1 /*
   2  * Copyright (c) 2008, 2024, 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     if (slot_offset >= 4096 - BytesPerWord) {
 156       __ add_slow(R2, OSR_buf, slot_offset);
 157       __ ldr(R1, Address(R2, 0*BytesPerWord));
 158       __ ldr(R2, Address(R2, 1*BytesPerWord));
 159     } else {
 160       __ ldr(R1, Address(OSR_buf, slot_offset + 0*BytesPerWord));
 161       __ ldr(R2, Address(OSR_buf, slot_offset + 1*BytesPerWord));
 162     }
 163     __ str(R1, frame_map()->address_for_monitor_lock(i));
 164     __ str(R2, frame_map()->address_for_monitor_object(i));
 165   }
 166 }
 167 
 168 
 169 int LIR_Assembler::check_icache() {
 170   return __ ic_check(CodeEntryAlignment);
 171 }
 172 
 173 void LIR_Assembler::clinit_barrier(ciMethod* method) {
 174   ShouldNotReachHere(); // not implemented
 175 }
 176 
 177 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
 178   jobject o = (jobject)Universe::non_oop_word();
 179   int index = __ oop_recorder()->allocate_oop_index(o);
 180 
 181   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), index);
 182 
 183   __ patchable_mov_oop(reg, o, index);
 184   patching_epilog(patch, lir_patch_normal, reg, info);
 185 }
 186 
 187 
 188 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
 189   Metadata* o = (Metadata*)Universe::non_oop_word();
 190   int index = __ oop_recorder()->allocate_metadata_index(o);
 191   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 192 
 193   __ patchable_mov_metadata(reg, o, index);
 194   patching_epilog(patch, lir_patch_normal, reg, info);
 195 }
 196 
 197 
 198 int LIR_Assembler::initial_frame_size_in_bytes() const {
 199   // Subtracts two words to account for return address and link
 200   return frame_map()->framesize()*VMRegImpl::stack_slot_size - 2*wordSize;
 201 }
 202 
 203 
 204 int LIR_Assembler::emit_exception_handler() {
 205   address handler_base = __ start_a_stub(exception_handler_size());
 206   if (handler_base == nullptr) {
 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(C1StubId::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   // Perform needed unlocking
 246   MonitorExitStub* stub = nullptr;
 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, *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(C1StubId::unwind_exception_id), relocInfo::runtime_call_type, Rtemp);
 257 
 258   // Emit the slow path assembly
 259   if (stub != nullptr) {
 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 == nullptr) {
 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() == nullptr),"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 != nullptr) {
 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   LIR_Address* to_addr = dest->as_address_ptr();
 499   Register base_reg = to_addr->base()->as_pointer_register();
 500   const bool needs_patching = (patch_code != lir_patch_none);
 501 
 502   PatchingStub* patch = nullptr;
 503   if (needs_patching) {
 504     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 505   }
 506 
 507   int null_check_offset = code_offset();
 508 
 509   switch (type) {
 510     case T_ARRAY:
 511     case T_OBJECT:
 512       if (UseCompressedOops && !wide) {
 513         ShouldNotReachHere();
 514       } else {
 515         __ str(src->as_register(), as_Address(to_addr));
 516       }
 517       break;
 518 
 519     case T_ADDRESS:
 520       __ str(src->as_pointer_register(), as_Address(to_addr));
 521       break;
 522 
 523     case T_BYTE:
 524     case T_BOOLEAN:
 525       __ strb(src->as_register(), as_Address(to_addr));
 526       break;
 527 
 528     case T_CHAR:
 529     case T_SHORT:
 530       __ strh(src->as_register(), as_Address(to_addr));
 531       break;
 532 
 533     case T_INT:
 534 #ifdef __SOFTFP__
 535     case T_FLOAT:
 536 #endif // __SOFTFP__
 537       __ str_32(src->as_register(), as_Address(to_addr));
 538       break;
 539 
 540 
 541 #ifdef __SOFTFP__
 542     case T_DOUBLE:
 543 #endif // __SOFTFP__
 544     case T_LONG: {
 545       Register from_lo = src->as_register_lo();
 546       Register from_hi = src->as_register_hi();
 547       if (to_addr->index()->is_register()) {
 548         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 549         assert(to_addr->disp() == 0, "Not yet supporting both");
 550         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 551         base_reg = Rtemp;
 552         __ str(from_lo, Address(Rtemp));
 553         if (patch != nullptr) {
 554           __ nop(); // see comment before patching_epilog for 2nd str
 555           patching_epilog(patch, lir_patch_low, base_reg, info);
 556           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 557           patch_code = lir_patch_high;
 558         }
 559         __ str(from_hi, Address(Rtemp, BytesPerWord));
 560       } else if (base_reg == from_lo) {
 561         __ str(from_hi, as_Address_hi(to_addr));
 562         if (patch != nullptr) {
 563           __ nop(); // see comment before patching_epilog for 2nd str
 564           patching_epilog(patch, lir_patch_high, base_reg, info);
 565           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 566           patch_code = lir_patch_low;
 567         }
 568         __ str(from_lo, as_Address_lo(to_addr));
 569       } else {
 570         __ str(from_lo, as_Address_lo(to_addr));
 571         if (patch != nullptr) {
 572           __ nop(); // see comment before patching_epilog for 2nd str
 573           patching_epilog(patch, lir_patch_low, base_reg, info);
 574           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 575           patch_code = lir_patch_high;
 576         }
 577         __ str(from_hi, as_Address_hi(to_addr));
 578       }
 579       break;
 580     }
 581 
 582 #ifndef __SOFTFP__
 583     case T_FLOAT:
 584       if (to_addr->index()->is_register()) {
 585         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 586         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 587         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 588         __ fsts(src->as_float_reg(), Address(Rtemp, to_addr->disp()));
 589       } else {
 590         __ fsts(src->as_float_reg(), as_Address(to_addr));
 591       }
 592       break;
 593 
 594     case T_DOUBLE:
 595       if (to_addr->index()->is_register()) {
 596         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 597         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 598         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 599         __ fstd(src->as_double_reg(), Address(Rtemp, to_addr->disp()));
 600       } else {
 601         __ fstd(src->as_double_reg(), as_Address(to_addr));
 602       }
 603       break;
 604 #endif // __SOFTFP__
 605 
 606 
 607     default:
 608       ShouldNotReachHere();
 609   }
 610 
 611   if (info != nullptr) {
 612     add_debug_info_for_null_check(null_check_offset, info);
 613   }
 614 
 615   if (patch != nullptr) {
 616     // Offset embedded into LDR/STR instruction may appear not enough
 617     // to address a field. So, provide a space for one more instruction
 618     // that will deal with larger offsets.
 619     __ nop();
 620     patching_epilog(patch, patch_code, base_reg, info);
 621   }
 622 }
 623 
 624 
 625 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 626   assert(src->is_stack(), "should not call otherwise");
 627   assert(dest->is_register(), "should not call otherwise");
