1 /*
   2  * Copyright (c) 2014, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2013, 2017 SAP SE. 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_zputfield:
 174     case Bytecodes::_fast_cputfield:
 175     case Bytecodes::_fast_dputfield:
 176     case Bytecodes::_fast_fputfield:
 177     case Bytecodes::_fast_iputfield:
 178     case Bytecodes::_fast_lputfield:
 179     case Bytecodes::_fast_sputfield:
 180     {
 181       // We skip bytecode quickening for putfield instructions when
 182       // the put_code written to the constant pool cache is zero.
 183       // This is required so that every execution of this instruction
 184       // calls out to InterpreterRuntime::resolve_get_put to do
 185       // additional, required work.
 186       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
 187       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
 188       __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
 189       // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
 190 #if defined(VM_LITTLE_ENDIAN)
 191       __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
 192 #else
 193       __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
 194 #endif
 195       __ cmpwi(CCR0, Rnew_bc, 0);
 196       __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
 197       __ beq(CCR0, L_patch_done);
 198       // __ isync(); // acquire not needed
 199       break;
 200     }
 201 
 202     default:
 203       assert(byte_no == -1, "sanity");
 204       if (load_bc_into_bc_reg) {
 205         __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
 206       }
 207   }
 208 
 209   if (JvmtiExport::can_post_breakpoint()) {
 210     Label L_fast_patch;
 211     __ lbz(Rtemp, 0, R14_bcp);
 212     __ cmpwi(CCR0, Rtemp, (unsigned int)(unsigned char)Bytecodes::_breakpoint);
 213     __ bne(CCR0, L_fast_patch);
 214     // Perform the quickening, slowly, in the bowels of the breakpoint table.
 215     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), R19_method, R14_bcp, Rnew_bc);
 216     __ b(L_patch_done);
 217     __ bind(L_fast_patch);
 218   }
 219 
 220   // Patch bytecode.
 221   __ stb(Rnew_bc, 0, R14_bcp);
 222 
 223   __ bind(L_patch_done);
 224 }
 225 
 226 // ============================================================================
 227 // Individual instructions
 228 
 229 void TemplateTable::nop() {
 230   transition(vtos, vtos);
 231   // Nothing to do.
 232 }
 233 
 234 void TemplateTable::shouldnotreachhere() {
 235   transition(vtos, vtos);
 236   __ stop("shouldnotreachhere bytecode");
 237 }
 238 
 239 void TemplateTable::aconst_null() {
 240   transition(vtos, atos);
 241   __ li(R17_tos, 0);
 242 }
 243 
 244 void TemplateTable::iconst(int value) {
 245   transition(vtos, itos);
 246   assert(value >= -1 && value <= 5, "");
 247   __ li(R17_tos, value);
 248 }
 249 
 250 void TemplateTable::lconst(int value) {
 251   transition(vtos, ltos);
 252   assert(value >= -1 && value <= 5, "");
 253   __ li(R17_tos, value);
 254 }
 255 
 256 void TemplateTable::fconst(int value) {
 257   transition(vtos, ftos);
 258   static float zero = 0.0;
 259   static float one  = 1.0;
 260   static float two  = 2.0;
 261   switch (value) {
 262     default: ShouldNotReachHere();
 263     case 0: {
 264       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
 265       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 266       break;
 267     }
 268     case 1: {
 269       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
 270       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 271       break;
 272     }
 273     case 2: {
 274       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&two, R0, true);
 275       __ lfs(F15_ftos, simm16_offset, R11_scratch1);
 276       break;
 277     }
 278   }
 279 }
 280 
 281 void TemplateTable::dconst(int value) {
 282   transition(vtos, dtos);
 283   static double zero = 0.0;
 284   static double one  = 1.0;
 285   switch (value) {
 286     case 0: {
 287       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&zero, R0, true);
 288       __ lfd(F15_ftos, simm16_offset, R11_scratch1);
 289       break;
 290     }
 291     case 1: {
 292       int simm16_offset = __ load_const_optimized(R11_scratch1, (address*)&one, R0, true);
 293       __ lfd(F15_ftos, simm16_offset, R11_scratch1);
 294       break;
 295     }
 296     default: ShouldNotReachHere();
 297   }
 298 }
 299 
 300 void TemplateTable::bipush() {
 301   transition(vtos, itos);
 302   __ lbz(R17_tos, 1, R14_bcp);
 303   __ extsb(R17_tos, R17_tos);
 304 }
 305 
 306 void TemplateTable::sipush() {
 307   transition(vtos, itos);
 308   __ get_2_byte_integer_at_bcp(1, R17_tos, InterpreterMacroAssembler::Signed);
 309 }
 310 
 311 void TemplateTable::ldc(bool wide) {
 312   Register Rscratch1 = R11_scratch1,
 313            Rscratch2 = R12_scratch2,
 314            Rcpool    = R3_ARG1;
 315 
 316   transition(vtos, vtos);
 317   Label notInt, notClass, exit;
 318 
 319   __ get_cpool_and_tags(Rcpool, Rscratch2); // Set Rscratch2 = &tags.
 320   if (wide) { // Read index.
 321     __ get_2_byte_integer_at_bcp(1, Rscratch1, InterpreterMacroAssembler::Unsigned);
 322   } else {
 323     __ lbz(Rscratch1, 1, R14_bcp);
 324   }
 325 
 326   const int base_offset = ConstantPool::header_size() * wordSize;
 327   const int tags_offset = Array<u1>::base_offset_in_bytes();
 328 
 329   // Get type from tags.
 330   __ addi(Rscratch2, Rscratch2, tags_offset);
 331   __ lbzx(Rscratch2, Rscratch2, Rscratch1);
 332 
 333   __ cmpwi(CCR0, Rscratch2, JVM_CONSTANT_UnresolvedClass); // Unresolved class?
 334   __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_UnresolvedClassInError); // Unresolved class in error state?
 335   __ cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
 336 
 337   // Resolved class - need to call vm to get java mirror of the class.
 338   __ cmpwi(CCR1, Rscratch2, JVM_CONSTANT_Class);
 339   __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal); // Neither resolved class nor unresolved case from above?
 340   __ beq(CCR0, notClass);
 341 
 342   __ li(R4, wide ? 1 : 0);
 343   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), R4);
 344   __ push(atos);
 345   __ b(exit);
 346 
 347   __ align(32, 12);
 348   __ bind(notClass);
 349   __ addi(Rcpool, Rcpool, base_offset);
 350   __ sldi(Rscratch1, Rscratch1, LogBytesPerWord);
 351   __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Integer);
 352   __ bne(CCR0, notInt);
 353   __ lwax(R17_tos, Rcpool, Rscratch1);
 354   __ push(itos);
 355   __ b(exit);
 356 
 357   __ align(32, 12);
 358   __ bind(notInt);
 359 #ifdef ASSERT
 360   // String and Object are rewritten to fast_aldc
 361   __ cmpdi(CCR0, Rscratch2, JVM_CONSTANT_Float);
 362   __ asm_assert_eq("unexpected type", 0x8765);
 363 #endif
 364   __ lfsx(F15_ftos, Rcpool, Rscratch1);
 365   __ push(ftos);
 366 
 367   __ align(32, 12);
 368   __ bind(exit);
 369 }
 370 
 371 // Fast path for caching oop constants.
 372 void TemplateTable::fast_aldc(bool wide) {
 373   transition(vtos, atos);
 374 
 375   int index_size = wide ? sizeof(u2) : sizeof(u1);
 376   const Register Rscratch = R11_scratch1;
 377   Label is_null;
 378 
 379   // We are resolved if the resolved reference cache entry contains a
 380   // non-null object (CallSite, etc.)
 381   __ get_cache_index_at_bcp(Rscratch, 1, index_size);  // Load index.
 382   __ load_resolved_reference_at_index(R17_tos, Rscratch, &is_null);
 383   __ verify_oop(R17_tos);
 384   __ dispatch_epilog(atos, Bytecodes::length_for(bytecode()));
 385 
 386   __ bind(is_null);
 387   __ load_const_optimized(R3_ARG1, (int)bytecode());
 388 
 389   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
 390 
 391   // First time invocation - must resolve first.
 392   __ call_VM(R17_tos, entry, R3_ARG1);
 393   __ verify_oop(R17_tos);
 394 }
 395 
 396 void TemplateTable::ldc2_w() {
 397   transition(vtos, vtos);
 398   Label Llong, Lexit;
 399 
 400   Register Rindex = R11_scratch1,
 401            Rcpool = R12_scratch2,
 402            Rtag   = R3_ARG1;
 403   __ get_cpool_and_tags(Rcpool, Rtag);
 404   __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
 405 
 406   const int base_offset = ConstantPool::header_size() * wordSize;
 407   const int tags_offset = Array<u1>::base_offset_in_bytes();
 408   // Get type from tags.
 409   __ addi(Rcpool, Rcpool, base_offset);
 410   __ addi(Rtag, Rtag, tags_offset);
 411 
 412   __ lbzx(Rtag, Rtag, Rindex);
 413 
 414   __ sldi(Rindex, Rindex, LogBytesPerWord);
 415   __ cmpdi(CCR0, Rtag, JVM_CONSTANT_Double);
 416   __ bne(CCR0, Llong);
 417   // A double can be placed at word-aligned locations in the constant pool.
 418   // Check out Conversions.java for an example.
 419   // Also ConstantPool::header_size() is 20, which makes it very difficult
 420   // to double-align double on the constant pool. SG, 11/7/97
 421   __ lfdx(F15_ftos, Rcpool, Rindex);
 422   __ push(dtos);
 423   __ b(Lexit);
 424 
 425   __ bind(Llong);
 426   __ ldx(R17_tos, Rcpool, Rindex);
 427   __ push(ltos);
 428 
 429   __ bind(Lexit);
 430 }
 431 
 432 // Get the locals index located in the bytecode stream at bcp + offset.
 433 void TemplateTable::locals_index(Register Rdst, int offset) {
 434   __ lbz(Rdst, offset, R14_bcp);
 435 }
 436 
 437 void TemplateTable::iload() {
 438   iload_internal();
 439 }
 440 
 441 void TemplateTable::nofast_iload() {
 442   iload_internal(may_not_rewrite);
 443 }
 444 
 445 void TemplateTable::iload_internal(RewriteControl rc) {
 446   transition(vtos, itos);
 447 
 448   // Get the local value into tos
 449   const Register Rindex = R22_tmp2;
 450   locals_index(Rindex);
 451 
 452   // Rewrite iload,iload  pair into fast_iload2
 453   //         iload,caload pair into fast_icaload
 454   if (RewriteFrequentPairs && rc == may_rewrite) {
 455     Label Lrewrite, Ldone;
 456     Register Rnext_byte  = R3_ARG1,
 457              Rrewrite_to = R6_ARG4,
 458              Rscratch    = R11_scratch1;
 459 
 460     // get next byte
 461     __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_iload), R14_bcp);
 462 
 463     // if _iload, wait to rewrite to iload2. We only want to rewrite the
 464     // last two iloads in a pair. Comparing against fast_iload means that
 465     // the next bytecode is neither an iload or a caload, and therefore
 466     // an iload pair.
 467     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_iload);
 468     __ beq(CCR0, Ldone);
 469 
 470     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
 471     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload2);
 472     __ beq(CCR1, Lrewrite);
 473 
 474     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_caload);
 475     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_icaload);
 476     __ beq(CCR0, Lrewrite);
 477 
 478     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iload);
 479 
 480     __ bind(Lrewrite);
 481     patch_bytecode(Bytecodes::_iload, Rrewrite_to, Rscratch, false);
 482     __ bind(Ldone);
 483   }
 484 
 485   __ load_local_int(R17_tos, Rindex, Rindex);
 486 }
 487 
 488 // Load 2 integers in a row without dispatching
 489 void TemplateTable::fast_iload2() {
 490   transition(vtos, itos);
 491 
 492   __ lbz(R3_ARG1, 1, R14_bcp);
 493   __ lbz(R17_tos, Bytecodes::length_for(Bytecodes::_iload) + 1, R14_bcp);
 494 
 495   __ load_local_int(R3_ARG1, R11_scratch1, R3_ARG1);
 496   __ load_local_int(R17_tos, R12_scratch2, R17_tos);
 497   __ push_i(R3_ARG1);
 498 }
 499 
 500 void TemplateTable::fast_iload() {
 501   transition(vtos, itos);
 502   // Get the local value into tos
 503 
 504   const Register Rindex = R11_scratch1;
 505   locals_index(Rindex);
 506   __ load_local_int(R17_tos, Rindex, Rindex);
 507 }
 508 
 509 // Load a local variable type long from locals area to TOS cache register.
 510 // Local index resides in bytecodestream.
 511 void TemplateTable::lload() {
 512   transition(vtos, ltos);
 513 
 514   const Register Rindex = R11_scratch1;
 515   locals_index(Rindex);
 516   __ load_local_long(R17_tos, Rindex, Rindex);
 517 }
 518 
 519 void TemplateTable::fload() {
 520   transition(vtos, ftos);
 521 
 522   const Register Rindex = R11_scratch1;
 523   locals_index(Rindex);
 524   __ load_local_float(F15_ftos, Rindex, Rindex);
 525 }
 526 
 527 void TemplateTable::dload() {
 528   transition(vtos, dtos);
 529 
 530   const Register Rindex = R11_scratch1;
 531   locals_index(Rindex);
 532   __ load_local_double(F15_ftos, Rindex, Rindex);
 533 }
 534 
 535 void TemplateTable::aload() {
 536   transition(vtos, atos);
 537 
 538   const Register Rindex = R11_scratch1;
 539   locals_index(Rindex);
 540   __ load_local_ptr(R17_tos, Rindex, Rindex);
 541 }
 542 
 543 void TemplateTable::locals_index_wide(Register Rdst) {
 544   // Offset is 2, not 1, because Lbcp points to wide prefix code.
 545   __ get_2_byte_integer_at_bcp(2, Rdst, InterpreterMacroAssembler::Unsigned);
 546 }
 547 
 548 void TemplateTable::wide_iload() {
 549   // Get the local value into tos.
 550 
 551   const Register Rindex = R11_scratch1;
 552   locals_index_wide(Rindex);
 553   __ load_local_int(R17_tos, Rindex, Rindex);
 554 }
 555 
 556 void TemplateTable::wide_lload() {
 557   transition(vtos, ltos);
 558 
 559   const Register Rindex = R11_scratch1;
 560   locals_index_wide(Rindex);
 561   __ load_local_long(R17_tos, Rindex, Rindex);
 562 }
 563 
 564 void TemplateTable::wide_fload() {
 565   transition(vtos, ftos);
 566 
 567   const Register Rindex = R11_scratch1;
 568   locals_index_wide(Rindex);
 569   __ load_local_float(F15_ftos, Rindex, Rindex);
 570 }
 571 
 572 void TemplateTable::wide_dload() {
 573   transition(vtos, dtos);
 574 
 575   const Register Rindex = R11_scratch1;
 576   locals_index_wide(Rindex);
 577   __ load_local_double(F15_ftos, Rindex, Rindex);
 578 }
 579 
 580 void TemplateTable::wide_aload() {
 581   transition(vtos, atos);
 582 
 583   const Register Rindex = R11_scratch1;
 584   locals_index_wide(Rindex);
 585   __ load_local_ptr(R17_tos, Rindex, Rindex);
 586 }
 587 
 588 void TemplateTable::iaload() {
 589   transition(itos, itos);
 590 
 591   const Register Rload_addr = R3_ARG1,
 592                  Rarray     = R4_ARG2,
 593                  Rtemp      = R5_ARG3;
 594   __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
 595   __ lwa(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rload_addr);
 596 }
 597 
 598 void TemplateTable::laload() {
 599   transition(itos, ltos);
 600 
 601   const Register Rload_addr = R3_ARG1,
 602                  Rarray     = R4_ARG2,
 603                  Rtemp      = R5_ARG3;
 604   __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
 605   __ ld(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rload_addr);
 606 }
 607 
 608 void TemplateTable::faload() {
 609   transition(itos, ftos);
 610 
 611   const Register Rload_addr = R3_ARG1,
 612                  Rarray     = R4_ARG2,
 613                  Rtemp      = R5_ARG3;
 614   __ index_check(Rarray, R17_tos /* index */, LogBytesPerInt, Rtemp, Rload_addr);
 615   __ lfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rload_addr);
 616 }
 617 
 618 void TemplateTable::daload() {
 619   transition(itos, dtos);
 620 
 621   const Register Rload_addr = R3_ARG1,
 622                  Rarray     = R4_ARG2,
 623                  Rtemp      = R5_ARG3;
 624   __ index_check(Rarray, R17_tos /* index */, LogBytesPerLong, Rtemp, Rload_addr);
 625   __ lfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rload_addr);
 626 }
 627 
 628 void TemplateTable::aaload() {
 629   transition(itos, atos);
 630 
 631   // tos: index
 632   // result tos: array
 633   const Register Rload_addr = R3_ARG1,
 634                  Rarray     = R4_ARG2,
 635                  Rtemp      = R5_ARG3;
 636   __ index_check(Rarray, R17_tos /* index */, UseCompressedOops ? 2 : LogBytesPerWord, Rtemp, Rload_addr);
 637   __ load_heap_oop(R17_tos, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rload_addr);
 638   __ verify_oop(R17_tos);
 639   //__ dcbt(R17_tos); // prefetch
 640 }
 641 
 642 void TemplateTable::baload() {
 643   transition(itos, itos);
 644 
 645   const Register Rload_addr = R3_ARG1,
 646                  Rarray     = R4_ARG2,
 647                  Rtemp      = R5_ARG3;
 648   __ index_check(Rarray, R17_tos /* index */, 0, Rtemp, Rload_addr);
 649   __ lbz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rload_addr);
 650   __ extsb(R17_tos, R17_tos);
 651 }
 652 
 653 void TemplateTable::caload() {
 654   transition(itos, itos);
 655 
 656   const Register Rload_addr = R3_ARG1,
 657                  Rarray     = R4_ARG2,
 658                  Rtemp      = R5_ARG3;
 659   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 660   __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
 661 }
 662 
 663 // Iload followed by caload frequent pair.
 664 void TemplateTable::fast_icaload() {
 665   transition(vtos, itos);
 666 
 667   const Register Rload_addr = R3_ARG1,
 668                  Rarray     = R4_ARG2,
 669                  Rtemp      = R11_scratch1;
 670 
 671   locals_index(R17_tos);
 672   __ load_local_int(R17_tos, Rtemp, R17_tos);
 673   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 674   __ lhz(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rload_addr);
 675 }
 676 
 677 void TemplateTable::saload() {
 678   transition(itos, itos);
 679 
 680   const Register Rload_addr = R11_scratch1,
 681                  Rarray     = R12_scratch2,
 682                  Rtemp      = R3_ARG1;
 683   __ index_check(Rarray, R17_tos /* index */, LogBytesPerShort, Rtemp, Rload_addr);
 684   __ lha(R17_tos, arrayOopDesc::base_offset_in_bytes(T_SHORT), Rload_addr);
 685 }
 686 
 687 void TemplateTable::iload(int n) {
 688   transition(vtos, itos);
 689 
 690   __ lwz(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
 691 }
 692 
 693 void TemplateTable::lload(int n) {
 694   transition(vtos, ltos);
 695 
 696   __ ld(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
 697 }
 698 
 699 void TemplateTable::fload(int n) {
 700   transition(vtos, ftos);
 701 
 702   __ lfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
 703 }
 704 
 705 void TemplateTable::dload(int n) {
 706   transition(vtos, dtos);
 707 
 708   __ lfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
 709 }
 710 
 711 void TemplateTable::aload(int n) {
 712   transition(vtos, atos);
 713 
 714   __ ld(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
 715 }
 716 
 717 void TemplateTable::aload_0() {
 718   aload_0_internal();
 719 }
 720 
 721 void TemplateTable::nofast_aload_0() {
 722   aload_0_internal(may_not_rewrite);
 723 }
 724 
 725 void TemplateTable::aload_0_internal(RewriteControl rc) {
 726   transition(vtos, atos);
 727   // According to bytecode histograms, the pairs:
 728   //
 729   // _aload_0, _fast_igetfield
 730   // _aload_0, _fast_agetfield
 731   // _aload_0, _fast_fgetfield
 732   //
 733   // occur frequently. If RewriteFrequentPairs is set, the (slow)
 734   // _aload_0 bytecode checks if the next bytecode is either
 735   // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
 736   // rewrites the current bytecode into a pair bytecode; otherwise it
 737   // rewrites the current bytecode into _0 that doesn't do
 738   // the pair check anymore.
