1 /*
   2  * Copyright (c) 2008, 2021, 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 "asm/macroAssembler.inline.hpp"
  27 #include "c1/c1_Compilation.hpp"
  28 #include "c1/c1_LIRAssembler.hpp"
  29 #include "c1/c1_MacroAssembler.hpp"
  30 #include "c1/c1_Runtime1.hpp"
  31 #include "c1/c1_ValueStack.hpp"
  32 #include "ci/ciArrayKlass.hpp"
  33 #include "ci/ciInstance.hpp"
  34 #include "gc/shared/collectedHeap.hpp"
  35 #include "memory/universe.hpp"
  36 #include "nativeInst_arm.hpp"
  37 #include "oops/objArrayKlass.hpp"
  38 #include "runtime/frame.inline.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/stubRoutines.hpp"
  41 #include "utilities/powerOfTwo.hpp"
  42 #include "vmreg_arm.inline.hpp"
  43 
  44 #define __ _masm->
  45 
  46 // Note: Rtemp usage is this file should not impact C2 and should be
  47 // correct as long as it is not implicitly used in lower layers (the
  48 // arm [macro]assembler) and used with care in the other C1 specific
  49 // files.
  50 
  51 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  52   ShouldNotCallThis(); // Not used on ARM
  53   return false;
  54 }
  55 
  56 
  57 LIR_Opr LIR_Assembler::receiverOpr() {
  58   // The first register in Java calling conventions
  59   return FrameMap::R0_oop_opr;
  60 }
  61 
  62 LIR_Opr LIR_Assembler::osrBufferPointer() {
  63   return FrameMap::as_pointer_opr(R0);
  64 }
  65 
  66 #ifndef PRODUCT
  67 void LIR_Assembler::verify_reserved_argument_area_size(int args_count) {
  68   assert(args_count * wordSize <= frame_map()->reserved_argument_area_size(), "not enough space for arguments");
  69 }
  70 #endif // !PRODUCT
  71 
  72 void LIR_Assembler::store_parameter(jint c, int offset_from_sp_in_words) {
  73   assert(offset_from_sp_in_words >= 0, "invalid offset from sp");
  74   int offset_from_sp_in_bytes = offset_from_sp_in_words * BytesPerWord;
  75   assert(offset_from_sp_in_bytes < frame_map()->reserved_argument_area_size(), "not enough space");
  76   __ mov_slow(Rtemp, c);
  77   __ str(Rtemp, Address(SP, offset_from_sp_in_bytes));
  78 }
  79 
  80 void LIR_Assembler::store_parameter(Metadata* m, int offset_from_sp_in_words) {
  81   assert(offset_from_sp_in_words >= 0, "invalid offset from sp");
  82   int offset_from_sp_in_bytes = offset_from_sp_in_words * BytesPerWord;
  83   assert(offset_from_sp_in_bytes < frame_map()->reserved_argument_area_size(), "not enough space");
  84   __ mov_metadata(Rtemp, m);
  85   __ str(Rtemp, Address(SP, offset_from_sp_in_bytes));
  86 }
  87 
  88 //--------------fpu register translations-----------------------
  89 
  90 
  91 void LIR_Assembler::breakpoint() {
  92   __ breakpoint();
  93 }
  94 
  95 void LIR_Assembler::push(LIR_Opr opr) {
  96   Unimplemented();
  97 }
  98 
  99 void LIR_Assembler::pop(LIR_Opr opr) {
 100   Unimplemented();
 101 }
 102 
 103 //-------------------------------------------
 104 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 105   Register base = addr->base()->as_pointer_register();
 106 
 107 
 108   if (addr->index()->is_illegal() || addr->index()->is_constant()) {
 109     int offset = addr->disp();
 110     if (addr->index()->is_constant()) {
 111       offset += addr->index()->as_constant_ptr()->as_jint() << addr->scale();
 112     }
 113 
 114     if ((offset <= -4096) || (offset >= 4096)) {
 115       BAILOUT_("offset not in range", Address(base));
 116     }
 117 
 118     return Address(base, offset);
 119 
 120   } else {
 121     assert(addr->disp() == 0, "can't have both");
 122     int scale = addr->scale();
 123 
 124     assert(addr->index()->is_single_cpu(), "should be");
 125     return scale >= 0 ? Address(base, addr->index()->as_register(), lsl, scale) :
 126                         Address(base, addr->index()->as_register(), lsr, -scale);
 127   }
 128 }
 129 
 130 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 131   Address base = as_Address(addr);
 132   assert(base.index() == noreg, "must be");
 133   if (base.disp() + BytesPerWord >= 4096) { BAILOUT_("offset not in range", Address(base.base(),0)); }
 134   return Address(base.base(), base.disp() + BytesPerWord);
 135 }
 136 
 137 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 138   return as_Address(addr);
 139 }
 140 
 141 
 142 void LIR_Assembler::osr_entry() {
 143   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 144   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 145   ValueStack* entry_state = osr_entry->end()->state();
 146   int number_of_locks = entry_state->locks_size();
 147 
 148   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 149   Register OSR_buf = osrBufferPointer()->as_pointer_register();
 150 
 151   assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 152   int monitor_offset = (method()->max_locals() + 2 * (number_of_locks - 1)) * BytesPerWord;
 153   for (int i = 0; i < number_of_locks; i++) {
 154     int slot_offset = monitor_offset - (i * 2 * BytesPerWord);
 155     __ ldr(R1, Address(OSR_buf, slot_offset + 0*BytesPerWord));
 156     __ ldr(R2, Address(OSR_buf, slot_offset + 1*BytesPerWord));
 157     __ str(R1, frame_map()->address_for_monitor_lock(i));
 158     __ str(R2, frame_map()->address_for_monitor_object(i));
 159   }
 160 }
 161 
 162 
 163 int LIR_Assembler::check_icache() {
 164   Register receiver = LIR_Assembler::receiverOpr()->as_register();
 165   int offset = __ offset();
 166   __ inline_cache_check(receiver, Ricklass);
 167   return offset;
 168 }
 169 
 170 void LIR_Assembler::clinit_barrier(ciMethod* method) {
 171   ShouldNotReachHere(); // not implemented
 172 }
 173 
 174 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
 175   jobject o = (jobject)Universe::non_oop_word();
 176   int index = __ oop_recorder()->allocate_oop_index(o);
 177 
 178   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), index);
 179 
 180   __ patchable_mov_oop(reg, o, index);
 181   patching_epilog(patch, lir_patch_normal, reg, info);
 182 }
 183 
 184 
 185 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
 186   Metadata* o = (Metadata*)Universe::non_oop_word();
 187   int index = __ oop_recorder()->allocate_metadata_index(o);
 188   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 189 
 190   __ patchable_mov_metadata(reg, o, index);
 191   patching_epilog(patch, lir_patch_normal, reg, info);
 192 }
 193 
 194 
 195 int LIR_Assembler::initial_frame_size_in_bytes() const {
 196   // Subtracts two words to account for return address and link
 197   return frame_map()->framesize()*VMRegImpl::stack_slot_size - 2*wordSize;
 198 }
 199 
 200 
 201 int LIR_Assembler::emit_exception_handler() {
 202   // TODO: ARM
 203   __ nop(); // See comments in other ports
 204 
 205   address handler_base = __ start_a_stub(exception_handler_size());
 206   if (handler_base == NULL) {
 207     bailout("exception handler overflow");
 208     return -1;
 209   }
 210 
 211   int offset = code_offset();
 212 
 213   // check that there is really an exception
 214   __ verify_not_null_oop(Rexception_obj);
 215 
 216   __ call(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id), relocInfo::runtime_call_type);
 217   __ should_not_reach_here();
 218 
 219   assert(code_offset() - offset <= exception_handler_size(), "overflow");
 220   __ end_a_stub();
 221 
 222   return offset;
 223 }
 224 
 225 // Emit the code to remove the frame from the stack in the exception
 226 // unwind path.
 227 int LIR_Assembler::emit_unwind_handler() {
 228 #ifndef PRODUCT
 229   if (CommentedAssembly) {
 230     _masm->block_comment("Unwind handler");
 231   }
 232 #endif
 233 
 234   int offset = code_offset();
 235 
 236   // Fetch the exception from TLS and clear out exception related thread state
 237   Register zero = __ zero_register(Rtemp);
 238   __ ldr(Rexception_obj, Address(Rthread, JavaThread::exception_oop_offset()));
 239   __ str(zero, Address(Rthread, JavaThread::exception_oop_offset()));
 240   __ str(zero, Address(Rthread, JavaThread::exception_pc_offset()));
 241 
 242   __ bind(_unwind_handler_entry);
 243   __ verify_not_null_oop(Rexception_obj);
 244 
 245   // Preform needed unlocking
 246   MonitorExitStub* stub = NULL;
 247   if (method()->is_synchronized()) {
 248     monitor_address(0, FrameMap::R0_opr);
 249     stub = new MonitorExitStub(FrameMap::R0_opr, true, 0);
 250     __ unlock_object(R2, R1, R0, Rtemp, *stub->entry());
 251     __ bind(*stub->continuation());
 252   }
 253 
 254   // remove the activation and dispatch to the unwind handler
 255   __ remove_frame(initial_frame_size_in_bytes()); // restores FP and LR
 256   __ jump(Runtime1::entry_for(Runtime1::unwind_exception_id), relocInfo::runtime_call_type, Rtemp);
 257 
 258   // Emit the slow path assembly
 259   if (stub != NULL) {
 260     stub->emit_code(this);
 261   }
 262 
 263   return offset;
 264 }
 265 
 266 
 267 int LIR_Assembler::emit_deopt_handler() {
 268   address handler_base = __ start_a_stub(deopt_handler_size());
 269   if (handler_base == NULL) {
 270     bailout("deopt handler overflow");
 271     return -1;
 272   }
 273 
 274   int offset = code_offset();
 275 
 276   __ mov_relative_address(LR, __ pc());
 277   __ push(LR); // stub expects LR to be saved
 278   __ jump(SharedRuntime::deopt_blob()->unpack(), relocInfo::runtime_call_type, noreg);
 279 
 280   assert(code_offset() - offset <= deopt_handler_size(), "overflow");
 281   __ end_a_stub();
 282 
 283   return offset;
 284 }
 285 
 286 
 287 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
 288   // Pop the frame before safepoint polling
 289   __ remove_frame(initial_frame_size_in_bytes());
 290   __ read_polling_page(Rtemp, relocInfo::poll_return_type);
 291   __ ret();
 292 }
 293 
 294 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
 295 
 296   int offset = __ offset();
 297   __ get_polling_page(Rtemp);
 298   __ relocate(relocInfo::poll_type);
 299   add_debug_info_for_branch(info); // help pc_desc_at to find correct scope for current PC
 300   __ ldr(Rtemp, Address(Rtemp));
 301 
 302   return offset;
 303 }
 304 
 305 
 306 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 307   if (from_reg != to_reg) {
 308     __ mov(to_reg, from_reg);
 309   }
 310 }
 311 
 312 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 313   assert(src->is_constant() && dest->is_register(), "must be");
 314   LIR_Const* c = src->as_constant_ptr();
 315 
 316   switch (c->type()) {
 317     case T_ADDRESS:
 318     case T_INT:
 319       assert(patch_code == lir_patch_none, "no patching handled here");
 320       __ mov_slow(dest->as_register(), c->as_jint());
 321       break;
 322 
 323     case T_LONG:
 324       assert(patch_code == lir_patch_none, "no patching handled here");
 325       __ mov_slow(dest->as_register_lo(), c->as_jint_lo());
 326       __ mov_slow(dest->as_register_hi(), c->as_jint_hi());
 327       break;
 328 
 329     case T_OBJECT:
 330       if (patch_code == lir_patch_none) {
 331         __ mov_oop(dest->as_register(), c->as_jobject());
 332       } else {
 333         jobject2reg_with_patching(dest->as_register(), info);
 334       }
 335       break;
 336 
 337     case T_METADATA:
 338       if (patch_code == lir_patch_none) {
 339         __ mov_metadata(dest->as_register(), c->as_metadata());
 340       } else {
 341         klass2reg_with_patching(dest->as_register(), info);
 342       }
 343       break;
 344 
 345     case T_FLOAT:
 346       if (dest->is_single_fpu()) {
 347         __ mov_float(dest->as_float_reg(), c->as_jfloat());
 348       } else {
 349         // Simple getters can return float constant directly into r0
 350         __ mov_slow(dest->as_register(), c->as_jint_bits());
 351       }
 352       break;
 353 
 354     case T_DOUBLE:
 355       if (dest->is_double_fpu()) {
 356         __ mov_double(dest->as_double_reg(), c->as_jdouble());
 357       } else {
 358         // Simple getters can return double constant directly into r1r0
 359         __ mov_slow(dest->as_register_lo(), c->as_jint_lo_bits());
 360         __ mov_slow(dest->as_register_hi(), c->as_jint_hi_bits());
 361       }
 362       break;
 363 
 364     default:
 365       ShouldNotReachHere();
 366   }
 367 }
 368 
 369 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 370   assert(src->is_constant(), "must be");
 371   assert(dest->is_stack(), "must be");
 372   LIR_Const* c = src->as_constant_ptr();
 373 
 374   switch (c->type()) {
 375     case T_INT:  // fall through
 376     case T_FLOAT:
 377       __ mov_slow(Rtemp, c->as_jint_bits());
 378       __ str_32(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 379       break;
 380 
 381     case T_ADDRESS:
 382       __ mov_slow(Rtemp, c->as_jint());
 383       __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 384       break;
 385 
 386     case T_OBJECT:
 387       __ mov_oop(Rtemp, c->as_jobject());
 388       __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 389       break;
 390 
 391     case T_LONG:  // fall through
 392     case T_DOUBLE:
 393       __ mov_slow(Rtemp, c->as_jint_lo_bits());
 394       __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes));
 395       if (c->as_jint_hi_bits() != c->as_jint_lo_bits()) {
 396         __ mov_slow(Rtemp, c->as_jint_hi_bits());
 397       }
 398       __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 399       break;
 400 
 401     default:
 402       ShouldNotReachHere();
 403   }
 404 }
 405 
 406 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type,
 407                               CodeEmitInfo* info, bool wide) {
 408   assert((src->as_constant_ptr()->type() == T_OBJECT && src->as_constant_ptr()->as_jobject() == NULL),"cannot handle otherwise");
 409   __ mov(Rtemp, 0);
 410 
 411   int null_check_offset = code_offset();
 412   __ str(Rtemp, as_Address(dest->as_address_ptr()));
 413 
 414   if (info != NULL) {
 415     assert(false, "arm32 didn't support this before, investigate if bug");
 416     add_debug_info_for_null_check(null_check_offset, info);
 417   }
 418 }
 419 
 420 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
 421   assert(src->is_register() && dest->is_register(), "must be");
 422 
 423   if (src->is_single_cpu()) {
 424     if (dest->is_single_cpu()) {
 425       move_regs(src->as_register(), dest->as_register());
 426     } else if (dest->is_single_fpu()) {
 427       __ fmsr(dest->as_float_reg(), src->as_register());
 428     } else {
 429       ShouldNotReachHere();
 430     }
 431   } else if (src->is_double_cpu()) {
 432     if (dest->is_double_cpu()) {
 433       __ long_move(dest->as_register_lo(), dest->as_register_hi(), src->as_register_lo(), src->as_register_hi());
 434     } else {
 435       __ fmdrr(dest->as_double_reg(), src->as_register_lo(), src->as_register_hi());
 436     }
 437   } else if (src->is_single_fpu()) {
 438     if (dest->is_single_fpu()) {
 439       __ mov_float(dest->as_float_reg(), src->as_float_reg());
 440     } else if (dest->is_single_cpu()) {
 441       __ mov_fpr2gpr_float(dest->as_register(), src->as_float_reg());
 442     } else {
 443       ShouldNotReachHere();
 444     }
 445   } else if (src->is_double_fpu()) {
 446     if (dest->is_double_fpu()) {
 447       __ mov_double(dest->as_double_reg(), src->as_double_reg());
 448     } else if (dest->is_double_cpu()) {
 449       __ fmrrd(dest->as_register_lo(), dest->as_register_hi(), src->as_double_reg());
 450     } else {
 451       ShouldNotReachHere();
 452     }
 453   } else {
 454     ShouldNotReachHere();
 455   }
 456 }
 457 
 458 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
 459   assert(src->is_register(), "should not call otherwise");
 460   assert(dest->is_stack(), "should not call otherwise");