 628 
 629   Address addr = src->is_single_word() ?
 630     frame_map()->address_for_slot(src->single_stack_ix()) :
 631     frame_map()->address_for_slot(src->double_stack_ix());
 632 
 633   assert(lo_word_offset_in_bytes == 0 && hi_word_offset_in_bytes == 4, "little ending");
 634   if (dest->is_single_fpu() || dest->is_double_fpu()) {
 635     if (addr.disp() >= 1024) { BAILOUT("Too exotic case to handle here"); }
 636   }
 637 
 638   if (dest->is_single_cpu()) {
 639     switch (type) {
 640       case T_OBJECT:
 641       case T_ARRAY:
 642       case T_ADDRESS:
 643       case T_METADATA: __ ldr(dest->as_register(), addr); break;
 644       case T_FLOAT:    // used in floatToRawIntBits intrinsic implementation
 645       case T_INT:      __ ldr_u32(dest->as_register(), addr); break;
 646       default:
 647         ShouldNotReachHere();
 648     }
 649     if ((type == T_OBJECT) || (type == T_ARRAY)) {
 650       __ verify_oop(dest->as_register());
 651     }
 652   } else if (dest->is_double_cpu()) {
 653     __ ldr(dest->as_register_lo(), addr);
 654     __ ldr(dest->as_register_hi(), frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 655   } else if (dest->is_single_fpu()) {
 656     __ ldr_float(dest->as_float_reg(), addr);
 657   } else if (dest->is_double_fpu()) {
 658     __ ldr_double(dest->as_double_reg(), addr);
 659   } else {
 660     ShouldNotReachHere();
 661   }
 662 }
 663 
 664 
 665 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 666   if (src->is_single_stack()) {
 667     switch (src->type()) {
 668       case T_OBJECT:
 669       case T_ARRAY:
 670       case T_ADDRESS:
 671       case T_METADATA:
 672         __ ldr(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 673         __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 674         break;
 675 
 676       case T_INT:
 677       case T_FLOAT:
 678         __ ldr_u32(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 679         __ str_32(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 680         break;
 681 
 682       default:
 683         ShouldNotReachHere();
 684     }
 685   } else {
 686     assert(src->is_double_stack(), "must be");
 687     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes));
 688     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes));
 689     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 690     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 691   }
 692 }
 693 
 694 
 695 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type,
 696                             LIR_PatchCode patch_code, CodeEmitInfo* info,
 697                             bool wide) {
 698   assert(src->is_address(), "should not call otherwise");
 699   assert(dest->is_register(), "should not call otherwise");
 700   LIR_Address* addr = src->as_address_ptr();
 701 
 702   Register base_reg = addr->base()->as_pointer_register();
 703 
 704   PatchingStub* patch = nullptr;
 705   if (patch_code != lir_patch_none) {
 706     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 707   }
 708   if (info != nullptr) {
 709     add_debug_info_for_null_check_here(info);
 710   }
 711 
 712   switch (type) {
 713     case T_OBJECT:  // fall through
 714     case T_ARRAY:
 715       if (UseCompressedOops && !wide) {
 716         __ ldr_u32(dest->as_register(), as_Address(addr));
 717       } else {
 718         __ ldr(dest->as_register(), as_Address(addr));
 719       }
 720       break;
 721 
 722     case T_ADDRESS:
 723       __ ldr(dest->as_pointer_register(), as_Address(addr));
 724       break;
 725 
 726     case T_INT:
 727 #ifdef __SOFTFP__
 728     case T_FLOAT:
 729 #endif // __SOFTFP__
 730       __ ldr(dest->as_pointer_register(), as_Address(addr));
 731       break;
 732 
 733     case T_BOOLEAN:
 734       __ ldrb(dest->as_register(), as_Address(addr));
 735       break;
 736 
 737     case T_BYTE:
 738       __ ldrsb(dest->as_register(), as_Address(addr));
 739       break;
 740 
 741     case T_CHAR:
 742       __ ldrh(dest->as_register(), as_Address(addr));
 743       break;
 744 
 745     case T_SHORT:
 746       __ ldrsh(dest->as_register(), as_Address(addr));
 747       break;
 748 
 749 
 750 #ifdef __SOFTFP__
 751     case T_DOUBLE:
 752 #endif // __SOFTFP__
 753     case T_LONG: {
 754       Register to_lo = dest->as_register_lo();
 755       Register to_hi = dest->as_register_hi();
 756       if (addr->index()->is_register()) {
 757         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 758         assert(addr->disp() == 0, "Not yet supporting both");
 759         __ add(Rtemp, base_reg, addr->index()->as_register());
 760         base_reg = Rtemp;
 761         __ ldr(to_lo, Address(Rtemp));
 762         if (patch != nullptr) {
 763           __ nop(); // see comment before patching_epilog for 2nd ldr
 764           patching_epilog(patch, lir_patch_low, base_reg, info);
 765           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 766           patch_code = lir_patch_high;
 767         }
 768         __ ldr(to_hi, Address(Rtemp, BytesPerWord));
 769       } else if (base_reg == to_lo) {
 770         __ ldr(to_hi, as_Address_hi(addr));
 771         if (patch != nullptr) {
 772           __ nop(); // see comment before patching_epilog for 2nd ldr
 773           patching_epilog(patch, lir_patch_high, base_reg, info);
 774           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 775           patch_code = lir_patch_low;
 776         }
 777         __ ldr(to_lo, as_Address_lo(addr));
 778       } else {
 779         __ ldr(to_lo, as_Address_lo(addr));
 780         if (patch != nullptr) {
 781           __ nop(); // see comment before patching_epilog for 2nd ldr
 782           patching_epilog(patch, lir_patch_low, base_reg, info);
 783           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 784           patch_code = lir_patch_high;
 785         }
 786         __ ldr(to_hi, as_Address_hi(addr));
 787       }
 788       break;
 789     }
 790 
 791 #ifndef __SOFTFP__
 792     case T_FLOAT:
 793       if (addr->index()->is_register()) {
 794         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 795         __ add(Rtemp, base_reg, addr->index()->as_register());
 796         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 797         __ flds(dest->as_float_reg(), Address(Rtemp, addr->disp()));
 798       } else {
 799         __ flds(dest->as_float_reg(), as_Address(addr));
 800       }
 801       break;
 802 
 803     case T_DOUBLE:
 804       if (addr->index()->is_register()) {
 805         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 806         __ add(Rtemp, base_reg, addr->index()->as_register());
 807         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 808         __ fldd(dest->as_double_reg(), Address(Rtemp, addr->disp()));
 809       } else {
 810         __ fldd(dest->as_double_reg(), as_Address(addr));
 811       }
 812       break;
 813 #endif // __SOFTFP__
 814 
 815 
 816     default:
 817       ShouldNotReachHere();
 818   }
 819 
 820   if (patch != nullptr) {
 821     // Offset embedded into LDR/STR instruction may appear not enough
 822     // to address a field. So, provide a space for one more instruction
 823     // that will deal with larger offsets.
 824     __ nop();
 825     patching_epilog(patch, patch_code, base_reg, info);
 826   }
 827 
 828 }
 829 
 830 
 831 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 832   bool is_32 = op->result_opr()->is_single_cpu();
 833 
 834   if (op->code() == lir_idiv && op->in_opr2()->is_constant() && is_32) {
 835     int c = op->in_opr2()->as_constant_ptr()->as_jint();
 836     assert(is_power_of_2(c), "non power-of-2 constant should be put in a register");
 837 
 838     Register left = op->in_opr1()->as_register();
 839     Register dest = op->result_opr()->as_register();
 840     if (c == 1) {
 841       __ mov(dest, left);
 842     } else if (c == 2) {
 843       __ add_32(dest, left, AsmOperand(left, lsr, 31));
 844       __ asr_32(dest, dest, 1);
 845     } else if (c != (int) 0x80000000) {
 846       int power = log2i_exact(c);
 847       __ asr_32(Rtemp, left, 31);
 848       __ add_32(dest, left, AsmOperand(Rtemp, lsr, 32-power)); // dest = left + (left < 0 ? 2^power - 1 : 0);
 849       __ asr_32(dest, dest, power);                            // dest = dest >>> power;
 850     } else {
 851       // x/0x80000000 is a special case, since dividend is a power of two, but is negative.
 852       // The only possible result values are 0 and 1, with 1 only for dividend == divisor == 0x80000000.
 853       __ cmp_32(left, c);
 854       __ mov(dest, 0, ne);
 855       __ mov(dest, 1, eq);
 856     }
 857   } else {
 858     assert(op->code() == lir_idiv || op->code() == lir_irem, "unexpected op3");
 859     __ call(StubRoutines::Arm::idiv_irem_entry(), relocInfo::runtime_call_type);
 860     add_debug_info_for_div0_here(op->info());
 861   }
 862 }
 863 
 864 
 865 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 866 #ifdef ASSERT
 867   assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
 868   if (op->block() != nullptr)  _branch_target_blocks.append(op->block());
 869   if (op->ublock() != nullptr) _branch_target_blocks.append(op->ublock());
 870   assert(op->info() == nullptr, "CodeEmitInfo?");
 871 #endif // ASSERT
 872 
 873 #ifdef __SOFTFP__
 874   assert (op->code() != lir_cond_float_branch, "this should be impossible");
 875 #else
 876   if (op->code() == lir_cond_float_branch) {
 877     __ fmstat();
 878     __ b(*(op->ublock()->label()), vs);
 879   }
 880 #endif // __SOFTFP__
 881 
 882   AsmCondition acond = al;
 883   switch (op->cond()) {
 884     case lir_cond_equal:        acond = eq; break;
 885     case lir_cond_notEqual:     acond = ne; break;
 886     case lir_cond_less:         acond = lt; break;
 887     case lir_cond_lessEqual:    acond = le; break;
 888     case lir_cond_greaterEqual: acond = ge; break;
 889     case lir_cond_greater:      acond = gt; break;
 890     case lir_cond_aboveEqual:   acond = hs; break;
 891     case lir_cond_belowEqual:   acond = ls; break;
 892     default: assert(op->cond() == lir_cond_always, "must be");
 893   }
 894   __ b(*(op->label()), acond);
 895 }
 896 
 897 
 898 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 899   LIR_Opr src  = op->in_opr();
 900   LIR_Opr dest = op->result_opr();
 901 
 902   switch (op->bytecode()) {
 903     case Bytecodes::_i2l:
 904       move_regs(src->as_register(), dest->as_register_lo());
 905       __ mov(dest->as_register_hi(), AsmOperand(src->as_register(), asr, 31));
 906       break;
 907     case Bytecodes::_l2i:
 908       move_regs(src->as_register_lo(), dest->as_register());
 909       break;
 910     case Bytecodes::_i2b:
 911       __ sign_extend(dest->as_register(), src->as_register(), 8);
 912       break;
 913     case Bytecodes::_i2s:
 914       __ sign_extend(dest->as_register(), src->as_register(), 16);
 915       break;
 916     case Bytecodes::_i2c:
 917       __ zero_extend(dest->as_register(), src->as_register(), 16);
 918       break;
 919     case Bytecodes::_f2d:
 920       __ convert_f2d(dest->as_double_reg(), src->as_float_reg());
 921       break;
 922     case Bytecodes::_d2f:
 923       __ convert_d2f(dest->as_float_reg(), src->as_double_reg());
 924       break;
 925     case Bytecodes::_i2f:
 926       __ fmsr(Stemp, src->as_register());
 927       __ fsitos(dest->as_float_reg(), Stemp);
 928       break;
 929     case Bytecodes::_i2d:
 930       __ fmsr(Stemp, src->as_register());
 931       __ fsitod(dest->as_double_reg(), Stemp);
 932       break;
 933     case Bytecodes::_f2i:
 934       __ ftosizs(Stemp, src->as_float_reg());
 935       __ fmrs(dest->as_register(), Stemp);
 936       break;
 937     case Bytecodes::_d2i:
 938       __ ftosizd(Stemp, src->as_double_reg());
 939       __ fmrs(dest->as_register(), Stemp);
 940       break;
 941     default:
 942       ShouldNotReachHere();
 943   }
 944 }
 945 
 946 
 947 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
 948   if (op->init_check()) {
 949     Register tmp = op->tmp1()->as_register();
 950     __ ldrb(tmp, Address(op->klass()->as_register(), InstanceKlass::init_state_offset()));
 951     __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::LoadLoad | MacroAssembler::LoadStore), Rtemp);
 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::base_offset_in_bytes(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 != nullptr, "Sanity");
1029   data = md->bci_to_data(bci);
1030   assert(data != nullptr,       "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 != nullptr, "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(C1StubId::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(C1StubId::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(C1StubId::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(C1StubId::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(C1StubId::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_Opr::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     Register cmp_value_lo = op->cmp_value()->as_register_lo();
1393     Register cmp_value_hi = op->cmp_value()->as_register_hi();
1394     Register new_value_lo = op->new_value()->as_register_lo();
1395     Register new_value_hi = op->new_value()->as_register_hi();
1396     Register dest = op->result_opr()->as_register();
1397     Register tmp_lo = op->tmp1()->as_register_lo();
1398     Register tmp_hi = op->tmp1()->as_register_hi();
1399 
1400     assert_different_registers(tmp_lo, tmp_hi, cmp_value_lo, cmp_value_hi, dest, new_value_lo, new_value_hi, addr);
1401     assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
1402     assert(new_value_hi->encoding() == new_value_lo->encoding() + 1, "non aligned register pair");
1403     assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
1404     assert((new_value_lo->encoding() & 0x1) == 0, "misaligned register pair");
1405     __ atomic_cas64(tmp_lo, tmp_hi, dest, cmp_value_lo, cmp_value_hi,
1406                     new_value_lo, new_value_hi, addr, 0);
1407   } else {
1408     Unimplemented();
1409   }
1410   // FIXME: is full membar really needed instead of just membar_acquire?