 739   //
 740   // Note: If the next bytecode is _getfield, the rewrite must be
 741   //       delayed, otherwise we may miss an opportunity for a pair.
 742   //
 743   // Also rewrite frequent pairs
 744   //   aload_0, aload_1
 745   //   aload_0, iload_1
 746   // These bytecodes with a small amount of code are most profitable
 747   // to rewrite.
 748 
 749   if (RewriteFrequentPairs && rc == may_rewrite) {
 750 
 751     Label Lrewrite, Ldont_rewrite;
 752     Register Rnext_byte  = R3_ARG1,
 753              Rrewrite_to = R6_ARG4,
 754              Rscratch    = R11_scratch1;
 755 
 756     // Get next byte.
 757     __ lbz(Rnext_byte, Bytecodes::length_for(Bytecodes::_aload_0), R14_bcp);
 758 
 759     // If _getfield, wait to rewrite. We only want to rewrite the last two bytecodes in a pair.
 760     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_getfield);
 761     __ beq(CCR0, Ldont_rewrite);
 762 
 763     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_igetfield);
 764     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_iaccess_0);
 765     __ beq(CCR1, Lrewrite);
 766 
 767     __ cmpwi(CCR0, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_agetfield);
 768     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aaccess_0);
 769     __ beq(CCR0, Lrewrite);
 770 
 771     __ cmpwi(CCR1, Rnext_byte, (unsigned int)(unsigned char)Bytecodes::_fast_fgetfield);
 772     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_faccess_0);
 773     __ beq(CCR1, Lrewrite);
 774 
 775     __ li(Rrewrite_to, (unsigned int)(unsigned char)Bytecodes::_fast_aload_0);
 776 
 777     __ bind(Lrewrite);
 778     patch_bytecode(Bytecodes::_aload_0, Rrewrite_to, Rscratch, false);
 779     __ bind(Ldont_rewrite);
 780   }
 781 
 782   // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
 783   aload(0);
 784 }
 785 
 786 void TemplateTable::istore() {
 787   transition(itos, vtos);
 788 
 789   const Register Rindex = R11_scratch1;
 790   locals_index(Rindex);
 791   __ store_local_int(R17_tos, Rindex);
 792 }
 793 
 794 void TemplateTable::lstore() {
 795   transition(ltos, vtos);
 796   const Register Rindex = R11_scratch1;
 797   locals_index(Rindex);
 798   __ store_local_long(R17_tos, Rindex);
 799 }
 800 
 801 void TemplateTable::fstore() {
 802   transition(ftos, vtos);
 803 
 804   const Register Rindex = R11_scratch1;
 805   locals_index(Rindex);
 806   __ store_local_float(F15_ftos, Rindex);
 807 }
 808 
 809 void TemplateTable::dstore() {
 810   transition(dtos, vtos);
 811 
 812   const Register Rindex = R11_scratch1;
 813   locals_index(Rindex);
 814   __ store_local_double(F15_ftos, Rindex);
 815 }
 816 
 817 void TemplateTable::astore() {
 818   transition(vtos, vtos);
 819 
 820   const Register Rindex = R11_scratch1;
 821   __ pop_ptr();
 822   __ verify_oop_or_return_address(R17_tos, Rindex);
 823   locals_index(Rindex);
 824   __ store_local_ptr(R17_tos, Rindex);
 825 }
 826 
 827 void TemplateTable::wide_istore() {
 828   transition(vtos, vtos);
 829 
 830   const Register Rindex = R11_scratch1;
 831   __ pop_i();
 832   locals_index_wide(Rindex);
 833   __ store_local_int(R17_tos, Rindex);
 834 }
 835 
 836 void TemplateTable::wide_lstore() {
 837   transition(vtos, vtos);
 838 
 839   const Register Rindex = R11_scratch1;
 840   __ pop_l();
 841   locals_index_wide(Rindex);
 842   __ store_local_long(R17_tos, Rindex);
 843 }
 844 
 845 void TemplateTable::wide_fstore() {
 846   transition(vtos, vtos);
 847 
 848   const Register Rindex = R11_scratch1;
 849   __ pop_f();
 850   locals_index_wide(Rindex);
 851   __ store_local_float(F15_ftos, Rindex);
 852 }
 853 
 854 void TemplateTable::wide_dstore() {
 855   transition(vtos, vtos);
 856 
 857   const Register Rindex = R11_scratch1;
 858   __ pop_d();
 859   locals_index_wide(Rindex);
 860   __ store_local_double(F15_ftos, Rindex);
 861 }
 862 
 863 void TemplateTable::wide_astore() {
 864   transition(vtos, vtos);
 865 
 866   const Register Rindex = R11_scratch1;
 867   __ pop_ptr();
 868   __ verify_oop_or_return_address(R17_tos, Rindex);
 869   locals_index_wide(Rindex);
 870   __ store_local_ptr(R17_tos, Rindex);
 871 }
 872 
 873 void TemplateTable::iastore() {
 874   transition(itos, vtos);
 875 
 876   const Register Rindex      = R3_ARG1,
 877                  Rstore_addr = R4_ARG2,
 878                  Rarray      = R5_ARG3,
 879                  Rtemp       = R6_ARG4;
 880   __ pop_i(Rindex);
 881   __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
 882   __ stw(R17_tos, arrayOopDesc::base_offset_in_bytes(T_INT), Rstore_addr);
 883   }
 884 
 885 void TemplateTable::lastore() {
 886   transition(ltos, vtos);
 887 
 888   const Register Rindex      = R3_ARG1,
 889                  Rstore_addr = R4_ARG2,
 890                  Rarray      = R5_ARG3,
 891                  Rtemp       = R6_ARG4;
 892   __ pop_i(Rindex);
 893   __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
 894   __ std(R17_tos, arrayOopDesc::base_offset_in_bytes(T_LONG), Rstore_addr);
 895   }
 896 
 897 void TemplateTable::fastore() {
 898   transition(ftos, vtos);
 899 
 900   const Register Rindex      = R3_ARG1,
 901                  Rstore_addr = R4_ARG2,
 902                  Rarray      = R5_ARG3,
 903                  Rtemp       = R6_ARG4;
 904   __ pop_i(Rindex);
 905   __ index_check(Rarray, Rindex, LogBytesPerInt, Rtemp, Rstore_addr);
 906   __ stfs(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Rstore_addr);
 907   }
 908 
 909 void TemplateTable::dastore() {
 910   transition(dtos, vtos);
 911 
 912   const Register Rindex      = R3_ARG1,
 913                  Rstore_addr = R4_ARG2,
 914                  Rarray      = R5_ARG3,
 915                  Rtemp       = R6_ARG4;
 916   __ pop_i(Rindex);
 917   __ index_check(Rarray, Rindex, LogBytesPerLong, Rtemp, Rstore_addr);
 918   __ stfd(F15_ftos, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Rstore_addr);
 919   }
 920 
 921 // Pop 3 values from the stack and...
 922 void TemplateTable::aastore() {
 923   transition(vtos, vtos);
 924 
 925   Label Lstore_ok, Lis_null, Ldone;
 926   const Register Rindex    = R3_ARG1,
 927                  Rarray    = R4_ARG2,
 928                  Rscratch  = R11_scratch1,
 929                  Rscratch2 = R12_scratch2,
 930                  Rarray_klass = R5_ARG3,
 931                  Rarray_element_klass = Rarray_klass,
 932                  Rvalue_klass = R6_ARG4,
 933                  Rstore_addr = R31;    // Use register which survives VM call.
 934 
 935   __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp); // Get value to store.
 936   __ lwz(Rindex, Interpreter::expr_offset_in_bytes(1), R15_esp); // Get index.
 937   __ ld(Rarray, Interpreter::expr_offset_in_bytes(2), R15_esp);  // Get array.
 938 
 939   __ verify_oop(R17_tos);
 940   __ index_check_without_pop(Rarray, Rindex, UseCompressedOops ? 2 : LogBytesPerWord, Rscratch, Rstore_addr);
 941   // Rindex is dead!
 942   Register Rscratch3 = Rindex;
 943 
 944   // Do array store check - check for NULL value first.
 945   __ cmpdi(CCR0, R17_tos, 0);
 946   __ beq(CCR0, Lis_null);
 947 
 948   __ load_klass(Rarray_klass, Rarray);
 949   __ load_klass(Rvalue_klass, R17_tos);
 950 
 951   // Do fast instanceof cache test.
 952   __ ld(Rarray_element_klass, in_bytes(ObjArrayKlass::element_klass_offset()), Rarray_klass);
 953 
 954   // Generate a fast subtype check. Branch to store_ok if no failure. Throw if failure.
 955   __ gen_subtype_check(Rvalue_klass /*subklass*/, Rarray_element_klass /*superklass*/, Rscratch, Rscratch2, Rscratch3, Lstore_ok);
 956 
 957   // Fell through: subtype check failed => throw an exception.
 958   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArrayStoreException_entry);
 959   __ mtctr(R11_scratch1);
 960   __ bctr();
 961 
 962   __ bind(Lis_null);
 963   do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), noreg /* 0 */,
 964                Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
 965   __ profile_null_seen(Rscratch, Rscratch2);
 966   __ b(Ldone);
 967 
 968   // Store is OK.
 969   __ bind(Lstore_ok);
 970   do_oop_store(_masm, Rstore_addr, arrayOopDesc::base_offset_in_bytes(T_OBJECT), R17_tos /* value */,
 971                Rscratch, Rscratch2, Rscratch3, _bs->kind(), true /* precise */, false /* check_null */);
 972 
 973   __ bind(Ldone);
 974   // Adjust sp (pops array, index and value).
 975   __ addi(R15_esp, R15_esp, 3 * Interpreter::stackElementSize);
 976 }
 977 
 978 void TemplateTable::bastore() {
 979   transition(itos, vtos);
 980 
 981   const Register Rindex   = R11_scratch1,
 982                  Rarray   = R12_scratch2,
 983                  Rscratch = R3_ARG1;
 984   __ pop_i(Rindex);
 985   __ pop_ptr(Rarray);
 986   // tos: val
 987 
 988   // Need to check whether array is boolean or byte
 989   // since both types share the bastore bytecode.
 990   __ load_klass(Rscratch, Rarray);
 991   __ lwz(Rscratch, in_bytes(Klass::layout_helper_offset()), Rscratch);
 992   int diffbit = exact_log2(Klass::layout_helper_boolean_diffbit());
 993   __ testbitdi(CCR0, R0, Rscratch, diffbit);
 994   Label L_skip;
 995   __ bfalse(CCR0, L_skip);
 996   __ andi(R17_tos, R17_tos, 1);  // if it is a T_BOOLEAN array, mask the stored value to 0/1
 997   __ bind(L_skip);
 998 
 999   __ index_check_without_pop(Rarray, Rindex, 0, Rscratch, Rarray);
1000   __ stb(R17_tos, arrayOopDesc::base_offset_in_bytes(T_BYTE), Rarray);
1001 }
1002 
1003 void TemplateTable::castore() {
1004   transition(itos, vtos);
1005 
1006   const Register Rindex   = R11_scratch1,
1007                  Rarray   = R12_scratch2,
1008                  Rscratch = R3_ARG1;
1009   __ pop_i(Rindex);
1010   // tos: val
1011   // Rarray: array ptr (popped by index_check)
1012   __ index_check(Rarray, Rindex, LogBytesPerShort, Rscratch, Rarray);
1013   __ sth(R17_tos, arrayOopDesc::base_offset_in_bytes(T_CHAR), Rarray);
1014 }
1015 
1016 void TemplateTable::sastore() {
1017   castore();
1018 }
1019 
1020 void TemplateTable::istore(int n) {
1021   transition(itos, vtos);
1022   __ stw(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1023 }
1024 
1025 void TemplateTable::lstore(int n) {
1026   transition(ltos, vtos);
1027   __ std(R17_tos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1028 }
1029 
1030 void TemplateTable::fstore(int n) {
1031   transition(ftos, vtos);
1032   __ stfs(F15_ftos, Interpreter::local_offset_in_bytes(n), R18_locals);
1033 }
1034 
1035 void TemplateTable::dstore(int n) {
1036   transition(dtos, vtos);
1037   __ stfd(F15_ftos, Interpreter::local_offset_in_bytes(n + 1), R18_locals);
1038 }
1039 
1040 void TemplateTable::astore(int n) {
1041   transition(vtos, vtos);
1042 
1043   __ pop_ptr();
1044   __ verify_oop_or_return_address(R17_tos, R11_scratch1);
1045   __ std(R17_tos, Interpreter::local_offset_in_bytes(n), R18_locals);
1046 }
1047 
1048 void TemplateTable::pop() {
1049   transition(vtos, vtos);
1050 
1051   __ addi(R15_esp, R15_esp, Interpreter::stackElementSize);
1052 }
1053 
1054 void TemplateTable::pop2() {
1055   transition(vtos, vtos);
1056 
1057   __ addi(R15_esp, R15_esp, Interpreter::stackElementSize * 2);
1058 }
1059 
1060 void TemplateTable::dup() {
1061   transition(vtos, vtos);
1062 
1063   __ ld(R11_scratch1, Interpreter::stackElementSize, R15_esp);
1064   __ push_ptr(R11_scratch1);
1065 }
1066 
1067 void TemplateTable::dup_x1() {
1068   transition(vtos, vtos);
1069 
1070   Register Ra = R11_scratch1,
1071            Rb = R12_scratch2;
1072   // stack: ..., a, b
1073   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1074   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1075   __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1076   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1077   __ push_ptr(Rb);
1078   // stack: ..., b, a, b
1079 }
1080 
1081 void TemplateTable::dup_x2() {
1082   transition(vtos, vtos);
1083 
1084   Register Ra = R11_scratch1,
1085            Rb = R12_scratch2,
1086            Rc = R3_ARG1;
1087 
1088   // stack: ..., a, b, c
1089   __ ld(Rc, Interpreter::stackElementSize,     R15_esp);  // load c
1090   __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);  // load a
1091   __ std(Rc, Interpreter::stackElementSize * 3, R15_esp); // store c in a
1092   __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);  // load b
1093   // stack: ..., c, b, c
1094   __ std(Ra, Interpreter::stackElementSize * 2, R15_esp); // store a in b
1095   // stack: ..., c, a, c
1096   __ std(Rb, Interpreter::stackElementSize,     R15_esp); // store b in c
1097   __ push_ptr(Rc);                                        // push c
1098   // stack: ..., c, a, b, c
1099 }
1100 
1101 void TemplateTable::dup2() {
1102   transition(vtos, vtos);
1103 
1104   Register Ra = R11_scratch1,
1105            Rb = R12_scratch2;
1106   // stack: ..., a, b
1107   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1108   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1109   __ push_2ptrs(Ra, Rb);
1110   // stack: ..., a, b, a, b
1111 }
1112 
1113 void TemplateTable::dup2_x1() {
1114   transition(vtos, vtos);
1115 
1116   Register Ra = R11_scratch1,
1117            Rb = R12_scratch2,
1118            Rc = R3_ARG1;
1119   // stack: ..., a, b, c
1120   __ ld(Rc, Interpreter::stackElementSize,     R15_esp);
1121   __ ld(Rb, Interpreter::stackElementSize * 2, R15_esp);
1122   __ std(Rc, Interpreter::stackElementSize * 2, R15_esp);
1123   __ ld(Ra, Interpreter::stackElementSize * 3, R15_esp);
1124   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1125   __ std(Rb, Interpreter::stackElementSize * 3, R15_esp);
1126   // stack: ..., b, c, a
1127   __ push_2ptrs(Rb, Rc);
1128   // stack: ..., b, c, a, b, c
1129 }
1130 
1131 void TemplateTable::dup2_x2() {
1132   transition(vtos, vtos);
1133 
1134   Register Ra = R11_scratch1,
1135            Rb = R12_scratch2,
1136            Rc = R3_ARG1,
1137            Rd = R4_ARG2;
1138   // stack: ..., a, b, c, d
1139   __ ld(Rb, Interpreter::stackElementSize * 3, R15_esp);
1140   __ ld(Rd, Interpreter::stackElementSize,     R15_esp);
1141   __ std(Rb, Interpreter::stackElementSize,     R15_esp);  // store b in d
1142   __ std(Rd, Interpreter::stackElementSize * 3, R15_esp);  // store d in b
1143   __ ld(Ra, Interpreter::stackElementSize * 4, R15_esp);
1144   __ ld(Rc, Interpreter::stackElementSize * 2, R15_esp);
1145   __ std(Ra, Interpreter::stackElementSize * 2, R15_esp);  // store a in c
1146   __ std(Rc, Interpreter::stackElementSize * 4, R15_esp);  // store c in a
1147   // stack: ..., c, d, a, b
1148   __ push_2ptrs(Rc, Rd);
1149   // stack: ..., c, d, a, b, c, d
1150 }
1151 
1152 void TemplateTable::swap() {
1153   transition(vtos, vtos);
1154   // stack: ..., a, b
1155 
1156   Register Ra = R11_scratch1,
1157            Rb = R12_scratch2;
1158   // stack: ..., a, b
1159   __ ld(Rb, Interpreter::stackElementSize,     R15_esp);
1160   __ ld(Ra, Interpreter::stackElementSize * 2, R15_esp);
1161   __ std(Rb, Interpreter::stackElementSize * 2, R15_esp);
1162   __ std(Ra, Interpreter::stackElementSize,     R15_esp);
1163   // stack: ..., b, a
1164 }
1165 
1166 void TemplateTable::iop2(Operation op) {
1167   transition(itos, itos);
1168 
1169   Register Rscratch = R11_scratch1;
1170 
1171   __ pop_i(Rscratch);
1172   // tos  = number of bits to shift
1173   // Rscratch = value to shift
1174   switch (op) {
1175     case  add:   __ add(R17_tos, Rscratch, R17_tos); break;
1176     case  sub:   __ sub(R17_tos, Rscratch, R17_tos); break;
1177     case  mul:   __ mullw(R17_tos, Rscratch, R17_tos); break;
1178     case  _and:  __ andr(R17_tos, Rscratch, R17_tos); break;
1179     case  _or:   __ orr(R17_tos, Rscratch, R17_tos); break;
1180     case  _xor:  __ xorr(R17_tos, Rscratch, R17_tos); break;
1181     case  shl:   __ rldicl(R17_tos, R17_tos, 0, 64-5); __ slw(R17_tos, Rscratch, R17_tos); break;
1182     case  shr:   __ rldicl(R17_tos, R17_tos, 0, 64-5); __ sraw(R17_tos, Rscratch, R17_tos); break;
1183     case  ushr:  __ rldicl(R17_tos, R17_tos, 0, 64-5); __ srw(R17_tos, Rscratch, R17_tos); break;
1184     default:     ShouldNotReachHere();
1185   }
1186 }
1187 
1188 void TemplateTable::lop2(Operation op) {
1189   transition(ltos, ltos);
1190 
1191   Register Rscratch = R11_scratch1;
1192   __ pop_l(Rscratch);
1193   switch (op) {
1194     case  add:   __ add(R17_tos, Rscratch, R17_tos); break;
1195     case  sub:   __ sub(R17_tos, Rscratch, R17_tos); break;
1196     case  _and:  __ andr(R17_tos, Rscratch, R17_tos); break;
1197     case  _or:   __ orr(R17_tos, Rscratch, R17_tos); break;
1198     case  _xor:  __ xorr(R17_tos, Rscratch, R17_tos); break;
1199     default:     ShouldNotReachHere();
1200   }
1201 }
1202 
1203 void TemplateTable::idiv() {
1204   transition(itos, itos);
1205 
1206   Label Lnormal, Lexception, Ldone;
1207   Register Rdividend = R11_scratch1; // Used by irem.
1208 
1209   __ addi(R0, R17_tos, 1);
1210   __ cmplwi(CCR0, R0, 2);
1211   __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1212 
1213   __ cmpwi(CCR1, R17_tos, 0);
1214   __ beq(CCR1, Lexception); // divisor == 0
1215 
1216   __ pop_i(Rdividend);
1217   __ mullw(R17_tos, Rdividend, R17_tos); // div by +/-1
1218   __ b(Ldone);
1219 
1220   __ bind(Lexception);
1221   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1222   __ mtctr(R11_scratch1);
1223   __ bctr();
1224 
1225   __ align(32, 12);
1226   __ bind(Lnormal);
1227   __ pop_i(Rdividend);
1228   __ divw(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1229   __ bind(Ldone);
1230 }
1231 
1232 void TemplateTable::irem() {
1233   transition(itos, itos);
1234 
1235   __ mr(R12_scratch2, R17_tos);
1236   idiv();
1237   __ mullw(R17_tos, R17_tos, R12_scratch2);
1238   __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by idiv.