 461 
 462   Address addr = dest->is_single_word() ?
 463     frame_map()->address_for_slot(dest->single_stack_ix()) :
 464     frame_map()->address_for_slot(dest->double_stack_ix());
 465 
 466   assert(lo_word_offset_in_bytes == 0 && hi_word_offset_in_bytes == 4, "little ending");
 467   if (src->is_single_fpu() || src->is_double_fpu()) {
 468     if (addr.disp() >= 1024) { BAILOUT("Too exotic case to handle here"); }
 469   }
 470 
 471   if (src->is_single_cpu()) {
 472     switch (type) {
 473       case T_OBJECT:
 474       case T_ARRAY:    __ verify_oop(src->as_register());   // fall through
 475       case T_ADDRESS:
 476       case T_METADATA: __ str(src->as_register(), addr);    break;
 477       case T_FLOAT:    // used in intBitsToFloat intrinsic implementation, fall through
 478       case T_INT:      __ str_32(src->as_register(), addr); break;
 479       default:
 480         ShouldNotReachHere();
 481     }
 482   } else if (src->is_double_cpu()) {
 483     __ str(src->as_register_lo(), addr);
 484     __ str(src->as_register_hi(), frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 485   } else if (src->is_single_fpu()) {
 486     __ str_float(src->as_float_reg(), addr);
 487   } else if (src->is_double_fpu()) {
 488     __ str_double(src->as_double_reg(), addr);
 489   } else {
 490     ShouldNotReachHere();
 491   }
 492 }
 493 
 494 
 495 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type,
 496                             LIR_PatchCode patch_code, CodeEmitInfo* info,
 497                             bool pop_fpu_stack, bool wide,
 498                             bool unaligned) {
 499   LIR_Address* to_addr = dest->as_address_ptr();
 500   Register base_reg = to_addr->base()->as_pointer_register();
 501   const bool needs_patching = (patch_code != lir_patch_none);
 502 
 503   PatchingStub* patch = NULL;
 504   if (needs_patching) {
 505     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 506   }
 507 
 508   int null_check_offset = code_offset();
 509 
 510   switch (type) {
 511     case T_ARRAY:
 512     case T_OBJECT:
 513       if (UseCompressedOops && !wide) {
 514         ShouldNotReachHere();
 515       } else {
 516         __ str(src->as_register(), as_Address(to_addr));
 517       }
 518       break;
 519 
 520     case T_ADDRESS:
 521       __ str(src->as_pointer_register(), as_Address(to_addr));
 522       break;
 523 
 524     case T_BYTE:
 525     case T_BOOLEAN:
 526       __ strb(src->as_register(), as_Address(to_addr));
 527       break;
 528 
 529     case T_CHAR:
 530     case T_SHORT:
 531       __ strh(src->as_register(), as_Address(to_addr));
 532       break;
 533 
 534     case T_INT:
 535 #ifdef __SOFTFP__
 536     case T_FLOAT:
 537 #endif // __SOFTFP__
 538       __ str_32(src->as_register(), as_Address(to_addr));
 539       break;
 540 
 541 
 542 #ifdef __SOFTFP__
 543     case T_DOUBLE:
 544 #endif // __SOFTFP__
 545     case T_LONG: {
 546       Register from_lo = src->as_register_lo();
 547       Register from_hi = src->as_register_hi();
 548       if (to_addr->index()->is_register()) {
 549         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 550         assert(to_addr->disp() == 0, "Not yet supporting both");
 551         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 552         base_reg = Rtemp;
 553         __ str(from_lo, Address(Rtemp));
 554         if (patch != NULL) {
 555           __ nop(); // see comment before patching_epilog for 2nd str
 556           patching_epilog(patch, lir_patch_low, base_reg, info);
 557           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 558           patch_code = lir_patch_high;
 559         }
 560         __ str(from_hi, Address(Rtemp, BytesPerWord));
 561       } else if (base_reg == from_lo) {
 562         __ str(from_hi, as_Address_hi(to_addr));
 563         if (patch != NULL) {
 564           __ nop(); // see comment before patching_epilog for 2nd str
 565           patching_epilog(patch, lir_patch_high, base_reg, info);
 566           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 567           patch_code = lir_patch_low;
 568         }
 569         __ str(from_lo, as_Address_lo(to_addr));
 570       } else {
 571         __ str(from_lo, as_Address_lo(to_addr));
 572         if (patch != NULL) {
 573           __ nop(); // see comment before patching_epilog for 2nd str
 574           patching_epilog(patch, lir_patch_low, base_reg, info);
 575           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 576           patch_code = lir_patch_high;
 577         }
 578         __ str(from_hi, as_Address_hi(to_addr));
 579       }
 580       break;
 581     }
 582 
 583 #ifndef __SOFTFP__
 584     case T_FLOAT:
 585       if (to_addr->index()->is_register()) {
 586         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 587         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 588         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 589         __ fsts(src->as_float_reg(), Address(Rtemp, to_addr->disp()));
 590       } else {
 591         __ fsts(src->as_float_reg(), as_Address(to_addr));
 592       }
 593       break;
 594 
 595     case T_DOUBLE:
 596       if (to_addr->index()->is_register()) {
 597         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 598         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 599         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 600         __ fstd(src->as_double_reg(), Address(Rtemp, to_addr->disp()));
 601       } else {
 602         __ fstd(src->as_double_reg(), as_Address(to_addr));
 603       }
 604       break;
 605 #endif // __SOFTFP__
 606 
 607 
 608     default:
 609       ShouldNotReachHere();
 610   }
 611 
 612   if (info != NULL) {
 613     add_debug_info_for_null_check(null_check_offset, info);
 614   }
 615 
 616   if (patch != NULL) {
 617     // Offset embedded into LDR/STR instruction may appear not enough
 618     // to address a field. So, provide a space for one more instruction
 619     // that will deal with larger offsets.
 620     __ nop();
 621     patching_epilog(patch, patch_code, base_reg, info);
 622   }
 623 }
 624 
 625 
 626 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 627   assert(src->is_stack(), "should not call otherwise");
 628   assert(dest->is_register(), "should not call otherwise");
 629 
 630   Address addr = src->is_single_word() ?
 631     frame_map()->address_for_slot(src->single_stack_ix()) :
 632     frame_map()->address_for_slot(src->double_stack_ix());
 633 
 634   assert(lo_word_offset_in_bytes == 0 && hi_word_offset_in_bytes == 4, "little ending");
 635   if (dest->is_single_fpu() || dest->is_double_fpu()) {
 636     if (addr.disp() >= 1024) { BAILOUT("Too exotic case to handle here"); }
 637   }
 638 
 639   if (dest->is_single_cpu()) {
 640     switch (type) {
 641       case T_OBJECT:
 642       case T_ARRAY:
 643       case T_ADDRESS:
 644       case T_METADATA: __ ldr(dest->as_register(), addr); break;
 645       case T_FLOAT:    // used in floatToRawIntBits intrinsic implemenation
 646       case T_INT:      __ ldr_u32(dest->as_register(), addr); break;
 647       default:
 648         ShouldNotReachHere();
 649     }
 650     if ((type == T_OBJECT) || (type == T_ARRAY)) {
 651       __ verify_oop(dest->as_register());
 652     }
 653   } else if (dest->is_double_cpu()) {
 654     __ ldr(dest->as_register_lo(), addr);
 655     __ ldr(dest->as_register_hi(), frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 656   } else if (dest->is_single_fpu()) {
 657     __ ldr_float(dest->as_float_reg(), addr);
 658   } else if (dest->is_double_fpu()) {
 659     __ ldr_double(dest->as_double_reg(), addr);
 660   } else {
 661     ShouldNotReachHere();
 662   }
 663 }
 664 
 665 
 666 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 667   if (src->is_single_stack()) {
 668     switch (src->type()) {
 669       case T_OBJECT:
 670       case T_ARRAY:
 671       case T_ADDRESS:
 672       case T_METADATA:
 673         __ ldr(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 674         __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 675         break;
 676 
 677       case T_INT:
 678       case T_FLOAT:
 679         __ ldr_u32(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 680         __ str_32(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 681         break;
 682 
 683       default:
 684         ShouldNotReachHere();
 685     }
 686   } else {
 687     assert(src->is_double_stack(), "must be");
 688     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes));
 689     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes));
 690     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 691     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 692   }
 693 }
 694 
 695 
 696 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type,
 697                             LIR_PatchCode patch_code, CodeEmitInfo* info,
 698                             bool wide, bool unaligned) {
 699   assert(src->is_address(), "should not call otherwise");
 700   assert(dest->is_register(), "should not call otherwise");
 701   LIR_Address* addr = src->as_address_ptr();
 702 
 703   Register base_reg = addr->base()->as_pointer_register();
 704 
 705   PatchingStub* patch = NULL;
 706   if (patch_code != lir_patch_none) {
 707     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 708   }
 709   if (info != NULL) {
 710     add_debug_info_for_null_check_here(info);
 711   }
 712 
 713   switch (type) {
 714     case T_OBJECT:  // fall through
 715     case T_ARRAY:
 716       if (UseCompressedOops && !wide) {
 717         __ ldr_u32(dest->as_register(), as_Address(addr));
 718       } else {
 719         __ ldr(dest->as_register(), as_Address(addr));
 720       }
 721       break;
 722 
 723     case T_ADDRESS:
 724       __ ldr(dest->as_pointer_register(), as_Address(addr));
 725       break;
 726 
 727     case T_INT:
 728 #ifdef __SOFTFP__
 729     case T_FLOAT:
 730 #endif // __SOFTFP__
 731       __ ldr(dest->as_pointer_register(), as_Address(addr));
 732       break;
 733 
 734     case T_BOOLEAN:
 735       __ ldrb(dest->as_register(), as_Address(addr));
 736       break;
 737 
 738     case T_BYTE:
 739       __ ldrsb(dest->as_register(), as_Address(addr));
 740       break;
 741 
 742     case T_CHAR:
 743       __ ldrh(dest->as_register(), as_Address(addr));
 744       break;
 745 
 746     case T_SHORT:
 747       __ ldrsh(dest->as_register(), as_Address(addr));
 748       break;
 749 
 750 
 751 #ifdef __SOFTFP__
 752     case T_DOUBLE:
 753 #endif // __SOFTFP__
 754     case T_LONG: {
 755       Register to_lo = dest->as_register_lo();
 756       Register to_hi = dest->as_register_hi();
 757       if (addr->index()->is_register()) {
 758         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 759         assert(addr->disp() == 0, "Not yet supporting both");
 760         __ add(Rtemp, base_reg, addr->index()->as_register());
 761         base_reg = Rtemp;
 762         __ ldr(to_lo, Address(Rtemp));
 763         if (patch != NULL) {
 764           __ nop(); // see comment before patching_epilog for 2nd ldr
 765           patching_epilog(patch, lir_patch_low, base_reg, info);
 766           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 767           patch_code = lir_patch_high;
 768         }
 769         __ ldr(to_hi, Address(Rtemp, BytesPerWord));
 770       } else if (base_reg == to_lo) {
 771         __ ldr(to_hi, as_Address_hi(addr));
 772         if (patch != NULL) {
 773           __ nop(); // see comment before patching_epilog for 2nd ldr
 774           patching_epilog(patch, lir_patch_high, base_reg, info);
 775           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 776           patch_code = lir_patch_low;
 777         }
 778         __ ldr(to_lo, as_Address_lo(addr));
 779       } else {
 780         __ ldr(to_lo, as_Address_lo(addr));
 781         if (patch != NULL) {
 782           __ nop(); // see comment before patching_epilog for 2nd ldr
 783           patching_epilog(patch, lir_patch_low, base_reg, info);
 784           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 785           patch_code = lir_patch_high;
 786         }
 787         __ ldr(to_hi, as_Address_hi(addr));
 788       }
 789       break;
 790     }
 791 
 792 #ifndef __SOFTFP__
 793     case T_FLOAT:
 794       if (addr->index()->is_register()) {
 795         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 796         __ add(Rtemp, base_reg, addr->index()->as_register());
 797         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 798         __ flds(dest->as_float_reg(), Address(Rtemp, addr->disp()));
 799       } else {
 800         __ flds(dest->as_float_reg(), as_Address(addr));
 801       }
 802       break;
 803 
 804     case T_DOUBLE:
 805       if (addr->index()->is_register()) {
 806         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 807         __ add(Rtemp, base_reg, addr->index()->as_register());
 808         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 809         __ fldd(dest->as_double_reg(), Address(Rtemp, addr->disp()));
 810       } else {
 811         __ fldd(dest->as_double_reg(), as_Address(addr));
 812       }
 813       break;
 814 #endif // __SOFTFP__
 815 
 816 
 817     default:
 818       ShouldNotReachHere();
 819   }
 820 
 821   if (patch != NULL) {
 822     // Offset embedded into LDR/STR instruction may appear not enough
 823     // to address a field. So, provide a space for one more instruction
 824     // that will deal with larger offsets.