1411   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
1412 }
1413 
1414 
1415 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1416                           LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1417   assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on arm");
1418 
1419   AsmCondition acond = al;
1420   AsmCondition ncond = nv;
1421   if (opr1 != opr2) {
1422     switch (condition) {
1423       case lir_cond_equal:        acond = eq; ncond = ne; break;
1424       case lir_cond_notEqual:     acond = ne; ncond = eq; break;
1425       case lir_cond_less:         acond = lt; ncond = ge; break;
1426       case lir_cond_lessEqual:    acond = le; ncond = gt; break;
1427       case lir_cond_greaterEqual: acond = ge; ncond = lt; break;
1428       case lir_cond_greater:      acond = gt; ncond = le; break;
1429       case lir_cond_aboveEqual:   acond = hs; ncond = lo; break;
1430       case lir_cond_belowEqual:   acond = ls; ncond = hi; break;
1431       default: ShouldNotReachHere();
1432     }
1433   }
1434 
1435   for (;;) {                         // two iterations only
1436     if (opr1 == result) {
1437       // do nothing
1438     } else if (opr1->is_single_cpu()) {
1439       __ mov(result->as_register(), opr1->as_register(), acond);
1440     } else if (opr1->is_double_cpu()) {
1441       __ long_move(result->as_register_lo(), result->as_register_hi(),
1442                    opr1->as_register_lo(), opr1->as_register_hi(), acond);
1443     } else if (opr1->is_single_stack()) {
1444       __ ldr(result->as_register(), frame_map()->address_for_slot(opr1->single_stack_ix()), acond);
1445     } else if (opr1->is_double_stack()) {
1446       __ ldr(result->as_register_lo(),
1447              frame_map()->address_for_slot(opr1->double_stack_ix(), lo_word_offset_in_bytes), acond);
1448       __ ldr(result->as_register_hi(),
1449              frame_map()->address_for_slot(opr1->double_stack_ix(), hi_word_offset_in_bytes), acond);
1450     } else if (opr1->is_illegal()) {
1451       // do nothing: this part of the cmove has been optimized away in the peephole optimizer
1452     } else {
1453       assert(opr1->is_constant(), "must be");
1454       LIR_Const* c = opr1->as_constant_ptr();
1455 
1456       switch (c->type()) {
1457         case T_INT:
1458           __ mov_slow(result->as_register(), c->as_jint(), acond);
1459           break;
1460         case T_LONG:
1461           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1462           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1463           break;
1464         case T_OBJECT:
1465           __ mov_oop(result->as_register(), c->as_jobject(), 0, acond);
1466           break;
1467         case T_FLOAT:
1468 #ifdef __SOFTFP__
1469           // not generated now.
1470           __ mov_slow(result->as_register(), c->as_jint(), acond);
1471 #else
1472           __ mov_float(result->as_float_reg(), c->as_jfloat(), acond);
1473 #endif // __SOFTFP__
1474           break;
1475         case T_DOUBLE:
1476 #ifdef __SOFTFP__
1477           // not generated now.
1478           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1479           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1480 #else
1481           __ mov_double(result->as_double_reg(), c->as_jdouble(), acond);
1482 #endif // __SOFTFP__
1483           break;
1484         case T_METADATA:
1485           __ mov_metadata(result->as_register(), c->as_metadata(), acond);
1486           break;
1487         default:
1488           ShouldNotReachHere();
1489       }
1490     }
1491 
1492     // Negate the condition and repeat the algorithm with the second operand
1493     if (opr1 == opr2) { break; }
1494     opr1 = opr2;
1495     acond = ncond;
1496   }
1497 }
1498 
1499 #ifdef ASSERT
1500 static int reg_size(LIR_Opr op) {
1501   switch (op->type()) {
1502   case T_FLOAT:
1503   case T_INT:      return BytesPerInt;
1504   case T_LONG:
1505   case T_DOUBLE:   return BytesPerLong;
1506   case T_OBJECT:
1507   case T_ARRAY:
1508   case T_METADATA: return BytesPerWord;
1509   case T_ADDRESS:
1510   case T_ILLEGAL:  // fall through
1511   default: ShouldNotReachHere(); return -1;
1512   }
1513 }
1514 #endif
1515 
1516 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1517   assert(info == nullptr, "unused on this code path");
1518   assert(dest->is_register(), "wrong items state");
1519 
1520   if (right->is_address()) {
1521     // special case for adding shifted/extended register
1522     const Register res = dest->as_pointer_register();
1523     const Register lreg = left->as_pointer_register();
1524     const LIR_Address* addr = right->as_address_ptr();
1525 
1526     assert(addr->base()->as_pointer_register() == lreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1527 
1528     int scale = addr->scale();
1529     AsmShift shift = lsl;
1530 
1531 
1532     assert(reg_size(addr->base()) == reg_size(addr->index()), "should be");
1533     assert(reg_size(addr->base()) == reg_size(dest), "should be");
1534     assert(reg_size(dest) == wordSize, "should be");
1535 
1536     AsmOperand operand(addr->index()->as_pointer_register(), shift, scale);
1537     switch (code) {
1538       case lir_add: __ add(res, lreg, operand); break;
1539       case lir_sub: __ sub(res, lreg, operand); break;
1540       default: ShouldNotReachHere();
1541     }
1542 
1543   } else if (left->is_address()) {
1544     assert(code == lir_sub && right->is_single_cpu(), "special case used by strength_reduce_multiply()");
1545     const LIR_Address* addr = left->as_address_ptr();
1546     const Register res = dest->as_register();
1547     const Register rreg = right->as_register();
1548     assert(addr->base()->as_register() == rreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1549     __ rsb(res, rreg, AsmOperand(addr->index()->as_register(), lsl, addr->scale()));
1550 
1551   } else if (dest->is_single_cpu()) {
1552     assert(left->is_single_cpu(), "unexpected left operand");
1553 
1554     const Register res = dest->as_register();
1555     const Register lreg = left->as_register();
1556 
1557     if (right->is_single_cpu()) {
1558       const Register rreg = right->as_register();
1559       switch (code) {
1560         case lir_add: __ add_32(res, lreg, rreg); break;
1561         case lir_sub: __ sub_32(res, lreg, rreg); break;
1562         case lir_mul: __ mul_32(res, lreg, rreg); break;
1563         default: ShouldNotReachHere();
1564       }
1565     } else {
1566       assert(right->is_constant(), "must be");
1567       const jint c = right->as_constant_ptr()->as_jint();
1568       if (!Assembler::is_arith_imm_in_range(c)) {
1569         BAILOUT("illegal arithmetic operand");
1570       }
1571       switch (code) {
1572         case lir_add: __ add_32(res, lreg, c); break;
1573         case lir_sub: __ sub_32(res, lreg, c); break;
1574         default: ShouldNotReachHere();
1575       }
1576     }
1577 
1578   } else if (dest->is_double_cpu()) {
1579     Register res_lo = dest->as_register_lo();
1580     Register res_hi = dest->as_register_hi();
1581     Register lreg_lo = left->as_register_lo();
1582     Register lreg_hi = left->as_register_hi();
1583     if (right->is_double_cpu()) {
1584       Register rreg_lo = right->as_register_lo();
1585       Register rreg_hi = right->as_register_hi();
1586       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1587         res_lo = Rtemp;
1588       }
1589       switch (code) {
1590         case lir_add:
1591           __ adds(res_lo, lreg_lo, rreg_lo);
1592           __ adc(res_hi, lreg_hi, rreg_hi);
1593           break;
1594         case lir_sub:
1595           __ subs(res_lo, lreg_lo, rreg_lo);
1596           __ sbc(res_hi, lreg_hi, rreg_hi);
1597           break;
1598         default:
1599           ShouldNotReachHere();
1600       }
1601     } else {
1602       assert(right->is_constant(), "must be");
1603       assert((right->as_constant_ptr()->as_jlong() >> 32) == 0, "out of range");
1604       const jint c = (jint) right->as_constant_ptr()->as_jlong();
1605       if (res_lo == lreg_hi) {
1606         res_lo = Rtemp;
1607       }
1608       switch (code) {
1609         case lir_add:
1610           __ adds(res_lo, lreg_lo, c);
1611           __ adc(res_hi, lreg_hi, 0);
1612           break;
1613         case lir_sub:
1614           __ subs(res_lo, lreg_lo, c);
1615           __ sbc(res_hi, lreg_hi, 0);
1616           break;
1617         default:
1618           ShouldNotReachHere();
1619       }
1620     }
1621     move_regs(res_lo, dest->as_register_lo());
1622 
1623   } else if (dest->is_single_fpu()) {
1624     assert(left->is_single_fpu(), "must be");
1625     assert(right->is_single_fpu(), "must be");
1626     const FloatRegister res = dest->as_float_reg();
1627     const FloatRegister lreg = left->as_float_reg();
1628     const FloatRegister rreg = right->as_float_reg();
1629     switch (code) {