1239 }
1240 
1241 void TemplateTable::lmul() {
1242   transition(ltos, ltos);
1243 
1244   __ pop_l(R11_scratch1);
1245   __ mulld(R17_tos, R11_scratch1, R17_tos);
1246 }
1247 
1248 void TemplateTable::ldiv() {
1249   transition(ltos, ltos);
1250 
1251   Label Lnormal, Lexception, Ldone;
1252   Register Rdividend = R11_scratch1; // Used by lrem.
1253 
1254   __ addi(R0, R17_tos, 1);
1255   __ cmpldi(CCR0, R0, 2);
1256   __ bgt(CCR0, Lnormal); // divisor <-1 or >1
1257 
1258   __ cmpdi(CCR1, R17_tos, 0);
1259   __ beq(CCR1, Lexception); // divisor == 0
1260 
1261   __ pop_l(Rdividend);
1262   __ mulld(R17_tos, Rdividend, R17_tos); // div by +/-1
1263   __ b(Ldone);
1264 
1265   __ bind(Lexception);
1266   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ArithmeticException_entry);
1267   __ mtctr(R11_scratch1);
1268   __ bctr();
1269 
1270   __ align(32, 12);
1271   __ bind(Lnormal);
1272   __ pop_l(Rdividend);
1273   __ divd(R17_tos, Rdividend, R17_tos); // Can't divide minint/-1.
1274   __ bind(Ldone);
1275 }
1276 
1277 void TemplateTable::lrem() {
1278   transition(ltos, ltos);
1279 
1280   __ mr(R12_scratch2, R17_tos);
1281   ldiv();
1282   __ mulld(R17_tos, R17_tos, R12_scratch2);
1283   __ subf(R17_tos, R17_tos, R11_scratch1); // Dividend set by ldiv.
1284 }
1285 
1286 void TemplateTable::lshl() {
1287   transition(itos, ltos);
1288 
1289   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1290   __ pop_l(R11_scratch1);
1291   __ sld(R17_tos, R11_scratch1, R17_tos);
1292 }
1293 
1294 void TemplateTable::lshr() {
1295   transition(itos, ltos);
1296 
1297   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1298   __ pop_l(R11_scratch1);
1299   __ srad(R17_tos, R11_scratch1, R17_tos);
1300 }
1301 
1302 void TemplateTable::lushr() {
1303   transition(itos, ltos);
1304 
1305   __ rldicl(R17_tos, R17_tos, 0, 64-6); // Extract least significant bits.
1306   __ pop_l(R11_scratch1);
1307   __ srd(R17_tos, R11_scratch1, R17_tos);
1308 }
1309 
1310 void TemplateTable::fop2(Operation op) {
1311   transition(ftos, ftos);
1312 
1313   switch (op) {
1314     case add: __ pop_f(F0_SCRATCH); __ fadds(F15_ftos, F0_SCRATCH, F15_ftos); break;
1315     case sub: __ pop_f(F0_SCRATCH); __ fsubs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1316     case mul: __ pop_f(F0_SCRATCH); __ fmuls(F15_ftos, F0_SCRATCH, F15_ftos); break;
1317     case div: __ pop_f(F0_SCRATCH); __ fdivs(F15_ftos, F0_SCRATCH, F15_ftos); break;
1318     case rem:
1319       __ pop_f(F1_ARG1);
1320       __ fmr(F2_ARG2, F15_ftos);
1321       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1322       __ fmr(F15_ftos, F1_RET);
1323       break;
1324 
1325     default: ShouldNotReachHere();
1326   }
1327 }
1328 
1329 void TemplateTable::dop2(Operation op) {
1330   transition(dtos, dtos);
1331 
1332   switch (op) {
1333     case add: __ pop_d(F0_SCRATCH); __ fadd(F15_ftos, F0_SCRATCH, F15_ftos); break;
1334     case sub: __ pop_d(F0_SCRATCH); __ fsub(F15_ftos, F0_SCRATCH, F15_ftos); break;
1335     case mul: __ pop_d(F0_SCRATCH); __ fmul(F15_ftos, F0_SCRATCH, F15_ftos); break;
1336     case div: __ pop_d(F0_SCRATCH); __ fdiv(F15_ftos, F0_SCRATCH, F15_ftos); break;
1337     case rem:
1338       __ pop_d(F1_ARG1);
1339       __ fmr(F2_ARG2, F15_ftos);
1340       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1341       __ fmr(F15_ftos, F1_RET);
1342       break;
1343 
1344     default: ShouldNotReachHere();
1345   }
1346 }
1347 
1348 // Negate the value in the TOS cache.
1349 void TemplateTable::ineg() {
1350   transition(itos, itos);
1351 
1352   __ neg(R17_tos, R17_tos);
1353 }
1354 
1355 // Negate the value in the TOS cache.
1356 void TemplateTable::lneg() {
1357   transition(ltos, ltos);
1358 
1359   __ neg(R17_tos, R17_tos);
1360 }
1361 
1362 void TemplateTable::fneg() {
1363   transition(ftos, ftos);
1364 
1365   __ fneg(F15_ftos, F15_ftos);
1366 }
1367 
1368 void TemplateTable::dneg() {
1369   transition(dtos, dtos);
1370 
1371   __ fneg(F15_ftos, F15_ftos);
1372 }
1373 
1374 // Increments a local variable in place.
1375 void TemplateTable::iinc() {
1376   transition(vtos, vtos);
1377 
1378   const Register Rindex     = R11_scratch1,
1379                  Rincrement = R0,
1380                  Rvalue     = R12_scratch2;
1381 
1382   locals_index(Rindex);              // Load locals index from bytecode stream.
1383   __ lbz(Rincrement, 2, R14_bcp);    // Load increment from the bytecode stream.
1384   __ extsb(Rincrement, Rincrement);
1385 
1386   __ load_local_int(Rvalue, Rindex, Rindex); // Puts address of local into Rindex.
1387 
1388   __ add(Rvalue, Rincrement, Rvalue);
1389   __ stw(Rvalue, 0, Rindex);
1390 }
1391 
1392 void TemplateTable::wide_iinc() {
1393   transition(vtos, vtos);
1394 
1395   Register Rindex       = R11_scratch1,
1396            Rlocals_addr = Rindex,
1397            Rincr        = R12_scratch2;
1398   locals_index_wide(Rindex);
1399   __ get_2_byte_integer_at_bcp(4, Rincr, InterpreterMacroAssembler::Signed);
1400   __ load_local_int(R17_tos, Rlocals_addr, Rindex);
1401   __ add(R17_tos, Rincr, R17_tos);
1402   __ stw(R17_tos, 0, Rlocals_addr);
1403 }
1404 
1405 void TemplateTable::convert() {
1406   // %%%%% Factor this first part accross platforms
1407 #ifdef ASSERT
1408   TosState tos_in  = ilgl;
1409   TosState tos_out = ilgl;
1410   switch (bytecode()) {
1411     case Bytecodes::_i2l: // fall through
1412     case Bytecodes::_i2f: // fall through
1413     case Bytecodes::_i2d: // fall through
1414     case Bytecodes::_i2b: // fall through
1415     case Bytecodes::_i2c: // fall through
1416     case Bytecodes::_i2s: tos_in = itos; break;
1417     case Bytecodes::_l2i: // fall through
1418     case Bytecodes::_l2f: // fall through
1419     case Bytecodes::_l2d: tos_in = ltos; break;
1420     case Bytecodes::_f2i: // fall through
1421     case Bytecodes::_f2l: // fall through
1422     case Bytecodes::_f2d: tos_in = ftos; break;
1423     case Bytecodes::_d2i: // fall through
1424     case Bytecodes::_d2l: // fall through
1425     case Bytecodes::_d2f: tos_in = dtos; break;
1426     default             : ShouldNotReachHere();
1427   }
1428   switch (bytecode()) {
1429     case Bytecodes::_l2i: // fall through
1430     case Bytecodes::_f2i: // fall through
1431     case Bytecodes::_d2i: // fall through
1432     case Bytecodes::_i2b: // fall through
1433     case Bytecodes::_i2c: // fall through
1434     case Bytecodes::_i2s: tos_out = itos; break;
1435     case Bytecodes::_i2l: // fall through
1436     case Bytecodes::_f2l: // fall through
1437     case Bytecodes::_d2l: tos_out = ltos; break;
1438     case Bytecodes::_i2f: // fall through
1439     case Bytecodes::_l2f: // fall through
1440     case Bytecodes::_d2f: tos_out = ftos; break;
1441     case Bytecodes::_i2d: // fall through
1442     case Bytecodes::_l2d: // fall through
1443     case Bytecodes::_f2d: tos_out = dtos; break;
1444     default             : ShouldNotReachHere();
1445   }
1446   transition(tos_in, tos_out);
1447 #endif
1448 
1449   // Conversion
1450   Label done;
1451   switch (bytecode()) {
1452     case Bytecodes::_i2l:
1453       __ extsw(R17_tos, R17_tos);
1454       break;
1455 
1456     case Bytecodes::_l2i:
1457       // Nothing to do, we'll continue to work with the lower bits.
1458       break;
1459 
1460     case Bytecodes::_i2b:
1461       __ extsb(R17_tos, R17_tos);
1462       break;
1463 
1464     case Bytecodes::_i2c:
1465       __ rldicl(R17_tos, R17_tos, 0, 64-2*8);
1466       break;
1467 
1468     case Bytecodes::_i2s:
1469       __ extsh(R17_tos, R17_tos);
1470       break;
1471 
1472     case Bytecodes::_i2d:
1473       __ extsw(R17_tos, R17_tos);
1474     case Bytecodes::_l2d:
1475       __ move_l_to_d();
1476       __ fcfid(F15_ftos, F15_ftos);
1477       break;
1478 
1479     case Bytecodes::_i2f:
1480       __ extsw(R17_tos, R17_tos);
1481       __ move_l_to_d();
1482       if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1483         // Comment: alternatively, load with sign extend could be done by lfiwax.
1484         __ fcfids(F15_ftos, F15_ftos);
1485       } else {
1486         __ fcfid(F15_ftos, F15_ftos);
1487         __ frsp(F15_ftos, F15_ftos);
1488       }
1489       break;
1490 
1491     case Bytecodes::_l2f:
1492       if (VM_Version::has_fcfids()) { // fcfids is >= Power7 only
1493         __ move_l_to_d();
1494         __ fcfids(F15_ftos, F15_ftos);
1495       } else {
1496         // Avoid rounding problem when result should be 0x3f800001: need fixup code before fcfid+frsp.
1497         __ mr(R3_ARG1, R17_tos);
1498         __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::l2f));
1499         __ fmr(F15_ftos, F1_RET);
1500       }
1501       break;
1502 
1503     case Bytecodes::_f2d:
1504       // empty
1505       break;
1506 
1507     case Bytecodes::_d2f:
1508       __ frsp(F15_ftos, F15_ftos);
1509       break;
1510 
1511     case Bytecodes::_d2i:
1512     case Bytecodes::_f2i:
1513       __ fcmpu(CCR0, F15_ftos, F15_ftos);
1514       __ li(R17_tos, 0); // 0 in case of NAN
1515       __ bso(CCR0, done);
1516       __ fctiwz(F15_ftos, F15_ftos);
1517       __ move_d_to_l();
1518       break;
1519 
1520     case Bytecodes::_d2l:
1521     case Bytecodes::_f2l:
1522       __ fcmpu(CCR0, F15_ftos, F15_ftos);
1523       __ li(R17_tos, 0); // 0 in case of NAN
1524       __ bso(CCR0, done);
1525       __ fctidz(F15_ftos, F15_ftos);
1526       __ move_d_to_l();
1527       break;
1528 
1529     default: ShouldNotReachHere();
1530   }
1531   __ bind(done);
1532 }
1533 
1534 // Long compare
1535 void TemplateTable::lcmp() {
1536   transition(ltos, itos);
1537 
1538   const Register Rscratch = R11_scratch1;
1539   __ pop_l(Rscratch); // first operand, deeper in stack
1540 
1541   __ cmpd(CCR0, Rscratch, R17_tos); // compare
1542   __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1543   __ srwi(Rscratch, R17_tos, 30);
1544   __ srawi(R17_tos, R17_tos, 31);
1545   __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1546 }
1547 
1548 // fcmpl/fcmpg and dcmpl/dcmpg bytecodes
1549 // unordered_result == -1 => fcmpl or dcmpl
1550 // unordered_result ==  1 => fcmpg or dcmpg
1551 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1552   const FloatRegister Rfirst  = F0_SCRATCH,
1553                       Rsecond = F15_ftos;
1554   const Register Rscratch = R11_scratch1;
1555 
1556   if (is_float) {
1557     __ pop_f(Rfirst);
1558   } else {
1559     __ pop_d(Rfirst);
1560   }
1561 
1562   Label Lunordered, Ldone;
1563   __ fcmpu(CCR0, Rfirst, Rsecond); // compare
1564   if (unordered_result) {
1565     __ bso(CCR0, Lunordered);
1566   }
1567   __ mfcr(R17_tos); // set bit 32..33 as follows: <: 0b10, =: 0b00, >: 0b01
1568   __ srwi(Rscratch, R17_tos, 30);
1569   __ srawi(R17_tos, R17_tos, 31);
1570   __ orr(R17_tos, Rscratch, R17_tos); // set result as follows: <: -1, =: 0, >: 1
1571   if (unordered_result) {
1572     __ b(Ldone);
1573     __ bind(Lunordered);
1574     __ load_const_optimized(R17_tos, unordered_result);
1575   }
1576   __ bind(Ldone);
1577 }
1578 
1579 // Branch_conditional which takes TemplateTable::Condition.
1580 void TemplateTable::branch_conditional(ConditionRegister crx, TemplateTable::Condition cc, Label& L, bool invert) {
1581   bool positive = false;
1582   Assembler::Condition cond = Assembler::equal;
1583   switch (cc) {
1584     case TemplateTable::equal:         positive = true ; cond = Assembler::equal  ; break;
1585     case TemplateTable::not_equal:     positive = false; cond = Assembler::equal  ; break;
1586     case TemplateTable::less:          positive = true ; cond = Assembler::less   ; break;
1587     case TemplateTable::less_equal:    positive = false; cond = Assembler::greater; break;
1588     case TemplateTable::greater:       positive = true ; cond = Assembler::greater; break;
1589     case TemplateTable::greater_equal: positive = false; cond = Assembler::less   ; break;
1590     default: ShouldNotReachHere();
1591   }
1592   int bo = (positive != invert) ? Assembler::bcondCRbiIs1 : Assembler::bcondCRbiIs0;
1593   int bi = Assembler::bi0(crx, cond);
1594   __ bc(bo, bi, L);
1595 }
1596 
1597 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1598 
1599   // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1600   __ verify_thread();
1601 
1602   const Register Rscratch1    = R11_scratch1,
1603                  Rscratch2    = R12_scratch2,
1604                  Rscratch3    = R3_ARG1,
1605                  R4_counters  = R4_ARG2,
1606                  bumped_count = R31,
1607                  Rdisp        = R22_tmp2;
1608 
1609   __ profile_taken_branch(Rscratch1, bumped_count);
1610 
1611   // Get (wide) offset.
1612   if (is_wide) {
1613     __ get_4_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1614   } else {
1615     __ get_2_byte_integer_at_bcp(1, Rdisp, InterpreterMacroAssembler::Signed);
1616   }
1617 
1618   // --------------------------------------------------------------------------
1619   // Handle all the JSR stuff here, then exit.
1620   // It's much shorter and cleaner than intermingling with the
1621   // non-JSR normal-branch stuff occurring below.
1622   if (is_jsr) {
1623     // Compute return address as bci in Otos_i.
1624     __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1625     __ addi(Rscratch2, R14_bcp, -in_bytes(ConstMethod::codes_offset()) + (is_wide ? 5 : 3));
1626     __ subf(R17_tos, Rscratch1, Rscratch2);
1627 
1628     // Bump bcp to target of JSR.
1629     __ add(R14_bcp, Rdisp, R14_bcp);
1630     // Push returnAddress for "ret" on stack.
1631     __ push_ptr(R17_tos);
1632     // And away we go!
1633     __ dispatch_next(vtos);
1634     return;
1635   }
1636 
1637   // --------------------------------------------------------------------------
1638   // Normal (non-jsr) branch handling
1639 
1640   // Bump bytecode pointer by displacement (take the branch).
1641   __ add(R14_bcp, Rdisp, R14_bcp); // Add to bc addr.
1642 
1643   const bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1644   if (increment_invocation_counter_for_backward_branches) {
1645     Label Lforward;
1646     __ dispatch_prolog(vtos);
1647 
1648     // Check branch direction.
1649     __ cmpdi(CCR0, Rdisp, 0);
1650     __ bgt(CCR0, Lforward);
1651 
1652     __ get_method_counters(R19_method, R4_counters, Lforward);
1653 
1654     if (TieredCompilation) {
1655       Label Lno_mdo, Loverflow;
1656       const int increment = InvocationCounter::count_increment;
1657       if (ProfileInterpreter) {
1658         Register Rmdo = Rscratch1;
1659 
1660         // If no method data exists, go to profile_continue.
1661         __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
1662         __ cmpdi(CCR0, Rmdo, 0);
1663         __ beq(CCR0, Lno_mdo);
1664 
1665         // Increment backedge counter in the MDO.
1666         const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1667         __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
1668         __ lwz(Rscratch3, in_bytes(MethodData::backedge_mask_offset()), Rmdo);
1669         __ addi(Rscratch2, Rscratch2, increment);
1670         __ stw(Rscratch2, mdo_bc_offs, Rmdo);
1671         if (UseOnStackReplacement) {
1672           __ and_(Rscratch3, Rscratch2, Rscratch3);
1673           __ bne(CCR0, Lforward);
1674           __ b(Loverflow);
1675         } else {
1676           __ b(Lforward);
1677         }
1678       }
1679 
1680       // If there's no MDO, increment counter in method.
1681       const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
1682       __ bind(Lno_mdo);
1683       __ lwz(Rscratch2, mo_bc_offs, R4_counters);
1684       __ lwz(Rscratch3, in_bytes(MethodCounters::backedge_mask_offset()), R4_counters);
1685       __ addi(Rscratch2, Rscratch2, increment);
1686       __ stw(Rscratch2, mo_bc_offs, R4_counters);
1687       if (UseOnStackReplacement) {
1688         __ and_(Rscratch3, Rscratch2, Rscratch3);
1689         __ bne(CCR0, Lforward);
1690       } else {
1691         __ b(Lforward);
1692       }
1693       __ bind(Loverflow);
1694 
1695       // Notify point for loop, pass branch bytecode.