 825     __ nop();
 826     patching_epilog(patch, patch_code, base_reg, info);
 827   }
 828 
 829 }
 830 
 831 
 832 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 833   bool is_32 = op->result_opr()->is_single_cpu();
 834 
 835   if (op->code() == lir_idiv && op->in_opr2()->is_constant() && is_32) {
 836     int c = op->in_opr2()->as_constant_ptr()->as_jint();
 837     assert(is_power_of_2(c), "non power-of-2 constant should be put in a register");
 838 
 839     Register left = op->in_opr1()->as_register();
 840     Register dest = op->result_opr()->as_register();
 841     if (c == 1) {
 842       __ mov(dest, left);
 843     } else if (c == 2) {
 844       __ add_32(dest, left, AsmOperand(left, lsr, 31));
 845       __ asr_32(dest, dest, 1);
 846     } else if (c != (int) 0x80000000) {
 847       int power = log2i_exact(c);
 848       __ asr_32(Rtemp, left, 31);
 849       __ add_32(dest, left, AsmOperand(Rtemp, lsr, 32-power)); // dest = left + (left < 0 ? 2^power - 1 : 0);
 850       __ asr_32(dest, dest, power);                            // dest = dest >>> power;
 851     } else {
 852       // x/0x80000000 is a special case, since dividend is a power of two, but is negative.
 853       // The only possible result values are 0 and 1, with 1 only for dividend == divisor == 0x80000000.
 854       __ cmp_32(left, c);
 855       __ mov(dest, 0, ne);
 856       __ mov(dest, 1, eq);
 857     }
 858   } else {
 859     assert(op->code() == lir_idiv || op->code() == lir_irem, "unexpected op3");
 860     __ call(StubRoutines::Arm::idiv_irem_entry(), relocInfo::runtime_call_type);
 861     add_debug_info_for_div0_here(op->info());
 862   }
 863 }
 864 
 865 
 866 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 867 #ifdef ASSERT
 868   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 869   if (op->block() != NULL)  _branch_target_blocks.append(op->block());
 870   if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
 871   assert(op->info() == NULL, "CodeEmitInfo?");
 872 #endif // ASSERT
 873 
 874 #ifdef __SOFTFP__
 875   assert (op->code() != lir_cond_float_branch, "this should be impossible");
 876 #else
 877   if (op->code() == lir_cond_float_branch) {
 878     __ fmstat();
 879     __ b(*(op->ublock()->label()), vs);
 880   }
 881 #endif // __SOFTFP__
 882 
 883   AsmCondition acond = al;
 884   switch (op->cond()) {
 885     case lir_cond_equal:        acond = eq; break;
 886     case lir_cond_notEqual:     acond = ne; break;
 887     case lir_cond_less:         acond = lt; break;
 888     case lir_cond_lessEqual:    acond = le; break;
 889     case lir_cond_greaterEqual: acond = ge; break;
 890     case lir_cond_greater:      acond = gt; break;
 891     case lir_cond_aboveEqual:   acond = hs; break;
 892     case lir_cond_belowEqual:   acond = ls; break;
 893     default: assert(op->cond() == lir_cond_always, "must be");
 894   }
 895   __ b(*(op->label()), acond);
 896 }
 897 
 898 
 899 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 900   LIR_Opr src  = op->in_opr();
 901   LIR_Opr dest = op->result_opr();
 902 
 903   switch (op->bytecode()) {
 904     case Bytecodes::_i2l:
 905       move_regs(src->as_register(), dest->as_register_lo());
 906       __ mov(dest->as_register_hi(), AsmOperand(src->as_register(), asr, 31));
 907       break;
 908     case Bytecodes::_l2i:
 909       move_regs(src->as_register_lo(), dest->as_register());
 910       break;
 911     case Bytecodes::_i2b:
 912       __ sign_extend(dest->as_register(), src->as_register(), 8);
 913       break;
 914     case Bytecodes::_i2s:
 915       __ sign_extend(dest->as_register(), src->as_register(), 16);
 916       break;
 917     case Bytecodes::_i2c:
 918       __ zero_extend(dest->as_register(), src->as_register(), 16);
 919       break;
 920     case Bytecodes::_f2d:
 921       __ convert_f2d(dest->as_double_reg(), src->as_float_reg());
 922       break;
 923     case Bytecodes::_d2f:
 924       __ convert_d2f(dest->as_float_reg(), src->as_double_reg());
 925       break;
 926     case Bytecodes::_i2f:
 927       __ fmsr(Stemp, src->as_register());
 928       __ fsitos(dest->as_float_reg(), Stemp);
 929       break;
 930     case Bytecodes::_i2d:
 931       __ fmsr(Stemp, src->as_register());
 932       __ fsitod(dest->as_double_reg(), Stemp);
 933       break;
 934     case Bytecodes::_f2i:
 935       __ ftosizs(Stemp, src->as_float_reg());
 936       __ fmrs(dest->as_register(), Stemp);
 937       break;
 938     case Bytecodes::_d2i:
 939       __ ftosizd(Stemp, src->as_double_reg());
 940       __ fmrs(dest->as_register(), Stemp);
 941       break;
 942     default:
 943       ShouldNotReachHere();
 944   }
 945 }
 946 
 947 
 948 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
 949   if (op->init_check()) {
 950     Register tmp = op->tmp1()->as_register();
 951     __ ldrb(tmp, Address(op->klass()->as_register(), InstanceKlass::init_state_offset()));
 952     add_debug_info_for_null_check_here(op->stub()->info());
 953     __ cmp(tmp, InstanceKlass::fully_initialized);
 954     __ b(*op->stub()->entry(), ne);
 955   }
 956   __ allocate_object(op->obj()->as_register(),
 957                      op->tmp1()->as_register(),
 958                      op->tmp2()->as_register(),
 959                      op->tmp3()->as_register(),
 960                      op->header_size(),
 961                      op->object_size(),
 962                      op->klass()->as_register(),
 963                      *op->stub()->entry());
 964   __ bind(*op->stub()->continuation());
 965 }
 966 
 967 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
 968   if (UseSlowPath ||
 969       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
 970       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
 971     __ b(*op->stub()->entry());
 972   } else {
 973     __ allocate_array(op->obj()->as_register(),
 974                       op->len()->as_register(),
 975                       op->tmp1()->as_register(),
 976                       op->tmp2()->as_register(),
 977                       op->tmp3()->as_register(),
 978                       arrayOopDesc::header_size(op->type()),
 979                       type2aelembytes(op->type()),
 980                       op->klass()->as_register(),
 981                       *op->stub()->entry());
 982   }
 983   __ bind(*op->stub()->continuation());
 984 }
 985 
 986 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
 987                                         ciMethodData *md, ciProfileData *data,
 988                                         Register recv, Register tmp1, Label* update_done) {
 989   assert_different_registers(mdo, recv, tmp1);
 990   uint i;
 991   for (i = 0; i < VirtualCallData::row_limit(); i++) {
 992     Label next_test;
 993     // See if the receiver is receiver[n].
 994     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
 995                           mdo_offset_bias);
 996     __ ldr(tmp1, receiver_addr);
 997     __ verify_klass_ptr(tmp1);
 998     __ cmp(recv, tmp1);
 999     __ b(next_test, ne);
1000     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
1001                       mdo_offset_bias);
1002     __ ldr(tmp1, data_addr);
1003     __ add(tmp1, tmp1, DataLayout::counter_increment);
1004     __ str(tmp1, data_addr);
1005     __ b(*update_done);
1006     __ bind(next_test);
1007   }
1008 
1009   // Didn't find receiver; find next empty slot and fill it in
1010   for (i = 0; i < VirtualCallData::row_limit(); i++) {
1011     Label next_test;
1012     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
1013                       mdo_offset_bias);
1014     __ ldr(tmp1, recv_addr);
1015     __ cbnz(tmp1, next_test);
1016     __ str(recv, recv_addr);
1017     __ mov(tmp1, DataLayout::counter_increment);
1018     __ str(tmp1, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
1019                          mdo_offset_bias));
1020     __ b(*update_done);
1021     __ bind(next_test);
1022   }
1023 }
1024 
1025 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
1026                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
1027   md = method->method_data_or_null();
1028   assert(md != NULL, "Sanity");
1029   data = md->bci_to_data(bci);
1030   assert(data != NULL,       "need data for checkcast");
1031   assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1032   if (md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes() >= 4096) {
1033     // The offset is large so bias the mdo by the base of the slot so
1034     // that the ldr can use an immediate offset to reference the slots of the data
1035     mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
1036   }
1037 }
1038 
1039 // On 32-bit ARM, code before this helper should test obj for null (ZF should be set if obj is null).
1040 void LIR_Assembler::typecheck_profile_helper1(ciMethod* method, int bci,
1041                                               ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias,
1042                                               Register obj, Register mdo, Register data_val, Label* obj_is_null) {
1043   assert(method != NULL, "Should have method");
1044   assert_different_registers(obj, mdo, data_val);
1045   setup_md_access(method, bci, md, data, mdo_offset_bias);
1046   Label not_null;
1047   __ b(not_null, ne);
1048   __ mov_metadata(mdo, md->constant_encoding());
1049   if (mdo_offset_bias > 0) {
1050     __ mov_slow(data_val, mdo_offset_bias);
1051     __ add(mdo, mdo, data_val);
1052   }
1053   Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
1054   __ ldrb(data_val, flags_addr);
1055   __ orr(data_val, data_val, (uint)BitData::null_seen_byte_constant());
1056   __ strb(data_val, flags_addr);
1057   __ b(*obj_is_null);
1058   __ bind(not_null);
1059 }
1060 
1061 void LIR_Assembler::typecheck_profile_helper2(ciMethodData* md, ciProfileData* data, int mdo_offset_bias,
1062                                               Register mdo, Register recv, Register value, Register tmp1,
1063                                               Label* profile_cast_success, Label* profile_cast_failure,
1064                                               Label* success, Label* failure) {
1065   assert_different_registers(mdo, value, tmp1);
1066   __ bind(*profile_cast_success);
1067   __ mov_metadata(mdo, md->constant_encoding());
1068   if (mdo_offset_bias > 0) {
1069     __ mov_slow(tmp1, mdo_offset_bias);
1070     __ add(mdo, mdo, tmp1);
1071   }
1072   __ load_klass(recv, value);
1073   type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, success);
1074   __ b(*success);
1075   // Cast failure case
1076   __ bind(*profile_cast_failure);
1077   __ mov_metadata(mdo, md->constant_encoding());
1078   if (mdo_offset_bias > 0) {
1079     __ mov_slow(tmp1, mdo_offset_bias);
1080     __ add(mdo, mdo, tmp1);
1081   }
1082   Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
1083   __ ldr(tmp1, data_addr);
1084   __ sub(tmp1, tmp1, DataLayout::counter_increment);
1085   __ str(tmp1, data_addr);
1086   __ b(*failure);
1087 }
1088 
1089 // Sets `res` to true, if `cond` holds.
