1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_sparc.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "logging/log.hpp"
  30 #include "oops/arrayOop.hpp"
  31 #include "oops/markOop.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/method.hpp"
  34 #include "oops/methodCounters.hpp"
  35 #include "prims/jvmtiExport.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/thread.inline.hpp"
  41 #include "utilities/align.hpp"
  42 
  43 // Implementation of InterpreterMacroAssembler
  44 
  45 // This file specializes the assember with interpreter-specific macros
  46 
  47 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
  48 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
  49 
  50 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  51   assert(entry, "Entry must have been generated by now");
  52   AddressLiteral al(entry);
  53   jump_to(al, G3_scratch);
  54   delayed()->nop();
  55 }
  56 
  57 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
  58   // Note: this algorithm is also used by C1's OSR entry sequence.
  59   // Any changes should also be applied to CodeEmitter::emit_osr_entry().
  60   assert_different_registers(args_size, locals_size);
  61   // max_locals*2 for TAGS.  Assumes that args_size has already been adjusted.
  62   subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
  63   // Use br/mov combination because it works on both V8 and V9 and is
  64   // faster.
  65   Label skip_move;
  66   br(Assembler::negative, true, Assembler::pt, skip_move);
  67   delayed()->mov(G0, delta);
  68   bind(skip_move);
  69   align_up(delta, WordsPerLong);       // make multiple of 2 (SP must be 2-word aligned)
  70   sll(delta, LogBytesPerWord, delta);  // extra space for locals in bytes
  71 }
  72 
  73 // Dispatch code executed in the prolog of a bytecode which does not do it's
  74 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
  75 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
  76   assert_not_delayed();
  77   ldub( Lbcp, bcp_incr, Lbyte_code);                    // load next bytecode
  78   // dispatch table to use
  79   AddressLiteral tbl(Interpreter::dispatch_table(state));
  80   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);         // multiply by wordSize
  81   set(tbl, G3_scratch);                                 // compute addr of table
  82   ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress);     // get entry addr
  83 }
  84 
  85 
  86 // Dispatch code executed in the epilog of a bytecode which does not do it's
  87 // own dispatch. The dispatch address in IdispatchAddress is used for the
  88 // dispatch.
  89 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
  90   assert_not_delayed();
  91   verify_FPU(1, state);
  92   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
  93   jmp( IdispatchAddress, 0 );
  94   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
  95   else                delayed()->nop();
  96 }
  97 
  98 
  99 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
 100   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 101   assert_not_delayed();
 102   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 103   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
 104 }
 105 
 106 
 107 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
 108   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 109   assert_not_delayed();
 110   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 111   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
 112 }
 113 
 114 
 115 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 116   // load current bytecode
 117   assert_not_delayed();
 118   ldub( Lbcp, 0, Lbyte_code);               // load next bytecode
 119   dispatch_base(state, table);
 120 }
 121 
 122 
 123 void InterpreterMacroAssembler::call_VM_leaf_base(
 124   Register java_thread,
 125   address  entry_point,
 126   int      number_of_arguments
 127 ) {
 128   if (!java_thread->is_valid())
 129     java_thread = L7_thread_cache;
 130   // super call
 131   MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
 132 }
 133 
 134 
 135 void InterpreterMacroAssembler::call_VM_base(
 136   Register        oop_result,
 137   Register        java_thread,
 138   Register        last_java_sp,
 139   address         entry_point,
 140   int             number_of_arguments,
 141   bool            check_exception
 142 ) {
 143   if (!java_thread->is_valid())
 144     java_thread = L7_thread_cache;
 145   // See class ThreadInVMfromInterpreter, which assumes that the interpreter
 146   // takes responsibility for setting its own thread-state on call-out.
 147   // However, ThreadInVMfromInterpreter resets the state to "in_Java".
 148 
 149   //save_bcp();                                  // save bcp
 150   MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
 151   //restore_bcp();                               // restore bcp
 152   //restore_locals();                            // restore locals pointer
 153 }
 154 
 155 
 156 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
 157   if (JvmtiExport::can_pop_frame()) {
 158     Label L;
 159 
 160     // Check the "pending popframe condition" flag in the current thread
 161     ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
 162 
 163     // Initiate popframe handling only if it is not already being processed.  If the flag
 164     // has the popframe_processing bit set, it means that this code is called *during* popframe
 165     // handling - we don't want to reenter.
 166     btst(JavaThread::popframe_pending_bit, scratch_reg);
 167     br(zero, false, pt, L);
 168     delayed()->nop();
 169     btst(JavaThread::popframe_processing_bit, scratch_reg);
 170     br(notZero, false, pt, L);
 171     delayed()->nop();
 172 
 173     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 174     // address of the same-named entrypoint in the generated interpreter code.
 175     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 176 
 177     // Jump to Interpreter::_remove_activation_preserving_args_entry
 178     jmpl(O0, G0, G0);
 179     delayed()->nop();
 180     bind(L);
 181   }
 182 }
 183 
 184 
 185 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 186   Register thr_state = G4_scratch;
 187   ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 188   const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
 189   const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
 190   const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
 191   switch (state) {
 192   case ltos: ld_long(val_addr, Otos_l);                   break;
 193   case atos: ld_ptr(oop_addr, Otos_l);
 194              st_ptr(G0, oop_addr);                        break;
 195   case btos:                                           // fall through
 196   case ztos:                                           // fall through
 197   case ctos:                                           // fall through
 198   case stos:                                           // fall through
 199   case itos: ld(val_addr, Otos_l1);                       break;
 200   case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
 201   case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
 202   case vtos: /* nothing to do */                          break;
 203   default  : ShouldNotReachHere();
 204   }
 205   // Clean up tos value in the jvmti thread state
 206   or3(G0, ilgl, G3_scratch);
 207   stw(G3_scratch, tos_addr);
 208   st_long(G0, val_addr);
 209   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 210 }
 211 
 212 
 213 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
 214   if (JvmtiExport::can_force_early_return()) {
 215     Label L;
 216     Register thr_state = G3_scratch;
 217     ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 218     br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
 219 
 220     // Initiate earlyret handling only if it is not already being processed.
 221     // If the flag has the earlyret_processing bit set, it means that this code
 222     // is called *during* earlyret handling - we don't want to reenter.
 223     ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
 224     cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
 225 
 226     // Call Interpreter::remove_activation_early_entry() to get the address of the
 227     // same-named entrypoint in the generated interpreter code
 228     ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
 229     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
 230 
 231     // Jump to Interpreter::_remove_activation_early_entry
 232     jmpl(O0, G0, G0);
 233     delayed()->nop();
 234     bind(L);
 235   }
 236 }
 237 
 238 
 239 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
 240   mov(arg_1, O0);
 241   mov(arg_2, O1);
 242   MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
 243 }
 244 
 245 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
 246   assert_not_delayed();
 247   dispatch_Lbyte_code(state, table);
 248 }
 249 
 250 
 251 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
 252   dispatch_base(state, Interpreter::normal_table(state));
 253 }
 254 
 255 
 256 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 257   dispatch_base(state, Interpreter::dispatch_table(state));
 258 }
 259 
 260 
 261 // common code to dispatch and dispatch_only
 262 // dispatch value in Lbyte_code and increment Lbcp
 263 
 264 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
 265   verify_FPU(1, state);
 266   // %%%%% maybe implement +VerifyActivationFrameSize here
 267   //verify_thread(); //too slow; we will just verify on method entry & exit
 268   if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 269   // dispatch table to use
 270   AddressLiteral tbl(table);
 271   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);       // multiply by wordSize
 272   set(tbl, G3_scratch);                               // compute addr of table
 273   ld_ptr(G3_scratch, Lbyte_code, G3_scratch);         // get entry addr
 274   jmp( G3_scratch, 0 );
 275   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
 276   else                delayed()->nop();
 277 }
 278 
 279 
 280 // Helpers for expression stack
 281 
 282 // Longs and doubles are Category 2 computational types in the
 283 // JVM specification (section 3.11.1) and take 2 expression stack or
 284 // local slots.
 285 // Aligning them on 32 bit with tagged stacks is hard because the code generated
 286 // for the dup* bytecodes depends on what types are already on the stack.
 287 // If the types are split into the two stack/local slots, that is much easier
 288 // (and we can use 0 for non-reference tags).
 289 
 290 // Known good alignment in _LP64 but unknown otherwise
 291 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
 292   assert_not_delayed();
 293 
 294   ldf(FloatRegisterImpl::D, r1, offset, d);
 295 }
 296 
 297 // Known good alignment in _LP64 but unknown otherwise
 298 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
 299   assert_not_delayed();
 300 
 301   stf(FloatRegisterImpl::D, d, r1, offset);
 302   // store something more useful here
 303   debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
 304 }
 305 
 306 
 307 // Known good alignment in _LP64 but unknown otherwise
 308 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
 309   assert_not_delayed();
 310   ldx(r1, offset, rd);
 311 }
 312 
 313 // Known good alignment in _LP64 but unknown otherwise
 314 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
 315   assert_not_delayed();
 316 
 317   stx(l, r1, offset);
 318   // store something more useful here
 319   stx(G0, r1, offset+Interpreter::stackElementSize);
 320 }
 321 
 322 void InterpreterMacroAssembler::pop_i(Register r) {
 323   assert_not_delayed();
 324   ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 325   inc(Lesp, Interpreter::stackElementSize);
 326   debug_only(verify_esp(Lesp));
 327 }
 328 
 329 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
 330   assert_not_delayed();
 331   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 332   inc(Lesp, Interpreter::stackElementSize);
 333   debug_only(verify_esp(Lesp));
 334 }
 335 
 336 void InterpreterMacroAssembler::pop_l(Register r) {
 337   assert_not_delayed();
 338   load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 339   inc(Lesp, 2*Interpreter::stackElementSize);
 340   debug_only(verify_esp(Lesp));
 341 }
 342 
 343 
 344 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
 345   assert_not_delayed();
 346   ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
 347   inc(Lesp, Interpreter::stackElementSize);
 348   debug_only(verify_esp(Lesp));
 349 }
 350 
 351 
 352 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
 353   assert_not_delayed();
 354   load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
 355   inc(Lesp, 2*Interpreter::stackElementSize);
 356   debug_only(verify_esp(Lesp));
 357 }
 358 
 359 
 360 void InterpreterMacroAssembler::push_i(Register r) {
 361   assert_not_delayed();
 362   debug_only(verify_esp(Lesp));
 363   st(r, Lesp, 0);
 364   dec(Lesp, Interpreter::stackElementSize);
 365 }
 366 
 367 void InterpreterMacroAssembler::push_ptr(Register r) {
 368   assert_not_delayed();
 369   st_ptr(r, Lesp, 0);
 370   dec(Lesp, Interpreter::stackElementSize);
 371 }
 372 
 373 // remember: our convention for longs in SPARC is:
 374 // O0 (Otos_l1) has high-order part in first word,
 375 // O1 (Otos_l2) has low-order part in second word
 376 
 377 void InterpreterMacroAssembler::push_l(Register r) {
 378   assert_not_delayed();
 379   debug_only(verify_esp(Lesp));
 380   // Longs are stored in memory-correct order, even if unaligned.