1696       __ subf(R4_ARG2, Rdisp, R14_bcp); // Compute branch bytecode (previous bcp).
1697       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
1698 
1699       // Was an OSR adapter generated?
1700       __ cmpdi(CCR0, R3_RET, 0);
1701       __ beq(CCR0, Lforward);
1702 
1703       // Has the nmethod been invalidated already?
1704       __ lbz(R0, nmethod::state_offset(), R3_RET);
1705       __ cmpwi(CCR0, R0, nmethod::in_use);
1706       __ bne(CCR0, Lforward);
1707 
1708       // Migrate the interpreter frame off of the stack.
1709       // We can use all registers because we will not return to interpreter from this point.
1710 
1711       // Save nmethod.
1712       const Register osr_nmethod = R31;
1713       __ mr(osr_nmethod, R3_RET);
1714       __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1);
1715       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread);
1716       __ reset_last_Java_frame();
1717       // OSR buffer is in ARG1.
1718 
1719       // Remove the interpreter frame.
1720       __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1721 
1722       // Jump to the osr code.
1723       __ ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod);
1724       __ mtlr(R0);
1725       __ mtctr(R11_scratch1);
1726       __ bctr();
1727 
1728     } else {
1729 
1730       const Register invoke_ctr = Rscratch1;
1731       // Update Backedge branch separately from invocations.
1732       __ increment_backedge_counter(R4_counters, invoke_ctr, Rscratch2, Rscratch3);
1733 
1734       if (ProfileInterpreter) {
1735         __ test_invocation_counter_for_mdp(invoke_ctr, R4_counters, Rscratch2, Lforward);
1736         if (UseOnStackReplacement) {
1737           __ test_backedge_count_for_osr(bumped_count, R4_counters, R14_bcp, Rdisp, Rscratch2);
1738         }
1739       } else {
1740         if (UseOnStackReplacement) {
1741           __ test_backedge_count_for_osr(invoke_ctr, R4_counters, R14_bcp, Rdisp, Rscratch2);
1742         }
1743       }
1744     }
1745 
1746     __ bind(Lforward);
1747     __ dispatch_epilog(vtos);
1748 
1749   } else {
1750     __ dispatch_next(vtos);
1751   }
1752 }
1753 
1754 // Helper function for if_cmp* methods below.
1755 // Factored out common compare and branch code.
1756 void TemplateTable::if_cmp_common(Register Rfirst, Register Rsecond, Register Rscratch1, Register Rscratch2, Condition cc, bool is_jint, bool cmp0) {
1757   Label Lnot_taken;
1758   // Note: The condition code we get is the condition under which we
1759   // *fall through*! So we have to inverse the CC here.
1760 
1761   if (is_jint) {
1762     if (cmp0) {
1763       __ cmpwi(CCR0, Rfirst, 0);
1764     } else {
1765       __ cmpw(CCR0, Rfirst, Rsecond);
1766     }
1767   } else {
1768     if (cmp0) {
1769       __ cmpdi(CCR0, Rfirst, 0);
1770     } else {
1771       __ cmpd(CCR0, Rfirst, Rsecond);
1772     }
1773   }
1774   branch_conditional(CCR0, cc, Lnot_taken, /*invert*/ true);
1775 
1776   // Conition is false => Jump!
1777   branch(false, false);
1778 
1779   // Condition is not true => Continue.
1780   __ align(32, 12);
1781   __ bind(Lnot_taken);
1782   __ profile_not_taken_branch(Rscratch1, Rscratch2);
1783 }
1784 
1785 // Compare integer values with zero and fall through if CC holds, branch away otherwise.
1786 void TemplateTable::if_0cmp(Condition cc) {
1787   transition(itos, vtos);
1788 
1789   if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, true, true);
1790 }
1791 
1792 // Compare integer values and fall through if CC holds, branch away otherwise.
1793 //
1794 // Interface:
1795 //  - Rfirst: First operand  (older stack value)
1796 //  - tos:    Second operand (younger stack value)
1797 void TemplateTable::if_icmp(Condition cc) {
1798   transition(itos, vtos);
1799 
1800   const Register Rfirst  = R0,
1801                  Rsecond = R17_tos;
1802 
1803   __ pop_i(Rfirst);
1804   if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, true, false);
1805 }
1806 
1807 void TemplateTable::if_nullcmp(Condition cc) {
1808   transition(atos, vtos);
1809 
1810   if_cmp_common(R17_tos, noreg, R11_scratch1, R12_scratch2, cc, false, true);
1811 }
1812 
1813 void TemplateTable::if_acmp(Condition cc) {
1814   transition(atos, vtos);
1815 
1816   const Register Rfirst  = R0,
1817                  Rsecond = R17_tos;
1818 
1819   __ pop_ptr(Rfirst);
1820   if_cmp_common(Rfirst, Rsecond, R11_scratch1, R12_scratch2, cc, false, false);
1821 }
1822 
1823 void TemplateTable::ret() {
1824   locals_index(R11_scratch1);
1825   __ load_local_ptr(R17_tos, R11_scratch1, R11_scratch1);
1826 
1827   __ profile_ret(vtos, R17_tos, R11_scratch1, R12_scratch2);
1828 
1829   __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
1830   __ add(R11_scratch1, R17_tos, R11_scratch1);
1831   __ addi(R14_bcp, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
1832   __ dispatch_next(vtos);
1833 }
1834 
1835 void TemplateTable::wide_ret() {
1836   transition(vtos, vtos);
1837 
1838   const Register Rindex = R3_ARG1,
1839                  Rscratch1 = R11_scratch1,
1840                  Rscratch2 = R12_scratch2;
1841 
1842   locals_index_wide(Rindex);
1843   __ load_local_ptr(R17_tos, R17_tos, Rindex);
1844   __ profile_ret(vtos, R17_tos, Rscratch1, R12_scratch2);
1845   // Tos now contains the bci, compute the bcp from that.
1846   __ ld(Rscratch1, in_bytes(Method::const_offset()), R19_method);
1847   __ addi(Rscratch2, R17_tos, in_bytes(ConstMethod::codes_offset()));
1848   __ add(R14_bcp, Rscratch1, Rscratch2);
1849   __ dispatch_next(vtos);
1850 }
1851 
1852 void TemplateTable::tableswitch() {
1853   transition(itos, vtos);
1854 
1855   Label Ldispatch, Ldefault_case;
1856   Register Rlow_byte         = R3_ARG1,
1857            Rindex            = Rlow_byte,
1858            Rhigh_byte        = R4_ARG2,
1859            Rdef_offset_addr  = R5_ARG3, // is going to contain address of default offset
1860            Rscratch1         = R11_scratch1,
1861            Rscratch2         = R12_scratch2,
1862            Roffset           = R6_ARG4;
1863 
1864   // Align bcp.
1865   __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1866   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1867 
1868   // Load lo & hi.
1869   __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1870   __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
1871 
1872   // Check for default case (=index outside [low,high]).
1873   __ cmpw(CCR0, R17_tos, Rlow_byte);
1874   __ cmpw(CCR1, R17_tos, Rhigh_byte);
1875   __ blt(CCR0, Ldefault_case);
1876   __ bgt(CCR1, Ldefault_case);
1877 
1878   // Lookup dispatch offset.
1879   __ sub(Rindex, R17_tos, Rlow_byte);
1880   __ extsw(Rindex, Rindex);
1881   __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
1882   __ sldi(Rindex, Rindex, LogBytesPerInt);
1883   __ addi(Rindex, Rindex, 3 * BytesPerInt);
1884 #if defined(VM_LITTLE_ENDIAN)
1885   __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
1886   __ extsw(Roffset, Roffset);
1887 #else
1888   __ lwax(Roffset, Rdef_offset_addr, Rindex);
1889 #endif
1890   __ b(Ldispatch);
1891 
1892   __ bind(Ldefault_case);
1893   __ profile_switch_default(Rhigh_byte, Rscratch1);
1894   __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1895 
1896   __ bind(Ldispatch);
1897 
1898   __ add(R14_bcp, Roffset, R14_bcp);
1899   __ dispatch_next(vtos);
1900 }
1901 
1902 void TemplateTable::lookupswitch() {
1903   transition(itos, itos);
1904   __ stop("lookupswitch bytecode should have been rewritten");
1905 }
1906 
1907 // Table switch using linear search through cases.
1908 // Bytecode stream format:
1909 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1910 // Note: Everything is big-endian format here.
1911 void TemplateTable::fast_linearswitch() {
1912   transition(itos, vtos);
1913 
1914   Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
1915   Register Rcount           = R3_ARG1,
1916            Rcurrent_pair    = R4_ARG2,
1917            Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
1918            Roffset          = R31,     // Might need to survive C call.
1919            Rvalue           = R12_scratch2,
1920            Rscratch         = R11_scratch1,
1921            Rcmp_value       = R17_tos;
1922 
1923   // Align bcp.
1924   __ addi(Rdef_offset_addr, R14_bcp, BytesPerInt);
1925   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
1926 
1927   // Setup loop counter and limit.
1928   __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
1929   __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
1930 
1931   __ mtctr(Rcount);
1932   __ cmpwi(CCR0, Rcount, 0);
1933   __ bne(CCR0, Lloop_entry);
1934 
1935   // Default case
1936   __ bind(Ldefault_case);
1937   __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
1938   if (ProfileInterpreter) {
1939     __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
1940   }
1941   __ b(Lcontinue_execution);
1942 
1943   // Next iteration
1944   __ bind(Lsearch_loop);
1945   __ bdz(Ldefault_case);
1946   __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
1947   __ bind(Lloop_entry);
1948   __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
1949   __ cmpw(CCR0, Rvalue, Rcmp_value);
1950   __ bne(CCR0, Lsearch_loop);
1951 
1952   // Found, load offset.
1953   __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
1954   // Calculate case index and profile
1955   __ mfctr(Rcurrent_pair);
1956   if (ProfileInterpreter) {
1957     __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
1958     __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
1959   }
1960 
1961   __ bind(Lcontinue_execution);
1962   __ add(R14_bcp, Roffset, R14_bcp);
1963   __ dispatch_next(vtos);
1964 }
1965 
1966 // Table switch using binary search (value/offset pairs are ordered).
1967 // Bytecode stream format:
1968 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
1969 // Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
1970 void TemplateTable::fast_binaryswitch() {
1971 
1972   transition(itos, vtos);
1973   // Implementation using the following core algorithm: (copied from Intel)
1974   //
1975   // int binary_search(int key, LookupswitchPair* array, int n) {
1976   //   // Binary search according to "Methodik des Programmierens" by
1977   //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1978   //   int i = 0;
1979   //   int j = n;
1980   //   while (i+1 < j) {
1981   //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1982   //     // with      Q: for all i: 0 <= i < n: key < a[i]
1983   //     // where a stands for the array and assuming that the (inexisting)
1984   //     // element a[n] is infinitely big.
1985   //     int h = (i + j) >> 1;
1986   //     // i < h < j
1987   //     if (key < array[h].fast_match()) {
1988   //       j = h;
1989   //     } else {
1990   //       i = h;
1991   //     }
1992   //   }
1993   //   // R: a[i] <= key < a[i+1] or Q
1994   //   // (i.e., if key is within array, i is the correct index)
1995   //   return i;
1996   // }
1997 
1998   // register allocation
1999   const Register Rkey     = R17_tos;          // already set (tosca)
2000   const Register Rarray   = R3_ARG1;
2001   const Register Ri       = R4_ARG2;
2002   const Register Rj       = R5_ARG3;
2003   const Register Rh       = R6_ARG4;
2004   const Register Rscratch = R11_scratch1;
2005 
2006   const int log_entry_size = 3;
2007   const int entry_size = 1 << log_entry_size;
2008 
2009   Label found;
2010 
2011   // Find Array start,
2012   __ addi(Rarray, R14_bcp, 3 * BytesPerInt);
2013   __ clrrdi(Rarray, Rarray, log2_long((jlong)BytesPerInt));
2014 
2015   // initialize i & j
2016   __ li(Ri,0);
2017   __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
2018 
2019   // and start.
2020   Label entry;
2021   __ b(entry);
2022 
2023   // binary search loop
2024   { Label loop;
2025     __ bind(loop);
2026     // int h = (i + j) >> 1;
2027     __ srdi(Rh, Rh, 1);
2028     // if (key < array[h].fast_match()) {
2029     //   j = h;
2030     // } else {
2031     //   i = h;
2032     // }
2033     __ sldi(Rscratch, Rh, log_entry_size);
2034 #if defined(VM_LITTLE_ENDIAN)
2035     __ lwbrx(Rscratch, Rscratch, Rarray);
2036 #else
2037     __ lwzx(Rscratch, Rscratch, Rarray);
2038 #endif
2039 
2040     // if (key < current value)
2041     //   Rh = Rj
2042     // else
2043     //   Rh = Ri
2044     Label Lgreater;
2045     __ cmpw(CCR0, Rkey, Rscratch);
2046     __ bge(CCR0, Lgreater);
2047     __ mr(Rj, Rh);
2048     __ b(entry);
2049     __ bind(Lgreater);
2050     __ mr(Ri, Rh);
2051 
2052     // while (i+1 < j)
2053     __ bind(entry);
2054     __ addi(Rscratch, Ri, 1);
2055     __ cmpw(CCR0, Rscratch, Rj);
2056     __ add(Rh, Ri, Rj); // start h = i + j >> 1;
2057 
2058     __ blt(CCR0, loop);
2059   }
2060 
2061   // End of binary search, result index is i (must check again!).
2062   Label default_case;
2063   Label continue_execution;
2064   if (ProfileInterpreter) {
2065     __ mr(Rh, Ri);              // Save index in i for profiling.
2066   }
2067   // Ri = value offset
2068   __ sldi(Ri, Ri, log_entry_size);
2069   __ add(Ri, Ri, Rarray);
2070   __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
2071 
2072   Label not_found;
2073   // Ri = offset offset
2074   __ cmpw(CCR0, Rkey, Rscratch);
2075   __ beq(CCR0, not_found);
2076   // entry not found -> j = default offset
2077   __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
2078   __ b(default_case);
2079 
2080   __ bind(not_found);
2081   // entry found -> j = offset
2082   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
2083   __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
2084 
2085   if (ProfileInterpreter) {
2086     __ b(continue_execution);
2087   }
2088 
2089   __ bind(default_case); // fall through (if not profiling)
2090   __ profile_switch_default(Ri, Rscratch);
2091 
2092   __ bind(continue_execution);
2093 
2094   __ extsw(Rj, Rj);
2095   __ add(R14_bcp, Rj, R14_bcp);
2096   __ dispatch_next(vtos);
2097 }
2098 
2099 void TemplateTable::_return(TosState state) {
2100   transition(state, state);
2101   assert(_desc->calls_vm(),
2102          "inconsistent calls_vm information"); // call in remove_activation
2103 
2104   if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2105 
2106     Register Rscratch     = R11_scratch1,
2107              Rklass       = R12_scratch2,
2108              Rklass_flags = Rklass;
2109     Label Lskip_register_finalizer;
2110 
2111     // Check if the method has the FINALIZER flag set and call into the VM to finalize in this case.
2112     assert(state == vtos, "only valid state");
2113     __ ld(R17_tos, 0, R18_locals);
2114 
2115     // Load klass of this obj.
2116     __ load_klass(Rklass, R17_tos);
2117     __ lwz(Rklass_flags, in_bytes(Klass::access_flags_offset()), Rklass);
2118     __ testbitdi(CCR0, R0, Rklass_flags, exact_log2(JVM_ACC_HAS_FINALIZER));
2119     __ bfalse(CCR0, Lskip_register_finalizer);
2120 
2121     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), R17_tos /* obj */);
2122 
2123     __ align(32, 12);
2124     __ bind(Lskip_register_finalizer);
2125   }
2126 
2127   // Move the result value into the correct register and remove memory stack frame.
2128   __ remove_activation(state, /* throw_monitor_exception */ true);
2129   // Restoration of lr done by remove_activation.
2130   switch (state) {
2131     // Narrow result if state is itos but result type is smaller.
2132     // Need to narrow in the return bytecode rather than in generate_return_entry
2133     // since compiled code callers expect the result to already be narrowed.
2134     case itos: __ narrow(R17_tos); /* fall through */
2135     case ltos:
2136     case atos: __ mr(R3_RET, R17_tos); break;
2137     case ftos:
2138     case dtos: __ fmr(F1_RET, F15_ftos); break;
2139     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2140                // to get visible before the reference to the object gets stored anywhere.
2141                __ membar(Assembler::StoreStore); break;
2142     default  : ShouldNotReachHere();
2143   }
2144   __ blr();
2145 }
2146 
2147 // ============================================================================
2148 // Constant pool cache access
2149 //
2150 // Memory ordering:
2151 //
2152 // Like done in C++ interpreter, we load the fields
2153 //   - _indices
2154 //   - _f12_oop
2155 // acquired, because these are asked if the cache is already resolved. We don't
2156 // want to float loads above this check.
2157 // See also comments in ConstantPoolCacheEntry::bytecode_1(),
2158 // ConstantPoolCacheEntry::bytecode_2() and ConstantPoolCacheEntry::f1();
2159 
2160 // Call into the VM if call site is not yet resolved
2161 //
2162 // Input regs:
2163 //   - None, all passed regs are outputs.
2164 //
2165 // Returns:
2166 //   - Rcache:  The const pool cache entry that contains the resolved result.
2167 //   - Rresult: Either noreg or output for f1/f2.
2168 //
2169 // Kills:
2170 //   - Rscratch
2171 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register Rscratch, size_t index_size) {
2172 
2173   __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2174   Label Lresolved, Ldone;
2175 
2176   Bytecodes::Code code = bytecode();
2177   switch (code) {
2178   case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2179   case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2180   }
2181 
2182   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2183   // We are resolved if the indices offset contains the current bytecode.
2184 #if defined(VM_LITTLE_ENDIAN)
2185   __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
2186 #else
2187   __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
2188 #endif
2189   // Acquire by cmp-br-isync (see below).
2190   __ cmpdi(CCR0, Rscratch, (int)code);
2191   __ beq(CCR0, Lresolved);
2192 
2193   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2194   __ li(R4_ARG2, code);
2195   __ call_VM(noreg, entry, R4_ARG2, true);
2196 
2197   // Update registers with resolved info.
2198   __ get_cache_and_index_at_bcp(Rcache, 1, index_size);
2199   __ b(Ldone);
2200 
2201   __ bind(Lresolved);
2202   __ isync(); // Order load wrt. succeeding loads.
2203   __ bind(Ldone);
2204 }
2205 
2206 // Load the constant pool cache entry at field accesses into registers.
2207 // The Rcache and Rindex registers must be set before call.
2208 // Input:
2209 //   - Rcache, Rindex
2210 // Output:
2211 //   - Robj, Roffset, Rflags
2212 void TemplateTable::load_field_cp_cache_entry(Register Robj,
2213                                               Register Rcache,
2214                                               Register Rindex /* unused on PPC64 */,
2215                                               Register Roffset,
2216                                               Register Rflags,
2217                                               bool is_static = false) {
2218   assert_different_registers(Rcache, Rflags, Roffset);
2219   // assert(Rindex == noreg, "parameter not used on PPC64");
2220 
2221   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2222   __ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache);
2223   __ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcache);
2224   if (is_static) {
2225     __ ld(Robj, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f1_offset()), Rcache);
2226     __ ld(Robj, in_bytes(Klass::java_mirror_offset()), Robj);
2227     __ resolve_oop_handle(Robj);
2228     // Acquire not needed here. Following access has an address dependency on this value.