1090 static void set_instanceof_result(MacroAssembler* _masm, Register res, AsmCondition cond) {
1091   __ mov(res, 1, cond);
1092 }
1093 
1094 
1095 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1096   // TODO: ARM - can be more effective with one more register
1097   switch (op->code()) {
1098     case lir_store_check: {
1099       CodeStub* stub = op->stub();
1100       Register value = op->object()->as_register();
1101       Register array = op->array()->as_register();
1102       Register klass_RInfo = op->tmp1()->as_register();
1103       Register k_RInfo = op->tmp2()->as_register();
1104       assert_different_registers(klass_RInfo, k_RInfo, Rtemp);
1105       if (op->should_profile()) {
1106         assert_different_registers(value, klass_RInfo, k_RInfo, Rtemp);
1107       }
1108 
1109       // check if it needs to be profiled
1110       ciMethodData* md;
1111       ciProfileData* data;
1112       int mdo_offset_bias = 0;
1113       Label profile_cast_success, profile_cast_failure, done;
1114       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1115       Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1116 
1117       if (op->should_profile()) {
1118         __ cmp(value, 0);
1119         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, value, k_RInfo, Rtemp, &done);
1120       } else {
1121         __ cbz(value, done);
1122       }
1123       assert_different_registers(k_RInfo, value);
1124       add_debug_info_for_null_check_here(op->info_for_exception());
1125       __ load_klass(k_RInfo, array);
1126       __ load_klass(klass_RInfo, value);
1127       __ ldr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
1128       __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1129       // check for immediate positive hit
1130       __ ldr(Rtemp, Address(klass_RInfo, Rtemp));
1131       __ cmp(klass_RInfo, k_RInfo);
1132       __ cond_cmp(Rtemp, k_RInfo, ne);
1133       __ b(*success_target, eq);
1134       // check for immediate negative hit
1135       __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1136       __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1137       __ b(*failure_target, ne);
1138       // slow case
1139       assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1140       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1141       __ cbz(R0, *failure_target);
1142       if (op->should_profile()) {
1143         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1144         if (mdo == value) {
1145           mdo = k_RInfo;
1146           recv = klass_RInfo;
1147         }
1148         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, value, tmp1,
1149                                   &profile_cast_success, &profile_cast_failure,
1150                                   &done, stub->entry());
1151       }
1152       __ bind(done);
1153       break;
1154     }
1155 
1156     case lir_checkcast: {
1157       CodeStub* stub = op->stub();
1158       Register obj = op->object()->as_register();
1159       Register res = op->result_opr()->as_register();
1160       Register klass_RInfo = op->tmp1()->as_register();
1161       Register k_RInfo = op->tmp2()->as_register();
1162       ciKlass* k = op->klass();
1163       assert_different_registers(res, k_RInfo, klass_RInfo, Rtemp);
1164 
1165       if (stub->is_simple_exception_stub()) {
1166       // TODO: ARM - Late binding is used to prevent confusion of register allocator
1167       assert(stub->is_exception_throw_stub(), "must be");
1168       ((SimpleExceptionStub*)stub)->set_obj(op->result_opr());
1169       }
1170       ciMethodData* md;
1171       ciProfileData* data;
1172       int mdo_offset_bias = 0;
1173 
1174       Label done;
1175 
1176       Label profile_cast_failure, profile_cast_success;
1177       Label *failure_target = op->should_profile() ? &profile_cast_failure : op->stub()->entry();
1178       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1179 
1180 
1181       __ movs(res, obj);
1182       if (op->should_profile()) {
1183         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, res, klass_RInfo, Rtemp, &done);
1184       } else {
1185         __ b(done, eq);
1186       }
1187       if (k->is_loaded()) {
1188         __ mov_metadata(k_RInfo, k->constant_encoding());
1189       } else if (k_RInfo != obj) {
1190         klass2reg_with_patching(k_RInfo, op->info_for_patch());
1191         __ movs(res, obj);
1192       } else {
1193         // Patching doesn't update "res" register after GC, so do patching first
1194         klass2reg_with_patching(Rtemp, op->info_for_patch());
1195         __ movs(res, obj);
1196         __ mov(k_RInfo, Rtemp);
1197       }
1198       __ load_klass(klass_RInfo, res, ne);
1199 
1200       if (op->fast_check()) {
1201         __ cmp(klass_RInfo, k_RInfo, ne);
1202         __ b(*failure_target, ne);
1203       } else if (k->is_loaded()) {
1204         __ b(*success_target, eq);
1205         __ ldr(Rtemp, Address(klass_RInfo, k->super_check_offset()));
1206         if (in_bytes(Klass::secondary_super_cache_offset()) != (int) k->super_check_offset()) {
1207           __ cmp(Rtemp, k_RInfo);
1208           __ b(*failure_target, ne);
1209         } else {
1210           __ cmp(klass_RInfo, k_RInfo);
1211           __ cmp(Rtemp, k_RInfo, ne);
1212           __ b(*success_target, eq);
1213           assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1214           __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1215           __ cbz(R0, *failure_target);
1216         }
1217       } else {
1218         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1219         __ b(*success_target, eq);
1220         // check for immediate positive hit
1221         __ ldr(Rtemp, Address(klass_RInfo, Rtemp));
1222         __ cmp(klass_RInfo, k_RInfo);
1223         __ cmp(Rtemp, k_RInfo, ne);
1224         __ b(*success_target, eq);
1225         // check for immediate negative hit
1226         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1227         __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1228         __ b(*failure_target, ne);
1229         // slow case
1230         assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1231         __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1232         __ cbz(R0, *failure_target);
1233       }
1234 
1235       if (op->should_profile()) {
1236         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1237         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, res, tmp1,
1238                                   &profile_cast_success, &profile_cast_failure,
1239                                   &done, stub->entry());
1240       }
1241       __ bind(done);
1242       break;
1243     }
1244 
1245     case lir_instanceof: {
1246       Register obj = op->object()->as_register();
1247       Register res = op->result_opr()->as_register();
1248       Register klass_RInfo = op->tmp1()->as_register();
1249       Register k_RInfo = op->tmp2()->as_register();
1250       ciKlass* k = op->klass();
1251       assert_different_registers(res, klass_RInfo, k_RInfo, Rtemp);
1252 
1253       ciMethodData* md;
1254       ciProfileData* data;
1255       int mdo_offset_bias = 0;
1256 
1257       Label done;
1258 
1259       Label profile_cast_failure, profile_cast_success;
1260       Label *failure_target = op->should_profile() ? &profile_cast_failure : &done;
1261       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1262 
1263       __ movs(res, obj);
1264 
1265       if (op->should_profile()) {
1266         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, res, klass_RInfo, Rtemp, &done);
1267       } else {
1268         __ b(done, eq);
1269       }
1270 
1271       if (k->is_loaded()) {
1272         __ mov_metadata(k_RInfo, k->constant_encoding());
1273       } else {
1274         op->info_for_patch()->add_register_oop(FrameMap::as_oop_opr(res));
1275         klass2reg_with_patching(k_RInfo, op->info_for_patch());
1276       }
1277       __ load_klass(klass_RInfo, res);
1278 
1279       if (!op->should_profile()) {
1280         __ mov(res, 0);
1281       }
1282 
1283       if (op->fast_check()) {
1284         __ cmp(klass_RInfo, k_RInfo);
1285         if (!op->should_profile()) {
1286           set_instanceof_result(_masm, res, eq);
1287         } else {
1288           __ b(profile_cast_failure, ne);
1289         }
1290       } else if (k->is_loaded()) {
1291         __ ldr(Rtemp, Address(klass_RInfo, k->super_check_offset()));
1292         if (in_bytes(Klass::secondary_super_cache_offset()) != (int) k->super_check_offset()) {
1293           __ cmp(Rtemp, k_RInfo);
1294           if (!op->should_profile()) {
1295             set_instanceof_result(_masm, res, eq);
1296           } else {
1297             __ b(profile_cast_failure, ne);
1298           }
1299         } else {
1300           __ cmp(klass_RInfo, k_RInfo);
1301           __ cond_cmp(Rtemp, k_RInfo, ne);
1302           if (!op->should_profile()) {
1303             set_instanceof_result(_masm, res, eq);
1304           }
1305           __ b(*success_target, eq);
1306           assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1307           __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1308           if (!op->should_profile()) {
1309             move_regs(R0, res);
1310           } else {
1311             __ cbz(R0, *failure_target);
1312           }
1313         }
1314       } else {
1315         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1316         // check for immediate positive hit
1317         __ cmp(klass_RInfo, k_RInfo);
1318         if (!op->should_profile()) {
1319           __ ldr(res, Address(klass_RInfo, Rtemp), ne);
1320           __ cond_cmp(res, k_RInfo, ne);
1321           set_instanceof_result(_masm, res, eq);
1322         } else {
1323           __ ldr(Rtemp, Address(klass_RInfo, Rtemp), ne);
1324           __ cond_cmp(Rtemp, k_RInfo, ne);
1325         }
1326         __ b(*success_target, eq);
1327         // check for immediate negative hit
1328         if (op->should_profile()) {
1329           __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1330         }
1331         __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1332         if (!op->should_profile()) {
1333           __ mov(res, 0, ne);
1334         }
1335         __ b(*failure_target, ne);
1336         // slow case
1337         assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1338         __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1339         if (!op->should_profile()) {
1340           move_regs(R0, res);
1341         }
1342         if (op->should_profile()) {
1343           __ cbz(R0, *failure_target);
1344         }
1345       }
1346 
1347       if (op->should_profile()) {
1348         Label done_ok, done_failure;
1349         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1350         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, res, tmp1,
1351                                   &profile_cast_success, &profile_cast_failure,
1352                                   &done_ok, &done_failure);
1353         __ bind(done_failure);
1354         __ mov(res, 0);
1355         __ b(done);
1356         __ bind(done_ok);
1357         __ mov(res, 1);
1358       }
1359       __ bind(done);
1360       break;
1361     }
1362     default:
1363       ShouldNotReachHere();
1364   }
1365 }
1366 
1367 
1368 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1369   //   if (*addr == cmpval) {
1370   //     *addr = newval;
1371   //     dest = 1;
1372   //   } else {
1373   //     dest = 0;
1374   //   }
1375   // FIXME: membar_release
1376   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
1377   Register addr = op->addr()->is_register() ?
1378     op->addr()->as_pointer_register() :
1379     op->addr()->as_address_ptr()->base()->as_pointer_register();
1380   assert(op->addr()->is_register() || op->addr()->as_address_ptr()->disp() == 0, "unexpected disp");
1381   assert(op->addr()->is_register() || op->addr()->as_address_ptr()->index() == LIR_OprDesc::illegalOpr(), "unexpected index");
1382   if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
1383     Register cmpval = op->cmp_value()->as_register();
1384     Register newval = op->new_value()->as_register();
1385     Register dest = op->result_opr()->as_register();
1386     assert_different_registers(dest, addr, cmpval, newval, Rtemp);
1387 
1388     __ atomic_cas_bool(cmpval, newval, addr, 0, Rtemp); // Rtemp free by default at C1 LIR layer
1389     __ mov(dest, 1, eq);
1390     __ mov(dest, 0, ne);
1391   } else if (op->code() == lir_cas_long) {
1392     assert(VM_Version::supports_cx8(), "wrong machine");
1393     Register cmp_value_lo = op->cmp_value()->as_register_lo();
1394     Register cmp_value_hi = op->cmp_value()->as_register_hi();
1395     Register new_value_lo = op->new_value()->as_register_lo();
1396     Register new_value_hi = op->new_value()->as_register_hi();
1397     Register dest = op->result_opr()->as_register();
1398     Register tmp_lo = op->tmp1()->as_register_lo();
1399     Register tmp_hi = op->tmp1()->as_register_hi();
1400 
1401     assert_different_registers(tmp_lo, tmp_hi, cmp_value_lo, cmp_value_hi, dest, new_value_lo, new_value_hi, addr);
1402     assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
1403     assert(new_value_hi->encoding() == new_value_lo->encoding() + 1, "non aligned register pair");
1404     assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
1405     assert((new_value_lo->encoding() & 0x1) == 0, "misaligned register pair");
1406     __ atomic_cas64(tmp_lo, tmp_hi, dest, cmp_value_lo, cmp_value_hi,
1407                     new_value_lo, new_value_hi, addr, 0);
1408   } else {
1409     Unimplemented();
1410   }
1411   // FIXME: is full membar really needed instead of just membar_acquire?
1412   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
1413 }
1414 
1415 
1416 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1417   AsmCondition acond = al;
1418   AsmCondition ncond = nv;
1419   if (opr1 != opr2) {
1420     switch (condition) {
1421       case lir_cond_equal:        acond = eq; ncond = ne; break;
1422       case lir_cond_notEqual:     acond = ne; ncond = eq; break;
1423       case lir_cond_less:         acond = lt; ncond = ge; break;
1424       case lir_cond_lessEqual:    acond = le; ncond = gt; break;
1425       case lir_cond_greaterEqual: acond = ge; ncond = lt; break;
1426       case lir_cond_greater:      acond = gt; ncond = le; break;
1427       case lir_cond_aboveEqual:   acond = hs; ncond = lo; break;
1428       case lir_cond_belowEqual:   acond = ls; ncond = hi; break;
1429       default: ShouldNotReachHere();
1430     }
1431   }
1432 
1433   for (;;) {                         // two iterations only
1434     if (opr1 == result) {
1435       // do nothing
1436     } else if (opr1->is_single_cpu()) {
1437       __ mov(result->as_register(), opr1->as_register(), acond);
1438     } else if (opr1->is_double_cpu()) {
1439       __ long_move(result->as_register_lo(), result->as_register_hi(),
1440                    opr1->as_register_lo(), opr1->as_register_hi(), acond);
1441     } else if (opr1->is_single_stack()) {
1442       __ ldr(result->as_register(), frame_map()->address_for_slot(opr1->single_stack_ix()), acond);
1443     } else if (opr1->is_double_stack()) {
1444       __ ldr(result->as_register_lo(),
1445              frame_map()->address_for_slot(opr1->double_stack_ix(), lo_word_offset_in_bytes), acond);
1446       __ ldr(result->as_register_hi(),
1447              frame_map()->address_for_slot(opr1->double_stack_ix(), hi_word_offset_in_bytes), acond);
1448     } else if (opr1->is_illegal()) {
1449       // do nothing: this part of the cmove has been optimized away in the peephole optimizer
1450     } else {
1451       assert(opr1->is_constant(), "must be");
1452       LIR_Const* c = opr1->as_constant_ptr();
1453 
1454       switch (c->type()) {
1455         case T_INT:
1456           __ mov_slow(result->as_register(), c->as_jint(), acond);
1457           break;
1458         case T_LONG:
1459           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1460           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1461           break;
1462         case T_OBJECT:
1463           __ mov_oop(result->as_register(), c->as_jobject(), 0, acond);
1464           break;
1465         case T_FLOAT:
1466 #ifdef __SOFTFP__
1467           // not generated now.