 381   int offset = -Interpreter::stackElementSize;
 382   store_unaligned_long(r, Lesp, offset);
 383   dec(Lesp, 2 * Interpreter::stackElementSize);
 384 }
 385 
 386 
 387 void InterpreterMacroAssembler::push_f(FloatRegister f) {
 388   assert_not_delayed();
 389   debug_only(verify_esp(Lesp));
 390   stf(FloatRegisterImpl::S, f, Lesp, 0);
 391   dec(Lesp, Interpreter::stackElementSize);
 392 }
 393 
 394 
 395 void InterpreterMacroAssembler::push_d(FloatRegister d)   {
 396   assert_not_delayed();
 397   debug_only(verify_esp(Lesp));
 398   // Longs are stored in memory-correct order, even if unaligned.
 399   int offset = -Interpreter::stackElementSize;
 400   store_unaligned_double(d, Lesp, offset);
 401   dec(Lesp, 2 * Interpreter::stackElementSize);
 402 }
 403 
 404 
 405 void InterpreterMacroAssembler::push(TosState state) {
 406   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 407   switch (state) {
 408     case atos: push_ptr();            break;
 409     case btos:                        // fall through
 410     case ztos:                        // fall through
 411     case ctos:                        // fall through
 412     case stos:                        // fall through
 413     case itos: push_i();              break;
 414     case ltos: push_l();              break;
 415     case ftos: push_f();              break;
 416     case dtos: push_d();              break;
 417     case vtos: /* nothing to do */    break;
 418     default  : ShouldNotReachHere();
 419   }
 420 }
 421 
 422 
 423 void InterpreterMacroAssembler::pop(TosState state) {
 424   switch (state) {
 425     case atos: pop_ptr();            break;
 426     case btos:                       // fall through
 427     case ztos:                       // fall through
 428     case ctos:                       // fall through
 429     case stos:                       // fall through
 430     case itos: pop_i();              break;
 431     case ltos: pop_l();              break;
 432     case ftos: pop_f();              break;
 433     case dtos: pop_d();              break;
 434     case vtos: /* nothing to do */   break;
 435     default  : ShouldNotReachHere();
 436   }
 437   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 438 }
 439 
 440 
 441 // Helpers for swap and dup
 442 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 443   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
 444 }
 445 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 446   st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
 447 }
 448 
 449 
 450 void InterpreterMacroAssembler::load_receiver(Register param_count,
 451                                               Register recv) {
 452   sll(param_count, Interpreter::logStackElementSize, param_count);
 453   ld_ptr(Lesp, param_count, recv);  // gets receiver oop
 454 }
 455 
 456 void InterpreterMacroAssembler::empty_expression_stack() {
 457   // Reset Lesp.
 458   sub( Lmonitors, wordSize, Lesp );
 459 
 460   // Reset SP by subtracting more space from Lesp.
 461   Label done;
 462   assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
 463 
 464   // A native does not need to do this, since its callee does not change SP.
 465   ld(Lmethod, Method::access_flags_offset(), Gframe_size);  // Load access flags.
 466   btst(JVM_ACC_NATIVE, Gframe_size);
 467   br(Assembler::notZero, false, Assembler::pt, done);
 468   delayed()->nop();
 469 
 470   // Compute max expression stack+register save area
 471   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
 472   lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size);  // Load max stack.
 473   add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size );
 474 
 475   //
 476   // now set up a stack frame with the size computed above
 477   //
 478   //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
 479   sll( Gframe_size, LogBytesPerWord, Gframe_size );
 480   sub( Lesp, Gframe_size, Gframe_size );
 481   and3( Gframe_size, -(2 * wordSize), Gframe_size );          // align SP (downwards) to an 8/16-byte boundary
 482   debug_only(verify_sp(Gframe_size, G4_scratch));
 483   sub(Gframe_size, STACK_BIAS, Gframe_size );
 484   mov(Gframe_size, SP);
 485 
 486   bind(done);
 487 }
 488 
 489 
 490 #ifdef ASSERT
 491 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
 492   Label Bad, OK;
 493 
 494   // Saved SP must be aligned.
 495   btst(2*BytesPerWord-1, Rsp);
 496   br(Assembler::notZero, false, Assembler::pn, Bad);
 497   delayed()->nop();
 498 
 499   // Saved SP, plus register window size, must not be above FP.
 500   add(Rsp, frame::register_save_words * wordSize, Rtemp);
 501   sub(Rtemp, STACK_BIAS, Rtemp);  // Bias Rtemp before cmp to FP
 502   cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
 503 
 504   // Saved SP must not be ridiculously below current SP.
 505   size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
 506   set(maxstack, Rtemp);
 507   sub(SP, Rtemp, Rtemp);
 508   add(Rtemp, STACK_BIAS, Rtemp);  // Unbias Rtemp before cmp to Rsp
 509   cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
 510 
 511   ba_short(OK);
 512 
 513   bind(Bad);
 514   stop("on return to interpreted call, restored SP is corrupted");
 515 
 516   bind(OK);
 517 }
 518 
 519 
 520 void InterpreterMacroAssembler::verify_esp(Register Resp) {
 521   // about to read or write Resp[0]
 522   // make sure it is not in the monitors or the register save area
 523   Label OK1, OK2;
 524 
 525   cmp(Resp, Lmonitors);
 526   brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
 527   delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
 528   stop("too many pops:  Lesp points into monitor area");
 529   bind(OK1);
 530   sub(Resp, STACK_BIAS, Resp);
 531   cmp(Resp, SP);
 532   brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
 533   delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
 534   stop("too many pushes:  Lesp points into register window");
 535   bind(OK2);
 536 }
 537 #endif // ASSERT
 538 
 539 // Load compiled (i2c) or interpreter entry when calling from interpreted and
 540 // do the call. Centralized so that all interpreter calls will do the same actions.
 541 // If jvmti single stepping is on for a thread we must not call compiled code.
 542 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
 543 
 544   // Assume we want to go compiled if available
 545 
 546   ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
 547 
 548   if (JvmtiExport::can_post_interpreter_events()) {
 549     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 550     // compiled code in threads for which the event is enabled.  Check here for
 551     // interp_only_mode if these events CAN be enabled.
 552     verify_thread();
 553     Label skip_compiled_code;
 554 
 555     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
 556     ld(interp_only, scratch);
 557     cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
 558     delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
 559     bind(skip_compiled_code);
 560   }
 561 
 562   // the i2c_adapters need Method* in G5_method (right? %%%)
 563   // do the call
 564 #ifdef ASSERT
 565   {
 566     Label ok;
 567     br_notnull_short(target, Assembler::pt, ok);
 568     stop("null entry point");
 569     bind(ok);
 570   }
 571 #endif // ASSERT
 572 
 573   // Adjust Rret first so Llast_SP can be same as Rret
 574   add(Rret, -frame::pc_return_offset, O7);
 575   add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
 576   // Record SP so we can remove any stack space allocated by adapter transition
 577   jmp(target, 0);
 578   delayed()->mov(SP, Llast_SP);
 579 }
 580 
 581 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
 582   assert_not_delayed();
 583 
 584   Label not_taken;
 585   if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
 586   else             br (cc, false, Assembler::pn, not_taken);
 587   delayed()->nop();
 588 
 589   TemplateTable::branch(false,false);
 590 
 591   bind(not_taken);
 592 
 593   profile_not_taken_branch(G3_scratch);
 594 }
 595 
 596 
 597 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
 598                                   int         bcp_offset,
 599                                   Register    Rtmp,
 600                                   Register    Rdst,
 601                                   signedOrNot is_signed,
 602                                   setCCOrNot  should_set_CC ) {
 603   assert(Rtmp != Rdst, "need separate temp register");
 604   assert_not_delayed();
 605   switch (is_signed) {
 606    default: ShouldNotReachHere();
 607 
 608    case   Signed:  ldsb( Lbcp, bcp_offset, Rdst  );  break; // high byte
 609    case Unsigned:  ldub( Lbcp, bcp_offset, Rdst  );  break; // high byte
 610   }
 611   ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
 612   sll( Rdst, BitsPerByte, Rdst);
 613   switch (should_set_CC ) {
 614    default: ShouldNotReachHere();
 615 
 616    case      set_CC:  orcc( Rdst, Rtmp, Rdst ); break;
 617    case dont_set_CC:  or3(  Rdst, Rtmp, Rdst ); break;
 618   }
 619 }
 620 
 621 
 622 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
 623                                   int        bcp_offset,
 624                                   Register   Rtmp,
 625                                   Register   Rdst,
 626                                   setCCOrNot should_set_CC ) {
 627   assert(Rtmp != Rdst, "need separate temp register");
 628   assert_not_delayed();
 629   add( Lbcp, bcp_offset, Rtmp);
 630   andcc( Rtmp, 3, G0);
 631   Label aligned;
 632   switch (should_set_CC ) {
 633    default: ShouldNotReachHere();
 634 
 635    case      set_CC: break;
 636    case dont_set_CC: break;
 637   }
 638 
 639   br(Assembler::zero, true, Assembler::pn, aligned);
 640   delayed()->ldsw(Rtmp, 0, Rdst);
 641 
 642   ldub(Lbcp, bcp_offset + 3, Rdst);
 643   ldub(Lbcp, bcp_offset + 2, Rtmp);  sll(Rtmp,  8, Rtmp);  or3(Rtmp, Rdst, Rdst);
 644   ldub(Lbcp, bcp_offset + 1, Rtmp);  sll(Rtmp, 16, Rtmp);  or3(Rtmp, Rdst, Rdst);
 645   ldsb(Lbcp, bcp_offset + 0, Rtmp);  sll(Rtmp, 24, Rtmp);
 646   or3(Rtmp, Rdst, Rdst );
 647 
 648   bind(aligned);
 649   if (should_set_CC == set_CC) tst(Rdst);
 650 }
 651 
 652 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
 653                                                        int bcp_offset, size_t index_size) {
 654   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 655   if (index_size == sizeof(u2)) {
 656     get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
 657   } else if (index_size == sizeof(u4)) {
 658     get_4_byte_integer_at_bcp(bcp_offset, temp, index);
 659     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 660     xor3(index, -1, index);  // convert to plain index
 661   } else if (index_size == sizeof(u1)) {
 662     ldub(Lbcp, bcp_offset, index);
 663   } else {
 664     ShouldNotReachHere();
 665   }
 666 }
 667 
 668 
 669 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
 670                                                            int bcp_offset, size_t index_size) {
 671   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 672   assert_different_registers(cache, tmp);
 673   assert_not_delayed();
 674   get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
 675   // convert from field index to ConstantPoolCacheEntry index and from
 676   // word index to byte offset
 677   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 678   add(LcpoolCache, tmp, cache);
 679 }
 680 
 681 
 682 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 683                                                                         Register temp,
 684                                                                         Register bytecode,