2229   }
2230 }
2231 
2232 // Load the constant pool cache entry at invokes into registers.
2233 // Resolve if necessary.
2234 
2235 // Input Registers:
2236 //   - None, bcp is used, though
2237 //
2238 // Return registers:
2239 //   - Rmethod       (f1 field or f2 if invokevirtual)
2240 //   - Ritable_index (f2 field)
2241 //   - Rflags        (flags field)
2242 //
2243 // Kills:
2244 //   - R21
2245 //
2246 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2247                                                Register Rmethod,
2248                                                Register Ritable_index,
2249                                                Register Rflags,
2250                                                bool is_invokevirtual,
2251                                                bool is_invokevfinal,
2252                                                bool is_invokedynamic) {
2253 
2254   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2255   // Determine constant pool cache field offsets.
2256   assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2257   const int method_offset = in_bytes(cp_base_offset + (is_invokevirtual ? ConstantPoolCacheEntry::f2_offset() : ConstantPoolCacheEntry::f1_offset()));
2258   const int flags_offset  = in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset());
2259   // Access constant pool cache fields.
2260   const int index_offset  = in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset());
2261 
2262   Register Rcache = R21_tmp1; // Note: same register as R21_sender_SP.
2263 
2264   if (is_invokevfinal) {
2265     assert(Ritable_index == noreg, "register not used");
2266     // Already resolved.
2267     __ get_cache_and_index_at_bcp(Rcache, 1);
2268   } else {
2269     resolve_cache_and_index(byte_no, Rcache, R0, is_invokedynamic ? sizeof(u4) : sizeof(u2));
2270   }
2271 
2272   __ ld(Rmethod, method_offset, Rcache);
2273   __ ld(Rflags, flags_offset, Rcache);
2274 
2275   if (Ritable_index != noreg) {
2276     __ ld(Ritable_index, index_offset, Rcache);
2277   }
2278 }
2279 
2280 // ============================================================================
2281 // Field access
2282 
2283 // Volatile variables demand their effects be made known to all CPU's
2284 // in order. Store buffers on most chips allow reads & writes to
2285 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2286 // without some kind of memory barrier (i.e., it's not sufficient that
2287 // the interpreter does not reorder volatile references, the hardware
2288 // also must not reorder them).
2289 //
2290 // According to the new Java Memory Model (JMM):
2291 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2292 //     writes act as aquire & release, so:
2293 // (2) A read cannot let unrelated NON-volatile memory refs that
2294 //     happen after the read float up to before the read. It's OK for
2295 //     non-volatile memory refs that happen before the volatile read to
2296 //     float down below it.
2297 // (3) Similar a volatile write cannot let unrelated NON-volatile
2298 //     memory refs that happen BEFORE the write float down to after the
2299 //     write. It's OK for non-volatile memory refs that happen after the
2300 //     volatile write to float up before it.
2301 //
2302 // We only put in barriers around volatile refs (they are expensive),
2303 // not _between_ memory refs (that would require us to track the
2304 // flavor of the previous memory refs). Requirements (2) and (3)
2305 // require some barriers before volatile stores and after volatile
2306 // loads. These nearly cover requirement (1) but miss the
2307 // volatile-store-volatile-load case.  This final case is placed after
2308 // volatile-stores although it could just as well go before
2309 // volatile-loads.
2310 
2311 // The registers cache and index expected to be set before call.
2312 // Correct values of the cache and index registers are preserved.
2313 // Kills:
2314 //   Rcache (if has_tos)
2315 //   Rscratch
2316 void TemplateTable::jvmti_post_field_access(Register Rcache, Register Rscratch, bool is_static, bool has_tos) {
2317 
2318   assert_different_registers(Rcache, Rscratch);
2319 
2320   if (JvmtiExport::can_post_field_access()) {
2321     ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2322     Label Lno_field_access_post;
2323 
2324     // Check if post field access in enabled.
2325     int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_access_count_addr(), R0, true);
2326     __ lwz(Rscratch, offs, Rscratch);
2327 
2328     __ cmpwi(CCR0, Rscratch, 0);
2329     __ beq(CCR0, Lno_field_access_post);
2330 
2331     // Post access enabled - do it!
2332     __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2333     if (is_static) {
2334       __ li(R17_tos, 0);
2335     } else {
2336       if (has_tos) {
2337         // The fast bytecode versions have obj ptr in register.
2338         // Thus, save object pointer before call_VM() clobbers it
2339         // put object on tos where GC wants it.
2340         __ push_ptr(R17_tos);
2341       } else {
2342         // Load top of stack (do not pop the value off the stack).
2343         __ ld(R17_tos, Interpreter::expr_offset_in_bytes(0), R15_esp);
2344       }
2345       __ verify_oop(R17_tos);
2346     }
2347     // tos:   object pointer or NULL if static
2348     // cache: cache entry pointer
2349     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), R17_tos, Rcache);
2350     if (!is_static && has_tos) {
2351       // Restore object pointer.
2352       __ pop_ptr(R17_tos);
2353       __ verify_oop(R17_tos);
2354     } else {
2355       // Cache is still needed to get class or obj.
2356       __ get_cache_and_index_at_bcp(Rcache, 1);
2357     }
2358 
2359     __ align(32, 12);
2360     __ bind(Lno_field_access_post);
2361   }
2362 }
2363 
2364 // kills R11_scratch1
2365 void TemplateTable::pop_and_check_object(Register Roop) {
2366   Register Rtmp = R11_scratch1;
2367 
2368   assert_different_registers(Rtmp, Roop);
2369   __ pop_ptr(Roop);
2370   // For field access must check obj.
2371   __ null_check_throw(Roop, -1, Rtmp);
2372   __ verify_oop(Roop);
2373 }
2374 
2375 // PPC64: implement volatile loads as fence-store-acquire.
2376 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2377   transition(vtos, vtos);
2378 
2379   Label Lacquire, Lisync;
2380 
2381   const Register Rcache        = R3_ARG1,
2382                  Rclass_or_obj = R22_tmp2,
2383                  Roffset       = R23_tmp3,
2384                  Rflags        = R31,
2385                  Rbtable       = R5_ARG3,
2386                  Rbc           = R6_ARG4,
2387                  Rscratch      = R12_scratch2;
2388 
2389   static address field_branch_table[number_of_states],
2390                  static_branch_table[number_of_states];
2391 
2392   address* branch_table = (is_static || rc == may_not_rewrite) ? static_branch_table : field_branch_table;
2393 
2394   // Get field offset.
2395   resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2396 
2397   // JVMTI support
2398   jvmti_post_field_access(Rcache, Rscratch, is_static, false);
2399 
2400   // Load after possible GC.
2401   load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2402 
2403   // Load pointer to branch table.
2404   __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2405 
2406   // Get volatile flag.
2407   __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2408   // Note: sync is needed before volatile load on PPC64.
2409 
2410   // Check field type.
2411   __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2412 
2413 #ifdef ASSERT
2414   Label LFlagInvalid;
2415   __ cmpldi(CCR0, Rflags, number_of_states);
2416   __ bge(CCR0, LFlagInvalid);
2417 #endif
2418 
2419   // Load from branch table and dispatch (volatile case: one instruction ahead).
2420   __ sldi(Rflags, Rflags, LogBytesPerWord);
2421   __ cmpwi(CCR6, Rscratch, 1); // Volatile?
2422   if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2423     __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile ? size of 1 instruction : 0.
2424   }
2425   __ ldx(Rbtable, Rbtable, Rflags);
2426 
2427   // Get the obj from stack.
2428   if (!is_static) {
2429     pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2430   } else {
2431     __ verify_oop(Rclass_or_obj);
2432   }
2433 
2434   if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2435     __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2436   }
2437   __ mtctr(Rbtable);
2438   __ bctr();
2439 
2440 #ifdef ASSERT
2441   __ bind(LFlagInvalid);
2442   __ stop("got invalid flag", 0x654);
2443 #endif
2444 
2445   if (!is_static && rc == may_not_rewrite) {
2446     // We reuse the code from is_static.  It's jumped to via the table above.
2447     return;
2448   }
2449 
2450 #ifdef ASSERT
2451   // __ bind(Lvtos);
2452   address pc_before_fence = __ pc();
2453   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2454   assert(__ pc() - pc_before_fence == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2455   assert(branch_table[vtos] == 0, "can't compute twice");
2456   branch_table[vtos] = __ pc(); // non-volatile_entry point
2457   __ stop("vtos unexpected", 0x655);
2458 #endif
2459 
2460   __ align(32, 28, 28); // Align load.
2461   // __ bind(Ldtos);
2462   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2463   assert(branch_table[dtos] == 0, "can't compute twice");
2464   branch_table[dtos] = __ pc(); // non-volatile_entry point
2465   __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
2466   __ push(dtos);
2467   if (!is_static && rc == may_rewrite) {
2468     patch_bytecode(Bytecodes::_fast_dgetfield, Rbc, Rscratch);
2469   }
2470   {
2471     Label acquire_double;
2472     __ beq(CCR6, acquire_double); // Volatile?
2473     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2474 
2475     __ bind(acquire_double);
2476     __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2477     __ beq_predict_taken(CCR0, Lisync);
2478     __ b(Lisync); // In case of NAN.
2479   }
2480 
2481   __ align(32, 28, 28); // Align load.
2482   // __ bind(Lftos);
2483   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2484   assert(branch_table[ftos] == 0, "can't compute twice");
2485   branch_table[ftos] = __ pc(); // non-volatile_entry point
2486   __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
2487   __ push(ftos);
2488   if (!is_static && rc == may_rewrite) {
2489     patch_bytecode(Bytecodes::_fast_fgetfield, Rbc, Rscratch);
2490   }
2491   {
2492     Label acquire_float;
2493     __ beq(CCR6, acquire_float); // Volatile?
2494     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2495 
2496     __ bind(acquire_float);
2497     __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
2498     __ beq_predict_taken(CCR0, Lisync);
2499     __ b(Lisync); // In case of NAN.
2500   }
2501 
2502   __ align(32, 28, 28); // Align load.
2503   // __ bind(Litos);
2504   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2505   assert(branch_table[itos] == 0, "can't compute twice");
2506   branch_table[itos] = __ pc(); // non-volatile_entry point
2507   __ lwax(R17_tos, Rclass_or_obj, Roffset);
2508   __ push(itos);
2509   if (!is_static && rc == may_rewrite) {
2510     patch_bytecode(Bytecodes::_fast_igetfield, Rbc, Rscratch);
2511   }
2512   __ beq(CCR6, Lacquire); // Volatile?
2513   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2514 
2515   __ align(32, 28, 28); // Align load.
2516   // __ bind(Lltos);
2517   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2518   assert(branch_table[ltos] == 0, "can't compute twice");
2519   branch_table[ltos] = __ pc(); // non-volatile_entry point
2520   __ ldx(R17_tos, Rclass_or_obj, Roffset);
2521   __ push(ltos);
2522   if (!is_static && rc == may_rewrite) {
2523     patch_bytecode(Bytecodes::_fast_lgetfield, Rbc, Rscratch);
2524   }
2525   __ beq(CCR6, Lacquire); // Volatile?
2526   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2527 
2528   __ align(32, 28, 28); // Align load.
2529   // __ bind(Lbtos);
2530   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2531   assert(branch_table[btos] == 0, "can't compute twice");
2532   branch_table[btos] = __ pc(); // non-volatile_entry point
2533   __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2534   __ extsb(R17_tos, R17_tos);
2535   __ push(btos);
2536   if (!is_static && rc == may_rewrite) {
2537     patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2538   }
2539   __ beq(CCR6, Lacquire); // Volatile?
2540   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2541 
2542   __ align(32, 28, 28); // Align load.
2543   // __ bind(Lztos); (same code as btos)
2544   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2545   assert(branch_table[ztos] == 0, "can't compute twice");
2546   branch_table[ztos] = __ pc(); // non-volatile_entry point
2547   __ lbzx(R17_tos, Rclass_or_obj, Roffset);
2548   __ push(ztos);
2549   if (!is_static && rc == may_rewrite) {
2550     // use btos rewriting, no truncating to t/f bit is needed for getfield.
2551     patch_bytecode(Bytecodes::_fast_bgetfield, Rbc, Rscratch);
2552   }
2553   __ beq(CCR6, Lacquire); // Volatile?
2554   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2555 
2556   __ align(32, 28, 28); // Align load.
2557   // __ bind(Lctos);
2558   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2559   assert(branch_table[ctos] == 0, "can't compute twice");
2560   branch_table[ctos] = __ pc(); // non-volatile_entry point
2561   __ lhzx(R17_tos, Rclass_or_obj, Roffset);
2562   __ push(ctos);
2563   if (!is_static && rc == may_rewrite) {
2564     patch_bytecode(Bytecodes::_fast_cgetfield, Rbc, Rscratch);
2565   }
2566   __ beq(CCR6, Lacquire); // Volatile?
2567   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2568 
2569   __ align(32, 28, 28); // Align load.
2570   // __ bind(Lstos);
2571   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2572   assert(branch_table[stos] == 0, "can't compute twice");
2573   branch_table[stos] = __ pc(); // non-volatile_entry point
2574   __ lhax(R17_tos, Rclass_or_obj, Roffset);
2575   __ push(stos);
2576   if (!is_static && rc == may_rewrite) {
2577     patch_bytecode(Bytecodes::_fast_sgetfield, Rbc, Rscratch);
2578   }
2579   __ beq(CCR6, Lacquire); // Volatile?
2580   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2581 
2582   __ align(32, 28, 28); // Align load.
2583   // __ bind(Latos);
2584   __ fence(); // Volatile entry point (one instruction before non-volatile_entry point).
2585   assert(branch_table[atos] == 0, "can't compute twice");
2586   branch_table[atos] = __ pc(); // non-volatile_entry point
2587   __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
2588   __ verify_oop(R17_tos);
2589   __ push(atos);
2590   //__ dcbt(R17_tos); // prefetch
2591   if (!is_static && rc == may_rewrite) {
2592     patch_bytecode(Bytecodes::_fast_agetfield, Rbc, Rscratch);
2593   }
2594   __ beq(CCR6, Lacquire); // Volatile?
2595   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2596 
2597   __ align(32, 12);
2598   __ bind(Lacquire);
2599   __ twi_0(R17_tos);
2600   __ bind(Lisync);
2601   __ isync(); // acquire
2602 
2603 #ifdef ASSERT
2604   for (int i = 0; i<number_of_states; ++i) {
2605     assert(branch_table[i], "get initialization");
2606     //tty->print_cr("get: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2607     //              is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2608   }
2609 #endif
2610 }
2611 
2612 void TemplateTable::getfield(int byte_no) {
2613   getfield_or_static(byte_no, false);
2614 }
2615 
2616 void TemplateTable::nofast_getfield(int byte_no) {
2617   getfield_or_static(byte_no, false, may_not_rewrite);
2618 }
2619 
2620 void TemplateTable::getstatic(int byte_no) {
2621   getfield_or_static(byte_no, true);
2622 }
2623 
2624 // The registers cache and index expected to be set before call.
2625 // The function may destroy various registers, just not the cache and index registers.
2626 void TemplateTable::jvmti_post_field_mod(Register Rcache, Register Rscratch, bool is_static) {
2627 
2628   assert_different_registers(Rcache, Rscratch, R6_ARG4);
2629 
2630   if (JvmtiExport::can_post_field_modification()) {
2631     Label Lno_field_mod_post;
2632 
2633     // Check if post field access in enabled.
2634     int offs = __ load_const_optimized(Rscratch, JvmtiExport::get_field_modification_count_addr(), R0, true);
2635     __ lwz(Rscratch, offs, Rscratch);
2636 
2637     __ cmpwi(CCR0, Rscratch, 0);
2638     __ beq(CCR0, Lno_field_mod_post);
2639 
2640     // Do the post
2641     ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2642     const Register Robj = Rscratch;
2643 
2644     __ addi(Rcache, Rcache, in_bytes(cp_base_offset));
2645     if (is_static) {
2646       // Life is simple. Null out the object pointer.
2647       __ li(Robj, 0);
2648     } else {
2649       // In case of the fast versions, value lives in registers => put it back on tos.
2650       int offs = Interpreter::expr_offset_in_bytes(0);
2651       Register base = R15_esp;
2652       switch(bytecode()) {
2653         case Bytecodes::_fast_aputfield: __ push_ptr(); offs+= Interpreter::stackElementSize; break;
2654         case Bytecodes::_fast_iputfield: // Fall through
2655         case Bytecodes::_fast_bputfield: // Fall through
2656         case Bytecodes::_fast_zputfield: // Fall through
2657         case Bytecodes::_fast_cputfield: // Fall through
2658         case Bytecodes::_fast_sputfield: __ push_i(); offs+=  Interpreter::stackElementSize; break;
2659         case Bytecodes::_fast_lputfield: __ push_l(); offs+=2*Interpreter::stackElementSize; break;
2660         case Bytecodes::_fast_fputfield: __ push_f(); offs+=  Interpreter::stackElementSize; break;
2661         case Bytecodes::_fast_dputfield: __ push_d(); offs+=2*Interpreter::stackElementSize; break;
2662         default: {
2663           offs = 0;
2664           base = Robj;
2665           const Register Rflags = Robj;
2666           Label is_one_slot;
2667           // Life is harder. The stack holds the value on top, followed by the
2668           // object. We don't know the size of the value, though; it could be
2669           // one or two words depending on its type. As a result, we must find
2670           // the type to determine where the object is.
2671           __ ld(Rflags, in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcache); // Big Endian
2672           __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2673 
2674           __ cmpwi(CCR0, Rflags, ltos);
2675           __ cmpwi(CCR1, Rflags, dtos);
2676           __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(1));
2677           __ crnor(CCR0, Assembler::equal, CCR1, Assembler::equal);
2678           __ beq(CCR0, is_one_slot);
2679           __ addi(base, R15_esp, Interpreter::expr_offset_in_bytes(2));
2680           __ bind(is_one_slot);
2681           break;
2682         }
2683       }
2684       __ ld(Robj, offs, base);
2685       __ verify_oop(Robj);
2686     }
2687 
2688     __ addi(R6_ARG4, R15_esp, Interpreter::expr_offset_in_bytes(0));
2689     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), Robj, Rcache, R6_ARG4);
2690     __ get_cache_and_index_at_bcp(Rcache, 1);
2691 
2692     // In case of the fast versions, value lives in registers => put it back on tos.
2693     switch(bytecode()) {
2694       case Bytecodes::_fast_aputfield: __ pop_ptr(); break;
2695       case Bytecodes::_fast_iputfield: // Fall through
2696       case Bytecodes::_fast_bputfield: // Fall through
2697       case Bytecodes::_fast_zputfield: // Fall through
2698       case Bytecodes::_fast_cputfield: // Fall through
2699       case Bytecodes::_fast_sputfield: __ pop_i(); break;
2700       case Bytecodes::_fast_lputfield: __ pop_l(); break;
2701       case Bytecodes::_fast_fputfield: __ pop_f(); break;
2702       case Bytecodes::_fast_dputfield: __ pop_d(); break;
2703       default: break; // Nothin' to do.
2704     }
2705 
2706     __ align(32, 12);
2707     __ bind(Lno_field_mod_post);
2708   }
2709 }
2710 
2711 // PPC64: implement volatile stores as release-store (return bytecode contains an additional release).
2712 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2713   Label Lvolatile;
2714 
2715   const Register Rcache        = R5_ARG3,  // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2716                  Rclass_or_obj = R31,      // Needs to survive C call.