1468           __ mov_slow(result->as_register(), c->as_jint(), acond);
1469 #else
1470           __ mov_float(result->as_float_reg(), c->as_jfloat(), acond);
1471 #endif // __SOFTFP__
1472           break;
1473         case T_DOUBLE:
1474 #ifdef __SOFTFP__
1475           // not generated now.
1476           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1477           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1478 #else
1479           __ mov_double(result->as_double_reg(), c->as_jdouble(), acond);
1480 #endif // __SOFTFP__
1481           break;
1482         case T_METADATA:
1483           __ mov_metadata(result->as_register(), c->as_metadata(), acond);
1484           break;
1485         default:
1486           ShouldNotReachHere();
1487       }
1488     }
1489 
1490     // Negate the condition and repeat the algorithm with the second operand
1491     if (opr1 == opr2) { break; }
1492     opr1 = opr2;
1493     acond = ncond;
1494   }
1495 }
1496 
1497 #ifdef ASSERT
1498 static int reg_size(LIR_Opr op) {
1499   switch (op->type()) {
1500   case T_FLOAT:
1501   case T_INT:      return BytesPerInt;
1502   case T_LONG:
1503   case T_DOUBLE:   return BytesPerLong;
1504   case T_OBJECT:
1505   case T_ARRAY:
1506   case T_METADATA: return BytesPerWord;
1507   case T_ADDRESS:
1508   case T_ILLEGAL:  // fall through
1509   default: ShouldNotReachHere(); return -1;
1510   }
1511 }
1512 #endif
1513 
1514 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1515   assert(info == NULL, "unused on this code path");
1516   assert(dest->is_register(), "wrong items state");
1517 
1518   if (right->is_address()) {
1519     // special case for adding shifted/extended register
1520     const Register res = dest->as_pointer_register();
1521     const Register lreg = left->as_pointer_register();
1522     const LIR_Address* addr = right->as_address_ptr();
1523 
1524     assert(addr->base()->as_pointer_register() == lreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1525 
1526     int scale = addr->scale();
1527     AsmShift shift = lsl;
1528 
1529 
1530     assert(reg_size(addr->base()) == reg_size(addr->index()), "should be");
1531     assert(reg_size(addr->base()) == reg_size(dest), "should be");
1532     assert(reg_size(dest) == wordSize, "should be");
1533 
1534     AsmOperand operand(addr->index()->as_pointer_register(), shift, scale);
1535     switch (code) {
1536       case lir_add: __ add(res, lreg, operand); break;
1537       case lir_sub: __ sub(res, lreg, operand); break;
1538       default: ShouldNotReachHere();
1539     }
1540 
1541   } else if (left->is_address()) {
1542     assert(code == lir_sub && right->is_single_cpu(), "special case used by strength_reduce_multiply()");
1543     const LIR_Address* addr = left->as_address_ptr();
1544     const Register res = dest->as_register();
1545     const Register rreg = right->as_register();
1546     assert(addr->base()->as_register() == rreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1547     __ rsb(res, rreg, AsmOperand(addr->index()->as_register(), lsl, addr->scale()));
1548 
1549   } else if (dest->is_single_cpu()) {
1550     assert(left->is_single_cpu(), "unexpected left operand");
1551 
1552     const Register res = dest->as_register();
1553     const Register lreg = left->as_register();
1554 
1555     if (right->is_single_cpu()) {
1556       const Register rreg = right->as_register();
1557       switch (code) {
1558         case lir_add: __ add_32(res, lreg, rreg); break;
1559         case lir_sub: __ sub_32(res, lreg, rreg); break;
1560         case lir_mul: __ mul_32(res, lreg, rreg); break;
1561         default: ShouldNotReachHere();
1562       }
1563     } else {
1564       assert(right->is_constant(), "must be");
1565       const jint c = right->as_constant_ptr()->as_jint();
1566       if (!Assembler::is_arith_imm_in_range(c)) {
1567         BAILOUT("illegal arithmetic operand");
1568       }
1569       switch (code) {
1570         case lir_add: __ add_32(res, lreg, c); break;
1571         case lir_sub: __ sub_32(res, lreg, c); break;
1572         default: ShouldNotReachHere();
1573       }
1574     }
1575 
1576   } else if (dest->is_double_cpu()) {
1577     Register res_lo = dest->as_register_lo();
1578     Register res_hi = dest->as_register_hi();
1579     Register lreg_lo = left->as_register_lo();
1580     Register lreg_hi = left->as_register_hi();
1581     if (right->is_double_cpu()) {
1582       Register rreg_lo = right->as_register_lo();
1583       Register rreg_hi = right->as_register_hi();
1584       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1585         res_lo = Rtemp;
1586       }
1587       switch (code) {
1588         case lir_add:
1589           __ adds(res_lo, lreg_lo, rreg_lo);
1590           __ adc(res_hi, lreg_hi, rreg_hi);
1591           break;
1592         case lir_sub:
1593           __ subs(res_lo, lreg_lo, rreg_lo);
1594           __ sbc(res_hi, lreg_hi, rreg_hi);
1595           break;
1596         default:
1597           ShouldNotReachHere();
1598       }
1599     } else {
1600       assert(right->is_constant(), "must be");
1601       assert((right->as_constant_ptr()->as_jlong() >> 32) == 0, "out of range");
1602       const jint c = (jint) right->as_constant_ptr()->as_jlong();
1603       if (res_lo == lreg_hi) {
1604         res_lo = Rtemp;
1605       }
1606       switch (code) {
1607         case lir_add:
1608           __ adds(res_lo, lreg_lo, c);
1609           __ adc(res_hi, lreg_hi, 0);
1610           break;
1611         case lir_sub:
1612           __ subs(res_lo, lreg_lo, c);
1613           __ sbc(res_hi, lreg_hi, 0);
1614           break;
1615         default:
1616           ShouldNotReachHere();
1617       }
1618     }
1619     move_regs(res_lo, dest->as_register_lo());
1620 
1621   } else if (dest->is_single_fpu()) {
1622     assert(left->is_single_fpu(), "must be");
1623     assert(right->is_single_fpu(), "must be");
1624     const FloatRegister res = dest->as_float_reg();
1625     const FloatRegister lreg = left->as_float_reg();
1626     const FloatRegister rreg = right->as_float_reg();
1627     switch (code) {
1628       case lir_add: __ add_float(res, lreg, rreg); break;
1629       case lir_sub: __ sub_float(res, lreg, rreg); break;
1630       case lir_mul: __ mul_float(res, lreg, rreg); break;
1631       case lir_div: __ div_float(res, lreg, rreg); break;
1632       default: ShouldNotReachHere();
1633     }
1634   } else if (dest->is_double_fpu()) {
1635     assert(left->is_double_fpu(), "must be");
1636     assert(right->is_double_fpu(), "must be");
1637     const FloatRegister res = dest->as_double_reg();
1638     const FloatRegister lreg = left->as_double_reg();
1639     const FloatRegister rreg = right->as_double_reg();
1640     switch (code) {
1641       case lir_add: __ add_double(res, lreg, rreg); break;
1642       case lir_sub: __ sub_double(res, lreg, rreg); break;
1643       case lir_mul: __ mul_double(res, lreg, rreg); break;
1644       case lir_div: __ div_double(res, lreg, rreg); break;
1645       default: ShouldNotReachHere();
1646     }
1647   } else {
1648     ShouldNotReachHere();
1649   }
1650 }
1651 
1652 
1653 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
1654   switch (code) {
1655     case lir_abs:
1656       __ abs_double(dest->as_double_reg(), value->as_double_reg());
1657       break;
1658     case lir_sqrt:
1659       __ sqrt_double(dest->as_double_reg(), value->as_double_reg());
1660       break;
1661     default:
1662       ShouldNotReachHere();
1663   }
1664 }
1665 
1666 
1667 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1668   assert(dest->is_register(), "wrong items state");
1669   assert(left->is_register(), "wrong items state");
1670 
1671   if (dest->is_single_cpu()) {
1672 
1673     const Register res = dest->as_register();
1674     const Register lreg = left->as_register();
1675 
1676     if (right->is_single_cpu()) {
1677       const Register rreg = right->as_register();
1678       switch (code) {
1679         case lir_logic_and: __ and_32(res, lreg, rreg); break;
1680         case lir_logic_or:  __ orr_32(res, lreg, rreg); break;
1681         case lir_logic_xor: __ eor_32(res, lreg, rreg); break;
1682         default: ShouldNotReachHere();
1683       }
1684     } else {
1685       assert(right->is_constant(), "must be");
1686       const uint c = (uint)right->as_constant_ptr()->as_jint();
1687       if (!Assembler::is_arith_imm_in_range(c)) {
1688         BAILOUT("illegal arithmetic operand");
1689       }
1690       switch (code) {
1691         case lir_logic_and: __ and_32(res, lreg, c); break;
1692         case lir_logic_or:  __ orr_32(res, lreg, c); break;
1693         case lir_logic_xor: __ eor_32(res, lreg, c); break;
1694         default: ShouldNotReachHere();
1695       }
1696     }
1697   } else {
1698     assert(dest->is_double_cpu(), "should be");
1699     Register res_lo = dest->as_register_lo();
1700 
1701     assert (dest->type() == T_LONG, "unexpected result type");
1702     assert (left->type() == T_LONG, "unexpected left type");
1703     assert (right->type() == T_LONG, "unexpected right type");
1704 
1705     const Register res_hi = dest->as_register_hi();
1706     const Register lreg_lo = left->as_register_lo();
1707     const Register lreg_hi = left->as_register_hi();
1708 
1709     if (right->is_register()) {
1710       const Register rreg_lo = right->as_register_lo();
1711       const Register rreg_hi = right->as_register_hi();
1712       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1713         res_lo = Rtemp; // Temp register helps to avoid overlap between result and input
1714       }
1715       switch (code) {
1716         case lir_logic_and:
1717           __ andr(res_lo, lreg_lo, rreg_lo);
1718           __ andr(res_hi, lreg_hi, rreg_hi);
1719           break;
1720         case lir_logic_or:
1721           __ orr(res_lo, lreg_lo, rreg_lo);
1722           __ orr(res_hi, lreg_hi, rreg_hi);
1723           break;
1724         case lir_logic_xor:
1725           __ eor(res_lo, lreg_lo, rreg_lo);
1726           __ eor(res_hi, lreg_hi, rreg_hi);
1727           break;
1728         default:
1729           ShouldNotReachHere();
1730       }
1731       move_regs(res_lo, dest->as_register_lo());
1732     } else {
1733       assert(right->is_constant(), "must be");
1734       const jint c_lo = (jint) right->as_constant_ptr()->as_jlong();
1735       const jint c_hi = (jint) (right->as_constant_ptr()->as_jlong() >> 32);
1736       // Case for logic_or from do_ClassIDIntrinsic()
1737       if (c_hi == 0 && AsmOperand::is_rotated_imm(c_lo)) {
1738         switch (code) {
1739           case lir_logic_and:
1740             __ andr(res_lo, lreg_lo, c_lo);
1741             __ mov(res_hi, 0);
1742             break;
1743           case lir_logic_or:
1744             __ orr(res_lo, lreg_lo, c_lo);
1745             break;
1746           case lir_logic_xor:
1747             __ eor(res_lo, lreg_lo, c_lo);
1748             break;
1749         default:
1750           ShouldNotReachHere();
1751         }
1752       } else if (code == lir_logic_and &&
1753                  c_hi == -1 &&
1754                  (AsmOperand::is_rotated_imm(c_lo) ||
1755                   AsmOperand::is_rotated_imm(~c_lo))) {
1756         // Another case which handles logic_and from do_ClassIDIntrinsic()
1757         if (AsmOperand::is_rotated_imm(c_lo)) {
1758           __ andr(res_lo, lreg_lo, c_lo);
1759         } else {
1760           __ bic(res_lo, lreg_lo, ~c_lo);
1761         }
1762         if (res_hi != lreg_hi) {
1763           __ mov(res_hi, lreg_hi);
1764         }
1765       } else {
1766         BAILOUT("64 bit constant cannot be inlined");
1767       }
1768     }
1769   }
1770 }
1771 
1772 
1773 
1774 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1775   if (opr1->is_single_cpu()) {
1776     if (opr2->is_constant()) {
1777       switch (opr2->as_constant_ptr()->type()) {
1778         case T_INT: {
1779           const jint c = opr2->as_constant_ptr()->as_jint();
1780           if (Assembler::is_arith_imm_in_range(c)) {
1781             __ cmp_32(opr1->as_register(), c);
1782           } else if (Assembler::is_arith_imm_in_range(-c)) {
1783             __ cmn_32(opr1->as_register(), -c);
1784           } else {
1785             // This can happen when compiling lookupswitch
1786             __ mov_slow(Rtemp, c);
1787             __ cmp_32(opr1->as_register(), Rtemp);
1788           }
1789           break;
1790         }
1791         case T_OBJECT:
1792           assert(opr2->as_constant_ptr()->as_jobject() == NULL, "cannot handle otherwise");
1793           __ cmp(opr1->as_register(), 0);
1794           break;
1795         case T_METADATA:
1796           assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "Only equality tests");
1797           assert(opr2->as_constant_ptr()->as_metadata() == NULL, "cannot handle otherwise");
1798           __ cmp(opr1->as_register(), 0);
1799           break;
1800         default:
1801           ShouldNotReachHere();
1802       }
1803     } else if (opr2->is_single_cpu()) {
1804       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1805         assert(opr2->type() == T_OBJECT || opr2->type() == T_ARRAY, "incompatibe type");
1806         __ cmpoop(opr1->as_register(), opr2->as_register());
1807       } else if (opr1->type() == T_METADATA || opr1->type() == T_ADDRESS) {
1808         assert(opr2->type() == T_METADATA || opr2->type() == T_ADDRESS, "incompatibe type");
1809         __ cmp(opr1->as_register(), opr2->as_register());