 685                                                                         int byte_no,
 686                                                                         int bcp_offset,
 687                                                                         size_t index_size) {
 688   get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
 689   ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
 690   const int shift_count = (1 + byte_no) * BitsPerByte;
 691   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 692          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 693          "correct shift count");
 694   srl(bytecode, shift_count, bytecode);
 695   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 696   and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
 697 }
 698 
 699 
 700 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
 701                                                                int bcp_offset, size_t index_size) {
 702   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 703   assert_different_registers(cache, tmp);
 704   assert_not_delayed();
 705   if (index_size == sizeof(u2)) {
 706     get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
 707   } else {
 708     ShouldNotReachHere();  // other sizes not supported here
 709   }
 710               // convert from field index to ConstantPoolCacheEntry index
 711               // and from word index to byte offset
 712   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 713               // skip past the header
 714   add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
 715               // construct pointer to cache entry
 716   add(LcpoolCache, tmp, cache);
 717 }
 718 
 719 
 720 // Load object from cpool->resolved_references(index)
 721 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 722                                            Register result, Register index) {
 723   assert_different_registers(result, index);
 724   assert_not_delayed();
 725   // convert from field index to resolved_references() index and from
 726   // word index to byte offset. Since this is a java object, it can be compressed
 727   Register tmp = index;  // reuse
 728   sll(index, LogBytesPerHeapOop, tmp);
 729   get_constant_pool(result);
 730   // load pointer for resolved_references[] objArray
 731   ld_ptr(result, ConstantPool::cache_offset_in_bytes(), result);
 732   ld_ptr(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
 733   // JNIHandles::resolve(result)
 734   ld_ptr(result, 0, result);
 735   // Add in the index
 736   add(result, tmp, result);
 737   load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
 738 }
 739 
 740 
 741 // load cpool->resolved_klass_at(index)
 742 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register Rcpool,
 743                                            Register Roffset, Register Rklass) {
 744   // int value = *this_cp->int_at_addr(which);
 745   // int resolved_klass_index = extract_low_short_from_int(value);
 746   //
 747   // Because SPARC is big-endian, the low_short is at (cpool->int_at_addr(which) + 2 bytes)
 748   add(Roffset, Rcpool, Roffset);
 749   lduh(Roffset, sizeof(ConstantPool) + 2, Roffset);  // Roffset = resolved_klass_index
 750 
 751   Register Rresolved_klasses = Rklass;
 752   ld_ptr(Rcpool, ConstantPool::resolved_klasses_offset_in_bytes(), Rresolved_klasses);
 753   sll(Roffset, LogBytesPerWord, Roffset);
 754   add(Roffset, Array<Klass*>::base_offset_in_bytes(), Roffset);
 755   ld_ptr(Rresolved_klasses, Roffset, Rklass);
 756 }
 757 
 758 
 759 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
 760 // a subtype of super_klass.  Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
 761 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 762                                                   Register Rsuper_klass,
 763                                                   Register Rtmp1,
 764                                                   Register Rtmp2,
 765                                                   Register Rtmp3,
 766                                                   Label &ok_is_subtype ) {
 767   Label not_subtype;
 768 
 769   // Profile the not-null value's klass.
 770   profile_typecheck(Rsub_klass, Rtmp1);
 771 
 772   check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
 773                                 Rtmp1, Rtmp2,
 774                                 &ok_is_subtype, &not_subtype, NULL);
 775 
 776   check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
 777                                 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
 778                                 &ok_is_subtype, NULL);
 779 
 780   bind(not_subtype);
 781   profile_typecheck_failed(Rtmp1);
 782 }
 783 
 784 // Separate these two to allow for delay slot in middle
 785 // These are used to do a test and full jump to exception-throwing code.
 786 
 787 // %%%%% Could possibly reoptimize this by testing to see if could use
 788 // a single conditional branch (i.e. if span is small enough.
 789 // If you go that route, than get rid of the split and give up
 790 // on the delay-slot hack.
 791 
 792 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
 793                                                     Label&    ok ) {
 794   assert_not_delayed();
 795   br(ok_condition, true, pt, ok);
 796   // DELAY SLOT
 797 }
 798 
 799 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
 800                                                     Label&    ok ) {
 801   assert_not_delayed();
 802   bp( ok_condition, true, Assembler::xcc, pt, ok);
 803   // DELAY SLOT
 804 }
 805 
 806 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
 807                                                   Label&    ok ) {
 808   assert_not_delayed();
 809   brx(ok_condition, true, pt, ok);
 810   // DELAY SLOT
 811 }
 812 
 813 void InterpreterMacroAssembler::throw_if_not_2( address  throw_entry_point,
 814                                                 Register Rscratch,
 815                                                 Label&   ok ) {
 816   assert(throw_entry_point != NULL, "entry point must be generated by now");
 817   AddressLiteral dest(throw_entry_point);
 818   jump_to(dest, Rscratch);
 819   delayed()->nop();
 820   bind(ok);
 821 }
 822 
 823 
 824 // And if you cannot use the delay slot, here is a shorthand:
 825 
 826 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
 827                                                   address   throw_entry_point,
 828                                                   Register  Rscratch ) {
 829   Label ok;
 830   if (ok_condition != never) {
 831     throw_if_not_1_icc( ok_condition, ok);
 832     delayed()->nop();
 833   }
 834   throw_if_not_2( throw_entry_point, Rscratch, ok);
 835 }
 836 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
 837                                                   address   throw_entry_point,
 838                                                   Register  Rscratch ) {
 839   Label ok;
 840   if (ok_condition != never) {
 841     throw_if_not_1_xcc( ok_condition, ok);
 842     delayed()->nop();
 843   }
 844   throw_if_not_2( throw_entry_point, Rscratch, ok);
 845 }
 846 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
 847                                                 address   throw_entry_point,
 848                                                 Register  Rscratch ) {
 849   Label ok;
 850   if (ok_condition != never) {
 851     throw_if_not_1_x( ok_condition, ok);
 852     delayed()->nop();
 853   }
 854   throw_if_not_2( throw_entry_point, Rscratch, ok);
 855 }
 856 
 857 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
 858 // Note: res is still shy of address by array offset into object.
 859 
 860 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
 861   assert_not_delayed();
 862 
 863   verify_oop(array);
 864   // sign extend since tos (index) can be a 32bit value
 865   sra(index, G0, index);
 866 
 867   // check array
 868   Label ptr_ok;
 869   tst(array);
 870   throw_if_not_1_x( notZero, ptr_ok );
 871   delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
 872   throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
 873 
 874   Label index_ok;
 875   cmp(index, tmp);
 876   throw_if_not_1_icc( lessUnsigned, index_ok );
 877   if (index_shift > 0)  delayed()->sll(index, index_shift, index);
 878   else                  delayed()->add(array, index, res); // addr - const offset in index
 879   // convention: move aberrant index into G3_scratch for exception message
 880   mov(index, G3_scratch);
 881   throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
 882 
 883   // add offset if didn't do it in delay slot
 884   if (index_shift > 0)   add(array, index, res); // addr - const offset in index
 885 }
 886 
 887 
 888 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
 889   assert_not_delayed();
 890 
 891   // pop array
 892   pop_ptr(array);
 893 
 894   // check array
 895   index_check_without_pop(array, index, index_shift, tmp, res);
 896 }
 897 
 898 
 899 void InterpreterMacroAssembler::get_const(Register Rdst) {
 900   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
 901 }
 902 
 903 
 904 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
 905   get_const(Rdst);
 906   ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
 907 }
 908 
 909 
 910 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
 911   get_constant_pool(Rdst);
 912   ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
 913 }
 914 
 915 
 916 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
 917   get_constant_pool(Rcpool);
 918   ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
 919 }
 920 
 921 
 922 // unlock if synchronized method
 923 //
 924 // Unlock the receiver if this is a synchronized method.
 925 // Unlock any Java monitors from syncronized blocks.
 926 //
 927 // If there are locked Java monitors
 928 //    If throw_monitor_exception
 929 //       throws IllegalMonitorStateException
 930 //    Else if install_monitor_exception
 931 //       installs IllegalMonitorStateException
 932 //    Else
 933 //       no error processing
 934 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
 935                                                               bool throw_monitor_exception,
 936                                                               bool install_monitor_exception) {
 937   Label unlocked, unlock, no_unlock;
 938 
 939   // get the value of _do_not_unlock_if_synchronized into G1_scratch
 940   const Address do_not_unlock_if_synchronized(G2_thread,
 941     JavaThread::do_not_unlock_if_synchronized_offset());
 942   ldbool(do_not_unlock_if_synchronized, G1_scratch);
 943   stbool(G0, do_not_unlock_if_synchronized); // reset the flag
 944 
 945   // check if synchronized method
 946   const Address access_flags(Lmethod, Method::access_flags_offset());
 947   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 948   push(state); // save tos
 949   ld(access_flags, G3_scratch); // Load access flags.
 950   btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
 951   br(zero, false, pt, unlocked);
 952   delayed()->nop();
 953 
 954   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 955   // is set.
 956   cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
 957   delayed()->nop();
 958 
 959   // BasicObjectLock will be first in list, since this is a synchronized method. However, need
 960   // to check that the object has not been unlocked by an explicit monitorexit bytecode.
 961 
 962   //Intel: if (throw_monitor_exception) ... else ...
 963   // Entry already unlocked, need to throw exception
 964   //...