1630       case lir_add: __ add_float(res, lreg, rreg); break;
1631       case lir_sub: __ sub_float(res, lreg, rreg); break;
1632       case lir_mul: __ mul_float(res, lreg, rreg); break;
1633       case lir_div: __ div_float(res, lreg, rreg); break;
1634       default: ShouldNotReachHere();
1635     }
1636   } else if (dest->is_double_fpu()) {
1637     assert(left->is_double_fpu(), "must be");
1638     assert(right->is_double_fpu(), "must be");
1639     const FloatRegister res = dest->as_double_reg();
1640     const FloatRegister lreg = left->as_double_reg();
1641     const FloatRegister rreg = right->as_double_reg();
1642     switch (code) {
1643       case lir_add: __ add_double(res, lreg, rreg); break;
1644       case lir_sub: __ sub_double(res, lreg, rreg); break;
1645       case lir_mul: __ mul_double(res, lreg, rreg); break;
1646       case lir_div: __ div_double(res, lreg, rreg); break;
1647       default: ShouldNotReachHere();
1648     }
1649   } else {
1650     ShouldNotReachHere();
1651   }
1652 }
1653 
1654 
1655 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
1656   switch (code) {
1657     case lir_abs:
1658       __ abs_double(dest->as_double_reg(), value->as_double_reg());
1659       break;
1660     case lir_sqrt:
1661       __ sqrt_double(dest->as_double_reg(), value->as_double_reg());
1662       break;
1663     default:
1664       ShouldNotReachHere();
1665   }
1666 }
1667 
1668 
1669 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1670   assert(dest->is_register(), "wrong items state");
1671   assert(left->is_register(), "wrong items state");
1672 
1673   if (dest->is_single_cpu()) {
1674 
1675     const Register res = dest->as_register();
1676     const Register lreg = left->as_register();
1677 
1678     if (right->is_single_cpu()) {
1679       const Register rreg = right->as_register();
1680       switch (code) {
1681         case lir_logic_and: __ and_32(res, lreg, rreg); break;
1682         case lir_logic_or:  __ orr_32(res, lreg, rreg); break;
1683         case lir_logic_xor: __ eor_32(res, lreg, rreg); break;
1684         default: ShouldNotReachHere();
1685       }
1686     } else {
1687       assert(right->is_constant(), "must be");
1688       const uint c = (uint)right->as_constant_ptr()->as_jint();
1689       if (!Assembler::is_arith_imm_in_range(c)) {
1690         BAILOUT("illegal arithmetic operand");
1691       }
1692       switch (code) {
1693         case lir_logic_and: __ and_32(res, lreg, c); break;
1694         case lir_logic_or:  __ orr_32(res, lreg, c); break;
1695         case lir_logic_xor: __ eor_32(res, lreg, c); break;
1696         default: ShouldNotReachHere();
1697       }
1698     }
1699   } else {
1700     assert(dest->is_double_cpu(), "should be");
1701     Register res_lo = dest->as_register_lo();
1702 
1703     assert (dest->type() == T_LONG, "unexpected result type");
1704     assert (left->type() == T_LONG, "unexpected left type");
1705     assert (right->type() == T_LONG, "unexpected right type");
1706 
1707     const Register res_hi = dest->as_register_hi();
1708     const Register lreg_lo = left->as_register_lo();
1709     const Register lreg_hi = left->as_register_hi();
1710 
1711     if (right->is_register()) {
1712       const Register rreg_lo = right->as_register_lo();
1713       const Register rreg_hi = right->as_register_hi();
1714       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1715         res_lo = Rtemp; // Temp register helps to avoid overlap between result and input
1716       }
1717       switch (code) {
1718         case lir_logic_and:
1719           __ andr(res_lo, lreg_lo, rreg_lo);
1720           __ andr(res_hi, lreg_hi, rreg_hi);
1721           break;
1722         case lir_logic_or:
1723           __ orr(res_lo, lreg_lo, rreg_lo);
1724           __ orr(res_hi, lreg_hi, rreg_hi);
1725           break;
1726         case lir_logic_xor:
1727           __ eor(res_lo, lreg_lo, rreg_lo);
1728           __ eor(res_hi, lreg_hi, rreg_hi);
1729           break;
1730         default:
1731           ShouldNotReachHere();
1732       }
1733       move_regs(res_lo, dest->as_register_lo());
1734     } else {
1735       assert(right->is_constant(), "must be");
1736       const jint c_lo = (jint) right->as_constant_ptr()->as_jlong();
1737       const jint c_hi = (jint) (right->as_constant_ptr()->as_jlong() >> 32);
1738       // Case for logic_or from do_ClassIDIntrinsic()
1739       if (c_hi == 0 && AsmOperand::is_rotated_imm(c_lo)) {
1740         switch (code) {
1741           case lir_logic_and:
1742             __ andr(res_lo, lreg_lo, c_lo);
1743             __ mov(res_hi, 0);
1744             break;
1745           case lir_logic_or:
1746             __ orr(res_lo, lreg_lo, c_lo);
1747             break;
1748           case lir_logic_xor:
1749             __ eor(res_lo, lreg_lo, c_lo);
1750             break;
1751         default:
1752           ShouldNotReachHere();
1753         }
1754       } else if (code == lir_logic_and &&
1755                  c_hi == -1 &&
1756                  (AsmOperand::is_rotated_imm(c_lo) ||
1757                   AsmOperand::is_rotated_imm(~c_lo))) {
1758         // Another case which handles logic_and from do_ClassIDIntrinsic()
1759         if (AsmOperand::is_rotated_imm(c_lo)) {
1760           __ andr(res_lo, lreg_lo, c_lo);
1761         } else {
1762           __ bic(res_lo, lreg_lo, ~c_lo);
1763         }
1764         if (res_hi != lreg_hi) {
1765           __ mov(res_hi, lreg_hi);
1766         }
1767       } else {
1768         BAILOUT("64 bit constant cannot be inlined");
1769       }
1770     }
1771   }
1772 }
1773 
1774 
1775 
1776 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1777   if (opr1->is_single_cpu()) {
1778     if (opr2->is_constant()) {
1779       switch (opr2->as_constant_ptr()->type()) {
1780         case T_INT: {
1781           const jint c = opr2->as_constant_ptr()->as_jint();
1782           if (Assembler::is_arith_imm_in_range(c)) {
1783             __ cmp_32(opr1->as_register(), c);
1784           } else if (Assembler::is_arith_imm_in_range(-c)) {
1785             __ cmn_32(opr1->as_register(), -c);
1786           } else {
1787             // This can happen when compiling lookupswitch
1788             __ mov_slow(Rtemp, c);
1789             __ cmp_32(opr1->as_register(), Rtemp);
1790           }
1791           break;
1792         }
1793         case T_OBJECT:
1794           assert(opr2->as_constant_ptr()->as_jobject() == nullptr, "cannot handle otherwise");
1795           __ cmp(opr1->as_register(), 0);
1796           break;
1797         case T_METADATA:
1798           assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "Only equality tests");
1799           assert(opr2->as_constant_ptr()->as_metadata() == nullptr, "cannot handle otherwise");
1800           __ cmp(opr1->as_register(), 0);
1801           break;
1802         default:
1803           ShouldNotReachHere();
1804       }
1805     } else if (opr2->is_single_cpu()) {
1806       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1807         assert(opr2->type() == T_OBJECT || opr2->type() == T_ARRAY, "incompatibe type");
1808         __ cmpoop(opr1->as_register(), opr2->as_register());
1809       } else if (opr1->type() == T_METADATA || opr1->type() == T_ADDRESS) {
1810         assert(opr2->type() == T_METADATA || opr2->type() == T_ADDRESS, "incompatibe type");
1811         __ cmp(opr1->as_register(), opr2->as_register());
1812       } else {
1813         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY && opr2->type() != T_METADATA && opr2->type() != T_ADDRESS, "incompatibe type");
1814         __ cmp_32(opr1->as_register(), opr2->as_register());
1815       }
1816     } else {
1817       ShouldNotReachHere();
1818     }
1819   } else if (opr1->is_double_cpu()) {
1820     Register xlo = opr1->as_register_lo();
1821     Register xhi = opr1->as_register_hi();
1822     if (opr2->is_constant() && opr2->as_jlong() == 0) {
1823       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "cannot handle otherwise");
1824       __ orrs(Rtemp, xlo, xhi);
1825     } else if (opr2->is_register()) {
1826       Register ylo = opr2->as_register_lo();
1827       Register yhi = opr2->as_register_hi();
1828       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1829         __ teq(xhi, yhi);
1830         __ teq(xlo, ylo, eq);
1831       } else {
1832         __ subs(Rtemp, xlo, ylo);
1833         __ sbcs(Rtemp, xhi, yhi);
1834       }
1835     } else {
1836       ShouldNotReachHere();
1837     }
1838   } else if (opr1->is_single_fpu()) {
1839     if (opr2->is_constant()) {
1840       assert(opr2->as_jfloat() == 0.0f, "cannot handle otherwise");
1841       __ cmp_zero_float(opr1->as_float_reg());
1842     } else {
1843       __ cmp_float(opr1->as_float_reg(), opr2->as_float_reg());
1844     }
1845   } else if (opr1->is_double_fpu()) {
1846     if (opr2->is_constant()) {