1810       } else {
1811         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY && opr2->type() != T_METADATA && opr2->type() != T_ADDRESS, "incompatibe type");
1812         __ cmp_32(opr1->as_register(), opr2->as_register());
1813       }
1814     } else {
1815       ShouldNotReachHere();
1816     }
1817   } else if (opr1->is_double_cpu()) {
1818     Register xlo = opr1->as_register_lo();
1819     Register xhi = opr1->as_register_hi();
1820     if (opr2->is_constant() && opr2->as_jlong() == 0) {
1821       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "cannot handle otherwise");
1822       __ orrs(Rtemp, xlo, xhi);
1823     } else if (opr2->is_register()) {
1824       Register ylo = opr2->as_register_lo();
1825       Register yhi = opr2->as_register_hi();
1826       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1827         __ teq(xhi, yhi);
1828         __ teq(xlo, ylo, eq);
1829       } else {
1830         __ subs(Rtemp, xlo, ylo);
1831         __ sbcs(Rtemp, xhi, yhi);
1832       }
1833     } else {
1834       ShouldNotReachHere();
1835     }
1836   } else if (opr1->is_single_fpu()) {
1837     if (opr2->is_constant()) {
1838       assert(opr2->as_jfloat() == 0.0f, "cannot handle otherwise");
1839       __ cmp_zero_float(opr1->as_float_reg());
1840     } else {
1841       __ cmp_float(opr1->as_float_reg(), opr2->as_float_reg());
1842     }
1843   } else if (opr1->is_double_fpu()) {
1844     if (opr2->is_constant()) {
1845       assert(opr2->as_jdouble() == 0.0, "cannot handle otherwise");
1846       __ cmp_zero_double(opr1->as_double_reg());
1847     } else {
1848       __ cmp_double(opr1->as_double_reg(), opr2->as_double_reg());
1849     }
1850   } else {
1851     ShouldNotReachHere();
1852   }
1853 }
1854 
1855 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1856   const Register res = dst->as_register();
1857   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1858     comp_op(lir_cond_unknown, left, right, op);
1859     __ fmstat();
1860     if (code == lir_ucmp_fd2i) {  // unordered is less
1861       __ mvn(res, 0, lt);
1862       __ mov(res, 1, ge);
1863     } else {                      // unordered is greater
1864       __ mov(res, 1, cs);
1865       __ mvn(res, 0, cc);
1866     }
1867     __ mov(res, 0, eq);
1868 
1869   } else {
1870     assert(code == lir_cmp_l2i, "must be");
1871 
1872     Label done;
1873     const Register xlo = left->as_register_lo();
1874     const Register xhi = left->as_register_hi();
1875     const Register ylo = right->as_register_lo();
1876     const Register yhi = right->as_register_hi();
1877     __ cmp(xhi, yhi);
1878     __ mov(res, 1, gt);
1879     __ mvn(res, 0, lt);
1880     __ b(done, ne);
1881     __ subs(res, xlo, ylo);
1882     __ mov(res, 1, hi);
1883     __ mvn(res, 0, lo);
1884     __ bind(done);
1885   }
1886 }
1887 
1888 
1889 void LIR_Assembler::align_call(LIR_Code code) {
1890   // Not needed
1891 }
1892 
1893 
1894 void LIR_Assembler::call(LIR_OpJavaCall *op, relocInfo::relocType rtype) {
1895   int ret_addr_offset = __ patchable_call(op->addr(), rtype);
1896   assert(ret_addr_offset == __ offset(), "embedded return address not allowed");
1897   add_call_info_here(op->info());
1898 }
1899 
1900 
1901 void LIR_Assembler::ic_call(LIR_OpJavaCall *op) {
1902   bool near_range = __ cache_fully_reachable();
1903   address oop_address = pc();
1904 
1905   bool use_movw = VM_Version::supports_movw();
1906 
1907   // Ricklass may contain something that is not a metadata pointer so
1908   // mov_metadata can't be used
1909   InlinedAddress value((address)Universe::non_oop_word());
1910   InlinedAddress addr(op->addr());
1911   if (use_movw) {
1912     __ movw(Ricklass, ((unsigned int)Universe::non_oop_word()) & 0xffff);
1913     __ movt(Ricklass, ((unsigned int)Universe::non_oop_word()) >> 16);
1914   } else {
1915     // No movw/movt, must be load a pc relative value but no
1916     // relocation so no metadata table to load from.
1917     // Use a b instruction rather than a bl, inline constant after the
1918     // branch, use a PC relative ldr to load the constant, arrange for
1919     // the call to return after the constant(s).
1920     __ ldr_literal(Ricklass, value);
1921   }
1922   __ relocate(virtual_call_Relocation::spec(oop_address));
1923   if (near_range && use_movw) {
1924     __ bl(op->addr());
1925   } else {
1926     Label call_return;
1927     __ adr(LR, call_return);
1928     if (near_range) {
1929       __ b(op->addr());
1930     } else {
1931       __ indirect_jump(addr, Rtemp);
1932       __ bind_literal(addr);
1933     }
1934     if (!use_movw) {
1935       __ bind_literal(value);
1936     }
1937     __ bind(call_return);
1938   }
1939   add_call_info(code_offset(), op->info());
1940 }
1941 
1942 void LIR_Assembler::emit_static_call_stub() {
1943   address call_pc = __ pc();
1944   address stub = __ start_a_stub(call_stub_size());
1945   if (stub == NULL) {
1946     BAILOUT("static call stub overflow");
1947   }
1948 
1949   DEBUG_ONLY(int offset = code_offset();)
1950 
1951   InlinedMetadata metadata_literal(NULL);
1952   __ relocate(static_stub_Relocation::spec(call_pc));
1953   // If not a single instruction, NativeMovConstReg::next_instruction_address()
1954   // must jump over the whole following ldr_literal.
1955   // (See CompiledStaticCall::set_to_interpreted())
1956 #ifdef ASSERT
1957   address ldr_site = __ pc();
1958 #endif
1959   __ ldr_literal(Rmethod, metadata_literal);
1960   assert(nativeMovConstReg_at(ldr_site)->next_instruction_address() == __ pc(), "Fix ldr_literal or its parsing");
1961   bool near_range = __ cache_fully_reachable();
1962   InlinedAddress dest((address)-1);
1963   if (near_range) {
1964     address branch_site = __ pc();
1965     __ b(branch_site); // b to self maps to special NativeJump -1 destination
1966   } else {
1967     __ indirect_jump(dest, Rtemp);
1968   }
1969   __ bind_literal(metadata_literal); // includes spec_for_immediate reloc
1970   if (!near_range) {
1971     __ bind_literal(dest); // special NativeJump -1 destination
1972   }
1973 
1974   assert(code_offset() - offset <= call_stub_size(), "overflow");
1975   __ end_a_stub();
1976 }
1977 
1978 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1979   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1980   assert(exceptionPC->as_register()  == Rexception_pc, "must match");
1981   info->add_register_oop(exceptionOop);
1982 
1983   Runtime1::StubID handle_id = compilation()->has_fpu_code() ?
1984                                Runtime1::handle_exception_id :
1985                                Runtime1::handle_exception_nofpu_id;
1986   Label return_address;
1987   __ adr(Rexception_pc, return_address);
1988   __ call(Runtime1::entry_for(handle_id), relocInfo::runtime_call_type);
1989   __ bind(return_address);
1990   add_call_info_here(info);  // for exception handler
1991 }
1992 
1993 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1994   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1995   __ b(_unwind_handler_entry);
1996 }
1997 
1998 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
1999   AsmShift shift = lsl;
2000   switch (code) {
2001     case lir_shl:  shift = lsl; break;
2002     case lir_shr:  shift = asr; break;
2003     case lir_ushr: shift = lsr; break;
2004     default: ShouldNotReachHere();
2005   }
2006 
2007   if (dest->is_single_cpu()) {
2008     __ andr(Rtemp, count->as_register(), 31);
2009     __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, Rtemp));
2010   } else if (dest->is_double_cpu()) {
2011     Register dest_lo = dest->as_register_lo();
2012     Register dest_hi = dest->as_register_hi();
2013     Register src_lo  = left->as_register_lo();
2014     Register src_hi  = left->as_register_hi();
2015     Register Rcount  = count->as_register();
2016     // Resolve possible register conflicts
2017     if (shift == lsl && dest_hi == src_lo) {
2018       dest_hi = Rtemp;
2019     } else if (shift != lsl && dest_lo == src_hi) {
2020       dest_lo = Rtemp;
2021     } else if (dest_lo == src_lo && dest_hi == src_hi) {
2022       dest_lo = Rtemp;
2023     } else if (dest_lo == Rcount || dest_hi == Rcount) {
2024       Rcount = Rtemp;
2025     }
2026     __ andr(Rcount, count->as_register(), 63);
2027     __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, Rcount);
2028     move_regs(dest_lo, dest->as_register_lo());
2029     move_regs(dest_hi, dest->as_register_hi());
2030   } else {
2031     ShouldNotReachHere();
2032   }
2033 }
2034 
2035 
2036 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2037   AsmShift shift = lsl;
2038   switch (code) {
2039     case lir_shl:  shift = lsl; break;
2040     case lir_shr:  shift = asr; break;
2041     case lir_ushr: shift = lsr; break;
2042     default: ShouldNotReachHere();
2043   }
2044 
2045   if (dest->is_single_cpu()) {
2046     count &= 31;
2047     if (count != 0) {
2048       __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, count));
2049     } else {
2050       move_regs(left->as_register(), dest->as_register());
2051     }
2052   } else if (dest->is_double_cpu()) {
2053     count &= 63;
2054     if (count != 0) {
2055       Register dest_lo = dest->as_register_lo();
2056       Register dest_hi = dest->as_register_hi();
2057       Register src_lo  = left->as_register_lo();
2058       Register src_hi  = left->as_register_hi();
2059       // Resolve possible register conflicts
2060       if (shift == lsl && dest_hi == src_lo) {
2061         dest_hi = Rtemp;
2062       } else if (shift != lsl && dest_lo == src_hi) {
2063         dest_lo = Rtemp;
2064       }
2065       __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, count);
2066       move_regs(dest_lo, dest->as_register_lo());
2067       move_regs(dest_hi, dest->as_register_hi());
2068     } else {
2069       __ long_move(dest->as_register_lo(), dest->as_register_hi(),
2070                    left->as_register_lo(), left->as_register_hi());
2071     }
2072   } else {
2073     ShouldNotReachHere();
2074   }
2075 }
2076 
2077 
2078 // Saves 4 given registers in reserved argument area.
2079 void LIR_Assembler::save_in_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2080   verify_reserved_argument_area_size(4);
2081   __ stmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4));
2082 }
2083 
2084 // Restores 4 given registers from reserved argument area.
2085 void LIR_Assembler::restore_from_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2086   __ ldmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4), no_writeback);
2087 }
2088 
2089 
2090 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2091   ciArrayKlass* default_type = op->expected_type();
2092   Register src = op->src()->as_register();
2093   Register src_pos = op->src_pos()->as_register();
2094   Register dst = op->dst()->as_register();
2095   Register dst_pos = op->dst_pos()->as_register();
2096   Register length  = op->length()->as_register();
2097   Register tmp = op->tmp()->as_register();
2098   Register tmp2 = Rtemp;
2099 
2100   assert(src == R0 && src_pos == R1 && dst == R2 && dst_pos == R3, "code assumption");
2101 
2102   CodeStub* stub = op->stub();
2103 
2104   int flags = op->flags();
2105   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
2106   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2107 
2108   // If we don't know anything or it's an object array, just go through the generic arraycopy
2109   if (default_type == NULL) {
2110 
2111     // save arguments, because they will be killed by a runtime call
2112     save_in_reserved_area(R0, R1, R2, R3);
2113 
2114     // pass length argument on SP[0]
2115     __ str(length, Address(SP, -2*wordSize, pre_indexed));  // 2 words for a proper stack alignment
2116 
2117     address copyfunc_addr = StubRoutines::generic_arraycopy();
2118     assert(copyfunc_addr != NULL, "generic arraycopy stub required");
2119 #ifndef PRODUCT
2120     if (PrintC1Statistics) {
2121       __ inc_counter((address)&Runtime1::_generic_arraycopystub_cnt, tmp, tmp2);
2122     }
2123 #endif // !PRODUCT
2124     // the stub is in the code cache so close enough
2125     __ call(copyfunc_addr, relocInfo::runtime_call_type);
2126 
2127     __ add(SP, SP, 2*wordSize);
2128 
2129     __ cbz_32(R0, *stub->continuation());
2130 
2131     __ mvn_32(tmp, R0);
2132     restore_from_reserved_area(R0, R1, R2, R3);  // load saved arguments in slow case only
2133     __ sub_32(length, length, tmp);
2134     __ add_32(src_pos, src_pos, tmp);
2135     __ add_32(dst_pos, dst_pos, tmp);
2136 
2137     __ b(*stub->entry());
2138 
2139     __ bind(*stub->continuation());
2140     return;
2141   }
2142 
2143   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(),
2144          "must be true at this point");
2145   int elem_size = type2aelembytes(basic_type);
2146   int shift = exact_log2(elem_size);
2147 
2148   // Check for NULL
2149   if (flags & LIR_OpArrayCopy::src_null_check) {
2150     if (flags & LIR_OpArrayCopy::dst_null_check) {
2151       __ cmp(src, 0);
2152       __ cond_cmp(dst, 0, ne);  // make one instruction shorter if both checks are needed
2153       __ b(*stub->entry(), eq);
2154     } else {
2155       __ cbz(src, *stub->entry());
2156     }
2157   } else if (flags & LIR_OpArrayCopy::dst_null_check) {
2158     __ cbz(dst, *stub->entry());
2159   }
2160 
2161   // If the compiler was not able to prove that exact type of the source or the destination
2162   // of the arraycopy is an array type, check at runtime if the source or the destination is
2163   // an instance type.