 965 
 966   // pass top-most monitor elem
 967   add( top_most_monitor(), O1 );
 968 
 969   ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
 970   br_notnull_short(G3_scratch, pt, unlock);
 971 
 972   if (throw_monitor_exception) {
 973     // Entry already unlocked need to throw an exception
 974     MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
 975     should_not_reach_here();
 976   } else {
 977     // Monitor already unlocked during a stack unroll.
 978     // If requested, install an illegal_monitor_state_exception.
 979     // Continue with stack unrolling.
 980     if (install_monitor_exception) {
 981       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
 982     }
 983     ba_short(unlocked);
 984   }
 985 
 986   bind(unlock);
 987 
 988   unlock_object(O1);
 989 
 990   bind(unlocked);
 991 
 992   // I0, I1: Might contain return value
 993 
 994   // Check that all monitors are unlocked
 995   { Label loop, exception, entry, restart;
 996 
 997     Register Rmptr   = O0;
 998     Register Rtemp   = O1;
 999     Register Rlimit  = Lmonitors;
1000     const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1001     assert( (delta & LongAlignmentMask) == 0,
1002             "sizeof BasicObjectLock must be even number of doublewords");
1003 
1004     #ifdef ASSERT
1005     add(top_most_monitor(), Rmptr, delta);
1006     { Label L;
1007       // ensure that Rmptr starts out above (or at) Rlimit
1008       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1009       stop("monitor stack has negative size");
1010       bind(L);
1011     }
1012     #endif
1013     bind(restart);
1014     ba(entry);
1015     delayed()->
1016     add(top_most_monitor(), Rmptr, delta);      // points to current entry, starting with bottom-most entry
1017 
1018     // Entry is still locked, need to throw exception
1019     bind(exception);
1020     if (throw_monitor_exception) {
1021       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1022       should_not_reach_here();
1023     } else {
1024       // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1025       // Unlock does not block, so don't have to worry about the frame
1026       unlock_object(Rmptr);
1027       if (install_monitor_exception) {
1028         MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1029       }
1030       ba_short(restart);
1031     }
1032 
1033     bind(loop);
1034     cmp(Rtemp, G0);                             // check if current entry is used
1035     brx(Assembler::notEqual, false, pn, exception);
1036     delayed()->
1037     dec(Rmptr, delta);                          // otherwise advance to next entry
1038     #ifdef ASSERT
1039     { Label L;
1040       // ensure that Rmptr has not somehow stepped below Rlimit
1041       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1042       stop("ran off the end of the monitor stack");
1043       bind(L);
1044     }
1045     #endif
1046     bind(entry);
1047     cmp(Rmptr, Rlimit);                         // check if bottom reached
1048     brx(Assembler::notEqual, true, pn, loop);   // if not at bottom then check this entry
1049     delayed()->
1050     ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1051   }
1052 
1053   bind(no_unlock);
1054   pop(state);
1055   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1056 }
1057 
1058 void InterpreterMacroAssembler::narrow(Register result) {
1059 
1060   ld_ptr(Address(Lmethod, Method::const_offset()), G3_scratch);
1061   ldub(G3_scratch, in_bytes(ConstMethod::result_type_offset()), G3_scratch);
1062 
1063   Label notBool, notByte, notChar, done;
1064 
1065   // common case first
1066   cmp(G3_scratch, T_INT);
1067   br(Assembler::equal, true, pn, done);
1068   delayed()->nop();
1069 
1070   cmp(G3_scratch, T_BOOLEAN);
1071   br(Assembler::notEqual, true, pn, notBool);
1072   delayed()->cmp(G3_scratch, T_BYTE);
1073   and3(result, 1, result);
1074   ba(done);
1075   delayed()->nop();
1076 
1077   bind(notBool);
1078   // cmp(G3_scratch, T_BYTE);
1079   br(Assembler::notEqual, true, pn, notByte);
1080   delayed()->cmp(G3_scratch, T_CHAR);
1081   sll(result, 24, result);
1082   sra(result, 24, result);
1083   ba(done);
1084   delayed()->nop();
1085 
1086   bind(notByte);
1087   // cmp(G3_scratch, T_CHAR);
1088   sll(result, 16, result);
1089   br(Assembler::notEqual, true, pn, done);
1090   delayed()->sra(result, 16, result);
1091   // sll(result, 16, result);
1092   srl(result, 16, result);
1093 
1094   // bind(notChar);
1095   // must be short, instructions already executed in delay slot
1096   // sll(result, 16, result);
1097   // sra(result, 16, result);
1098 
1099   bind(done);
1100 }
1101 
1102 // remove activation
1103 //
1104 // Unlock the receiver if this is a synchronized method.
1105 // Unlock any Java monitors from syncronized blocks.
1106 // Remove the activation from the stack.
1107 //
1108 // If there are locked Java monitors
1109 //    If throw_monitor_exception
1110 //       throws IllegalMonitorStateException
1111 //    Else if install_monitor_exception
1112 //       installs IllegalMonitorStateException
1113 //    Else
1114 //       no error processing
1115 void InterpreterMacroAssembler::remove_activation(TosState state,
1116                                                   bool throw_monitor_exception,
1117                                                   bool install_monitor_exception) {
1118 
1119   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1120 
1121   // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1122   notify_method_exit(false, state, NotifyJVMTI);
1123 
1124   if (StackReservedPages > 0) {
1125     // testing if Stack Reserved Area needs to be re-enabled
1126     Label no_reserved_zone_enabling;
1127     ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch);
1128     cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling);
1129 
1130     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread);
1131     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread);
1132     should_not_reach_here();
1133 
1134     bind(no_reserved_zone_enabling);
1135   }
1136 
1137   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1138   verify_thread();
1139 
1140   // return tos
1141   assert(Otos_l1 == Otos_i, "adjust code below");
1142   switch (state) {
1143   case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1144   case btos:                                      // fall through
1145   case ztos:                                      // fall through
1146   case ctos:
1147   case stos:                                      // fall through
1148   case atos:                                      // fall through
1149   case itos: mov(Otos_l1, Otos_l1->after_save());    break;        // O0 -> I0
1150   case ftos:                                      // fall through
1151   case dtos:                                      // fall through
1152   case vtos: /* nothing to do */                     break;
1153   default  : ShouldNotReachHere();
1154   }
1155 }
1156 
1157 // Lock object
1158 //
1159 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1160 //            it must be initialized with the object to lock
1161 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1162   if (UseHeavyMonitors) {
1163     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1164   }
1165   else {
1166     Register obj_reg = Object;
1167     Register mark_reg = G4_scratch;
1168     Register temp_reg = G1_scratch;
1169     Address  lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1170     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1171     Label    done;
1172 
1173     Label slow_case;
1174 
1175     assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1176 
1177     // load markOop from object into mark_reg
1178     ld_ptr(mark_addr, mark_reg);
1179 
1180     if (UseBiasedLocking) {
1181       biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1182     }
1183 
1184     // get the address of basicLock on stack that will be stored in the object
1185     // we need a temporary register here as we do not want to clobber lock_reg
1186     // (cas clobbers the destination register)
1187     mov(lock_reg, temp_reg);
1188     // set mark reg to be (markOop of object | UNLOCK_VALUE)
1189     or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1190     // initialize the box  (Must happen before we update the object mark!)
1191     st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1192     // compare and exchange object_addr, markOop | 1, stack address of basicLock
1193     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1194     cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1195 
1196     // if the compare and exchange succeeded we are done (we saw an unlocked object)
1197     cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1198 
1199     // We did not see an unlocked object so try the fast recursive case
1200 
1201     // Check if owner is self by comparing the value in the markOop of object
1202     // with the stack pointer
1203     sub(temp_reg, SP, temp_reg);
1204     sub(temp_reg, STACK_BIAS, temp_reg);
1205     assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1206 
1207     // Composite "andcc" test:
1208     // (a) %sp -vs- markword proximity check, and,
1209     // (b) verify mark word LSBs == 0 (Stack-locked).
1210     //
1211     // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1212     // Note that the page size used for %sp proximity testing is arbitrary and is
1213     // unrelated to the actual MMU page size.  We use a 'logical' page size of
1214     // 4096 bytes.   F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1215     // field of the andcc instruction.
1216     andcc (temp_reg, 0xFFFFF003, G0) ;
1217 
1218     // if condition is true we are done and hence we can store 0 in the displaced
1219     // header indicating it is a recursive lock and be done
1220     brx(Assembler::zero, true, Assembler::pt, done);
1221     delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1222 
1223     // none of the above fast optimizations worked so we have to get into the
1224     // slow case of monitor enter
1225     bind(slow_case);
1226     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1227 
1228     bind(done);
1229   }
1230 }
1231 
1232 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1233 //
1234 // Argument - lock_reg points to the BasicObjectLock for lock
1235 // Throw IllegalMonitorException if object is not locked by current thread
1236 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1237   if (UseHeavyMonitors) {
1238     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1239   } else {
1240     Register obj_reg = G3_scratch;
1241     Register mark_reg = G4_scratch;
1242     Register displaced_header_reg = G1_scratch;
1243     Address  lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1244     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1245     Label    done;
1246 
1247     if (UseBiasedLocking) {
1248       // load the object out of the BasicObjectLock
1249       ld_ptr(lockobj_addr, obj_reg);
1250       biased_locking_exit(mark_addr, mark_reg, done, true);
1251       st_ptr(G0, lockobj_addr);  // free entry
1252     }
1253 
1254     // Test first if we are in the fast recursive case
1255     Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1256     ld_ptr(lock_addr, displaced_header_reg);
1257     br_null(displaced_header_reg, true, Assembler::pn, done);
1258     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1259 
1260     // See if it is still a light weight lock, if so we just unlock
1261     // the object and we are done
1262 
1263     if (!UseBiasedLocking) {
1264       // load the object out of the BasicObjectLock
1265       ld_ptr(lockobj_addr, obj_reg);
1266     }
1267 
1268     // we have the displaced header in displaced_header_reg
1269     // we expect to see the stack address of the basicLock in case the
1270     // lock is still a light weight lock (lock_reg)
1271     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1272     cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1273     cmp(lock_reg, displaced_header_reg);
1274     brx(Assembler::equal, true, Assembler::pn, done);
1275     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1276 
1277     // The lock has been converted into a heavy lock and hence
1278     // we need to get into the slow case
1279 
1280     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1281 
1282     bind(done);
1283   }
1284 }
1285 
1286 // Get the method data pointer from the Method* and set the
1287 // specified register to its value.