2717                  Roffset       = R22_tmp2, // Needs to survive C call.
2718                  Rflags        = R3_ARG1,
2719                  Rbtable       = R4_ARG2,
2720                  Rscratch      = R11_scratch1,
2721                  Rscratch2     = R12_scratch2,
2722                  Rscratch3     = R6_ARG4,
2723                  Rbc           = Rscratch3;
2724   const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2725 
2726   static address field_rw_branch_table[number_of_states],
2727                  field_norw_branch_table[number_of_states],
2728                  static_branch_table[number_of_states];
2729 
2730   address* branch_table = is_static ? static_branch_table :
2731     (rc == may_rewrite ? field_rw_branch_table : field_norw_branch_table);
2732 
2733   // Stack (grows up):
2734   //  value
2735   //  obj
2736 
2737   // Load the field offset.
2738   resolve_cache_and_index(byte_no, Rcache, Rscratch, sizeof(u2));
2739   jvmti_post_field_mod(Rcache, Rscratch, is_static);
2740   load_field_cp_cache_entry(Rclass_or_obj, Rcache, noreg, Roffset, Rflags, is_static);
2741 
2742   // Load pointer to branch table.
2743   __ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
2744 
2745   // Get volatile flag.
2746   __ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2747 
2748   // Check the field type.
2749   __ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
2750 
2751 #ifdef ASSERT
2752   Label LFlagInvalid;
2753   __ cmpldi(CCR0, Rflags, number_of_states);
2754   __ bge(CCR0, LFlagInvalid);
2755 #endif
2756 
2757   // Load from branch table and dispatch (volatile case: one instruction ahead).
2758   __ sldi(Rflags, Rflags, LogBytesPerWord);
2759   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2760     __ cmpwi(CR_is_vol, Rscratch, 1);  // Volatile?
2761   }
2762   __ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // Volatile? size of instruction 1 : 0.
2763   __ ldx(Rbtable, Rbtable, Rflags);
2764 
2765   __ subf(Rbtable, Rscratch, Rbtable); // Point to volatile/non-volatile entry point.
2766   __ mtctr(Rbtable);
2767   __ bctr();
2768 
2769 #ifdef ASSERT
2770   __ bind(LFlagInvalid);
2771   __ stop("got invalid flag", 0x656);
2772 
2773   // __ bind(Lvtos);
2774   address pc_before_release = __ pc();
2775   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2776   assert(__ pc() - pc_before_release == (ptrdiff_t)BytesPerInstWord, "must be single instruction");
2777   assert(branch_table[vtos] == 0, "can't compute twice");
2778   branch_table[vtos] = __ pc(); // non-volatile_entry point
2779   __ stop("vtos unexpected", 0x657);
2780 #endif
2781 
2782   __ align(32, 28, 28); // Align pop.
2783   // __ bind(Ldtos);
2784   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2785   assert(branch_table[dtos] == 0, "can't compute twice");
2786   branch_table[dtos] = __ pc(); // non-volatile_entry point
2787   __ pop(dtos);
2788   if (!is_static) {
2789     pop_and_check_object(Rclass_or_obj);  // Kills R11_scratch1.
2790   }
2791   __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
2792   if (!is_static && rc == may_rewrite) {
2793     patch_bytecode(Bytecodes::_fast_dputfield, Rbc, Rscratch, true, byte_no);
2794   }
2795   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2796     __ beq(CR_is_vol, Lvolatile); // Volatile?
2797   }
2798   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2799 
2800   __ align(32, 28, 28); // Align pop.
2801   // __ bind(Lftos);
2802   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2803   assert(branch_table[ftos] == 0, "can't compute twice");
2804   branch_table[ftos] = __ pc(); // non-volatile_entry point
2805   __ pop(ftos);
2806   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2807   __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
2808   if (!is_static && rc == may_rewrite) {
2809     patch_bytecode(Bytecodes::_fast_fputfield, Rbc, Rscratch, true, byte_no);
2810   }
2811   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2812     __ beq(CR_is_vol, Lvolatile); // Volatile?
2813   }
2814   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2815 
2816   __ align(32, 28, 28); // Align pop.
2817   // __ bind(Litos);
2818   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2819   assert(branch_table[itos] == 0, "can't compute twice");
2820   branch_table[itos] = __ pc(); // non-volatile_entry point
2821   __ pop(itos);
2822   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2823   __ stwx(R17_tos, Rclass_or_obj, Roffset);
2824   if (!is_static && rc == may_rewrite) {
2825     patch_bytecode(Bytecodes::_fast_iputfield, Rbc, Rscratch, true, byte_no);
2826   }
2827   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2828     __ beq(CR_is_vol, Lvolatile); // Volatile?
2829   }
2830   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2831 
2832   __ align(32, 28, 28); // Align pop.
2833   // __ bind(Lltos);
2834   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2835   assert(branch_table[ltos] == 0, "can't compute twice");
2836   branch_table[ltos] = __ pc(); // non-volatile_entry point
2837   __ pop(ltos);
2838   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2839   __ stdx(R17_tos, Rclass_or_obj, Roffset);
2840   if (!is_static && rc == may_rewrite) {
2841     patch_bytecode(Bytecodes::_fast_lputfield, Rbc, Rscratch, true, byte_no);
2842   }
2843   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2844     __ beq(CR_is_vol, Lvolatile); // Volatile?
2845   }
2846   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2847 
2848   __ align(32, 28, 28); // Align pop.
2849   // __ bind(Lbtos);
2850   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2851   assert(branch_table[btos] == 0, "can't compute twice");
2852   branch_table[btos] = __ pc(); // non-volatile_entry point
2853   __ pop(btos);
2854   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2855   __ stbx(R17_tos, Rclass_or_obj, Roffset);
2856   if (!is_static && rc == may_rewrite) {
2857     patch_bytecode(Bytecodes::_fast_bputfield, Rbc, Rscratch, true, byte_no);
2858   }
2859   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2860     __ beq(CR_is_vol, Lvolatile); // Volatile?
2861   }
2862   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2863 
2864   __ align(32, 28, 28); // Align pop.
2865   // __ bind(Lztos);
2866   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2867   assert(branch_table[ztos] == 0, "can't compute twice");
2868   branch_table[ztos] = __ pc(); // non-volatile_entry point
2869   __ pop(ztos);
2870   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2871   __ andi(R17_tos, R17_tos, 0x1);
2872   __ stbx(R17_tos, Rclass_or_obj, Roffset);
2873   if (!is_static && rc == may_rewrite) {
2874     patch_bytecode(Bytecodes::_fast_zputfield, Rbc, Rscratch, true, byte_no);
2875   }
2876   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2877     __ beq(CR_is_vol, Lvolatile); // Volatile?
2878   }
2879   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2880 
2881   __ align(32, 28, 28); // Align pop.
2882   // __ bind(Lctos);
2883   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2884   assert(branch_table[ctos] == 0, "can't compute twice");
2885   branch_table[ctos] = __ pc(); // non-volatile_entry point
2886   __ pop(ctos);
2887   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1..
2888   __ sthx(R17_tos, Rclass_or_obj, Roffset);
2889   if (!is_static && rc == may_rewrite) {
2890     patch_bytecode(Bytecodes::_fast_cputfield, Rbc, Rscratch, true, byte_no);
2891   }
2892   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2893     __ beq(CR_is_vol, Lvolatile); // Volatile?
2894   }
2895   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2896 
2897   __ align(32, 28, 28); // Align pop.
2898   // __ bind(Lstos);
2899   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2900   assert(branch_table[stos] == 0, "can't compute twice");
2901   branch_table[stos] = __ pc(); // non-volatile_entry point
2902   __ pop(stos);
2903   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // Kills R11_scratch1.
2904   __ sthx(R17_tos, Rclass_or_obj, Roffset);
2905   if (!is_static && rc == may_rewrite) {
2906     patch_bytecode(Bytecodes::_fast_sputfield, Rbc, Rscratch, true, byte_no);
2907   }
2908   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2909     __ beq(CR_is_vol, Lvolatile); // Volatile?
2910   }
2911   __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2912 
2913   __ align(32, 28, 28); // Align pop.
2914   // __ bind(Latos);
2915   __ release(); // Volatile entry point (one instruction before non-volatile_entry point).
2916   assert(branch_table[atos] == 0, "can't compute twice");
2917   branch_table[atos] = __ pc(); // non-volatile_entry point
2918   __ pop(atos);
2919   if (!is_static) { pop_and_check_object(Rclass_or_obj); } // kills R11_scratch1
2920   do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2921   if (!is_static && rc == may_rewrite) {
2922     patch_bytecode(Bytecodes::_fast_aputfield, Rbc, Rscratch, true, byte_no);
2923   }
2924   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2925     __ beq(CR_is_vol, Lvolatile); // Volatile?
2926     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
2927 
2928     __ align(32, 12);
2929     __ bind(Lvolatile);
2930     __ fence();
2931   }
2932   // fallthru: __ b(Lexit);
2933 
2934 #ifdef ASSERT
2935   for (int i = 0; i<number_of_states; ++i) {
2936     assert(branch_table[i], "put initialization");
2937     //tty->print_cr("put: %s_branch_table[%d] = 0x%llx (opcode 0x%llx)",
2938     //              is_static ? "static" : "field", i, branch_table[i], *((unsigned int*)branch_table[i]));
2939   }
2940 #endif
2941 }
2942 
2943 void TemplateTable::putfield(int byte_no) {
2944   putfield_or_static(byte_no, false);
2945 }
2946 
2947 void TemplateTable::nofast_putfield(int byte_no) {
2948   putfield_or_static(byte_no, false, may_not_rewrite);
2949 }
2950 
2951 void TemplateTable::putstatic(int byte_no) {
2952   putfield_or_static(byte_no, true);
2953 }
2954 
2955 // See SPARC. On PPC64, we have a different jvmti_post_field_mod which does the job.
2956 void TemplateTable::jvmti_post_fast_field_mod() {
2957   __ should_not_reach_here();
2958 }
2959 
2960 void TemplateTable::fast_storefield(TosState state) {
2961   transition(state, vtos);
2962 
2963   const Register Rcache        = R5_ARG3,  // Do not use ARG1/2 (causes trouble in jvmti_post_field_mod).
2964                  Rclass_or_obj = R31,      // Needs to survive C call.
2965                  Roffset       = R22_tmp2, // Needs to survive C call.
2966                  Rflags        = R3_ARG1,
2967                  Rscratch      = R11_scratch1,
2968                  Rscratch2     = R12_scratch2,
2969                  Rscratch3     = R4_ARG2;
2970   const ConditionRegister CR_is_vol = CCR2; // Non-volatile condition register (survives runtime call in do_oop_store).
2971 
2972   // Constant pool already resolved => Load flags and offset of field.
2973   __ get_cache_and_index_at_bcp(Rcache, 1);
2974   jvmti_post_field_mod(Rcache, Rscratch, false /* not static */);
2975   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
2976 
2977   // Get the obj and the final store addr.
2978   pop_and_check_object(Rclass_or_obj); // Kills R11_scratch1.
2979 
2980   // Get volatile flag.
2981   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
2982   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) { __ cmpdi(CR_is_vol, Rscratch, 1); }
2983   {
2984     Label LnotVolatile;
2985     __ beq(CCR0, LnotVolatile);
2986     __ release();
2987     __ align(32, 12);
2988     __ bind(LnotVolatile);
2989   }
2990 
2991   // Do the store and fencing.
2992   switch(bytecode()) {
2993     case Bytecodes::_fast_aputfield:
2994       // Store into the field.
2995       do_oop_store(_masm, Rclass_or_obj, Roffset, R17_tos, Rscratch, Rscratch2, Rscratch3, _bs->kind(), false /* precise */, true /* check null */);
2996       break;
2997 
2998     case Bytecodes::_fast_iputfield:
2999       __ stwx(R17_tos, Rclass_or_obj, Roffset);
3000       break;
3001 
3002     case Bytecodes::_fast_lputfield:
3003       __ stdx(R17_tos, Rclass_or_obj, Roffset);
3004       break;
3005 
3006     case Bytecodes::_fast_zputfield:
3007       __ andi(R17_tos, R17_tos, 0x1);  // boolean is true if LSB is 1
3008       // fall through to bputfield
3009     case Bytecodes::_fast_bputfield:
3010       __ stbx(R17_tos, Rclass_or_obj, Roffset);
3011       break;
3012 
3013     case Bytecodes::_fast_cputfield:
3014     case Bytecodes::_fast_sputfield:
3015       __ sthx(R17_tos, Rclass_or_obj, Roffset);
3016       break;
3017 
3018     case Bytecodes::_fast_fputfield:
3019       __ stfsx(F15_ftos, Rclass_or_obj, Roffset);
3020       break;
3021 
3022     case Bytecodes::_fast_dputfield:
3023       __ stfdx(F15_ftos, Rclass_or_obj, Roffset);
3024       break;
3025 
3026     default: ShouldNotReachHere();
3027   }
3028 
3029   if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
3030     Label LVolatile;
3031     __ beq(CR_is_vol, LVolatile);
3032     __ dispatch_epilog(vtos, Bytecodes::length_for(bytecode()));
3033 
3034     __ align(32, 12);
3035     __ bind(LVolatile);
3036     __ fence();
3037   }
3038 }
3039 
3040 void TemplateTable::fast_accessfield(TosState state) {
3041   transition(atos, state);
3042 
3043   Label LisVolatile;
3044   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3045 
3046   const Register Rcache        = R3_ARG1,
3047                  Rclass_or_obj = R17_tos,
3048                  Roffset       = R22_tmp2,
3049                  Rflags        = R23_tmp3,
3050                  Rscratch      = R12_scratch2;
3051 
3052   // Constant pool already resolved. Get the field offset.
3053   __ get_cache_and_index_at_bcp(Rcache, 1);
3054   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3055 
3056   // JVMTI support
3057   jvmti_post_field_access(Rcache, Rscratch, false, true);
3058 
3059   // Get the load address.
3060   __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3061 
3062   // Get volatile flag.
3063   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3064   __ bne(CCR0, LisVolatile);
3065 
3066   switch(bytecode()) {
3067     case Bytecodes::_fast_agetfield:
3068     {
3069       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3070       __ verify_oop(R17_tos);
3071       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3072 
3073       __ bind(LisVolatile);
3074       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3075       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3076       __ verify_oop(R17_tos);
3077       __ twi_0(R17_tos);
3078       __ isync();
3079       break;
3080     }
3081     case Bytecodes::_fast_igetfield:
3082     {
3083       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3084       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3085 
3086       __ bind(LisVolatile);
3087       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3088       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3089       __ twi_0(R17_tos);
3090       __ isync();
3091       break;
3092     }
3093     case Bytecodes::_fast_lgetfield:
3094     {
3095       __ ldx(R17_tos, Rclass_or_obj, Roffset);
3096       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3097 
3098       __ bind(LisVolatile);
3099       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3100       __ ldx(R17_tos, Rclass_or_obj, Roffset);
3101       __ twi_0(R17_tos);
3102       __ isync();
3103       break;
3104     }
3105     case Bytecodes::_fast_bgetfield:
3106     {
3107       __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3108       __ extsb(R17_tos, R17_tos);
3109       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3110 
3111       __ bind(LisVolatile);
3112       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3113       __ lbzx(R17_tos, Rclass_or_obj, Roffset);
3114       __ twi_0(R17_tos);
3115       __ extsb(R17_tos, R17_tos);
3116       __ isync();
3117       break;
3118     }
3119     case Bytecodes::_fast_cgetfield:
3120     {
3121       __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3122       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3123 
3124       __ bind(LisVolatile);
3125       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3126       __ lhzx(R17_tos, Rclass_or_obj, Roffset);
3127       __ twi_0(R17_tos);
3128       __ isync();
3129       break;
3130     }
3131     case Bytecodes::_fast_sgetfield:
3132     {
3133       __ lhax(R17_tos, Rclass_or_obj, Roffset);
3134       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3135 
3136       __ bind(LisVolatile);
3137       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3138       __ lhax(R17_tos, Rclass_or_obj, Roffset);
3139       __ twi_0(R17_tos);
3140       __ isync();
3141       break;
3142     }
3143     case Bytecodes::_fast_fgetfield:
3144     {
3145       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3146       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3147 
3148       __ bind(LisVolatile);
3149       Label Ldummy;
3150       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3151       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3152       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3153       __ bne_predict_not_taken(CCR0, Ldummy);
3154       __ bind(Ldummy);
3155       __ isync();
3156       break;
3157     }
3158     case Bytecodes::_fast_dgetfield:
3159     {
3160       __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3161       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()));
3162 
3163       __ bind(LisVolatile);
3164       Label Ldummy;
3165       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3166       __ lfdx(F15_ftos, Rclass_or_obj, Roffset);
3167       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3168       __ bne_predict_not_taken(CCR0, Ldummy);
3169       __ bind(Ldummy);
3170       __ isync();
3171       break;
3172     }
3173     default: ShouldNotReachHere();
3174   }
3175 }
3176 
3177 void TemplateTable::fast_xaccess(TosState state) {
3178   transition(vtos, state);
3179 
3180   Label LisVolatile;
3181   ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3182   const Register Rcache        = R3_ARG1,
3183                  Rclass_or_obj = R17_tos,
3184                  Roffset       = R22_tmp2,
3185                  Rflags        = R23_tmp3,
3186                  Rscratch      = R12_scratch2;
3187 
3188   __ ld(Rclass_or_obj, 0, R18_locals);
3189 
3190   // Constant pool already resolved. Get the field offset.
3191   __ get_cache_and_index_at_bcp(Rcache, 2);
3192   load_field_cp_cache_entry(noreg, Rcache, noreg, Roffset, Rflags, false);
3193 
3194   // JVMTI support not needed, since we switch back to single bytecode as soon as debugger attaches.
3195 
3196   // Needed to report exception at the correct bcp.
3197   __ addi(R14_bcp, R14_bcp, 1);
3198 
3199   // Get the load address.
3200   __ null_check_throw(Rclass_or_obj, -1, Rscratch);
3201 
3202   // Get volatile flag.
3203   __ rldicl_(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // Extract volatile bit.
3204   __ bne(CCR0, LisVolatile);
3205 
3206   switch(state) {
3207   case atos:
3208     {
3209       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3210       __ verify_oop(R17_tos);
3211       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3212 
3213       __ bind(LisVolatile);
3214       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3215       __ load_heap_oop(R17_tos, (RegisterOrConstant)Roffset, Rclass_or_obj);
3216       __ verify_oop(R17_tos);
3217       __ twi_0(R17_tos);
3218       __ isync();
3219       break;
3220     }
3221   case itos:
3222     {
3223       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3224       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3225 
3226       __ bind(LisVolatile);
3227       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3228       __ lwax(R17_tos, Rclass_or_obj, Roffset);
3229       __ twi_0(R17_tos);
3230       __ isync();
3231       break;
3232     }
3233   case ftos:
3234     {
3235       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3236       __ dispatch_epilog(state, Bytecodes::length_for(bytecode()) - 1); // Undo bcp increment.
3237 
3238       __ bind(LisVolatile);
3239       Label Ldummy;
3240       if (support_IRIW_for_not_multiple_copy_atomic_cpu) { __ fence(); }
3241       __ lfsx(F15_ftos, Rclass_or_obj, Roffset);
3242       __ fcmpu(CCR0, F15_ftos, F15_ftos); // Acquire by cmp-br-isync.
3243       __ bne_predict_not_taken(CCR0, Ldummy);
3244       __ bind(Ldummy);
3245       __ isync();
3246       break;
3247     }
3248   default: ShouldNotReachHere();
3249   }
3250   __ addi(R14_bcp, R14_bcp, -1);
3251 }
3252 
3253 // ============================================================================
3254 // Calls
3255 
3256 // Common code for invoke
3257 //
3258 // Input:
3259 //   - byte_no
3260 //
3261 // Output:
3262 //   - Rmethod:        The method to invoke next.
3263 //   - Rret_addr:      The return address to return to.
3264 //   - Rindex:         MethodType (invokehandle) or CallSite obj (invokedynamic)
3265 //   - Rrecv:          Cache for "this" pointer, might be noreg if static call.
3266 //   - Rflags:         Method flags from const pool cache.