1847       assert(opr2->as_jdouble() == 0.0, "cannot handle otherwise");
1848       __ cmp_zero_double(opr1->as_double_reg());
1849     } else {
1850       __ cmp_double(opr1->as_double_reg(), opr2->as_double_reg());
1851     }
1852   } else {
1853     ShouldNotReachHere();
1854   }
1855 }
1856 
1857 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1858   const Register res = dst->as_register();
1859   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1860     comp_op(lir_cond_unknown, left, right, op);
1861     __ fmstat();
1862     if (code == lir_ucmp_fd2i) {  // unordered is less
1863       __ mvn(res, 0, lt);
1864       __ mov(res, 1, ge);
1865     } else {                      // unordered is greater
1866       __ mov(res, 1, cs);
1867       __ mvn(res, 0, cc);
1868     }
1869     __ mov(res, 0, eq);
1870 
1871   } else {
1872     assert(code == lir_cmp_l2i, "must be");
1873 
1874     Label done;
1875     const Register xlo = left->as_register_lo();
1876     const Register xhi = left->as_register_hi();
1877     const Register ylo = right->as_register_lo();
1878     const Register yhi = right->as_register_hi();
1879     __ cmp(xhi, yhi);
1880     __ mov(res, 1, gt);
1881     __ mvn(res, 0, lt);
1882     __ b(done, ne);
1883     __ subs(res, xlo, ylo);
1884     __ mov(res, 1, hi);
1885     __ mvn(res, 0, lo);
1886     __ bind(done);
1887   }
1888 }
1889 
1890 
1891 void LIR_Assembler::align_call(LIR_Code code) {
1892   // Not needed
1893 }
1894 
1895 
1896 void LIR_Assembler::call(LIR_OpJavaCall *op, relocInfo::relocType rtype) {
1897   int ret_addr_offset = __ patchable_call(op->addr(), rtype);
1898   assert(ret_addr_offset == __ offset(), "embedded return address not allowed");
1899   add_call_info_here(op->info());
1900 }
1901 
1902 
1903 void LIR_Assembler::ic_call(LIR_OpJavaCall *op) {
1904   bool near_range = __ cache_fully_reachable();
1905   address oop_address = pc();
1906 
1907   bool use_movw = VM_Version::supports_movw();
1908 
1909   // Ricklass may contain something that is not a metadata pointer so
1910   // mov_metadata can't be used
1911   InlinedAddress value((address)Universe::non_oop_word());
1912   InlinedAddress addr(op->addr());
1913   if (use_movw) {
1914     __ movw(Ricklass, ((unsigned int)Universe::non_oop_word()) & 0xffff);
1915     __ movt(Ricklass, ((unsigned int)Universe::non_oop_word()) >> 16);
1916   } else {
1917     // No movw/movt, must be load a pc relative value but no
1918     // relocation so no metadata table to load from.
1919     // Use a b instruction rather than a bl, inline constant after the
1920     // branch, use a PC relative ldr to load the constant, arrange for
1921     // the call to return after the constant(s).
1922     __ ldr_literal(Ricklass, value);
1923   }
1924   __ relocate(virtual_call_Relocation::spec(oop_address));
1925   if (near_range && use_movw) {
1926     __ bl(op->addr());
1927   } else {
1928     Label call_return;
1929     __ adr(LR, call_return);
1930     if (near_range) {
1931       __ b(op->addr());
1932     } else {
1933       __ indirect_jump(addr, Rtemp);
1934       __ bind_literal(addr);
1935     }
1936     if (!use_movw) {
1937       __ bind_literal(value);
1938     }
1939     __ bind(call_return);
1940   }
1941   add_call_info(code_offset(), op->info());
1942 }
1943 
1944 void LIR_Assembler::emit_static_call_stub() {
1945   address call_pc = __ pc();
1946   address stub = __ start_a_stub(call_stub_size());
1947   if (stub == nullptr) {
1948     BAILOUT("static call stub overflow");
1949   }
1950 
1951   DEBUG_ONLY(int offset = code_offset();)
1952 
1953   InlinedMetadata metadata_literal(nullptr);
1954   __ relocate(static_stub_Relocation::spec(call_pc));
1955   // If not a single instruction, NativeMovConstReg::next_instruction_address()
1956   // must jump over the whole following ldr_literal.
1957   // (See CompiledDirectCall::set_to_interpreted())
1958 #ifdef ASSERT
1959   address ldr_site = __ pc();
1960 #endif
1961   __ ldr_literal(Rmethod, metadata_literal);
1962   assert(nativeMovConstReg_at(ldr_site)->next_instruction_address() == __ pc(), "Fix ldr_literal or its parsing");
1963   bool near_range = __ cache_fully_reachable();
1964   InlinedAddress dest((address)-1);
1965   if (near_range) {
1966     address branch_site = __ pc();
1967     __ b(branch_site); // b to self maps to special NativeJump -1 destination
1968   } else {
1969     __ indirect_jump(dest, Rtemp);
1970   }
1971   __ bind_literal(metadata_literal); // includes spec_for_immediate reloc
1972   if (!near_range) {
1973     __ bind_literal(dest); // special NativeJump -1 destination
1974   }
1975 
1976   assert(code_offset() - offset <= call_stub_size(), "overflow");
1977   __ end_a_stub();
1978 }
1979 
1980 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1981   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1982   assert(exceptionPC->as_register()  == Rexception_pc, "must match");
1983   info->add_register_oop(exceptionOop);
1984 
1985   C1StubId handle_id = compilation()->has_fpu_code() ?
1986                                C1StubId::handle_exception_id :
1987                                C1StubId::handle_exception_nofpu_id;
1988   Label return_address;
1989   __ adr(Rexception_pc, return_address);
1990   __ call(Runtime1::entry_for(handle_id), relocInfo::runtime_call_type);
1991   __ bind(return_address);
1992   add_call_info_here(info);  // for exception handler
1993 }
1994 
1995 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1996   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1997   __ b(_unwind_handler_entry);
1998 }
1999 
2000 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2001   AsmShift shift = lsl;
2002   switch (code) {
2003     case lir_shl:  shift = lsl; break;
2004     case lir_shr:  shift = asr; break;
2005     case lir_ushr: shift = lsr; break;
2006     default: ShouldNotReachHere();
2007   }
2008 
2009   if (dest->is_single_cpu()) {
2010     __ andr(Rtemp, count->as_register(), 31);
2011     __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, Rtemp));
2012   } else if (dest->is_double_cpu()) {
2013     Register dest_lo = dest->as_register_lo();
2014     Register dest_hi = dest->as_register_hi();
2015     Register src_lo  = left->as_register_lo();
2016     Register src_hi  = left->as_register_hi();
2017     Register Rcount  = count->as_register();
2018     // Resolve possible register conflicts
2019     if (shift == lsl && dest_hi == src_lo) {
2020       dest_hi = Rtemp;
2021     } else if (shift != lsl && dest_lo == src_hi) {
2022       dest_lo = Rtemp;
2023     } else if (dest_lo == src_lo && dest_hi == src_hi) {
2024       dest_lo = Rtemp;
2025     } else if (dest_lo == Rcount || dest_hi == Rcount) {
2026       Rcount = Rtemp;
2027     }
2028     __ andr(Rcount, count->as_register(), 63);
2029     __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, Rcount);
2030     move_regs(dest_lo, dest->as_register_lo());
2031     move_regs(dest_hi, dest->as_register_hi());
2032   } else {
2033     ShouldNotReachHere();
2034   }
2035 }
2036 
2037 
2038 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2039   AsmShift shift = lsl;
2040   switch (code) {
2041     case lir_shl:  shift = lsl; break;
2042     case lir_shr:  shift = asr; break;
2043     case lir_ushr: shift = lsr; break;
2044     default: ShouldNotReachHere();
2045   }
2046 
2047   if (dest->is_single_cpu()) {
2048     count &= 31;
2049     if (count != 0) {
2050       __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, count));
2051     } else {
2052       move_regs(left->as_register(), dest->as_register());
2053     }
2054   } else if (dest->is_double_cpu()) {
2055     count &= 63;
2056     if (count != 0) {
2057       Register dest_lo = dest->as_register_lo();
2058       Register dest_hi = dest->as_register_hi();
2059       Register src_lo  = left->as_register_lo();
2060       Register src_hi  = left->as_register_hi();
2061       // Resolve possible register conflicts
2062       if (shift == lsl && dest_hi == src_lo) {
2063         dest_hi = Rtemp;
2064       } else if (shift != lsl && dest_lo == src_hi) {
2065         dest_lo = Rtemp;
2066       }
2067       __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, count);
2068       move_regs(dest_lo, dest->as_register_lo());
2069       move_regs(dest_hi, dest->as_register_hi());
2070     } else {
2071       __ long_move(dest->as_register_lo(), dest->as_register_hi(),
2072                    left->as_register_lo(), left->as_register_hi());
2073     }
2074   } else {
2075     ShouldNotReachHere();
2076   }
2077 }
2078 
2079 
2080 // Saves 4 given registers in reserved argument area.
2081 void LIR_Assembler::save_in_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2082   verify_reserved_argument_area_size(4);
2083   __ stmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4));
2084 }
2085 
2086 // Restores 4 given registers from reserved argument area.