1288 
1289 void InterpreterMacroAssembler::set_method_data_pointer() {
1290   assert(ProfileInterpreter, "must be profiling interpreter");
1291   Label get_continue;
1292 
1293   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1294   test_method_data_pointer(get_continue);
1295   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1296   bind(get_continue);
1297 }
1298 
1299 // Set the method data pointer for the current bcp.
1300 
1301 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1302   assert(ProfileInterpreter, "must be profiling interpreter");
1303   Label zero_continue;
1304 
1305   // Test MDO to avoid the call if it is NULL.
1306   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1307   test_method_data_pointer(zero_continue);
1308   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1309   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1310   add(ImethodDataPtr, O0, ImethodDataPtr);
1311   bind(zero_continue);
1312 }
1313 
1314 // Test ImethodDataPtr.  If it is null, continue at the specified label
1315 
1316 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1317   assert(ProfileInterpreter, "must be profiling interpreter");
1318   br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1319 }
1320 
1321 void InterpreterMacroAssembler::verify_method_data_pointer() {
1322   assert(ProfileInterpreter, "must be profiling interpreter");
1323 #ifdef ASSERT
1324   Label verify_continue;
1325   test_method_data_pointer(verify_continue);
1326 
1327   // If the mdp is valid, it will point to a DataLayout header which is
1328   // consistent with the bcp.  The converse is highly probable also.
1329   lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1330   ld_ptr(Lmethod, Method::const_offset(), O5);
1331   add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1332   add(G3_scratch, O5, G3_scratch);
1333   cmp(Lbcp, G3_scratch);
1334   brx(Assembler::equal, false, Assembler::pt, verify_continue);
1335 
1336   Register temp_reg = O5;
1337   delayed()->mov(ImethodDataPtr, temp_reg);
1338   // %%% should use call_VM_leaf here?
1339   //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1340   save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1341   Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1342   stf(FloatRegisterImpl::D, Ftos_d, d_save);
1343   mov(temp_reg->after_save(), O2);
1344   save_thread(L7_thread_cache);
1345   call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1346   delayed()->nop();
1347   restore_thread(L7_thread_cache);
1348   ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1349   restore();
1350   bind(verify_continue);
1351 #endif // ASSERT
1352 }
1353 
1354 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1355                                                                 Register method_counters,
1356                                                                 Register Rtmp,
1357                                                                 Label &profile_continue) {
1358   assert(ProfileInterpreter, "must be profiling interpreter");
1359   // Control will flow to "profile_continue" if the counter is less than the
1360   // limit or if we call profile_method()
1361 
1362   Label done;
1363 
1364   // if no method data exists, and the counter is high enough, make one
1365   br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1366 
1367   // Test to see if we should create a method data oop
1368   Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset());
1369   ld(profile_limit, Rtmp);
1370   cmp(invocation_count, Rtmp);
1371   // Use long branches because call_VM() code and following code generated by
1372   // test_backedge_count_for_osr() is large in debug VM.
1373   br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1374   delayed()->nop();
1375 
1376   // Build it now.
1377   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1378   set_method_data_pointer_for_bcp();
1379   ba(profile_continue);
1380   delayed()->nop();
1381   bind(done);
1382 }
1383 
1384 // Store a value at some constant offset from the method data pointer.
1385 
1386 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1387   assert(ProfileInterpreter, "must be profiling interpreter");
1388   st_ptr(value, ImethodDataPtr, constant);
1389 }
1390 
1391 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1392                                                       Register bumped_count,
1393                                                       bool decrement) {
1394   assert(ProfileInterpreter, "must be profiling interpreter");
1395 
1396   // Load the counter.
1397   ld_ptr(counter, bumped_count);
1398 
1399   if (decrement) {
1400     // Decrement the register.  Set condition codes.
1401     subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1402 
1403     // If the decrement causes the counter to overflow, stay negative
1404     Label L;
1405     brx(Assembler::negative, true, Assembler::pn, L);
1406 
1407     // Store the decremented counter, if it is still negative.
1408     delayed()->st_ptr(bumped_count, counter);
1409     bind(L);
1410   } else {
1411     // Increment the register.  Set carry flag.
1412     addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1413 
1414     // If the increment causes the counter to overflow, pull back by 1.
1415     assert(DataLayout::counter_increment == 1, "subc works");
1416     subc(bumped_count, G0, bumped_count);
1417 
1418     // Store the incremented counter.
1419     st_ptr(bumped_count, counter);
1420   }
1421 }
1422 
1423 // Increment the value at some constant offset from the method data pointer.
1424 
1425 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1426                                                       Register bumped_count,
1427                                                       bool decrement) {
1428   // Locate the counter at a fixed offset from the mdp:
1429   Address counter(ImethodDataPtr, constant);
1430   increment_mdp_data_at(counter, bumped_count, decrement);
1431 }
1432 
1433 // Increment the value at some non-fixed (reg + constant) offset from
1434 // the method data pointer.
1435 
1436 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1437                                                       int constant,
1438                                                       Register bumped_count,
1439                                                       Register scratch2,
1440                                                       bool decrement) {
1441   // Add the constant to reg to get the offset.
1442   add(ImethodDataPtr, reg, scratch2);
1443   Address counter(scratch2, constant);
1444   increment_mdp_data_at(counter, bumped_count, decrement);
1445 }
1446 
1447 // Set a flag value at the current method data pointer position.
1448 // Updates a single byte of the header, to avoid races with other header bits.
1449 
1450 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1451                                                 Register scratch) {
1452   assert(ProfileInterpreter, "must be profiling interpreter");
1453   // Load the data header
1454   ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1455 
1456   // Set the flag
1457   or3(scratch, flag_constant, scratch);
1458 
1459   // Store the modified header.
1460   stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1461 }
1462 
1463 // Test the location at some offset from the method data pointer.
1464 // If it is not equal to value, branch to the not_equal_continue Label.
1465 // Set condition codes to match the nullness of the loaded value.
1466 
1467 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1468                                                  Register value,
1469                                                  Label& not_equal_continue,
1470                                                  Register scratch) {
1471   assert(ProfileInterpreter, "must be profiling interpreter");
1472   ld_ptr(ImethodDataPtr, offset, scratch);
1473   cmp(value, scratch);
1474   brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1475   delayed()->tst(scratch);
1476 }
1477 
1478 // Update the method data pointer by the displacement located at some fixed
1479 // offset from the method data pointer.
1480 
1481 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1482                                                      Register scratch) {
1483   assert(ProfileInterpreter, "must be profiling interpreter");
1484   ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1485   add(ImethodDataPtr, scratch, ImethodDataPtr);
1486 }
1487 
1488 // Update the method data pointer by the displacement located at the
1489 // offset (reg + offset_of_disp).
1490 
1491 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1492                                                      int offset_of_disp,
1493                                                      Register scratch) {
1494   assert(ProfileInterpreter, "must be profiling interpreter");
1495   add(reg, offset_of_disp, scratch);
1496   ld_ptr(ImethodDataPtr, scratch, scratch);
1497   add(ImethodDataPtr, scratch, ImethodDataPtr);
1498 }
1499 
1500 // Update the method data pointer by a simple constant displacement.
1501 
1502 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1503   assert(ProfileInterpreter, "must be profiling interpreter");
1504   add(ImethodDataPtr, constant, ImethodDataPtr);
1505 }
1506 
1507 // Update the method data pointer for a _ret bytecode whose target
1508 // was not among our cached targets.
1509 
1510 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1511                                                    Register return_bci) {
1512   assert(ProfileInterpreter, "must be profiling interpreter");
1513   push(state);
1514   st_ptr(return_bci, l_tmp);  // protect return_bci, in case it is volatile
1515   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1516   ld_ptr(l_tmp, return_bci);
1517   pop(state);
1518 }
1519 
1520 // Count a taken branch in the bytecodes.
1521 
1522 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1523   if (ProfileInterpreter) {
1524     Label profile_continue;
1525 
1526     // If no method data exists, go to profile_continue.
1527     test_method_data_pointer(profile_continue);
1528 
1529     // We are taking a branch.  Increment the taken count.
1530     increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1531 
1532     // The method data pointer needs to be updated to reflect the new target.
1533     update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1534     bind (profile_continue);
1535   }
1536 }
1537 
1538 
1539 // Count a not-taken branch in the bytecodes.
1540 
1541 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1542   if (ProfileInterpreter) {
1543     Label profile_continue;
1544 
1545     // If no method data exists, go to profile_continue.
1546     test_method_data_pointer(profile_continue);
1547 
1548     // We are taking a branch.  Increment the not taken count.
1549     increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1550 
1551     // The method data pointer needs to be updated to correspond to the
1552     // next bytecode.
1553     update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1554     bind (profile_continue);
1555   }
1556 }
1557 
1558 
1559 // Count a non-virtual call in the bytecodes.
1560 
1561 void InterpreterMacroAssembler::profile_call(Register scratch) {
1562   if (ProfileInterpreter) {
1563     Label profile_continue;
1564 
1565     // If no method data exists, go to profile_continue.
1566     test_method_data_pointer(profile_continue);
1567 
1568     // We are making a call.  Increment the count.
1569     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1570 
1571     // The method data pointer needs to be updated to reflect the new target.
1572     update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1573     bind (profile_continue);
1574   }
1575 }
1576 
1577 
1578 // Count a final call in the bytecodes.
1579 
1580 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1581   if (ProfileInterpreter) {
1582     Label profile_continue;
1583 
1584     // If no method data exists, go to profile_continue.
1585     test_method_data_pointer(profile_continue);
1586 
1587     // We are making a call.  Increment the count.
1588     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1589 
1590     // The method data pointer needs to be updated to reflect the new target.
1591     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1592     bind (profile_continue);
1593   }
1594 }
1595 
1596 
1597 // Count a virtual call in the bytecodes.
1598 
1599 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1600                                                      Register scratch,
1601                                                      bool receiver_can_be_null) {
1602   if (ProfileInterpreter) {
1603     Label profile_continue;
1604 
1605     // If no method data exists, go to profile_continue.
1606     test_method_data_pointer(profile_continue);
1607 
1608 
1609     Label skip_receiver_profile;
1610     if (receiver_can_be_null) {
1611       Label not_null;
1612       br_notnull_short(receiver, Assembler::pt, not_null);
1613       // We are making a call.  Increment the count for null receiver.
1614       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1615       ba_short(skip_receiver_profile);
1616       bind(not_null);
1617     }
1618 
1619     // Record the receiver type.
