3678   // else: Always go the slow path.
3679 
3680   // --------------------------------------------------------------------------
3681   // slow case
3682   __ bind(Lslow_case);
3683   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3684 
3685   if (UseTLAB) {
3686     __ b(Ldone);
3687     // --------------------------------------------------------------------------
3688     // Init1: Zero out newly allocated memory.
3689 
3690     if (!ZeroTLAB || allow_shared_alloc) {
3691       // Clear object fields.
3692       __ bind(Linitialize_object);
3693 
3694       // Initialize remaining object fields.
3695       Register Rbase = Rtags;
3696       __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3697       __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3698       __ srdi(Rinstance_size, Rinstance_size, 3);
3699 
3700       // Clear out object skipping header. Takes also care of the zero length case.
3701       __ clear_memory_doubleword(Rbase, Rinstance_size);
3702       // fallthru: __ b(Linitialize_header);
3703     }
3704 
3705     // --------------------------------------------------------------------------
3706     // Init2: Initialize the header: mark, klass
3707     __ bind(Linitialize_header);
3708 
3709     // Init mark.
3710     if (UseBiasedLocking) {
3711       __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3712     } else {
3713       __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3714     }
3715     __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3716 
3717     // Init klass.
3718     __ store_klass_gap(RallocatedObject);
3719     __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3720 
3721     // Check and trigger dtrace event.
3722     {
3723       SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3724       __ push(atos);
3725       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3726       __ pop(atos);
3727     }
3728   }
3729 
3730   // continue
3731   __ bind(Ldone);
3732 
3733   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3734   __ membar(Assembler::StoreStore);
3735 }
3736 
3737 void TemplateTable::newarray() {
3738   transition(itos, atos);
3739 
3740   __ lbz(R4, 1, R14_bcp);
3741   __ extsw(R5, R17_tos);
3742   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3743 
3744   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3745   __ membar(Assembler::StoreStore);
3746 }
3747 
3748 void TemplateTable::anewarray() {
3749   transition(itos, atos);
3750 
3751   __ get_constant_pool(R4);
3752   __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3753   __ extsw(R6, R17_tos); // size
3754   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3755 
3756   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3757   __ membar(Assembler::StoreStore);
3758 }
3759 
3760 // Allocate a multi dimensional array
3761 void TemplateTable::multianewarray() {
3762   transition(vtos, atos);
3763 
3764   Register Rptr = R31; // Needs to survive C call.
3765 
3766   // Put ndims * wordSize into frame temp slot
3767   __ lbz(Rptr, 3, R14_bcp);
3768   __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3769   // Esp points past last_dim, so set to R4 to first_dim address.
3770   __ add(R4, Rptr, R15_esp);
3771   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3772   // Pop all dimensions off the stack.
3773   __ add(R15_esp, Rptr, R15_esp);
3774 
3775   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3776   __ membar(Assembler::StoreStore);
3777 }
3778 
3779 void TemplateTable::arraylength() {
3780   transition(atos, itos);
3781 
3782   Label LnoException;
3783   __ verify_oop(R17_tos);
3784   __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3785   __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3786 }
3787 
3788 // ============================================================================
3789 // Typechecks
3790 
3791 void TemplateTable::checkcast() {
3792   transition(atos, atos);
3793 
3794   Label Ldone, Lis_null, Lquicked, Lresolved;
3795   Register Roffset         = R6_ARG4,
3796            RobjKlass       = R4_ARG2,
3797            RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3798            Rcpool          = R11_scratch1,
3799            Rtags           = R12_scratch2;
3800 
3801   // Null does not pass.
3802   __ cmpdi(CCR0, R17_tos, 0);
3803   __ beq(CCR0, Lis_null);
3804 
3805   // Get constant pool tag to find out if the bytecode has already been "quickened".
3806   __ get_cpool_and_tags(Rcpool, Rtags);
3807 
3808   __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3809 
3810   __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3811   __ lbzx(Rtags, Rtags, Roffset);
3812 
3813   __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3814   __ beq(CCR0, Lquicked);
3815 
3816   // Call into the VM to "quicken" instanceof.
3817   __ push_ptr();  // for GC
3818   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3819   __ get_vm_result_2(RspecifiedKlass);
3820   __ pop_ptr();   // Restore receiver.
3821   __ b(Lresolved);
3822 
3823   // Extract target class from constant pool.
3824   __ bind(Lquicked);
3825   __ sldi(Roffset, Roffset, LogBytesPerWord);
3826   __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3827   __ isync(); // Order load of specified Klass wrt. tags.
3828   __ ldx(RspecifiedKlass, Rcpool, Roffset);
3829 
3830   // Do the checkcast.
3831   __ bind(Lresolved);
3832   // Get value klass in RobjKlass.
3833   __ load_klass(RobjKlass, R17_tos);
3834   // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3835   __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3836 
3837   // Not a subtype; so must throw exception
3838   // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3839   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3840   __ mtctr(R11_scratch1);
3841   __ bctr();
3842 
3843   // Profile the null case.