3267 //
3268 //  Kills:
3269 //   - Rscratch1
3270 //
3271 void TemplateTable::prepare_invoke(int byte_no,
3272                                    Register Rmethod,  // linked method (or i-klass)
3273                                    Register Rret_addr,// return address
3274                                    Register Rindex,   // itable index, MethodType, etc.
3275                                    Register Rrecv,    // If caller wants to see it.
3276                                    Register Rflags,   // If caller wants to test it.
3277                                    Register Rscratch
3278                                    ) {
3279   // Determine flags.
3280   const Bytecodes::Code code = bytecode();
3281   const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
3282   const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
3283   const bool is_invokehandle     = code == Bytecodes::_invokehandle;
3284   const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
3285   const bool is_invokespecial    = code == Bytecodes::_invokespecial;
3286   const bool load_receiver       = (Rrecv != noreg);
3287   assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3288 
3289   assert_different_registers(Rmethod, Rindex, Rflags, Rscratch);
3290   assert_different_registers(Rmethod, Rrecv, Rflags, Rscratch);
3291   assert_different_registers(Rret_addr, Rscratch);
3292 
3293   load_invoke_cp_cache_entry(byte_no, Rmethod, Rindex, Rflags, is_invokevirtual, false, is_invokedynamic);
3294 
3295   // Saving of SP done in call_from_interpreter.
3296 
3297   // Maybe push "appendix" to arguments.
3298   if (is_invokedynamic || is_invokehandle) {
3299     Label Ldone;
3300     __ rldicl_(R0, Rflags, 64-ConstantPoolCacheEntry::has_appendix_shift, 63);
3301     __ beq(CCR0, Ldone);
3302     // Push "appendix" (MethodType, CallSite, etc.).
3303     // This must be done before we get the receiver,
3304     // since the parameter_size includes it.
3305     __ load_resolved_reference_at_index(Rscratch, Rindex);
3306     __ verify_oop(Rscratch);
3307     __ push_ptr(Rscratch);
3308     __ bind(Ldone);
3309   }
3310 
3311   // Load receiver if needed (after appendix is pushed so parameter size is correct).
3312   if (load_receiver) {
3313     const Register Rparam_count = Rscratch;
3314     __ andi(Rparam_count, Rflags, ConstantPoolCacheEntry::parameter_size_mask);
3315     __ load_receiver(Rparam_count, Rrecv);
3316     __ verify_oop(Rrecv);
3317   }
3318 
3319   // Get return address.
3320   {
3321     Register Rtable_addr = Rscratch;
3322     Register Rret_type = Rret_addr;
3323     address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3324 
3325     // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3326     __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3327     __ load_dispatch_table(Rtable_addr, (address*)table_addr);
3328     __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3329     // Get return address.
3330     __ ldx(Rret_addr, Rtable_addr, Rret_type);
3331   }
3332 }
3333 
3334 // Helper for virtual calls. Load target out of vtable and jump off!
3335 // Kills all passed registers.
3336 void TemplateTable::generate_vtable_call(Register Rrecv_klass, Register Rindex, Register Rret, Register Rtemp) {
3337 
3338   assert_different_registers(Rrecv_klass, Rtemp, Rret);
3339   const Register Rtarget_method = Rindex;
3340 
3341   // Get target method & entry point.
3342   const int base = in_bytes(Klass::vtable_start_offset());
3343   // Calc vtable addr scale the vtable index by 8.
3344   __ sldi(Rindex, Rindex, exact_log2(vtableEntry::size_in_bytes()));
3345   // Load target.
3346   __ addi(Rrecv_klass, Rrecv_klass, base + vtableEntry::method_offset_in_bytes());
3347   __ ldx(Rtarget_method, Rindex, Rrecv_klass);
3348   // Argument and return type profiling.
3349   __ profile_arguments_type(Rtarget_method, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */, true);
3350   __ call_from_interpreter(Rtarget_method, Rret, Rrecv_klass /* scratch1 */, Rtemp /* scratch2 */);
3351 }
3352 
3353 // Virtual or final call. Final calls are rewritten on the fly to run through "fast_finalcall" next time.
3354 void TemplateTable::invokevirtual(int byte_no) {
3355   transition(vtos, vtos);
3356 
3357   Register Rtable_addr = R11_scratch1,
3358            Rret_type = R12_scratch2,
3359            Rret_addr = R5_ARG3,
3360            Rflags = R22_tmp2, // Should survive C call.
3361            Rrecv = R3_ARG1,
3362            Rrecv_klass = Rrecv,
3363            Rvtableindex_or_method = R31, // Should survive C call.
3364            Rnum_params = R4_ARG2,
3365            Rnew_bc = R6_ARG4;
3366 
3367   Label LnotFinal;
3368 
3369   load_invoke_cp_cache_entry(byte_no, Rvtableindex_or_method, noreg, Rflags, /*virtual*/ true, false, false);
3370 
3371   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3372   __ bfalse(CCR0, LnotFinal);
3373 
3374   if (RewriteBytecodes && !UseSharedSpaces) {
3375     patch_bytecode(Bytecodes::_fast_invokevfinal, Rnew_bc, R12_scratch2);
3376   }
3377   invokevfinal_helper(Rvtableindex_or_method, Rflags, R11_scratch1, R12_scratch2);
3378 
3379   __ align(32, 12);
3380   __ bind(LnotFinal);
3381   // Load "this" pointer (receiver).
3382   __ rldicl(Rnum_params, Rflags, 64, 48);
3383   __ load_receiver(Rnum_params, Rrecv);
3384   __ verify_oop(Rrecv);
3385 
3386   // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3387   __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3388   __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3389   __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3390   __ ldx(Rret_addr, Rret_type, Rtable_addr);
3391   __ null_check_throw(Rrecv, oopDesc::klass_offset_in_bytes(), R11_scratch1);
3392   __ load_klass(Rrecv_klass, Rrecv);
3393   __ verify_klass_ptr(Rrecv_klass);
3394   __ profile_virtual_call(Rrecv_klass, R11_scratch1, R12_scratch2, false);
3395 
3396   generate_vtable_call(Rrecv_klass, Rvtableindex_or_method, Rret_addr, R11_scratch1);
3397 }
3398 
3399 void TemplateTable::fast_invokevfinal(int byte_no) {
3400   transition(vtos, vtos);
3401 
3402   assert(byte_no == f2_byte, "use this argument");
3403   Register Rflags  = R22_tmp2,
3404            Rmethod = R31;
3405   load_invoke_cp_cache_entry(byte_no, Rmethod, noreg, Rflags, /*virtual*/ true, /*is_invokevfinal*/ true, false);
3406   invokevfinal_helper(Rmethod, Rflags, R11_scratch1, R12_scratch2);
3407 }
3408 
3409 void TemplateTable::invokevfinal_helper(Register Rmethod, Register Rflags, Register Rscratch1, Register Rscratch2) {
3410 
3411   assert_different_registers(Rmethod, Rflags, Rscratch1, Rscratch2);
3412 
3413   // Load receiver from stack slot.
3414   Register Rrecv = Rscratch2;
3415   Register Rnum_params = Rrecv;
3416 
3417   __ ld(Rnum_params, in_bytes(Method::const_offset()), Rmethod);
3418   __ lhz(Rnum_params /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), Rnum_params);
3419 
3420   // Get return address.
3421   Register Rtable_addr = Rscratch1,
3422            Rret_addr   = Rflags,
3423            Rret_type   = Rret_addr;
3424   // Get return type. It's coded into the upper 4 bits of the lower half of the 64 bit value.
3425   __ rldicl(Rret_type, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
3426   __ load_dispatch_table(Rtable_addr, Interpreter::invoke_return_entry_table());
3427   __ sldi(Rret_type, Rret_type, LogBytesPerWord);
3428   __ ldx(Rret_addr, Rret_type, Rtable_addr);
3429 
3430   // Load receiver and receiver NULL check.
3431   __ load_receiver(Rnum_params, Rrecv);
3432   __ null_check_throw(Rrecv, -1, Rscratch1);
3433 
3434   __ profile_final_call(Rrecv, Rscratch1);
3435   // Argument and return type profiling.
3436   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3437 
3438   // Do the call.
3439   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1, Rscratch2);
3440 }
3441 
3442 void TemplateTable::invokespecial(int byte_no) {
3443   assert(byte_no == f1_byte, "use this argument");
3444   transition(vtos, vtos);
3445 
3446   Register Rtable_addr = R3_ARG1,
3447            Rret_addr   = R4_ARG2,
3448            Rflags      = R5_ARG3,
3449            Rreceiver   = R6_ARG4,
3450            Rmethod     = R31;
3451 
3452   prepare_invoke(byte_no, Rmethod, Rret_addr, noreg, Rreceiver, Rflags, R11_scratch1);
3453 
3454   // Receiver NULL check.
3455   __ null_check_throw(Rreceiver, -1, R11_scratch1);
3456 
3457   __ profile_call(R11_scratch1, R12_scratch2);
3458   // Argument and return type profiling.
3459   __ profile_arguments_type(Rmethod, R11_scratch1, R12_scratch2, false);
3460   __ call_from_interpreter(Rmethod, Rret_addr, R11_scratch1, R12_scratch2);
3461 }
3462 
3463 void TemplateTable::invokestatic(int byte_no) {
3464   assert(byte_no == f1_byte, "use this argument");
3465   transition(vtos, vtos);
3466 
3467   Register Rtable_addr = R3_ARG1,
3468            Rret_addr   = R4_ARG2,
3469            Rflags      = R5_ARG3;
3470 
3471   prepare_invoke(byte_no, R19_method, Rret_addr, noreg, noreg, Rflags, R11_scratch1);
3472 
3473   __ profile_call(R11_scratch1, R12_scratch2);
3474   // Argument and return type profiling.
3475   __ profile_arguments_type(R19_method, R11_scratch1, R12_scratch2, false);
3476   __ call_from_interpreter(R19_method, Rret_addr, R11_scratch1, R12_scratch2);
3477 }
3478 
3479 void TemplateTable::invokeinterface_object_method(Register Rrecv_klass,
3480                                                   Register Rret,
3481                                                   Register Rflags,
3482                                                   Register Rindex,
3483                                                   Register Rtemp1,
3484                                                   Register Rtemp2) {
3485 
3486   assert_different_registers(Rindex, Rret, Rrecv_klass, Rflags, Rtemp1, Rtemp2);
3487   Label LnotFinal;
3488 
3489   // Check for vfinal.
3490   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_vfinal_shift);
3491   __ bfalse(CCR0, LnotFinal);
3492 
3493   Register Rscratch = Rflags; // Rflags is dead now.
3494 
3495   // Final call case.
3496   __ profile_final_call(Rtemp1, Rscratch);
3497   // Argument and return type profiling.
3498   __ profile_arguments_type(Rindex, Rscratch, Rrecv_klass /* scratch */, true);
3499   // Do the final call - the index (f2) contains the method.
3500   __ call_from_interpreter(Rindex, Rret, Rscratch, Rrecv_klass /* scratch */);
3501 
3502   // Non-final callc case.
3503   __ bind(LnotFinal);
3504   __ profile_virtual_call(Rrecv_klass, Rtemp1, Rscratch, false);
3505   generate_vtable_call(Rrecv_klass, Rindex, Rret, Rscratch);
3506 }
3507 
3508 void TemplateTable::invokeinterface(int byte_no) {
3509   assert(byte_no == f1_byte, "use this argument");
3510   transition(vtos, vtos);
3511 
3512   const Register Rscratch1        = R11_scratch1,
3513                  Rscratch2        = R12_scratch2,
3514                  Rscratch3        = R9_ARG7,
3515                  Rscratch4        = R10_ARG8,
3516                  Rtable_addr      = Rscratch2,
3517                  Rinterface_klass = R5_ARG3,
3518                  Rret_type        = R8_ARG6,
3519                  Rret_addr        = Rret_type,
3520                  Rindex           = R6_ARG4,
3521                  Rreceiver        = R4_ARG2,
3522                  Rrecv_klass      = Rreceiver,
3523                  Rflags           = R7_ARG5;
3524 
3525   prepare_invoke(byte_no, Rinterface_klass, Rret_addr, Rindex, Rreceiver, Rflags, Rscratch1);
3526 
3527   // Get receiver klass.
3528   __ null_check_throw(Rreceiver, oopDesc::klass_offset_in_bytes(), Rscratch3);
3529   __ load_klass(Rrecv_klass, Rreceiver);
3530 
3531   // Check corner case object method.
3532   Label LobjectMethod;
3533 
3534   __ testbitdi(CCR0, R0, Rflags, ConstantPoolCacheEntry::is_forced_virtual_shift);
3535   __ btrue(CCR0, LobjectMethod);
3536 
3537   // Fallthrough: The normal invokeinterface case.
3538   __ profile_virtual_call(Rrecv_klass, Rscratch1, Rscratch2, false);
3539 
3540   // Find entry point to call.
3541   Label Lthrow_icc, Lthrow_ame;
3542   // Result will be returned in Rindex.
3543   __ mr(Rscratch4, Rrecv_klass);
3544   __ mr(Rscratch3, Rindex);
3545   __ lookup_interface_method(Rrecv_klass, Rinterface_klass, Rindex, Rindex, Rscratch1, Rscratch2, Lthrow_icc);
3546 
3547   __ cmpdi(CCR0, Rindex, 0);
3548   __ beq(CCR0, Lthrow_ame);
3549   // Found entry. Jump off!
3550   // Argument and return type profiling.
3551   __ profile_arguments_type(Rindex, Rscratch1, Rscratch2, true);
3552   __ call_from_interpreter(Rindex, Rret_addr, Rscratch1, Rscratch2);
3553 
3554   // Vtable entry was NULL => Throw abstract method error.
3555   __ bind(Lthrow_ame);
3556   __ mr(Rrecv_klass, Rscratch4);
3557   __ mr(Rindex, Rscratch3);
3558   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3559 
3560   // Interface was not found => Throw incompatible class change error.
3561   __ bind(Lthrow_icc);
3562   __ mr(Rrecv_klass, Rscratch4);
3563   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3564 
3565   __ should_not_reach_here();
3566 
3567   // Special case of invokeinterface called for virtual method of
3568   // java.lang.Object. See ConstantPoolCacheEntry::set_method() for details:
3569   // The invokeinterface was rewritten to a invokevirtual, hence we have
3570   // to handle this corner case. This code isn't produced by javac, but could
3571   // be produced by another compliant java compiler.
3572   __ bind(LobjectMethod);
3573   invokeinterface_object_method(Rrecv_klass, Rret_addr, Rflags, Rindex, Rscratch1, Rscratch2);
3574 }
3575 
3576 void TemplateTable::invokedynamic(int byte_no) {
3577   transition(vtos, vtos);
3578 
3579   const Register Rret_addr = R3_ARG1,
3580                  Rflags    = R4_ARG2,
3581                  Rmethod   = R22_tmp2,
3582                  Rscratch1 = R11_scratch1,
3583                  Rscratch2 = R12_scratch2;
3584 
3585   prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, noreg, Rflags, Rscratch2);
3586 
3587   // Profile this call.
3588   __ profile_call(Rscratch1, Rscratch2);
3589 
3590   // Off we go. With the new method handles, we don't jump to a method handle
3591   // entry any more. Instead, we pushed an "appendix" in prepare invoke, which happens
3592   // to be the callsite object the bootstrap method returned. This is passed to a
3593   // "link" method which does the dispatch (Most likely just grabs the MH stored
3594   // inside the callsite and does an invokehandle).
3595   // Argument and return type profiling.
3596   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, false);
3597   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3598 }
3599 
3600 void TemplateTable::invokehandle(int byte_no) {
3601   transition(vtos, vtos);
3602 
3603   const Register Rret_addr = R3_ARG1,
3604                  Rflags    = R4_ARG2,
3605                  Rrecv     = R5_ARG3,
3606                  Rmethod   = R22_tmp2,
3607                  Rscratch1 = R11_scratch1,
3608                  Rscratch2 = R12_scratch2;
3609 
3610   prepare_invoke(byte_no, Rmethod, Rret_addr, Rscratch1, Rrecv, Rflags, Rscratch2);
3611   __ verify_method_ptr(Rmethod);
3612   __ null_check_throw(Rrecv, -1, Rscratch2);
3613 
3614   __ profile_final_call(Rrecv, Rscratch1);
3615 
3616   // Still no call from handle => We call the method handle interpreter here.
3617   // Argument and return type profiling.
3618   __ profile_arguments_type(Rmethod, Rscratch1, Rscratch2, true);
3619   __ call_from_interpreter(Rmethod, Rret_addr, Rscratch1 /* scratch1 */, Rscratch2 /* scratch2 */);
3620 }
3621 
3622 // =============================================================================
3623 // Allocation
3624 
3625 // Puts allocated obj ref onto the expression stack.
3626 void TemplateTable::_new() {
3627   transition(vtos, atos);
3628 
3629   Label Lslow_case,
3630         Ldone,
3631         Linitialize_header,
3632         Lallocate_shared,
3633         Linitialize_object;  // Including clearing the fields.
3634 
3635   const Register RallocatedObject = R17_tos,
3636                  RinstanceKlass   = R9_ARG7,
3637                  Rscratch         = R11_scratch1,
3638                  Roffset          = R8_ARG6,
3639                  Rinstance_size   = Roffset,
3640                  Rcpool           = R4_ARG2,
3641                  Rtags            = R3_ARG1,
3642                  Rindex           = R5_ARG3;
3643 
3644   const bool allow_shared_alloc = Universe::heap()->supports_inline_contig_alloc();
3645 
3646   // --------------------------------------------------------------------------
3647   // Check if fast case is possible.
3648 
3649   // Load pointers to const pool and const pool's tags array.
3650   __ get_cpool_and_tags(Rcpool, Rtags);
3651   // Load index of constant pool entry.
3652   __ get_2_byte_integer_at_bcp(1, Rindex, InterpreterMacroAssembler::Unsigned);
3653 
3654   if (UseTLAB) {
3655     // Make sure the class we're about to instantiate has been resolved
3656     // This is done before loading instanceKlass to be consistent with the order
3657     // how Constant Pool is updated (see ConstantPoolCache::klass_at_put).
3658     __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3659     __ lbzx(Rtags, Rindex, Rtags);
3660 
3661     __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3662     __ bne(CCR0, Lslow_case);
3663 
3664     // Get instanceKlass
3665     __ sldi(Roffset, Rindex, LogBytesPerWord);
3666     __ load_resolved_klass_at_offset(Rcpool, Roffset, RinstanceKlass);
3667 
3668     // Make sure klass is fully initialized and get instance_size.
3669     __ lbz(Rscratch, in_bytes(InstanceKlass::init_state_offset()), RinstanceKlass);
3670     __ lwz(Rinstance_size, in_bytes(Klass::layout_helper_offset()), RinstanceKlass);
3671 
3672     __ cmpdi(CCR1, Rscratch, InstanceKlass::fully_initialized);
3673     // Make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class.
3674     __ andi_(R0, Rinstance_size, Klass::_lh_instance_slow_path_bit); // slow path bit equals 0?
3675 
3676     __ crnand(CCR0, Assembler::equal, CCR1, Assembler::equal); // slow path bit set or not fully initialized?
3677     __ beq(CCR0, Lslow_case);
3678 
3679     // --------------------------------------------------------------------------
3680     // Fast case:
3681     // Allocate the instance.
3682     // 1) Try to allocate in the TLAB.
3683     // 2) If fail, and the TLAB is not full enough to discard, allocate in the shared Eden.
3684     // 3) If the above fails (or is not applicable), go to a slow case (creates a new TLAB, etc.).
3685 
3686     Register RoldTopValue = RallocatedObject; // Object will be allocated here if it fits.
3687     Register RnewTopValue = R6_ARG4;
3688     Register RendValue    = R7_ARG5;
3689 
3690     // Check if we can allocate in the TLAB.