1620     record_klass_in_profile(receiver, scratch, true);
1621     bind(skip_receiver_profile);
1622 
1623     // The method data pointer needs to be updated to reflect the new target.
1624 #if INCLUDE_JVMCI
1625     if (MethodProfileWidth == 0) {
1626       update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1627     }
1628 #else
1629     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1630 #endif
1631     bind(profile_continue);
1632   }
1633 }
1634 
1635 #if INCLUDE_JVMCI
1636 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) {
1637   assert_different_registers(method, scratch);
1638   if (ProfileInterpreter && MethodProfileWidth > 0) {
1639     Label profile_continue;
1640 
1641     // If no method data exists, go to profile_continue.
1642     test_method_data_pointer(profile_continue);
1643 
1644     Label done;
1645     record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth,
1646       &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1647     bind(done);
1648 
1649     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1650     bind(profile_continue);
1651   }
1652 }
1653 #endif // INCLUDE_JVMCI
1654 
1655 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch,
1656                                                                Label& done, bool is_virtual_call) {
1657   if (TypeProfileWidth == 0) {
1658     if (is_virtual_call) {
1659       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1660     }
1661 #if INCLUDE_JVMCI
1662     else if (EnableJVMCI) {
1663       increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch);
1664     }
1665 #endif
1666   } else {
1667     int non_profiled_offset = -1;
1668     if (is_virtual_call) {
1669       non_profiled_offset = in_bytes(CounterData::count_offset());
1670     }
1671 #if INCLUDE_JVMCI
1672     else if (EnableJVMCI) {
1673       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1674     }
1675 #endif
1676 
1677     record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth,
1678       &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1679   }
1680 }
1681 
1682 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item,
1683                                           Register scratch, int start_row, Label& done, int total_rows,
1684                                           OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1685                                           int non_profiled_offset) {
1686   int last_row = total_rows - 1;
1687   assert(start_row <= last_row, "must be work left to do");
1688   // Test this row for both the item and for null.
1689   // Take any of three different outcomes:
1690   //   1. found item => increment count and goto done
1691   //   2. found null => keep looking for case 1, maybe allocate this cell
1692   //   3. found something else => keep looking for cases 1 and 2
1693   // Case 3 is handled by a recursive call.
1694   for (int row = start_row; row <= last_row; row++) {
1695     Label next_test;
1696     bool test_for_null_also = (row == start_row);
1697 
1698     // See if the item is item[n].
1699     int item_offset = in_bytes(item_offset_fn(row));
1700     test_mdp_data_at(item_offset, item, next_test, scratch);
1701     // delayed()->tst(scratch);
1702 
1703     // The receiver is item[n].  Increment count[n].
1704     int count_offset = in_bytes(item_count_offset_fn(row));
1705     increment_mdp_data_at(count_offset, scratch);
1706     ba_short(done);
1707     bind(next_test);
1708 
1709     if (test_for_null_also) {
1710       Label found_null;
1711       // Failed the equality check on item[n]...  Test for null.
1712       if (start_row == last_row) {
1713         // The only thing left to do is handle the null case.
1714         if (non_profiled_offset >= 0) {
1715           brx(Assembler::zero, false, Assembler::pn, found_null);
1716           delayed()->nop();
1717           // Item did not match any saved item and there is no empty row for it.
1718           // Increment total counter to indicate polymorphic case.
1719           increment_mdp_data_at(non_profiled_offset, scratch);
1720           ba_short(done);
1721           bind(found_null);
1722         } else {
1723           brx(Assembler::notZero, false, Assembler::pt, done);
1724           delayed()->nop();
1725         }
1726         break;
1727       }
1728       // Since null is rare, make it be the branch-taken case.
1729       brx(Assembler::zero, false, Assembler::pn, found_null);
1730       delayed()->nop();
1731 
1732       // Put all the "Case 3" tests here.
1733       record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows,
1734         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1735 
1736       // Found a null.  Keep searching for a matching item,
1737       // but remember that this is an empty (unused) slot.
1738       bind(found_null);
1739     }
1740   }
1741 
1742   // In the fall-through case, we found no matching item, but we
1743   // observed the item[start_row] is NULL.
1744 
1745   // Fill in the item field and increment the count.
1746   int item_offset = in_bytes(item_offset_fn(start_row));
1747   set_mdp_data_at(item_offset, item);
1748   int count_offset = in_bytes(item_count_offset_fn(start_row));
1749   mov(DataLayout::counter_increment, scratch);
1750   set_mdp_data_at(count_offset, scratch);
1751   if (start_row > 0) {
1752     ba_short(done);
1753   }
1754 }
1755 
1756 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1757                                                         Register scratch, bool is_virtual_call) {
1758   assert(ProfileInterpreter, "must be profiling");
1759   Label done;
1760 
1761   record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call);
1762 
1763   bind (done);
1764 }
1765 
1766 
1767 // Count a ret in the bytecodes.
1768 
1769 void InterpreterMacroAssembler::profile_ret(TosState state,
1770                                             Register return_bci,
1771                                             Register scratch) {
1772   if (ProfileInterpreter) {
1773     Label profile_continue;
1774     uint row;
1775 
1776     // If no method data exists, go to profile_continue.
1777     test_method_data_pointer(profile_continue);
1778 
1779     // Update the total ret count.
1780     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1781 
1782     for (row = 0; row < RetData::row_limit(); row++) {
1783       Label next_test;
1784 
1785       // See if return_bci is equal to bci[n]:
1786       test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1787                        return_bci, next_test, scratch);
1788 
1789       // return_bci is equal to bci[n].  Increment the count.
1790       increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1791 
1792       // The method data pointer needs to be updated to reflect the new target.
1793       update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1794       ba_short(profile_continue);
1795       bind(next_test);
1796     }
1797 
1798     update_mdp_for_ret(state, return_bci);
1799 
1800     bind (profile_continue);
1801   }
1802 }
1803 
1804 // Profile an unexpected null in the bytecodes.
1805 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1806   if (ProfileInterpreter) {
1807     Label profile_continue;
1808 
1809     // If no method data exists, go to profile_continue.
1810     test_method_data_pointer(profile_continue);
1811 
1812     set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1813 
1814     // The method data pointer needs to be updated.
1815     int mdp_delta = in_bytes(BitData::bit_data_size());
1816     if (TypeProfileCasts) {
1817       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1818     }
1819     update_mdp_by_constant(mdp_delta);
1820 
1821     bind (profile_continue);
1822   }
1823 }
1824 
1825 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1826                                                   Register scratch) {
1827   if (ProfileInterpreter) {
1828     Label profile_continue;
1829 
1830     // If no method data exists, go to profile_continue.
1831     test_method_data_pointer(profile_continue);
1832 
1833     int mdp_delta = in_bytes(BitData::bit_data_size());
1834     if (TypeProfileCasts) {
1835       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1836 
1837       // Record the object type.
1838       record_klass_in_profile(klass, scratch, false);
1839     }
1840 
1841     // The method data pointer needs to be updated.
1842     update_mdp_by_constant(mdp_delta);
1843 
1844     bind (profile_continue);
1845   }
1846 }
1847 
1848 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1849   if (ProfileInterpreter && TypeProfileCasts) {
1850     Label profile_continue;
1851 
1852     // If no method data exists, go to profile_continue.
1853     test_method_data_pointer(profile_continue);
1854 
1855     int count_offset = in_bytes(CounterData::count_offset());
1856     // Back up the address, since we have already bumped the mdp.
1857     count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size());
1858 
1859     // *Decrement* the counter.  We expect to see zero or small negatives.
1860     increment_mdp_data_at(count_offset, scratch, true);
1861 
1862     bind (profile_continue);
1863   }
1864 }
1865 
1866 // Count the default case of a switch construct.
1867 
1868 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1869   if (ProfileInterpreter) {
1870     Label profile_continue;
1871 
1872     // If no method data exists, go to profile_continue.
1873     test_method_data_pointer(profile_continue);
1874 
1875     // Update the default case count
1876     increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1877                           scratch);
1878 
1879     // The method data pointer needs to be updated.
1880     update_mdp_by_offset(
1881                     in_bytes(MultiBranchData::default_displacement_offset()),
1882                     scratch);
1883 
1884     bind (profile_continue);
1885   }
1886 }
1887 
1888 // Count the index'th case of a switch construct.
1889 
1890 void InterpreterMacroAssembler::profile_switch_case(Register index,
1891                                                     Register scratch,
1892                                                     Register scratch2,
1893                                                     Register scratch3) {
1894   if (ProfileInterpreter) {
1895     Label profile_continue;
1896 
1897     // If no method data exists, go to profile_continue.
1898     test_method_data_pointer(profile_continue);
1899 
1900     // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1901     set(in_bytes(MultiBranchData::per_case_size()), scratch);
1902     smul(index, scratch, scratch);
1903     add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1904 
1905     // Update the case count
1906     increment_mdp_data_at(scratch,
1907                           in_bytes(MultiBranchData::relative_count_offset()),
1908                           scratch2,
1909                           scratch3);
1910 
1911     // The method data pointer needs to be updated.
1912     update_mdp_by_offset(scratch,
1913                      in_bytes(MultiBranchData::relative_displacement_offset()),
1914                      scratch2);
1915 
1916     bind (profile_continue);
1917   }
1918 }
1919 
1920 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1921   Label not_null, do_nothing, do_update;
1922 
1923   assert_different_registers(obj, mdo_addr.base(), tmp);
1924 
1925   verify_oop(obj);
1926 
1927   ld_ptr(mdo_addr, tmp);
1928 
1929   br_notnull_short(obj, pt, not_null);
1930   or3(tmp, TypeEntries::null_seen, tmp);
1931   ba_short(do_update);
1932 
1933   bind(not_null);
1934   load_klass(obj, obj);
1935 
1936   xor3(obj, tmp, obj);
1937   btst(TypeEntries::type_klass_mask, obj);
1938   // klass seen before, nothing to do. The unknown bit may have been
1939   // set already but no need to check.
1940   brx(zero, false, pt, do_nothing);
1941   delayed()->
1942 
1943   btst(TypeEntries::type_unknown, obj);
1944   // already unknown. Nothing to do anymore.
1945   brx(notZero, false, pt, do_nothing);
1946   delayed()->
1947 
1948   btst(TypeEntries::type_mask, tmp);
1949   brx(zero, true, pt, do_update);
1950   // first time here. Set profile type.
1951   delayed()->or3(tmp, obj, tmp);
1952 
1953   // different than before. Cannot keep accurate profile.