2164   if (flags & LIR_OpArrayCopy::type_check) {
2165     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::dst_objarray)) {
2166       __ load_klass(tmp, dst);
2167       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2168       __ mov_slow(tmp, Klass::_lh_neutral_value);
2169       __ cmp_32(tmp2, tmp);
2170       __ b(*stub->entry(), ge);
2171     }
2172 
2173     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::src_objarray)) {
2174       __ load_klass(tmp, src);
2175       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2176       __ mov_slow(tmp, Klass::_lh_neutral_value);
2177       __ cmp_32(tmp2, tmp);
2178       __ b(*stub->entry(), ge);
2179     }
2180   }
2181 
2182   // Check if negative
2183   const int all_positive_checks = LIR_OpArrayCopy::src_pos_positive_check |
2184                                   LIR_OpArrayCopy::dst_pos_positive_check |
2185                                   LIR_OpArrayCopy::length_positive_check;
2186   switch (flags & all_positive_checks) {
2187     case LIR_OpArrayCopy::src_pos_positive_check:
2188       __ branch_if_negative_32(src_pos, *stub->entry());
2189       break;
2190     case LIR_OpArrayCopy::dst_pos_positive_check:
2191       __ branch_if_negative_32(dst_pos, *stub->entry());
2192       break;
2193     case LIR_OpArrayCopy::length_positive_check:
2194       __ branch_if_negative_32(length, *stub->entry());
2195       break;
2196     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::dst_pos_positive_check:
2197       __ branch_if_any_negative_32(src_pos, dst_pos, tmp, *stub->entry());
2198       break;
2199     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2200       __ branch_if_any_negative_32(src_pos, length, tmp, *stub->entry());
2201       break;
2202     case LIR_OpArrayCopy::dst_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2203       __ branch_if_any_negative_32(dst_pos, length, tmp, *stub->entry());
2204       break;
2205     case all_positive_checks:
2206       __ branch_if_any_negative_32(src_pos, dst_pos, length, tmp, *stub->entry());
2207       break;
2208     default:
2209       assert((flags & all_positive_checks) == 0, "the last option");
2210   }
2211 
2212   // Range checks
2213   if (flags & LIR_OpArrayCopy::src_range_check) {
2214     __ ldr_s32(tmp2, Address(src, arrayOopDesc::length_offset_in_bytes()));
2215     __ add_32(tmp, src_pos, length);
2216     __ cmp_32(tmp, tmp2);
2217     __ b(*stub->entry(), hi);
2218   }
2219   if (flags & LIR_OpArrayCopy::dst_range_check) {
2220     __ ldr_s32(tmp2, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2221     __ add_32(tmp, dst_pos, length);
2222     __ cmp_32(tmp, tmp2);
2223     __ b(*stub->entry(), hi);
2224   }
2225 
2226   // Check if src and dst are of the same type
2227   if (flags & LIR_OpArrayCopy::type_check) {
2228     // We don't know the array types are compatible
2229     if (basic_type != T_OBJECT) {
2230       // Simple test for basic type arrays
2231       if (UseCompressedClassPointers) {
2232         // We don't need decode because we just need to compare
2233         __ ldr_u32(tmp, Address(src, oopDesc::klass_offset_in_bytes()));
2234         __ ldr_u32(tmp2, Address(dst, oopDesc::klass_offset_in_bytes()));
2235         __ cmp_32(tmp, tmp2);
2236       } else {
2237         __ load_klass(tmp, src);
2238         __ load_klass(tmp2, dst);
2239         __ cmp(tmp, tmp2);
2240       }
2241       __ b(*stub->entry(), ne);
2242     } else {
2243       // For object arrays, if src is a sub class of dst then we can
2244       // safely do the copy.
2245       Label cont, slow;
2246 
2247       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2248 
2249       __ load_klass(tmp, src);
2250       __ load_klass(tmp2, dst);
2251 
2252       // We are at a call so all live registers are saved before we
2253       // get here
2254       assert_different_registers(tmp, tmp2, R6, altFP_7_11);
2255 
2256       __ check_klass_subtype_fast_path(tmp, tmp2, R6, altFP_7_11, &cont, copyfunc_addr == NULL ? stub->entry() : &slow, NULL);
2257 
2258       __ mov(R6, R0);
2259       __ mov(altFP_7_11, R1);
2260       __ mov(R0, tmp);
2261       __ mov(R1, tmp2);
2262       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); // does not blow any registers except R0, LR and Rtemp
2263       __ cmp_32(R0, 0);
2264       __ mov(R0, R6);
2265       __ mov(R1, altFP_7_11);
2266 
2267       if (copyfunc_addr != NULL) { // use stub if available
2268         // src is not a sub class of dst so we have to do a
2269         // per-element check.
2270 
2271         __ b(cont, ne);
2272 
2273         __ bind(slow);
2274 
2275         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2276         if ((flags & mask) != mask) {
2277           // Check that at least both of them object arrays.
2278           assert(flags & mask, "one of the two should be known to be an object array");
2279 
2280           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2281             __ load_klass(tmp, src);
2282           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2283             __ load_klass(tmp, dst);
2284           }
2285           int lh_offset = in_bytes(Klass::layout_helper_offset());
2286 
2287           __ ldr_u32(tmp2, Address(tmp, lh_offset));
2288 
2289           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2290           __ mov_slow(tmp, objArray_lh);
2291           __ cmp_32(tmp, tmp2);
2292           __ b(*stub->entry(), ne);
2293         }
2294 
2295         save_in_reserved_area(R0, R1, R2, R3);
2296 
2297         Register src_ptr = R0;
2298         Register dst_ptr = R1;
2299         Register len     = R2;
2300         Register chk_off = R3;
2301         Register super_k = tmp;
2302 
2303         __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2304         __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2305 
2306         __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2307         __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2308         __ load_klass(tmp, dst);
2309 
2310         int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2311         int sco_offset = in_bytes(Klass::super_check_offset_offset());
2312 
2313         __ ldr(super_k, Address(tmp, ek_offset));
2314 
2315         __ mov(len, length);
2316         __ ldr_u32(chk_off, Address(super_k, sco_offset));
2317         __ push(super_k);
2318 
2319         __ call(copyfunc_addr, relocInfo::runtime_call_type);
2320 
2321 #ifndef PRODUCT
2322         if (PrintC1Statistics) {
2323           Label failed;
2324           __ cbnz_32(R0, failed);
2325           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_cnt, tmp, tmp2);
2326           __ bind(failed);
2327         }
2328 #endif // PRODUCT
2329 
2330         __ add(SP, SP, wordSize);  // Drop super_k argument
2331 
2332         __ cbz_32(R0, *stub->continuation());
2333         __ mvn_32(tmp, R0);
2334 
2335         // load saved arguments in slow case only
2336         restore_from_reserved_area(R0, R1, R2, R3);
2337 
2338         __ sub_32(length, length, tmp);
2339         __ add_32(src_pos, src_pos, tmp);
2340         __ add_32(dst_pos, dst_pos, tmp);
2341 
2342 #ifndef PRODUCT
2343         if (PrintC1Statistics) {
2344           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_attempt_cnt, tmp, tmp2);
2345         }
2346 #endif
2347 
2348         __ b(*stub->entry());
2349 
2350         __ bind(cont);
2351       } else {
2352         __ b(*stub->entry(), eq);
2353         __ bind(cont);
2354       }
2355     }
2356   }
2357 
2358 #ifndef PRODUCT
2359   if (PrintC1Statistics) {
2360     address counter = Runtime1::arraycopy_count_address(basic_type);
2361     __ inc_counter(counter, tmp, tmp2);
2362   }
2363 #endif // !PRODUCT
2364 
2365   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2366   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2367   const char *name;
2368   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2369 
2370   Register src_ptr = R0;
2371   Register dst_ptr = R1;
2372   Register len     = R2;
2373 
2374   __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2375   __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2376 
2377   __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2378   __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2379 
2380   __ mov(len, length);
2381 
2382   __ call(entry, relocInfo::runtime_call_type);
2383 
2384   __ bind(*stub->continuation());
2385 }
2386 
2387 #ifdef ASSERT
2388  // emit run-time assertion
2389 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2390   assert(op->code() == lir_assert, "must be");
2391 
2392   if (op->in_opr1()->is_valid()) {
2393     assert(op->in_opr2()->is_valid(), "both operands must be valid");
2394     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
2395   } else {
2396     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
2397     assert(op->condition() == lir_cond_always, "no other conditions allowed");
2398   }
2399 
2400   Label ok;
2401   if (op->condition() != lir_cond_always) {
2402     AsmCondition acond = al;
2403     switch (op->condition()) {
2404       case lir_cond_equal:        acond = eq; break;
2405       case lir_cond_notEqual:     acond = ne; break;
2406       case lir_cond_less:         acond = lt; break;
2407       case lir_cond_lessEqual:    acond = le; break;
2408       case lir_cond_greaterEqual: acond = ge; break;
2409       case lir_cond_greater:      acond = gt; break;
2410       case lir_cond_aboveEqual:   acond = hs; break;
2411       case lir_cond_belowEqual:   acond = ls; break;
2412       default:                    ShouldNotReachHere();
2413     }
2414     __ b(ok, acond);
2415   }
2416   if (op->halt()) {
2417     const char* str = __ code_string(op->msg());
2418     __ stop(str);
2419   } else {
2420     breakpoint();
2421   }
2422   __ bind(ok);
2423 }
2424 #endif // ASSERT
2425 
2426 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
2427   fatal("CRC32 intrinsic is not implemented on this platform");
2428 }
2429 
2430 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2431   Register obj = op->obj_opr()->as_pointer_register();
2432   Register hdr = op->hdr_opr()->as_pointer_register();
2433   Register lock = op->lock_opr()->as_pointer_register();
2434   Register tmp = op->scratch_opr()->is_illegal() ? noreg :
2435                  op->scratch_opr()->as_pointer_register();
2436 
2437   if (!UseFastLocking) {
2438     __ b(*op->stub()->entry());
2439   } else if (op->code() == lir_lock) {
2440     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2441     int null_check_offset = __ lock_object(hdr, obj, lock, tmp, *op->stub()->entry());
2442     if (op->info() != NULL) {
2443       add_debug_info_for_null_check(null_check_offset, op->info());
2444     }
2445   } else if (op->code() == lir_unlock) {
2446     __ unlock_object(hdr, obj, lock, tmp, *op->stub()->entry());
2447   } else {
2448     ShouldNotReachHere();
2449   }
2450   __ bind(*op->stub()->continuation());
2451 }
2452 
2453 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2454   Register obj = op->obj()->as_pointer_register();
2455   Register result = op->result_opr()->as_pointer_register();
2456 
2457   CodeEmitInfo* info = op->info();
2458   if (info != NULL) {
2459     add_debug_info_for_null_check_here(info);
2460   }
2461 
2462   if (UseCompressedClassPointers) { // On 32 bit arm??