3844   __ align(32, 12);
3845   __ bind(Lis_null);
3846   __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3847 
3848   __ align(32, 12);
3849   __ bind(Ldone);
3850 }
3851 
3852 // Output:
3853 //   - tos == 0: Obj was null or not an instance of class.
3854 //   - tos == 1: Obj was an instance of class.
3855 void TemplateTable::instanceof() {
3856   transition(atos, itos);
3857 
3858   Label Ldone, Lis_null, Lquicked, Lresolved;
3859   Register Roffset         = R6_ARG4,
3860            RobjKlass       = R4_ARG2,
3861            RspecifiedKlass = R5_ARG3,
3862            Rcpool          = R11_scratch1,
3863            Rtags           = R12_scratch2;
3864 
3865   // Null does not pass.
3866   __ cmpdi(CCR0, R17_tos, 0);
3867   __ beq(CCR0, Lis_null);
3868 
3869   // Get constant pool tag to find out if the bytecode has already been "quickened".
3870   __ get_cpool_and_tags(Rcpool, Rtags);
3871 
3872   __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3873 
3874   __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3875   __ lbzx(Rtags, Rtags, Roffset);
3876 
3877   __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3878   __ beq(CCR0, Lquicked);
3879 
3880   // Call into the VM to "quicken" instanceof.
3881   __ push_ptr();  // for GC
3882   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3883   __ get_vm_result_2(RspecifiedKlass);
3884   __ pop_ptr();   // Restore receiver.
3885   __ b(Lresolved);
3886 
3887   // Extract target class from constant pool.
3888   __ bind(Lquicked);
3889   __ sldi(Roffset, Roffset, LogBytesPerWord);
3890   __ addi(Rcpool, Rcpool, sizeof(ConstantPool));
3891   __ isync(); // Order load of specified Klass wrt. tags.
3892   __ ldx(RspecifiedKlass, Rcpool, Roffset);
3893 
3894   // Do the checkcast.
3895   __ bind(Lresolved);
3896   // Get value klass in RobjKlass.
3897   __ load_klass(RobjKlass, R17_tos);
3898   // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3899   __ li(R17_tos, 1);
3900   __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3901   __ li(R17_tos, 0);
3902 
3903   if (ProfileInterpreter) {
3904     __ b(Ldone);
3905   }
3906 
3907   // Profile the null case.
3908   __ align(32, 12);
3909   __ bind(Lis_null);
3910   __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3911 
3912   __ align(32, 12);
3913   __ bind(Ldone);
3914 }
3915 
3916 // =============================================================================
3917 // Breakpoints
3918 
3919 void TemplateTable::_breakpoint() {
3920   transition(vtos, vtos);
3921 
3922   // Get the unpatched byte code.
3923   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3924   __ mr(R31, R3_RET);
3925 
3926   // Post the breakpoint event.
3927   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3928 
3929   // Complete the execution of original bytecode.
3930   __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3931 }
3932 
3933 // =============================================================================
3934 // Exceptions
3935 
3936 void TemplateTable::athrow() {
3937   transition(atos, vtos);
3938 
3939   // Exception oop is in tos
3940   __ verify_oop(R17_tos);
3941 
3942   __ null_check_throw(R17_tos, -1, R11_scratch1);
3943 
3944   // Throw exception interpreter entry expects exception oop to be in R3.
3945   __ mr(R3_RET, R17_tos);
3946   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3947   __ mtctr(R11_scratch1);
3948   __ bctr();
3949 }
3950 
3951 // =============================================================================
3952 // Synchronization
3953 // Searches the basic object lock list on the stack for a free slot
3954 // and uses it to lock the obect in tos.
3955 //
3956 // Recursive locking is enabled by exiting the search if the same
3957 // object is already found in the list. Thus, a new basic lock obj lock
3958 // is allocated "higher up" in the stack and thus is found first
3959 // at next monitor exit.
3960 void TemplateTable::monitorenter() {
3961   transition(atos, vtos);
3962 
3963   __ verify_oop(R17_tos);
3964 
3965   Register Rcurrent_monitor  = R11_scratch1,
3966            Rcurrent_obj      = R12_scratch2,
3967            Robj_to_lock      = R17_tos,
3968            Rscratch1         = R3_ARG1,
3969            Rscratch2         = R4_ARG2,
3970            Rscratch3         = R5_ARG3,
3971            Rcurrent_obj_addr = R6_ARG4;
3972 
3973   // ------------------------------------------------------------------------------
3974   // Null pointer exception.
3975   __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
3976 
3977   // Try to acquire a lock on the object.
3978   // Repeat until succeeded (i.e., until monitorenter returns true).
3979 
3980   // ------------------------------------------------------------------------------
3981   // Find a free slot in the monitor block.
3982   Label Lfound, Lexit, Lallocate_new;
3983   ConditionRegister found_free_slot = CCR0,
3984                     found_same_obj  = CCR1,
3985                     reached_limit   = CCR6;
3986   {
3987     Label Lloop, Lentry;
3988     Register Rlimit = Rcurrent_monitor;
3989 
3990     // Set up search loop - start with topmost monitor.
3991     __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
3992 
3993     __ ld(Rlimit, 0, R1_SP);
3994     __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
3995 
3996     // Check if any slot is present => short cut to allocation if not.