1954   or3(tmp, TypeEntries::type_unknown, tmp);
1955 
1956   bind(do_update);
1957   // update profile
1958   st_ptr(tmp, mdo_addr);
1959 
1960   bind(do_nothing);
1961 }
1962 
1963 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1964   if (!ProfileInterpreter) {
1965     return;
1966   }
1967 
1968   assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
1969 
1970   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1971     Label profile_continue;
1972 
1973     test_method_data_pointer(profile_continue);
1974 
1975     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1976 
1977     ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
1978     cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
1979 
1980     if (MethodData::profile_arguments()) {
1981       Label done;
1982       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1983       add(ImethodDataPtr, off_to_args, ImethodDataPtr);
1984 
1985       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1986         if (i > 0 || MethodData::profile_return()) {
1987           // If return value type is profiled we may have no argument to profile
1988           ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
1989           sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
1990           cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
1991         }
1992         ld_ptr(Address(callee, Method::const_offset()), tmp1);
1993         lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
1994         // stack offset o (zero based) from the start of the argument
1995         // list, for n arguments translates into offset n - o - 1 from
1996         // the end of the argument list. But there's an extra slot at
1997         // the stop of the stack. So the offset is n - o from Lesp.
1998         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
1999         sub(tmp1, tmp2, tmp1);
2000 
2001         // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
2002         sll(tmp1, Interpreter::logStackElementSize, tmp1);
2003         ld_ptr(Lesp, tmp1, tmp1);
2004 
2005         Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
2006         profile_obj_type(tmp1, mdo_arg_addr, tmp2);
2007 
2008         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
2009         add(ImethodDataPtr, to_add, ImethodDataPtr);
2010         off_to_args += to_add;
2011       }
2012 
2013       if (MethodData::profile_return()) {
2014         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2015         sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
2016       }
2017 
2018       bind(done);
2019 
2020       if (MethodData::profile_return()) {
2021         // We're right after the type profile for the last
2022         // argument. tmp1 is the number of cells left in the
2023         // CallTypeData/VirtualCallTypeData to reach its end. Non null
2024         // if there's a return to profile.
2025         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2026         sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2027         add(ImethodDataPtr, tmp1, ImethodDataPtr);
2028       }
2029     } else {
2030       assert(MethodData::profile_return(), "either profile call args or call ret");
2031       update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
2032     }
2033 
2034     // mdp points right after the end of the
2035     // CallTypeData/VirtualCallTypeData, right after the cells for the
2036     // return value type if there's one.
2037 
2038     bind(profile_continue);
2039   }
2040 }
2041 
2042 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2043   assert_different_registers(ret, tmp1, tmp2);
2044   if (ProfileInterpreter && MethodData::profile_return()) {
2045     Label profile_continue, done;
2046 
2047     test_method_data_pointer(profile_continue);
2048 
2049     if (MethodData::profile_return_jsr292_only()) {
2050       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
2051 
2052       // If we don't profile all invoke bytecodes we must make sure
2053       // it's a bytecode we indeed profile. We can't go back to the
2054       // begining of the ProfileData we intend to update to check its
2055       // type because we're right after it and we don't known its
2056       // length.
2057       Label do_profile;
2058       ldub(Lbcp, 0, tmp1);
2059       cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2060       cmp(tmp1, Bytecodes::_invokehandle);
2061       br(equal, false, pn, do_profile);
2062       delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2063       cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2064 
2065       bind(do_profile);
2066     }
2067 
2068     Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2069     mov(ret, tmp1);
2070     profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2071 
2072     bind(profile_continue);
2073   }
2074 }
2075 
2076 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2077   if (ProfileInterpreter && MethodData::profile_parameters()) {
2078     Label profile_continue, done;
2079 
2080     test_method_data_pointer(profile_continue);
2081 
2082     // Load the offset of the area within the MDO used for
2083     // parameters. If it's negative we're not profiling any parameters.
2084     lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2085     cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2086 
2087     // Compute a pointer to the area for parameters from the offset
2088     // and move the pointer to the slot for the last
2089     // parameters. Collect profiling from last parameter down.
2090     // mdo start + parameters offset + array length - 1
2091 
2092     // Pointer to the parameter area in the MDO
2093     Register mdp = tmp1;
2094     add(ImethodDataPtr, tmp1, mdp);
2095 
2096     // offset of the current profile entry to update
2097     Register entry_offset = tmp2;
2098     // entry_offset = array len in number of cells
2099     ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2100 
2101     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2102     assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2103 
2104     // entry_offset (number of cells)  = array len - size of 1 entry + offset of the stack slot field
2105     sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2106     // entry_offset in bytes
2107     sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2108 
2109     Label loop;
2110     bind(loop);
2111 
2112     // load offset on the stack from the slot for this parameter
2113     ld_ptr(mdp, entry_offset, tmp3);
2114     sll(tmp3,Interpreter::logStackElementSize, tmp3);
2115     neg(tmp3);
2116     // read the parameter from the local area
2117     ld_ptr(Llocals, tmp3, tmp3);
2118 
2119     // make entry_offset now point to the type field for this parameter
2120     int type_base = in_bytes(ParametersTypeData::type_offset(0));
2121     assert(type_base > off_base, "unexpected");
2122     add(entry_offset, type_base - off_base, entry_offset);
2123 
2124     // profile the parameter
2125     Address arg_type(mdp, entry_offset);
2126     profile_obj_type(tmp3, arg_type, tmp4);
2127 
2128     // go to next parameter
2129     sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2130     cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2131 
2132     bind(profile_continue);
2133   }
2134 }
2135 
2136 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2137 
2138 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2139                                                       Register Rtemp,
2140                                                       Register Rtemp2 ) {
2141 
2142   Register Rlimit = Lmonitors;
2143   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2144   assert( (delta & LongAlignmentMask) == 0,
2145           "sizeof BasicObjectLock must be even number of doublewords");
2146 
2147   sub( SP,        delta, SP);
2148   sub( Lesp,      delta, Lesp);
2149   sub( Lmonitors, delta, Lmonitors);
2150 
2151   if (!stack_is_empty) {
2152 
2153     // must copy stack contents down
2154 
2155     Label start_copying, next;
2156 
2157     // untested("monitor stack expansion");
2158     compute_stack_base(Rtemp);
2159     ba(start_copying);
2160     delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2161 
2162     // note: must copy from low memory upwards
2163     // On entry to loop,
2164     // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2165     // Loop mutates Rtemp
2166 
2167     bind( next);
2168 
2169     st_ptr(Rtemp2, Rtemp, 0);
2170     inc(Rtemp, wordSize);
2171     cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2172 
2173     bind( start_copying );
2174 
2175     brx( notEqual, true, pn, next );
2176     delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2177 
2178     // done copying stack
2179   }
2180 }
2181 
2182 // Locals
2183 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2184   assert_not_delayed();
2185   sll(index, Interpreter::logStackElementSize, index);
2186   sub(Llocals, index, index);
2187   ld_ptr(index, 0, dst);
2188   // Note:  index must hold the effective address--the iinc template uses it
2189 }
2190 
2191 // Just like access_local_ptr but the tag is a returnAddress
2192 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2193                                                            Register dst ) {
2194   assert_not_delayed();
2195   sll(index, Interpreter::logStackElementSize, index);
2196   sub(Llocals, index, index);
2197   ld_ptr(index, 0, dst);
2198 }
2199 
2200 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2201   assert_not_delayed();
2202   sll(index, Interpreter::logStackElementSize, index);
2203   sub(Llocals, index, index);
2204   ld(index, 0, dst);
2205   // Note:  index must hold the effective address--the iinc template uses it
2206 }
2207 
2208 
2209 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2210   assert_not_delayed();
2211   sll(index, Interpreter::logStackElementSize, index);
2212   sub(Llocals, index, index);
2213   // First half stored at index n+1 (which grows down from Llocals[n])
2214   load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2215 }
2216 
2217 
2218 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2219   assert_not_delayed();
2220   sll(index, Interpreter::logStackElementSize, index);
2221   sub(Llocals, index, index);
2222   ldf(FloatRegisterImpl::S, index, 0, dst);
2223 }
2224 
2225 
2226 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2227   assert_not_delayed();
2228   sll(index, Interpreter::logStackElementSize, index);
2229   sub(Llocals, index, index);
2230   load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2231 }
2232 
2233 
2234 #ifdef ASSERT
2235 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2236   Label L;
2237 
2238   assert(Rindex != Rscratch, "Registers cannot be same");
2239   assert(Rindex != Rscratch1, "Registers cannot be same");
2240   assert(Rlimit != Rscratch, "Registers cannot be same");
2241   assert(Rlimit != Rscratch1, "Registers cannot be same");
2242   assert(Rscratch1 != Rscratch, "Registers cannot be same");
2243 
2244   // untested("reg area corruption");
2245   add(Rindex, offset, Rscratch);
2246   add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2247   cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2248   stop("regsave area is being clobbered");
2249   bind(L);
2250 }
2251 #endif // ASSERT
2252 
2253 
2254 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2255   assert_not_delayed();
2256   sll(index, Interpreter::logStackElementSize, index);
2257   sub(Llocals, index, index);
2258   debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2259   st(src, index, 0);
2260 }
2261 
2262 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2263   assert_not_delayed();
2264   sll(index, Interpreter::logStackElementSize, index);
2265   sub(Llocals, index, index);
2266 #ifdef ASSERT
2267   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2268 #endif
2269   st_ptr(src, index, 0);
2270 }
2271 
2272 
2273 
2274 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2275   st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2276 }
2277 
2278 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2279   assert_not_delayed();
2280   sll(index, Interpreter::logStackElementSize, index);
2281   sub(Llocals, index, index);
2282 #ifdef ASSERT
2283   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2284 #endif
2285   store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2286 }
2287 
2288 
2289 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2290   assert_not_delayed();
2291   sll(index, Interpreter::logStackElementSize, index);
2292   sub(Llocals, index, index);
2293 #ifdef ASSERT
2294   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2295 #endif
2296   stf(FloatRegisterImpl::S, src, index, 0);
2297 }
2298 
2299 
2300 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2301   assert_not_delayed();
2302   sll(index, Interpreter::logStackElementSize, index);
2303   sub(Llocals, index, index);
2304 #ifdef ASSERT
2305   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2306 #endif