3691     __ ld(RoldTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3692     __ ld(RendValue,    in_bytes(JavaThread::tlab_end_offset()), R16_thread);
3693 
3694     __ add(RnewTopValue, Rinstance_size, RoldTopValue);
3695 
3696     // If there is enough space, we do not CAS and do not clear.
3697     __ cmpld(CCR0, RnewTopValue, RendValue);
3698     __ bgt(CCR0, allow_shared_alloc ? Lallocate_shared : Lslow_case);
3699 
3700     __ std(RnewTopValue, in_bytes(JavaThread::tlab_top_offset()), R16_thread);
3701 
3702     if (ZeroTLAB) {
3703       // The fields have already been cleared.
3704       __ b(Linitialize_header);
3705     } else {
3706       // Initialize both the header and fields.
3707       __ b(Linitialize_object);
3708     }
3709 
3710     // Fall through: TLAB was too small.
3711     if (allow_shared_alloc) {
3712       Register RtlabWasteLimitValue = R10_ARG8;
3713       Register RfreeValue = RnewTopValue;
3714 
3715       __ bind(Lallocate_shared);
3716       // Check if tlab should be discarded (refill_waste_limit >= free).
3717       __ ld(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3718       __ subf(RfreeValue, RoldTopValue, RendValue);
3719       __ srdi(RfreeValue, RfreeValue, LogHeapWordSize); // in dwords
3720       __ cmpld(CCR0, RtlabWasteLimitValue, RfreeValue);
3721       __ bge(CCR0, Lslow_case);
3722 
3723       // Increment waste limit to prevent getting stuck on this slow path.
3724       __ add_const_optimized(RtlabWasteLimitValue, RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment());
3725       __ std(RtlabWasteLimitValue, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), R16_thread);
3726     }
3727     // else: No allocation in the shared eden. // fallthru: __ b(Lslow_case);
3728   }
3729   // else: Always go the slow path.
3730 
3731   // --------------------------------------------------------------------------
3732   // slow case
3733   __ bind(Lslow_case);
3734   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), Rcpool, Rindex);
3735 
3736   if (UseTLAB) {
3737     __ b(Ldone);
3738     // --------------------------------------------------------------------------
3739     // Init1: Zero out newly allocated memory.
3740 
3741     if (!ZeroTLAB || allow_shared_alloc) {
3742       // Clear object fields.
3743       __ bind(Linitialize_object);
3744 
3745       // Initialize remaining object fields.
3746       Register Rbase = Rtags;
3747       __ addi(Rinstance_size, Rinstance_size, 7 - (int)sizeof(oopDesc));
3748       __ addi(Rbase, RallocatedObject, sizeof(oopDesc));
3749       __ srdi(Rinstance_size, Rinstance_size, 3);
3750 
3751       // Clear out object skipping header. Takes also care of the zero length case.
3752       __ clear_memory_doubleword(Rbase, Rinstance_size);
3753       // fallthru: __ b(Linitialize_header);
3754     }
3755 
3756     // --------------------------------------------------------------------------
3757     // Init2: Initialize the header: mark, klass
3758     __ bind(Linitialize_header);
3759 
3760     // Init mark.
3761     if (UseBiasedLocking) {
3762       __ ld(Rscratch, in_bytes(Klass::prototype_header_offset()), RinstanceKlass);
3763     } else {
3764       __ load_const_optimized(Rscratch, markOopDesc::prototype(), R0);
3765     }
3766     __ std(Rscratch, oopDesc::mark_offset_in_bytes(), RallocatedObject);
3767 
3768     // Init klass.
3769     __ store_klass_gap(RallocatedObject);
3770     __ store_klass(RallocatedObject, RinstanceKlass, Rscratch); // klass (last for cms)
3771 
3772     // Check and trigger dtrace event.
3773     {
3774       SkipIfEqualZero skip_if(_masm, Rscratch, &DTraceAllocProbes);
3775       __ push(atos);
3776       __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc));
3777       __ pop(atos);
3778     }
3779   }
3780 
3781   // continue
3782   __ bind(Ldone);
3783 
3784   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3785   __ membar(Assembler::StoreStore);
3786 }
3787 
3788 void TemplateTable::newarray() {
3789   transition(itos, atos);
3790 
3791   __ lbz(R4, 1, R14_bcp);
3792   __ extsw(R5, R17_tos);
3793   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), R4, R5 /* size */);
3794 
3795   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3796   __ membar(Assembler::StoreStore);
3797 }
3798 
3799 void TemplateTable::anewarray() {
3800   transition(itos, atos);
3801 
3802   __ get_constant_pool(R4);
3803   __ get_2_byte_integer_at_bcp(1, R5, InterpreterMacroAssembler::Unsigned);
3804   __ extsw(R6, R17_tos); // size
3805   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), R4 /* pool */, R5 /* index */, R6 /* size */);
3806 
3807   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3808   __ membar(Assembler::StoreStore);
3809 }
3810 
3811 // Allocate a multi dimensional array
3812 void TemplateTable::multianewarray() {
3813   transition(vtos, atos);
3814 
3815   Register Rptr = R31; // Needs to survive C call.
3816 
3817   // Put ndims * wordSize into frame temp slot
3818   __ lbz(Rptr, 3, R14_bcp);
3819   __ sldi(Rptr, Rptr, Interpreter::logStackElementSize);
3820   // Esp points past last_dim, so set to R4 to first_dim address.
3821   __ add(R4, Rptr, R15_esp);
3822   call_VM(R17_tos, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), R4 /* first_size_address */);
3823   // Pop all dimensions off the stack.
3824   __ add(R15_esp, Rptr, R15_esp);
3825 
3826   // Must prevent reordering of stores for object initialization with stores that publish the new object.
3827   __ membar(Assembler::StoreStore);
3828 }
3829 
3830 void TemplateTable::arraylength() {
3831   transition(atos, itos);
3832 
3833   Label LnoException;
3834   __ verify_oop(R17_tos);
3835   __ null_check_throw(R17_tos, arrayOopDesc::length_offset_in_bytes(), R11_scratch1);
3836   __ lwa(R17_tos, arrayOopDesc::length_offset_in_bytes(), R17_tos);
3837 }
3838 
3839 // ============================================================================
3840 // Typechecks
3841 
3842 void TemplateTable::checkcast() {
3843   transition(atos, atos);
3844 
3845   Label Ldone, Lis_null, Lquicked, Lresolved;
3846   Register Roffset         = R6_ARG4,
3847            RobjKlass       = R4_ARG2,
3848            RspecifiedKlass = R5_ARG3, // Generate_ClassCastException_verbose_handler will read value from this register.
3849            Rcpool          = R11_scratch1,
3850            Rtags           = R12_scratch2;
3851 
3852   // Null does not pass.
3853   __ cmpdi(CCR0, R17_tos, 0);
3854   __ beq(CCR0, Lis_null);
3855 
3856   // Get constant pool tag to find out if the bytecode has already been "quickened".
3857   __ get_cpool_and_tags(Rcpool, Rtags);
3858 
3859   __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3860 
3861   __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3862   __ lbzx(Rtags, Rtags, Roffset);
3863 
3864   __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3865   __ beq(CCR0, Lquicked);
3866 
3867   // Call into the VM to "quicken" instanceof.
3868   __ push_ptr();  // for GC
3869   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3870   __ get_vm_result_2(RspecifiedKlass);
3871   __ pop_ptr();   // Restore receiver.
3872   __ b(Lresolved);
3873 
3874   // Extract target class from constant pool.
3875   __ bind(Lquicked);
3876   __ sldi(Roffset, Roffset, LogBytesPerWord);
3877   __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
3878 
3879   // Do the checkcast.
3880   __ bind(Lresolved);
3881   // Get value klass in RobjKlass.
3882   __ load_klass(RobjKlass, R17_tos);
3883   // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3884   __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3885 
3886   // Not a subtype; so must throw exception
3887   // Target class oop is in register R6_ARG4 == RspecifiedKlass by convention.
3888   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::_throw_ClassCastException_entry);
3889   __ mtctr(R11_scratch1);
3890   __ bctr();
3891 
3892   // Profile the null case.
3893   __ align(32, 12);
3894   __ bind(Lis_null);
3895   __ profile_null_seen(R11_scratch1, Rtags); // Rtags used as scratch.
3896 
3897   __ align(32, 12);
3898   __ bind(Ldone);
3899 }
3900 
3901 // Output:
3902 //   - tos == 0: Obj was null or not an instance of class.
3903 //   - tos == 1: Obj was an instance of class.
3904 void TemplateTable::instanceof() {
3905   transition(atos, itos);
3906 
3907   Label Ldone, Lis_null, Lquicked, Lresolved;
3908   Register Roffset         = R6_ARG4,
3909            RobjKlass       = R4_ARG2,
3910            RspecifiedKlass = R5_ARG3,
3911            Rcpool          = R11_scratch1,
3912            Rtags           = R12_scratch2;
3913 
3914   // Null does not pass.
3915   __ cmpdi(CCR0, R17_tos, 0);
3916   __ beq(CCR0, Lis_null);
3917 
3918   // Get constant pool tag to find out if the bytecode has already been "quickened".
3919   __ get_cpool_and_tags(Rcpool, Rtags);
3920 
3921   __ get_2_byte_integer_at_bcp(1, Roffset, InterpreterMacroAssembler::Unsigned);
3922 
3923   __ addi(Rtags, Rtags, Array<u1>::base_offset_in_bytes());
3924   __ lbzx(Rtags, Rtags, Roffset);
3925 
3926   __ cmpdi(CCR0, Rtags, JVM_CONSTANT_Class);
3927   __ beq(CCR0, Lquicked);
3928 
3929   // Call into the VM to "quicken" instanceof.
3930   __ push_ptr();  // for GC
3931   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3932   __ get_vm_result_2(RspecifiedKlass);
3933   __ pop_ptr();   // Restore receiver.
3934   __ b(Lresolved);
3935 
3936   // Extract target class from constant pool.
3937   __ bind(Lquicked);
3938   __ sldi(Roffset, Roffset, LogBytesPerWord);
3939   __ load_resolved_klass_at_offset(Rcpool, Roffset, RspecifiedKlass);
3940 
3941   // Do the checkcast.
3942   __ bind(Lresolved);
3943   // Get value klass in RobjKlass.
3944   __ load_klass(RobjKlass, R17_tos);
3945   // Generate a fast subtype check. Branch to cast_ok if no failure. Return 0 if failure.
3946   __ li(R17_tos, 1);
3947   __ gen_subtype_check(RobjKlass, RspecifiedKlass, /*3 temp regs*/ Roffset, Rcpool, Rtags, /*target if subtype*/ Ldone);
3948   __ li(R17_tos, 0);
3949 
3950   if (ProfileInterpreter) {
3951     __ b(Ldone);
3952   }
3953 
3954   // Profile the null case.
3955   __ align(32, 12);
3956   __ bind(Lis_null);
3957   __ profile_null_seen(Rcpool, Rtags); // Rcpool and Rtags used as scratch.
3958 
3959   __ align(32, 12);
3960   __ bind(Ldone);
3961 }
3962 
3963 // =============================================================================
3964 // Breakpoints
3965 
3966 void TemplateTable::_breakpoint() {
3967   transition(vtos, vtos);
3968 
3969   // Get the unpatched byte code.
3970   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), R19_method, R14_bcp);
3971   __ mr(R31, R3_RET);
3972 
3973   // Post the breakpoint event.
3974   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), R19_method, R14_bcp);
3975 
3976   // Complete the execution of original bytecode.
3977   __ dispatch_Lbyte_code(vtos, R31, Interpreter::normal_table(vtos));
3978 }
3979 
3980 // =============================================================================
3981 // Exceptions
3982 
3983 void TemplateTable::athrow() {
3984   transition(atos, vtos);
3985 
3986   // Exception oop is in tos
3987   __ verify_oop(R17_tos);
3988 
3989   __ null_check_throw(R17_tos, -1, R11_scratch1);
3990 
3991   // Throw exception interpreter entry expects exception oop to be in R3.
3992   __ mr(R3_RET, R17_tos);
3993   __ load_dispatch_table(R11_scratch1, (address*)Interpreter::throw_exception_entry());
3994   __ mtctr(R11_scratch1);
3995   __ bctr();
3996 }
3997 
3998 // =============================================================================
3999 // Synchronization
4000 // Searches the basic object lock list on the stack for a free slot
4001 // and uses it to lock the obect in tos.
4002 //
4003 // Recursive locking is enabled by exiting the search if the same
4004 // object is already found in the list. Thus, a new basic lock obj lock
4005 // is allocated "higher up" in the stack and thus is found first
4006 // at next monitor exit.
4007 void TemplateTable::monitorenter() {
4008   transition(atos, vtos);
4009 
4010   __ verify_oop(R17_tos);
4011 
4012   Register Rcurrent_monitor  = R11_scratch1,
4013            Rcurrent_obj      = R12_scratch2,
4014            Robj_to_lock      = R17_tos,
4015            Rscratch1         = R3_ARG1,
4016            Rscratch2         = R4_ARG2,
4017            Rscratch3         = R5_ARG3,
4018            Rcurrent_obj_addr = R6_ARG4;
4019 
4020   // ------------------------------------------------------------------------------
4021   // Null pointer exception.
4022   __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4023 
4024   // Try to acquire a lock on the object.
4025   // Repeat until succeeded (i.e., until monitorenter returns true).
4026 
4027   // ------------------------------------------------------------------------------
4028   // Find a free slot in the monitor block.
4029   Label Lfound, Lexit, Lallocate_new;
4030   ConditionRegister found_free_slot = CCR0,
4031                     found_same_obj  = CCR1,
4032                     reached_limit   = CCR6;
4033   {
4034     Label Lloop, Lentry;
4035     Register Rlimit = Rcurrent_monitor;
4036 
4037     // Set up search loop - start with topmost monitor.
4038     __ add(Rcurrent_obj_addr, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
4039 
4040     __ ld(Rlimit, 0, R1_SP);
4041     __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes() - BasicObjectLock::obj_offset_in_bytes())); // Monitor base
4042 
4043     // Check if any slot is present => short cut to allocation if not.
4044     __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4045     __ bgt(reached_limit, Lallocate_new);
4046 
4047     // Pre-load topmost slot.
4048     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4049     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4050     // The search loop.
4051     __ bind(Lloop);
4052     // Found free slot?
4053     __ cmpdi(found_free_slot, Rcurrent_obj, 0);
4054     // Is this entry for same obj? If so, stop the search and take the found
4055     // free slot or allocate a new one to enable recursive locking.
4056     __ cmpd(found_same_obj, Rcurrent_obj, Robj_to_lock);
4057     __ cmpld(reached_limit, Rcurrent_obj_addr, Rlimit);
4058     __ beq(found_free_slot, Lexit);
4059     __ beq(found_same_obj, Lallocate_new);
4060     __ bgt(reached_limit, Lallocate_new);
4061     // Check if last allocated BasicLockObj reached.
4062     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4063     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4064     // Next iteration if unchecked BasicObjectLocks exist on the stack.
4065     __ b(Lloop);
4066   }
4067 
4068   // ------------------------------------------------------------------------------
4069   // Check if we found a free slot.
4070   __ bind(Lexit);
4071 
4072   __ addi(Rcurrent_monitor, Rcurrent_obj_addr, -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4073   __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, - frame::interpreter_frame_monitor_size() * wordSize);
4074   __ b(Lfound);
4075 
4076   // We didn't find a free BasicObjLock => allocate one.
4077   __ align(32, 12);
4078   __ bind(Lallocate_new);
4079   __ add_monitor_to_stack(false, Rscratch1, Rscratch2);
4080   __ mr(Rcurrent_monitor, R26_monitor);
4081   __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4082 
4083   // ------------------------------------------------------------------------------
4084   // We now have a slot to lock.
4085   __ bind(Lfound);
4086 
4087   // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
4088   // The object has already been poped from the stack, so the expression stack looks correct.
4089   __ addi(R14_bcp, R14_bcp, 1);
4090 
4091   __ std(Robj_to_lock, 0, Rcurrent_obj_addr);
4092   __ lock_object(Rcurrent_monitor, Robj_to_lock);
4093 
4094   // Check if there's enough space on the stack for the monitors after locking.
4095   // This emits a single store.
4096   __ generate_stack_overflow_check(0);
4097 
4098   // The bcp has already been incremented. Just need to dispatch to next instruction.
4099   __ dispatch_next(vtos);
4100 }
4101 
4102 void TemplateTable::monitorexit() {
4103   transition(atos, vtos);
4104   __ verify_oop(R17_tos);
4105 
4106   Register Rcurrent_monitor  = R11_scratch1,
4107            Rcurrent_obj      = R12_scratch2,
4108            Robj_to_lock      = R17_tos,
4109            Rcurrent_obj_addr = R3_ARG1,
4110            Rlimit            = R4_ARG2;
4111   Label Lfound, Lillegal_monitor_state;
4112 
4113   // Check corner case: unbalanced monitorEnter / Exit.
4114   __ ld(Rlimit, 0, R1_SP);
4115   __ addi(Rlimit, Rlimit, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base
4116 
4117   // Null pointer check.
4118   __ null_check_throw(Robj_to_lock, -1, R11_scratch1);
4119 
4120   __ cmpld(CCR0, R26_monitor, Rlimit);
4121   __ bgt(CCR0, Lillegal_monitor_state);
4122 
4123   // Find the corresponding slot in the monitors stack section.
4124   {
4125     Label Lloop;
4126 
4127     // Start with topmost monitor.
4128     __ addi(Rcurrent_obj_addr, R26_monitor, BasicObjectLock::obj_offset_in_bytes());
4129     __ addi(Rlimit, Rlimit, BasicObjectLock::obj_offset_in_bytes());
4130     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4131     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4132 
4133     __ bind(Lloop);
4134     // Is this entry for same obj?
4135     __ cmpd(CCR0, Rcurrent_obj, Robj_to_lock);
4136     __ beq(CCR0, Lfound);
4137 
4138     // Check if last allocated BasicLockObj reached.
4139 
4140     __ ld(Rcurrent_obj, 0, Rcurrent_obj_addr);
4141     __ cmpld(CCR0, Rcurrent_obj_addr, Rlimit);
4142     __ addi(Rcurrent_obj_addr, Rcurrent_obj_addr, frame::interpreter_frame_monitor_size() * wordSize);
4143 
4144     // Next iteration if unchecked BasicObjectLocks exist on the stack.
4145     __ ble(CCR0, Lloop);
4146   }
4147 
4148   // Fell through without finding the basic obj lock => throw up!
4149   __ bind(Lillegal_monitor_state);
4150   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
4151   __ should_not_reach_here();
4152 
4153   __ align(32, 12);
4154   __ bind(Lfound);
4155   __ addi(Rcurrent_monitor, Rcurrent_obj_addr,
4156           -(frame::interpreter_frame_monitor_size() * wordSize) - BasicObjectLock::obj_offset_in_bytes());
4157   __ unlock_object(Rcurrent_monitor);
4158 }
4159 
4160 // ============================================================================
4161 // Wide bytecodes
4162 
4163 // Wide instructions. Simply redirects to the wide entry point for that instruction.
4164 void TemplateTable::wide() {
4165   transition(vtos, vtos);
4166 
4167   const Register Rtable = R11_scratch1,
4168                  Rindex = R12_scratch2,
4169                  Rtmp   = R0;
4170 
4171   __ lbz(Rindex, 1, R14_bcp);
4172 
4173   __ load_dispatch_table(Rtable, Interpreter::_wentry_point);
4174 
4175   __ slwi(Rindex, Rindex, LogBytesPerWord);
4176   __ ldx(Rtmp, Rtable, Rindex);
4177   __ mtctr(Rtmp);
4178   __ bctr();
4179   // Note: the bcp increment step is part of the individual wide bytecode implementations.
4180 }