2087 void LIR_Assembler::restore_from_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2088   __ ldmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4), no_writeback);
2089 }
2090 
2091 
2092 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2093   ciArrayKlass* default_type = op->expected_type();
2094   Register src = op->src()->as_register();
2095   Register src_pos = op->src_pos()->as_register();
2096   Register dst = op->dst()->as_register();
2097   Register dst_pos = op->dst_pos()->as_register();
2098   Register length  = op->length()->as_register();
2099   Register tmp = op->tmp()->as_register();
2100   Register tmp2 = Rtemp;
2101 
2102   assert(src == R0 && src_pos == R1 && dst == R2 && dst_pos == R3, "code assumption");
2103 
2104   CodeStub* stub = op->stub();
2105 
2106   int flags = op->flags();
2107   BasicType basic_type = default_type != nullptr ? default_type->element_type()->basic_type() : T_ILLEGAL;
2108   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2109 
2110   // If we don't know anything or it's an object array, just go through the generic arraycopy
2111   if (default_type == nullptr) {
2112 
2113     // save arguments, because they will be killed by a runtime call
2114     save_in_reserved_area(R0, R1, R2, R3);
2115 
2116     // pass length argument on SP[0]
2117     __ str(length, Address(SP, -2*wordSize, pre_indexed));  // 2 words for a proper stack alignment
2118 
2119     address copyfunc_addr = StubRoutines::generic_arraycopy();
2120     assert(copyfunc_addr != nullptr, "generic arraycopy stub required");
2121 #ifndef PRODUCT
2122     if (PrintC1Statistics) {
2123       __ inc_counter((address)&Runtime1::_generic_arraycopystub_cnt, tmp, tmp2);
2124     }
2125 #endif // !PRODUCT
2126     // the stub is in the code cache so close enough
2127     __ call(copyfunc_addr, relocInfo::runtime_call_type);
2128 
2129     __ add(SP, SP, 2*wordSize);
2130 
2131     __ cbz_32(R0, *stub->continuation());
2132 
2133     __ mvn_32(tmp, R0);
2134     restore_from_reserved_area(R0, R1, R2, R3);  // load saved arguments in slow case only
2135     __ sub_32(length, length, tmp);
2136     __ add_32(src_pos, src_pos, tmp);
2137     __ add_32(dst_pos, dst_pos, tmp);
2138 
2139     __ b(*stub->entry());
2140 
2141     __ bind(*stub->continuation());
2142     return;
2143   }
2144 
2145   assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(),
2146          "must be true at this point");
2147   int elem_size = type2aelembytes(basic_type);
2148   int shift = exact_log2(elem_size);
2149 
2150   // Check for null
2151   if (flags & LIR_OpArrayCopy::src_null_check) {
2152     if (flags & LIR_OpArrayCopy::dst_null_check) {
2153       __ cmp(src, 0);
2154       __ cond_cmp(dst, 0, ne);  // make one instruction shorter if both checks are needed
2155       __ b(*stub->entry(), eq);
2156     } else {
2157       __ cbz(src, *stub->entry());
2158     }
2159   } else if (flags & LIR_OpArrayCopy::dst_null_check) {
2160     __ cbz(dst, *stub->entry());
2161   }
2162 
2163   // If the compiler was not able to prove that exact type of the source or the destination
2164   // of the arraycopy is an array type, check at runtime if the source or the destination is
2165   // an instance type.
2166   if (flags & LIR_OpArrayCopy::type_check) {
2167     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::dst_objarray)) {
2168       __ load_klass(tmp, dst);
2169       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2170       __ mov_slow(tmp, Klass::_lh_neutral_value);
2171       __ cmp_32(tmp2, tmp);
2172       __ b(*stub->entry(), ge);
2173     }
2174 
2175     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::src_objarray)) {
2176       __ load_klass(tmp, src);
2177       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2178       __ mov_slow(tmp, Klass::_lh_neutral_value);
2179       __ cmp_32(tmp2, tmp);
2180       __ b(*stub->entry(), ge);
2181     }
2182   }
2183 
2184   // Check if negative
2185   const int all_positive_checks = LIR_OpArrayCopy::src_pos_positive_check |
2186                                   LIR_OpArrayCopy::dst_pos_positive_check |
2187                                   LIR_OpArrayCopy::length_positive_check;
2188   switch (flags & all_positive_checks) {
2189     case LIR_OpArrayCopy::src_pos_positive_check:
2190       __ branch_if_negative_32(src_pos, *stub->entry());
2191       break;
2192     case LIR_OpArrayCopy::dst_pos_positive_check:
2193       __ branch_if_negative_32(dst_pos, *stub->entry());
2194       break;
2195     case LIR_OpArrayCopy::length_positive_check:
2196       __ branch_if_negative_32(length, *stub->entry());
2197       break;
2198     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::dst_pos_positive_check:
2199       __ branch_if_any_negative_32(src_pos, dst_pos, tmp, *stub->entry());
2200       break;
2201     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2202       __ branch_if_any_negative_32(src_pos, length, tmp, *stub->entry());
2203       break;
2204     case LIR_OpArrayCopy::dst_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2205       __ branch_if_any_negative_32(dst_pos, length, tmp, *stub->entry());
2206       break;
2207     case all_positive_checks:
2208       __ branch_if_any_negative_32(src_pos, dst_pos, length, tmp, *stub->entry());
2209       break;
2210     default:
2211       assert((flags & all_positive_checks) == 0, "the last option");
2212   }
2213 
2214   // Range checks
2215   if (flags & LIR_OpArrayCopy::src_range_check) {
2216     __ ldr_s32(tmp2, Address(src, arrayOopDesc::length_offset_in_bytes()));
2217     __ add_32(tmp, src_pos, length);
2218     __ cmp_32(tmp, tmp2);
2219     __ b(*stub->entry(), hi);
2220   }
2221   if (flags & LIR_OpArrayCopy::dst_range_check) {
2222     __ ldr_s32(tmp2, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2223     __ add_32(tmp, dst_pos, length);
2224     __ cmp_32(tmp, tmp2);
2225     __ b(*stub->entry(), hi);
2226   }
2227 
2228   // Check if src and dst are of the same type
2229   if (flags & LIR_OpArrayCopy::type_check) {
2230     // We don't know the array types are compatible
2231     if (basic_type != T_OBJECT) {
2232       // Simple test for basic type arrays
2233       if (UseCompressedClassPointers) {
2234         // We don't need decode because we just need to compare
2235         __ ldr_u32(tmp, Address(src, oopDesc::klass_offset_in_bytes()));
2236         __ ldr_u32(tmp2, Address(dst, oopDesc::klass_offset_in_bytes()));
2237         __ cmp_32(tmp, tmp2);
2238       } else {
2239         __ load_klass(tmp, src);
2240         __ load_klass(tmp2, dst);
2241         __ cmp(tmp, tmp2);
2242       }
2243       __ b(*stub->entry(), ne);
2244     } else {
2245       // For object arrays, if src is a sub class of dst then we can
2246       // safely do the copy.
2247       Label cont, slow;
2248 
2249       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2250 
2251       __ load_klass(tmp, src);
2252       __ load_klass(tmp2, dst);
2253 
2254       // We are at a call so all live registers are saved before we
2255       // get here
2256       assert_different_registers(tmp, tmp2, R6, altFP_7_11);
2257 
2258       __ check_klass_subtype_fast_path(tmp, tmp2, R6, altFP_7_11, &cont, copyfunc_addr == nullptr ? stub->entry() : &slow, nullptr);
2259 
2260       __ mov(R6, R0);
2261       __ mov(altFP_7_11, R1);
2262       __ mov(R0, tmp);
2263       __ mov(R1, tmp2);
2264       __ call(Runtime1::entry_for(C1StubId::slow_subtype_check_id), relocInfo::runtime_call_type); // does not blow any registers except R0, LR and Rtemp
2265       __ cmp_32(R0, 0);
2266       __ mov(R0, R6);
2267       __ mov(R1, altFP_7_11);
2268 
2269       if (copyfunc_addr != nullptr) { // use stub if available
2270         // src is not a sub class of dst so we have to do a
2271         // per-element check.
2272 
2273         __ b(cont, ne);
2274 
2275         __ bind(slow);
2276 
2277         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2278         if ((flags & mask) != mask) {
2279           // Check that at least both of them object arrays.