3997     __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
3998     __ bgt(reached_limit, Lallocate_new);
3999 
4000     // Pre-load topmost slot.
4001     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4002     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4003     // The search loop.
4004     __ bind(Lloop);
4005     // Found free slot?
4006     __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4007     // Is this entry for same obj? If so, stop the search and take the found
4008     // free slot or allocate a new one to enable recursive locking.
4009     __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4010     __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4011     __ beq(found_free_slot, Lexit);
4012     __ beq(found_same_obj, Lallocate_new);
4013     __ bgt(reached_limit, Lallocate_new);
4014     // Check if last allocated BasicLockObj reached.
4015     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4016     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4017     // Next iteration if unchecked BasicObjectLocks exist on the stack.
4018     __ b(Lloop);
4019   }
4020 
4021   // ------------------------------------------------------------------------------
4022   // Check if we found a free slot.
4023   __ bind(Lexit);
4024 
4025   __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4026   __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4027   __ b(Lfound);
4028 
4029   // We didn't find a free BasicObjLock => allocate one.
4030   __ align(32, 12);
4031   __ bind(Lallocate_new);
4032   __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4033   __ mr(Rcurrent_monitor, R26_monitor);
4034   __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4035 
4036   // ------------------------------------------------------------------------------
4037   // We now have a slot to lock.
4038   __ bind(Lfound);
4039 
4040   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4041   // The object has already been poped from the stack, so the expression stack looks correct.
4042   __ addi(R14_bcp, R14_bcp, 1);
4043 
4044   __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4045   __ lock_object(Rcurrent_monitor, Robj_to_lock);
4046 
4047   // Check if there's enough space on the stack for the monitors after locking.
4048   Label Lskip_stack_check;
4049   // Optimization: If the monitors stack section is less then a std page size (4K) don't run
4050   // the stack check. There should be enough shadow pages to fit that in.
4051   __ ld(Rscratch3, 0, R1_SP);
4052   __ sub(Rscratch3, Rscratch3, R26_monitor);
4053   __ cmpdi(CCR0, Rscratch3, 4*K);
4054   __ blt(CCR0, Lskip_stack_check);
4055 
4056   DEBUG_ONLY(__ untested("stack overflow check during monitor enter");)
4057   __ li(Rscratch1, 0);
4058   __ generate_stack_overflow_check_with_compare_and_throw(Rscratch1, Rscratch2);
4059 
4060   __ align(32, 12);
4061   __ bind(Lskip_stack_check);
4062 
4063   // The bcp has already been incremented. Just need to dispatch to next instruction.
4064   __ dispatch_next(vtos);
4065 }
4066 
4067 void TemplateTable::monitorexit() {
4068   transition(atos, vtos);
4069   __ verify_oop(R17_tos);
4070 
4071   Register Rcurrent_monitor  = R11_scratch1,
4072            Rcurrent_obj      = R12_scratch2,
4073            Robj_to_lock      = R17_tos,
4074            Rcurrent_obj_addr = R3_ARG1,
4075            Rlimit            = R4_ARG2;
4076   Label Lfound, Lillegal_monitor_state;
4077 
4078   // Check corner case: unbalanced monitorEnter / Exit.
4079   __ ld(Rlimit, 0, R1_SP);
4080   __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4081 
4082   // Null pointer check.
4083   __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4084 
4085   __ cmpld(CCR0, R26_monitor, Rlimit);
4086   __ bgt(CCR0, Lillegal_monitor_state);
4087 
4088   // Find the corresponding slot in the monitors stack section.
4089   {
4090     Label Lloop;
4091 
4092     // Start with topmost monitor.
4093     __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4094     __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4095     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4096     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4097 
4098     __ bind(Lloop);
4099     // Is this entry for same obj?
4100     __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4101     __ beq(CCR0, Lfound);
4102 
4103     // Check if last allocated BasicLockObj reached.
4104 
4105     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4106     __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4107     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4108 
4109     // Next iteration if unchecked BasicObjectLocks exist on the stack.
4110     __ ble(CCR0, Lloop);
4111   }
4112 
4113   // Fell through without finding the basic obj lock => throw up!
4114   __ bind(Lillegal_monitor_state);
4115   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4116   __ should_not_reach_here();
4117 
4118   __ align(32, 12);
4119   __ bind(Lfound);
4120   __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4121           -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4122   __ unlock_object(Rcurrent_monitor);
4123 }
4124 
4125 // ============================================================================
4126 // Wide bytecodes
4127 
4128 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4129 void TemplateTable::wide() {
4130   transition(vtos, vtos);
4131 
4132   const Register Rtable = R11_scratch1,
4133                  Rindex = R12_scratch2,
4134                  Rtmp   = R0;
4135 
4136   __ lbz(Rindex, 1, R14_bcp);
4137 
4138   __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4139 
4140   __ slwi(Rindex, Rindex, LogBytesPerWord);
4141   __ ldx(Rtmp, Rtable, Rindex);
4142   __ mtctr(Rtmp);
4143   __ bctr();
4144   // Note: the bcp increment step is part of the individual wide bytecode implementations.
4145 }