2307   store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2308 }
2309 
2310 
2311 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2312   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2313   int rounded_vm_local_words = align_up((int)frame::interpreter_frame_vm_local_words, WordsPerLong);
2314   return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2315 }
2316 
2317 
2318 Address InterpreterMacroAssembler::top_most_monitor() {
2319   return Address(FP, top_most_monitor_byte_offset());
2320 }
2321 
2322 
2323 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2324   add( Lesp,      wordSize,                                    Rdest );
2325 }
2326 
2327 void InterpreterMacroAssembler::get_method_counters(Register method,
2328                                                     Register Rcounters,
2329                                                     Label& skip) {
2330   Label has_counters;
2331   Address method_counters(method, in_bytes(Method::method_counters_offset()));
2332   ld_ptr(method_counters, Rcounters);
2333   br_notnull_short(Rcounters, Assembler::pt, has_counters);
2334   call_VM(noreg, CAST_FROM_FN_PTR(address,
2335           InterpreterRuntime::build_method_counters), method);
2336   ld_ptr(method_counters, Rcounters);
2337   br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2338   delayed()->nop();
2339   bind(has_counters);
2340 }
2341 
2342 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2343   assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2344   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2345 
2346   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2347                                  InvocationCounter::counter_offset());
2348   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2349                                  InvocationCounter::counter_offset());
2350   int delta = InvocationCounter::count_increment;
2351 
2352   // Load each counter in a register
2353   ld( inv_counter, Rtmp );
2354   ld( be_counter, Rtmp2 );
2355 
2356   assert( is_simm13( delta ), " delta too large.");
2357 
2358   // Add the delta to the invocation counter and store the result
2359   add( Rtmp, delta, Rtmp );
2360 
2361   // Mask the backedge counter
2362   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2363 
2364   // Store value
2365   st( Rtmp, inv_counter);
2366 
2367   // Add invocation counter + backedge counter
2368   add( Rtmp, Rtmp2, Rtmp);
2369 
2370   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2371 }
2372 
2373 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2374   assert(UseCompiler, "incrementing must be useful");
2375   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2376 
2377   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2378                                  InvocationCounter::counter_offset());
2379   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2380                                  InvocationCounter::counter_offset());
2381 
2382   int delta = InvocationCounter::count_increment;
2383   // Load each counter in a register
2384   ld( be_counter, Rtmp );
2385   ld( inv_counter, Rtmp2 );
2386 
2387   // Add the delta to the backedge counter
2388   add( Rtmp, delta, Rtmp );
2389 
2390   // Mask the invocation counter, add to backedge counter
2391   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2392 
2393   // and store the result to memory
2394   st( Rtmp, be_counter );
2395 
2396   // Add backedge + invocation counter
2397   add( Rtmp, Rtmp2, Rtmp );
2398 
2399   // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2400 }
2401 
2402 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2403                                                              Register method_counters,
2404                                                              Register branch_bcp,
2405                                                              Register Rtmp ) {
2406   Label did_not_overflow;
2407   Label overflow_with_error;
2408   assert_different_registers(backedge_count, Rtmp, branch_bcp);
2409   assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2410 
2411   Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()));
2412   ld(limit, Rtmp);
2413   cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2414 
2415   // When ProfileInterpreter is on, the backedge_count comes from the
2416   // MethodData*, which value does not get reset on the call to
2417   // frequency_counter_overflow().  To avoid excessive calls to the overflow
2418   // routine while the method is being compiled, add a second test to make sure
2419   // the overflow function is called only once every overflow_frequency.
2420   if (ProfileInterpreter) {
2421     const int overflow_frequency = 1024;
2422     andcc(backedge_count, overflow_frequency-1, Rtmp);
2423     brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2424     delayed()->nop();
2425   }
2426 
2427   // overflow in loop, pass branch bytecode
2428   set(6,Rtmp);
2429   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2430 
2431   // Was an OSR adapter generated?
2432   // O0 = osr nmethod
2433   br_null_short(O0, Assembler::pn, overflow_with_error);
2434 
2435   // Has the nmethod been invalidated already?
2436   ldub(O0, nmethod::state_offset(), O2);
2437   cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error);
2438 
2439   // migrate the interpreter frame off of the stack
2440 
2441   mov(G2_thread, L7);
2442   // save nmethod
2443   mov(O0, L6);
2444   set_last_Java_frame(SP, noreg);
2445   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2446   reset_last_Java_frame();
2447   mov(L7, G2_thread);
2448 
2449   // move OSR nmethod to I1
2450   mov(L6, I1);
2451 
2452   // OSR buffer to I0
2453   mov(O0, I0);
2454 
2455   // remove the interpreter frame
2456   restore(I5_savedSP, 0, SP);
2457 
2458   // Jump to the osr code.
2459   ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2460   jmp(O2, G0);
2461   delayed()->nop();
2462 
2463   bind(overflow_with_error);
2464 
2465   bind(did_not_overflow);
2466 }
2467 
2468 
2469 
2470 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2471   if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2472 }
2473 
2474 
2475 // local helper function for the verify_oop_or_return_address macro
2476 static bool verify_return_address(Method* m, int bci) {
2477 #ifndef PRODUCT
2478   address pc = (address)(m->constMethod())
2479              + in_bytes(ConstMethod::codes_offset()) + bci;
2480   // assume it is a valid return address if it is inside m and is preceded by a jsr
2481   if (!m->contains(pc))                                          return false;
2482   address jsr_pc;
2483   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2484   if (*jsr_pc == Bytecodes::_jsr   && jsr_pc >= m->code_base())    return true;
2485   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2486   if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base())    return true;
2487 #endif // PRODUCT
2488   return false;
2489 }
2490 
2491 
2492 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2493   if (!VerifyOops)  return;
2494   // the VM documentation for the astore[_wide] bytecode allows
2495   // the TOS to be not only an oop but also a return address
2496   Label test;
2497   Label skip;
2498   // See if it is an address (in the current method):
2499 
2500   mov(reg, Rtmp);
2501   const int log2_bytecode_size_limit = 16;
2502   srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2503   br_notnull_short( Rtmp, pt, test );
2504 
2505   // %%% should use call_VM_leaf here?
2506   save_frame_and_mov(0, Lmethod, O0, reg, O1);
2507   save_thread(L7_thread_cache);
2508   call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2509   delayed()->nop();
2510   restore_thread(L7_thread_cache);
2511   br_notnull( O0, false, pt, skip );
2512   delayed()->restore();
2513 
2514   // Perform a more elaborate out-of-line call
2515   // Not an address; verify it:
2516   bind(test);
2517   verify_oop(reg);
2518   bind(skip);
2519 }
2520 
2521 
2522 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2523   if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2524 }
2525 
2526 
2527 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2528 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2529                                                         int increment, Address mask_addr,
2530                                                         Register scratch1, Register scratch2,
2531                                                         Condition cond, Label *where) {
2532   ld(counter_addr, scratch1);
2533   add(scratch1, increment, scratch1);
2534   ld(mask_addr, scratch2);
2535   andcc(scratch1, scratch2,  G0);
2536   br(cond, false, Assembler::pn, *where);
2537   delayed()->st(scratch1, counter_addr);
2538 }
2539 
2540 // Inline assembly for:
2541 //
2542 // if (thread is in interp_only_mode) {
2543 //   InterpreterRuntime::post_method_entry();
2544 // }
2545 // if (DTraceMethodProbes) {
2546 //   SharedRuntime::dtrace_method_entry(method, receiver);
2547 // }
2548 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2549 //   SharedRuntime::rc_trace_method_entry(method, receiver);
2550 // }
2551 
2552 void InterpreterMacroAssembler::notify_method_entry() {
2553 
2554   // Whenever JVMTI puts a thread in interp_only_mode, method
2555   // entry/exit events are sent for that thread to track stack
2556   // depth.  If it is possible to enter interp_only_mode we add
2557   // the code to check if the event should be sent.
2558   if (JvmtiExport::can_post_interpreter_events()) {
2559     Label L;
2560     Register temp_reg = O5;
2561     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2562     ld(interp_only, temp_reg);
2563     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2564     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2565     bind(L);
2566   }
2567 
2568   {
2569     Register temp_reg = O5;
2570     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2571     call_VM_leaf(noreg,
2572       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2573       G2_thread, Lmethod);
2574   }
2575 
2576   // RedefineClasses() tracing support for obsolete method entry
2577   if (log_is_enabled(Trace, redefine, class, obsolete)) {
2578     call_VM_leaf(noreg,
2579       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2580       G2_thread, Lmethod);
2581   }
2582 }
2583 
2584 
2585 // Inline assembly for:
2586 //
2587 // if (thread is in interp_only_mode) {
2588 //   // save result
2589 //   InterpreterRuntime::post_method_exit();
2590 //   // restore result
2591 // }
2592 // if (DTraceMethodProbes) {
2593 //   SharedRuntime::dtrace_method_exit(thread, method);
2594 // }
2595 //
2596 // Native methods have their result stored in d_tmp and l_tmp
2597 // Java methods have their result stored in the expression stack
2598 
2599 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2600                                                    TosState state,
2601                                                    NotifyMethodExitMode mode) {
2602 
2603   // Whenever JVMTI puts a thread in interp_only_mode, method
2604   // entry/exit events are sent for that thread to track stack
2605   // depth.  If it is possible to enter interp_only_mode we add
2606   // the code to check if the event should be sent.
2607   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2608     Label L;
2609     Register temp_reg = O5;
2610     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2611     ld(interp_only, temp_reg);
2612     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2613 
2614     // Note: frame::interpreter_frame_result has a dependency on how the
2615     // method result is saved across the call to post_method_exit. For
2616     // native methods it assumes the result registers are saved to
2617     // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2618     // implementation will need to be updated too.
2619 
2620     save_return_value(state, is_native_method);
2621     call_VM(noreg,
2622             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2623     restore_return_value(state, is_native_method);
2624     bind(L);
2625   }
2626 
2627   {
2628     Register temp_reg = O5;
2629     // Dtrace notification
2630     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2631     save_return_value(state, is_native_method);
2632     call_VM_leaf(
2633       noreg,
2634       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2635       G2_thread, Lmethod);
2636     restore_return_value(state, is_native_method);
2637   }
2638 }
2639 
2640 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2641   if (is_native_call) {
2642     stf(FloatRegisterImpl::D, F0, d_tmp);
2643     stx(O0, l_tmp);
2644   } else {
2645     push(state);
2646   }
2647 }
2648 
2649 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2650   if (is_native_call) {
2651     ldf(FloatRegisterImpl::D, d_tmp, F0);
2652     ldx(l_tmp, O0);
2653   } else {
2654     pop(state);
2655   }
2656 }