2280           assert(flags & mask, "one of the two should be known to be an object array");
2281 
2282           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2283             __ load_klass(tmp, src);
2284           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2285             __ load_klass(tmp, dst);
2286           }
2287           int lh_offset = in_bytes(Klass::layout_helper_offset());
2288 
2289           __ ldr_u32(tmp2, Address(tmp, lh_offset));
2290 
2291           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2292           __ mov_slow(tmp, objArray_lh);
2293           __ cmp_32(tmp, tmp2);
2294           __ b(*stub->entry(), ne);
2295         }
2296 
2297         save_in_reserved_area(R0, R1, R2, R3);
2298 
2299         Register src_ptr = R0;
2300         Register dst_ptr = R1;
2301         Register len     = R2;
2302         Register chk_off = R3;
2303         Register super_k = tmp;
2304 
2305         __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2306         __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2307 
2308         __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2309         __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2310         __ load_klass(tmp, dst);
2311 
2312         int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2313         int sco_offset = in_bytes(Klass::super_check_offset_offset());
2314 
2315         __ ldr(super_k, Address(tmp, ek_offset));
2316 
2317         __ mov(len, length);
2318         __ ldr_u32(chk_off, Address(super_k, sco_offset));
2319         __ push(super_k);
2320 
2321         __ call(copyfunc_addr, relocInfo::runtime_call_type);
2322 
2323 #ifndef PRODUCT
2324         if (PrintC1Statistics) {
2325           Label failed;
2326           __ cbnz_32(R0, failed);
2327           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_cnt, tmp, tmp2);
2328           __ bind(failed);
2329         }
2330 #endif // PRODUCT
2331 
2332         __ add(SP, SP, wordSize);  // Drop super_k argument
2333 
2334         __ cbz_32(R0, *stub->continuation());
2335         __ mvn_32(tmp, R0);
2336 
2337         // load saved arguments in slow case only
2338         restore_from_reserved_area(R0, R1, R2, R3);
2339 
2340         __ sub_32(length, length, tmp);
2341         __ add_32(src_pos, src_pos, tmp);
2342         __ add_32(dst_pos, dst_pos, tmp);
2343 
2344 #ifndef PRODUCT
2345         if (PrintC1Statistics) {
2346           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_attempt_cnt, tmp, tmp2);
2347         }
2348 #endif
2349 
2350         __ b(*stub->entry());
2351 
2352         __ bind(cont);
2353       } else {
2354         __ b(*stub->entry(), eq);
2355         __ bind(cont);
2356       }
2357     }
2358   }
2359 
2360 #ifndef PRODUCT
2361   if (PrintC1Statistics) {
2362     address counter = Runtime1::arraycopy_count_address(basic_type);
2363     __ inc_counter(counter, tmp, tmp2);
2364   }
2365 #endif // !PRODUCT
2366 
2367   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2368   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2369   const char *name;
2370   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2371 
2372   Register src_ptr = R0;
2373   Register dst_ptr = R1;
2374   Register len     = R2;
2375 
2376   __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2377   __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2378 
2379   __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2380   __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2381 
2382   __ mov(len, length);
2383 
2384   __ call(entry, relocInfo::runtime_call_type);
2385 
2386   __ bind(*stub->continuation());
2387 }
2388 
2389 #ifdef ASSERT
2390  // emit run-time assertion
2391 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2392   assert(op->code() == lir_assert, "must be");
2393 
2394   if (op->in_opr1()->is_valid()) {
2395     assert(op->in_opr2()->is_valid(), "both operands must be valid");
2396     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
2397   } else {
2398     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
2399     assert(op->condition() == lir_cond_always, "no other conditions allowed");
2400   }
2401 
2402   Label ok;
2403   if (op->condition() != lir_cond_always) {
2404     AsmCondition acond = al;
2405     switch (op->condition()) {
2406       case lir_cond_equal:        acond = eq; break;
2407       case lir_cond_notEqual:     acond = ne; break;
2408       case lir_cond_less:         acond = lt; break;
2409       case lir_cond_lessEqual:    acond = le; break;
2410       case lir_cond_greaterEqual: acond = ge; break;
2411       case lir_cond_greater:      acond = gt; break;
2412       case lir_cond_aboveEqual:   acond = hs; break;
2413       case lir_cond_belowEqual:   acond = ls; break;
2414       default:                    ShouldNotReachHere();
2415     }
2416     __ b(ok, acond);
2417   }
2418   if (op->halt()) {
2419     const char* str = __ code_string(op->msg());
2420     __ stop(str);
2421   } else {
2422     breakpoint();
2423   }
2424   __ bind(ok);
2425 }
2426 #endif // ASSERT
2427 
2428 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
2429   fatal("CRC32 intrinsic is not implemented on this platform");
2430 }
2431 
2432 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2433   Register obj = op->obj_opr()->as_pointer_register();
2434   Register hdr = op->hdr_opr()->as_pointer_register();
2435   Register lock = op->lock_opr()->as_pointer_register();
2436 
2437   if (LockingMode == LM_MONITOR) {
2438     if (op->info() != nullptr) {
2439       add_debug_info_for_null_check_here(op->info());
2440       __ null_check(obj);
2441     }
2442     __ b(*op->stub()->entry());
2443   } else if (op->code() == lir_lock) {
2444     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2445     int null_check_offset = __ lock_object(hdr, obj, lock, *op->stub()->entry());
2446     if (op->info() != nullptr) {
2447       add_debug_info_for_null_check(null_check_offset, op->info());
2448     }
2449   } else if (op->code() == lir_unlock) {
2450     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2451   } else {
2452     ShouldNotReachHere();
2453   }
2454   __ bind(*op->stub()->continuation());
2455 }
2456 
2457 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2458   Register obj = op->obj()->as_pointer_register();
2459   Register result = op->result_opr()->as_pointer_register();
2460 
2461   CodeEmitInfo* info = op->info();
2462   if (info != nullptr) {
2463     add_debug_info_for_null_check_here(info);
2464   }
2465 
2466   if (UseCompressedClassPointers) { // On 32 bit arm??
2467     __ ldr_u32(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2468   } else {
2469     __ ldr(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2470   }
2471 }
2472 
2473 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2474   ciMethod* method = op->profiled_method();
2475   int bci          = op->profiled_bci();
2476   ciMethod* callee = op->profiled_callee();
2477 
2478   // Update counter for all call types
2479   ciMethodData* md = method->method_data_or_null();
2480   assert(md != nullptr, "Sanity");
2481   ciProfileData* data = md->bci_to_data(bci);
2482   assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
2483   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2484   Register mdo  = op->mdo()->as_register();
2485   assert(op->tmp1()->is_register(), "tmp1 must be allocated");
2486   Register tmp1 = op->tmp1()->as_pointer_register();
2487   assert_different_registers(mdo, tmp1);
2488   __ mov_metadata(mdo, md->constant_encoding());
2489   int mdo_offset_bias = 0;
2490   int max_offset = 4096;
2491   if (md->byte_offset_of_slot(data, CounterData::count_offset()) + data->size_in_bytes() >= max_offset) {
2492     // The offset is large so bias the mdo by the base of the slot so
2493     // that the ldr can use an immediate offset to reference the slots of the data
2494     mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2495     __ mov_slow(tmp1, mdo_offset_bias);
2496     __ add(mdo, mdo, tmp1);
2497   }
2498 
2499   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2500   // Perform additional virtual call profiling for invokevirtual and
2501   // invokeinterface bytecodes
2502   if (op->should_profile_receiver_type()) {
2503     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2504     Register recv = op->recv()->as_register();
2505     assert_different_registers(mdo, tmp1, recv);
2506     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2507     ciKlass* known_klass = op->known_holder();
2508     if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
2509       // We know the type that will be seen at this call site; we can
2510       // statically update the MethodData* rather than needing to do
2511       // dynamic tests on the receiver type
2512 
2513       // NOTE: we should probably put a lock around this search to
2514       // avoid collisions by concurrent compilations
2515       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2516       uint i;
2517       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2518         ciKlass* receiver = vc_data->receiver(i);
2519         if (known_klass->equals(receiver)) {
2520           Address data_addr(mdo, md->byte_offset_of_slot(data,
2521                                                          VirtualCallData::receiver_count_offset(i)) -
2522                             mdo_offset_bias);
2523           __ ldr(tmp1, data_addr);
2524           __ add(tmp1, tmp1, DataLayout::counter_increment);
2525           __ str(tmp1, data_addr);
2526           return;
2527         }
2528       }
2529 
2530       // Receiver type not found in profile data; select an empty slot
2531 
2532       // Note that this is less efficient than it should be because it
2533       // always does a write to the receiver part of the
2534       // VirtualCallData rather than just the first time
2535       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2536         ciKlass* receiver = vc_data->receiver(i);
2537         if (receiver == nullptr) {
2538           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
2539                             mdo_offset_bias);
2540           __ mov_metadata(tmp1, known_klass->constant_encoding());
2541           __ str(tmp1, recv_addr);
2542           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
2543                             mdo_offset_bias);
2544           __ ldr(tmp1, data_addr);
2545           __ add(tmp1, tmp1, DataLayout::counter_increment);
2546           __ str(tmp1, data_addr);
2547           return;
2548         }
2549       }
2550     } else {
2551       __ load_klass(recv, recv);
2552       Label update_done;
2553       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
2554       // Receiver did not match any saved receiver and there is no empty row for it.
2555       // Increment total counter to indicate polymorphic case.