2463     __ ldr_u32(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2464   } else {
2465     __ ldr(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2466   }
2467 }
2468 
2469 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2470   ciMethod* method = op->profiled_method();
2471   int bci          = op->profiled_bci();
2472   ciMethod* callee = op->profiled_callee();
2473 
2474   // Update counter for all call types
2475   ciMethodData* md = method->method_data_or_null();
2476   assert(md != NULL, "Sanity");
2477   ciProfileData* data = md->bci_to_data(bci);
2478   assert(data != NULL && data->is_CounterData(), "need CounterData for calls");
2479   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2480   Register mdo  = op->mdo()->as_register();
2481   assert(op->tmp1()->is_register(), "tmp1 must be allocated");
2482   Register tmp1 = op->tmp1()->as_pointer_register();
2483   assert_different_registers(mdo, tmp1);
2484   __ mov_metadata(mdo, md->constant_encoding());
2485   int mdo_offset_bias = 0;
2486   int max_offset = 4096;
2487   if (md->byte_offset_of_slot(data, CounterData::count_offset()) + data->size_in_bytes() >= max_offset) {
2488     // The offset is large so bias the mdo by the base of the slot so
2489     // that the ldr can use an immediate offset to reference the slots of the data
2490     mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2491     __ mov_slow(tmp1, mdo_offset_bias);
2492     __ add(mdo, mdo, tmp1);
2493   }
2494 
2495   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2496   // Perform additional virtual call profiling for invokevirtual and
2497   // invokeinterface bytecodes
2498   if (op->should_profile_receiver_type()) {
2499     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2500     Register recv = op->recv()->as_register();
2501     assert_different_registers(mdo, tmp1, recv);
2502     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2503     ciKlass* known_klass = op->known_holder();
2504     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2505       // We know the type that will be seen at this call site; we can
2506       // statically update the MethodData* rather than needing to do
2507       // dynamic tests on the receiver type
2508 
2509       // NOTE: we should probably put a lock around this search to
2510       // avoid collisions by concurrent compilations
2511       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2512       uint i;
2513       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2514         ciKlass* receiver = vc_data->receiver(i);
2515         if (known_klass->equals(receiver)) {
2516           Address data_addr(mdo, md->byte_offset_of_slot(data,
2517                                                          VirtualCallData::receiver_count_offset(i)) -
2518                             mdo_offset_bias);
2519           __ ldr(tmp1, data_addr);
2520           __ add(tmp1, tmp1, DataLayout::counter_increment);
2521           __ str(tmp1, data_addr);
2522           return;
2523         }
2524       }
2525 
2526       // Receiver type not found in profile data; select an empty slot
2527 
2528       // Note that this is less efficient than it should be because it
2529       // always does a write to the receiver part of the
2530       // VirtualCallData rather than just the first time
2531       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2532         ciKlass* receiver = vc_data->receiver(i);
2533         if (receiver == NULL) {
2534           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
2535                             mdo_offset_bias);
2536           __ mov_metadata(tmp1, known_klass->constant_encoding());
2537           __ str(tmp1, recv_addr);
2538           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
2539                             mdo_offset_bias);
2540           __ ldr(tmp1, data_addr);
2541           __ add(tmp1, tmp1, DataLayout::counter_increment);
2542           __ str(tmp1, data_addr);
2543           return;
2544         }
2545       }
2546     } else {
2547       __ load_klass(recv, recv);
2548       Label update_done;
2549       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
2550       // Receiver did not match any saved receiver and there is no empty row for it.
2551       // Increment total counter to indicate polymorphic case.
2552       __ ldr(tmp1, counter_addr);
2553       __ add(tmp1, tmp1, DataLayout::counter_increment);
2554       __ str(tmp1, counter_addr);
2555 
2556       __ bind(update_done);
2557     }
2558   } else {
2559     // Static call
2560     __ ldr(tmp1, counter_addr);
2561     __ add(tmp1, tmp1, DataLayout::counter_increment);
2562     __ str(tmp1, counter_addr);
2563   }
2564 }
2565 
2566 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2567   fatal("Type profiling not implemented on this platform");
2568 }
2569 
2570 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
2571   Unimplemented();
2572 }
2573 
2574 
2575 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
2576   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2577   __ add_slow(dst->as_pointer_register(), mon_addr.base(), mon_addr.disp());
2578 }
2579 
2580 
2581 void LIR_Assembler::align_backward_branch_target() {
2582   // Some ARM processors do better with 8-byte branch target alignment
2583   __ align(8);
2584 }
2585 
2586 
2587 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2588   // tmp must be unused
2589   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2590 
2591   if (left->is_single_cpu()) {
2592     assert (dest->type() == T_INT, "unexpected result type");
2593     assert (left->type() == T_INT, "unexpected left type");
2594     __ neg_32(dest->as_register(), left->as_register());
2595   } else if (left->is_double_cpu()) {
2596     Register dest_lo = dest->as_register_lo();
2597     Register dest_hi = dest->as_register_hi();
2598     Register src_lo = left->as_register_lo();
2599     Register src_hi = left->as_register_hi();
2600     if (dest_lo == src_hi) {
2601       dest_lo = Rtemp;
2602     }
2603     __ rsbs(dest_lo, src_lo, 0);
2604     __ rsc(dest_hi, src_hi, 0);
2605     move_regs(dest_lo, dest->as_register_lo());
2606   } else if (left->is_single_fpu()) {
2607     __ neg_float(dest->as_float_reg(), left->as_float_reg());
2608   } else if (left->is_double_fpu()) {
2609     __ neg_double(dest->as_double_reg(), left->as_double_reg());
2610   } else {
2611     ShouldNotReachHere();
2612   }
2613 }
2614 
2615 
2616 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2617   assert(patch_code == lir_patch_none, "Patch code not supported");
2618   LIR_Address* addr = addr_opr->as_address_ptr();
2619   if (addr->index()->is_illegal()) {
2620     jint c = addr->disp();
2621     if (!Assembler::is_arith_imm_in_range(c)) {
2622       BAILOUT("illegal arithmetic operand");
2623     }
2624     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(), c);
2625   } else {
2626     assert(addr->disp() == 0, "cannot handle otherwise");
2627     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(),
2628            AsmOperand(addr->index()->as_pointer_register(), lsl, addr->scale()));
2629   }
2630 }
2631 
2632 
2633 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2634   assert(!tmp->is_valid(), "don't need temporary");
2635   __ call(dest);
2636   if (info != NULL) {
2637     add_call_info_here(info);
2638   }
2639 }
2640 
2641 
2642 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2643   assert(src->is_double_cpu() && dest->is_address() ||
2644          src->is_address() && dest->is_double_cpu(),
2645          "Simple move_op is called for all other cases");
2646 
2647   int null_check_offset;
2648   if (dest->is_address()) {
2649     // Store
2650     const LIR_Address* addr = dest->as_address_ptr();
2651     const Register src_lo = src->as_register_lo();
2652     const Register src_hi = src->as_register_hi();
2653     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2654 
2655     if (src_lo < src_hi) {
2656       null_check_offset = __ offset();
2657       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(src_hi));
2658     } else {
2659       assert(src_lo < Rtemp, "Rtemp is higher than any allocatable register");
2660       __ mov(Rtemp, src_hi);
2661       null_check_offset = __ offset();
2662       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(Rtemp));
2663     }
2664   } else {
2665     // Load
2666     const LIR_Address* addr = src->as_address_ptr();
2667     const Register dest_lo = dest->as_register_lo();
2668     const Register dest_hi = dest->as_register_hi();
2669     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2670 
2671     null_check_offset = __ offset();
2672     if (dest_lo < dest_hi) {
2673       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(dest_hi));
2674     } else {
2675       assert(dest_lo < Rtemp, "Rtemp is higher than any allocatable register");
2676       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(Rtemp));
2677       __ mov(dest_hi, Rtemp);
2678     }
2679   }
2680 
2681   if (info != NULL) {
2682     add_debug_info_for_null_check(null_check_offset, info);
2683   }
2684 }
2685 
2686 
2687 void LIR_Assembler::membar() {
2688   __ membar(MacroAssembler::StoreLoad, Rtemp);
2689 }
2690 
2691 void LIR_Assembler::membar_acquire() {
2692   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::LoadLoad | MacroAssembler::LoadStore), Rtemp);
2693 }
2694 
2695 void LIR_Assembler::membar_release() {
2696   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2697 }
2698 
2699 void LIR_Assembler::membar_loadload() {
2700   __ membar(MacroAssembler::LoadLoad, Rtemp);
2701 }
2702 
2703 void LIR_Assembler::membar_storestore() {
2704   __ membar(MacroAssembler::StoreStore, Rtemp);
2705 }
2706 
2707 void LIR_Assembler::membar_loadstore() {
2708   __ membar(MacroAssembler::LoadStore, Rtemp);
2709 }
2710 
2711 void LIR_Assembler::membar_storeload() {
2712   __ membar(MacroAssembler::StoreLoad, Rtemp);
2713 }
2714 
2715 void LIR_Assembler::on_spin_wait() {
2716   Unimplemented();
2717 }
2718 
2719 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2720   // Not used on ARM
2721   Unimplemented();
2722 }
2723 
2724 void LIR_Assembler::peephole(LIR_List* lir) {
2725   LIR_OpList* inst = lir->instructions_list();
2726   const int inst_length = inst->length();
2727   for (int i = 0; i < inst_length; i++) {
2728     LIR_Op* op = inst->at(i);
2729     switch (op->code()) {
2730       case lir_cmp: {
2731         // Replace:
2732         //   cmp rX, y
2733         //   cmove [EQ] y, z, rX
2734         // with
2735         //   cmp rX, y
2736         //   cmove [EQ] illegalOpr, z, rX
2737         //
2738         // or
2739         //   cmp rX, y
2740         //   cmove [NE] z, y, rX
2741         // with
2742         //   cmp rX, y
2743         //   cmove [NE] z, illegalOpr, rX
2744         //
2745         // moves from illegalOpr should be removed when converting LIR to native assembly
2746 
2747         LIR_Op2* cmp = op->as_Op2();
2748         assert(cmp != NULL, "cmp LIR instruction is not an op2");
2749 
2750         if (i + 1 < inst_length) {
2751           LIR_Op2* cmove = inst->at(i + 1)->as_Op2();
2752           if (cmove != NULL && cmove->code() == lir_cmove) {
2753             LIR_Opr cmove_res = cmove->result_opr();
2754             bool res_is_op1 = cmove_res == cmp->in_opr1();
2755             bool res_is_op2 = cmove_res == cmp->in_opr2();
2756             LIR_Opr cmp_res, cmp_arg;
2757             if (res_is_op1) {
2758               cmp_res = cmp->in_opr1();
2759               cmp_arg = cmp->in_opr2();
2760             } else if (res_is_op2) {
2761               cmp_res = cmp->in_opr2();
2762               cmp_arg = cmp->in_opr1();
2763             } else {
2764               cmp_res = LIR_OprFact::illegalOpr;
2765               cmp_arg = LIR_OprFact::illegalOpr;
2766             }
2767 
2768             if (cmp_res != LIR_OprFact::illegalOpr) {
2769               LIR_Condition cond = cmove->condition();
2770               if (cond == lir_cond_equal && cmove->in_opr1() == cmp_arg) {
2771                 cmove->set_in_opr1(LIR_OprFact::illegalOpr);
2772               } else if (cond == lir_cond_notEqual && cmove->in_opr2() == cmp_arg) {
2773                 cmove->set_in_opr2(LIR_OprFact::illegalOpr);
2774               }
2775             }
2776           }
2777         }
2778         break;
2779       }
2780 
2781       default:
2782         break;
2783     }
2784   }
2785 }
2786 
2787 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2788   assert(src->is_address(), "sanity");
2789   Address addr = as_Address(src->as_address_ptr());
2790 
2791   if (code == lir_xchg) {
2792   } else {
2793     assert (!data->is_oop(), "xadd for oops");
2794   }
2795 
2796   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2797 
2798   Label retry;
2799   __ bind(retry);
2800 
2801   if (data->type() == T_INT || data->is_oop()) {
2802     Register dst = dest->as_register();
2803     Register new_val = noreg;
2804     __ ldrex(dst, addr);
2805     if (code == lir_xadd) {
2806       Register tmp_reg = tmp->as_register();
2807       if (data->is_constant()) {
2808         assert_different_registers(dst, tmp_reg);
2809         __ add_32(tmp_reg, dst, data->as_constant_ptr()->as_jint());
2810       } else {
2811         assert_different_registers(dst, tmp_reg, data->as_register());
2812         __ add_32(tmp_reg, dst, data->as_register());
2813       }
2814       new_val = tmp_reg;
2815     } else {
2816       if (UseCompressedOops && data->is_oop()) {
2817         new_val = tmp->as_pointer_register();
2818       } else {
2819         new_val = data->as_register();
2820       }
2821       assert_different_registers(dst, new_val);
2822     }
2823     __ strex(Rtemp, new_val, addr);
2824 
2825   } else if (data->type() == T_LONG) {
2826     Register dst_lo = dest->as_register_lo();
2827     Register new_val_lo = noreg;
2828     Register dst_hi = dest->as_register_hi();
2829 
2830     assert(dst_hi->encoding() == dst_lo->encoding() + 1, "non aligned register pair");
2831     assert((dst_lo->encoding() & 0x1) == 0, "misaligned register pair");
2832 
2833     __ bind(retry);
2834     __ ldrexd(dst_lo, addr);
2835     if (code == lir_xadd) {
2836       Register tmp_lo = tmp->as_register_lo();
2837       Register tmp_hi = tmp->as_register_hi();
2838 
2839       assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
2840       assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
2841 
2842       if (data->is_constant()) {
2843         jlong c = data->as_constant_ptr()->as_jlong();
2844         assert((jlong)((jint)c) == c, "overflow");
2845         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi);
2846         __ adds(tmp_lo, dst_lo, (jint)c);
2847         __ adc(tmp_hi, dst_hi, 0);
2848       } else {
2849         Register new_val_lo = data->as_register_lo();
2850         Register new_val_hi = data->as_register_hi();
2851         __ adds(tmp_lo, dst_lo, new_val_lo);
2852         __ adc(tmp_hi, dst_hi, new_val_hi);
2853         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi, new_val_lo, new_val_hi);
2854       }
2855       new_val_lo = tmp_lo;
2856     } else {
2857       new_val_lo = data->as_register_lo();
2858       Register new_val_hi = data->as_register_hi();
2859 
2860       assert_different_registers(dst_lo, dst_hi, new_val_lo, new_val_hi);
2861       assert(new_val_hi->encoding() == new_val_lo->encoding() + 1, "non aligned register pair");
2862       assert((new_val_lo->encoding() & 0x1) == 0, "misaligned register pair");
2863     }
2864     __ strexd(Rtemp, new_val_lo, addr);
2865   } else {
2866     ShouldNotReachHere();
2867   }
2868 
2869   __ cbnz_32(Rtemp, retry);
2870   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
2871 
2872 }
2873 
2874 #undef __