2556       __ ldr(tmp1, counter_addr);
2557       __ add(tmp1, tmp1, DataLayout::counter_increment);
2558       __ str(tmp1, counter_addr);
2559 
2560       __ bind(update_done);
2561     }
2562   } else {
2563     // Static call
2564     __ ldr(tmp1, counter_addr);
2565     __ add(tmp1, tmp1, DataLayout::counter_increment);
2566     __ str(tmp1, counter_addr);
2567   }
2568 }
2569 
2570 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2571   fatal("Type profiling not implemented on this platform");
2572 }
2573 
2574 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
2575   Unimplemented();
2576 }
2577 
2578 
2579 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
2580   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2581   __ add_slow(dst->as_pointer_register(), mon_addr.base(), mon_addr.disp());
2582 }
2583 
2584 
2585 void LIR_Assembler::align_backward_branch_target() {
2586   // Some ARM processors do better with 8-byte branch target alignment
2587   __ align(8);
2588 }
2589 
2590 
2591 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2592   // tmp must be unused
2593   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2594 
2595   if (left->is_single_cpu()) {
2596     assert (dest->type() == T_INT, "unexpected result type");
2597     assert (left->type() == T_INT, "unexpected left type");
2598     __ neg_32(dest->as_register(), left->as_register());
2599   } else if (left->is_double_cpu()) {
2600     Register dest_lo = dest->as_register_lo();
2601     Register dest_hi = dest->as_register_hi();
2602     Register src_lo = left->as_register_lo();
2603     Register src_hi = left->as_register_hi();
2604     if (dest_lo == src_hi) {
2605       dest_lo = Rtemp;
2606     }
2607     __ rsbs(dest_lo, src_lo, 0);
2608     __ rsc(dest_hi, src_hi, 0);
2609     move_regs(dest_lo, dest->as_register_lo());
2610   } else if (left->is_single_fpu()) {
2611     __ neg_float(dest->as_float_reg(), left->as_float_reg());
2612   } else if (left->is_double_fpu()) {
2613     __ neg_double(dest->as_double_reg(), left->as_double_reg());
2614   } else {
2615     ShouldNotReachHere();
2616   }
2617 }
2618 
2619 
2620 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2621   assert(patch_code == lir_patch_none, "Patch code not supported");
2622   LIR_Address* addr = addr_opr->as_address_ptr();
2623   if (addr->index()->is_illegal()) {
2624     jint c = addr->disp();
2625     if (!Assembler::is_arith_imm_in_range(c)) {
2626       BAILOUT("illegal arithmetic operand");
2627     }
2628     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(), c);
2629   } else {
2630     assert(addr->disp() == 0, "cannot handle otherwise");
2631     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(),
2632            AsmOperand(addr->index()->as_pointer_register(), lsl, addr->scale()));
2633   }
2634 }
2635 
2636 
2637 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2638   assert(!tmp->is_valid(), "don't need temporary");
2639   __ call(dest);
2640   if (info != nullptr) {
2641     add_call_info_here(info);
2642   }
2643 }
2644 
2645 
2646 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2647   assert((src->is_double_cpu() && dest->is_address()) ||
2648          (src->is_address() && dest->is_double_cpu()),
2649          "Simple move_op is called for all other cases");
2650 
2651   int null_check_offset;
2652   if (dest->is_address()) {
2653     // Store
2654     const LIR_Address* addr = dest->as_address_ptr();
2655     const Register src_lo = src->as_register_lo();
2656     const Register src_hi = src->as_register_hi();
2657     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2658 
2659     if (src_lo < src_hi) {
2660       null_check_offset = __ offset();
2661       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(src_hi));
2662     } else {
2663       assert(src_lo < Rtemp, "Rtemp is higher than any allocatable register");
2664       __ mov(Rtemp, src_hi);
2665       null_check_offset = __ offset();
2666       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(Rtemp));
2667     }
2668   } else {
2669     // Load
2670     const LIR_Address* addr = src->as_address_ptr();
2671     const Register dest_lo = dest->as_register_lo();
2672     const Register dest_hi = dest->as_register_hi();
2673     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2674 
2675     null_check_offset = __ offset();
2676     if (dest_lo < dest_hi) {
2677       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(dest_hi));
2678     } else {
2679       assert(dest_lo < Rtemp, "Rtemp is higher than any allocatable register");
2680       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(Rtemp));
2681       __ mov(dest_hi, Rtemp);
2682     }
2683   }
2684 
2685   if (info != nullptr) {
2686     add_debug_info_for_null_check(null_check_offset, info);
2687   }
2688 }
2689 
2690 
2691 void LIR_Assembler::membar() {
2692   __ membar(MacroAssembler::StoreLoad, Rtemp);
2693 }
2694 
2695 void LIR_Assembler::membar_acquire() {
2696   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::LoadLoad | MacroAssembler::LoadStore), Rtemp);
2697 }
2698 
2699 void LIR_Assembler::membar_release() {
2700   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2701 }
2702 
2703 void LIR_Assembler::membar_loadload() {
2704   __ membar(MacroAssembler::LoadLoad, Rtemp);
2705 }
2706 
2707 void LIR_Assembler::membar_storestore() {
2708   __ membar(MacroAssembler::StoreStore, Rtemp);
2709 }
2710 
2711 void LIR_Assembler::membar_loadstore() {
2712   __ membar(MacroAssembler::LoadStore, Rtemp);
2713 }
2714 
2715 void LIR_Assembler::membar_storeload() {
2716   __ membar(MacroAssembler::StoreLoad, Rtemp);
2717 }
2718 
2719 void LIR_Assembler::on_spin_wait() {
2720   Unimplemented();
2721 }
2722 
2723 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2724   // Not used on ARM
2725   Unimplemented();
2726 }
2727 
2728 void LIR_Assembler::peephole(LIR_List* lir) {
2729   LIR_OpList* inst = lir->instructions_list();
2730   const int inst_length = inst->length();
2731   for (int i = 0; i < inst_length; i++) {
2732     LIR_Op* op = inst->at(i);
2733     switch (op->code()) {
2734       case lir_cmp: {
2735         // Replace:
2736         //   cmp rX, y
2737         //   cmove [EQ] y, z, rX
2738         // with
2739         //   cmp rX, y
2740         //   cmove [EQ] illegalOpr, z, rX
2741         //
2742         // or
2743         //   cmp rX, y
2744         //   cmove [NE] z, y, rX
2745         // with
2746         //   cmp rX, y
2747         //   cmove [NE] z, illegalOpr, rX
2748         //
2749         // moves from illegalOpr should be removed when converting LIR to native assembly
2750 
2751         LIR_Op2* cmp = op->as_Op2();
2752         assert(cmp != nullptr, "cmp LIR instruction is not an op2");
2753 
2754         if (i + 1 < inst_length) {
2755           LIR_Op2* cmove = inst->at(i + 1)->as_Op2();
2756           if (cmove != nullptr && cmove->code() == lir_cmove) {
2757             LIR_Opr cmove_res = cmove->result_opr();
2758             bool res_is_op1 = cmove_res == cmp->in_opr1();
2759             bool res_is_op2 = cmove_res == cmp->in_opr2();
2760             LIR_Opr cmp_res, cmp_arg;
2761             if (res_is_op1) {
2762               cmp_res = cmp->in_opr1();
2763               cmp_arg = cmp->in_opr2();
2764             } else if (res_is_op2) {
2765               cmp_res = cmp->in_opr2();
2766               cmp_arg = cmp->in_opr1();
2767             } else {
2768               cmp_res = LIR_OprFact::illegalOpr;
2769               cmp_arg = LIR_OprFact::illegalOpr;
2770             }
2771 
2772             if (cmp_res != LIR_OprFact::illegalOpr) {
2773               LIR_Condition cond = cmove->condition();
2774               if (cond == lir_cond_equal && cmove->in_opr1() == cmp_arg) {
2775                 cmove->set_in_opr1(LIR_OprFact::illegalOpr);
2776               } else if (cond == lir_cond_notEqual && cmove->in_opr2() == cmp_arg) {
2777                 cmove->set_in_opr2(LIR_OprFact::illegalOpr);
2778               }
2779             }
2780           }
2781         }
2782         break;
2783       }
2784 
2785       default:
2786         break;
2787     }
2788   }
2789 }
2790 
2791 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2792   assert(src->is_address(), "sanity");
2793   Address addr = as_Address(src->as_address_ptr());
2794 
2795   if (code == lir_xchg) {
2796   } else {
2797     assert (!data->is_oop(), "xadd for oops");
2798   }
2799 
2800   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2801 
2802   Label retry;
2803   __ bind(retry);
2804 
2805   if (data->type() == T_INT || data->is_oop()) {
2806     Register dst = dest->as_register();
2807     Register new_val = noreg;
2808     __ ldrex(dst, addr);
2809     if (code == lir_xadd) {
2810       Register tmp_reg = tmp->as_register();
2811       if (data->is_constant()) {
2812         assert_different_registers(dst, tmp_reg);
2813         __ add_32(tmp_reg, dst, data->as_constant_ptr()->as_jint());
2814       } else {
2815         assert_different_registers(dst, tmp_reg, data->as_register());
2816         __ add_32(tmp_reg, dst, data->as_register());
2817       }
2818       new_val = tmp_reg;
2819     } else {
2820       if (UseCompressedOops && data->is_oop()) {
2821         new_val = tmp->as_pointer_register();
2822       } else {
2823         new_val = data->as_register();
2824       }
2825       assert_different_registers(dst, new_val);
2826     }
2827     __ strex(Rtemp, new_val, addr);
2828 
2829   } else if (data->type() == T_LONG) {
2830     Register dst_lo = dest->as_register_lo();
2831     Register new_val_lo = noreg;
2832     Register dst_hi = dest->as_register_hi();
2833 
2834     assert(dst_hi->encoding() == dst_lo->encoding() + 1, "non aligned register pair");
2835     assert((dst_lo->encoding() & 0x1) == 0, "misaligned register pair");
2836 
2837     __ bind(retry);
2838     __ ldrexd(dst_lo, addr);
2839     if (code == lir_xadd) {
2840       Register tmp_lo = tmp->as_register_lo();
2841       Register tmp_hi = tmp->as_register_hi();
2842 
2843       assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
2844       assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
2845 
2846       if (data->is_constant()) {
2847         jlong c = data->as_constant_ptr()->as_jlong();
2848         assert((jlong)((jint)c) == c, "overflow");
2849         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi);
2850         __ adds(tmp_lo, dst_lo, (jint)c);
2851         __ adc(tmp_hi, dst_hi, 0);
2852       } else {
2853         Register new_val_lo = data->as_register_lo();
2854         Register new_val_hi = data->as_register_hi();
2855         __ adds(tmp_lo, dst_lo, new_val_lo);
2856         __ adc(tmp_hi, dst_hi, new_val_hi);
2857         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi, new_val_lo, new_val_hi);
2858       }
2859       new_val_lo = tmp_lo;
2860     } else {
2861       new_val_lo = data->as_register_lo();
2862       Register new_val_hi = data->as_register_hi();
2863 
2864       assert_different_registers(dst_lo, dst_hi, new_val_lo, new_val_hi);
2865       assert(new_val_hi->encoding() == new_val_lo->encoding() + 1, "non aligned register pair");
2866       assert((new_val_lo->encoding() & 0x1) == 0, "misaligned register pair");
2867     }
2868     __ strexd(Rtemp, new_val_lo, addr);
2869   } else {
2870     ShouldNotReachHere();
2871   }
2872 
2873   __ cbnz_32(Rtemp, retry);
2874   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
2875 
2876 }
2877 
2878 #undef __