1 /*
   2  * Copyright (c) 2008, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.inline.hpp"
  27 #include "c1/c1_Compilation.hpp"
  28 #include "c1/c1_LIRAssembler.hpp"
  29 #include "c1/c1_MacroAssembler.hpp"
  30 #include "c1/c1_Runtime1.hpp"
  31 #include "c1/c1_ValueStack.hpp"
  32 #include "ci/ciArrayKlass.hpp"
  33 #include "ci/ciInstance.hpp"
  34 #include "gc/shared/collectedHeap.hpp"
  35 #include "memory/universe.hpp"
  36 #include "nativeInst_arm.hpp"
  37 #include "oops/objArrayKlass.hpp"
  38 #include "runtime/frame.inline.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/stubRoutines.hpp"
  41 #include "utilities/powerOfTwo.hpp"
  42 #include "vmreg_arm.inline.hpp"
  43 
  44 #define __ _masm->
  45 
  46 // Note: Rtemp usage is this file should not impact C2 and should be
  47 // correct as long as it is not implicitly used in lower layers (the
  48 // arm [macro]assembler) and used with care in the other C1 specific
  49 // files.
  50 
  51 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  52   ShouldNotCallThis(); // Not used on ARM
  53   return false;
  54 }
  55 
  56 
  57 LIR_Opr LIR_Assembler::receiverOpr() {
  58   // The first register in Java calling conventions
  59   return FrameMap::R0_oop_opr;
  60 }
  61 
  62 LIR_Opr LIR_Assembler::osrBufferPointer() {
  63   return FrameMap::as_pointer_opr(R0);
  64 }
  65 
  66 #ifndef PRODUCT
  67 void LIR_Assembler::verify_reserved_argument_area_size(int args_count) {
  68   assert(args_count * wordSize <= frame_map()->reserved_argument_area_size(), "not enough space for arguments");
  69 }
  70 #endif // !PRODUCT
  71 
  72 void LIR_Assembler::store_parameter(jint c, int offset_from_sp_in_words) {
  73   assert(offset_from_sp_in_words >= 0, "invalid offset from sp");
  74   int offset_from_sp_in_bytes = offset_from_sp_in_words * BytesPerWord;
  75   assert(offset_from_sp_in_bytes < frame_map()->reserved_argument_area_size(), "not enough space");
  76   __ mov_slow(Rtemp, c);
  77   __ str(Rtemp, Address(SP, offset_from_sp_in_bytes));
  78 }
  79 
  80 void LIR_Assembler::store_parameter(Metadata* m, int offset_from_sp_in_words) {
  81   assert(offset_from_sp_in_words >= 0, "invalid offset from sp");
  82   int offset_from_sp_in_bytes = offset_from_sp_in_words * BytesPerWord;
  83   assert(offset_from_sp_in_bytes < frame_map()->reserved_argument_area_size(), "not enough space");
  84   __ mov_metadata(Rtemp, m);
  85   __ str(Rtemp, Address(SP, offset_from_sp_in_bytes));
  86 }
  87 
  88 //--------------fpu register translations-----------------------
  89 
  90 
  91 void LIR_Assembler::breakpoint() {
  92   __ breakpoint();
  93 }
  94 
  95 void LIR_Assembler::push(LIR_Opr opr) {
  96   Unimplemented();
  97 }
  98 
  99 void LIR_Assembler::pop(LIR_Opr opr) {
 100   Unimplemented();
 101 }
 102 
 103 //-------------------------------------------
 104 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 105   Register base = addr->base()->as_pointer_register();
 106 
 107 
 108   if (addr->index()->is_illegal() || addr->index()->is_constant()) {
 109     int offset = addr->disp();
 110     if (addr->index()->is_constant()) {
 111       offset += addr->index()->as_constant_ptr()->as_jint() << addr->scale();
 112     }
 113 
 114     if ((offset <= -4096) || (offset >= 4096)) {
 115       BAILOUT_("offset not in range", Address(base));
 116     }
 117 
 118     return Address(base, offset);
 119 
 120   } else {
 121     assert(addr->disp() == 0, "can't have both");
 122     int scale = addr->scale();
 123 
 124     assert(addr->index()->is_single_cpu(), "should be");
 125     return scale >= 0 ? Address(base, addr->index()->as_register(), lsl, scale) :
 126                         Address(base, addr->index()->as_register(), lsr, -scale);
 127   }
 128 }
 129 
 130 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 131   Address base = as_Address(addr);
 132   assert(base.index() == noreg, "must be");
 133   if (base.disp() + BytesPerWord >= 4096) { BAILOUT_("offset not in range", Address(base.base(),0)); }
 134   return Address(base.base(), base.disp() + BytesPerWord);
 135 }
 136 
 137 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 138   return as_Address(addr);
 139 }
 140 
 141 
 142 void LIR_Assembler::osr_entry() {
 143   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 144   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 145   ValueStack* entry_state = osr_entry->end()->state();
 146   int number_of_locks = entry_state->locks_size();
 147 
 148   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 149   Register OSR_buf = osrBufferPointer()->as_pointer_register();
 150 
 151   assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 152   int monitor_offset = (method()->max_locals() + 2 * (number_of_locks - 1)) * BytesPerWord;
 153   for (int i = 0; i < number_of_locks; i++) {
 154     int slot_offset = monitor_offset - (i * 2 * BytesPerWord);
 155     __ ldr(R1, Address(OSR_buf, slot_offset + 0*BytesPerWord));
 156     __ ldr(R2, Address(OSR_buf, slot_offset + 1*BytesPerWord));
 157     __ str(R1, frame_map()->address_for_monitor_lock(i));
 158     __ str(R2, frame_map()->address_for_monitor_object(i));
 159   }
 160 }
 161 
 162 
 163 int LIR_Assembler::check_icache() {
 164   Register receiver = LIR_Assembler::receiverOpr()->as_register();
 165   int offset = __ offset();
 166   __ inline_cache_check(receiver, Ricklass);
 167   return offset;
 168 }
 169 
 170 void LIR_Assembler::clinit_barrier(ciMethod* method) {
 171   ShouldNotReachHere(); // not implemented
 172 }
 173 
 174 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
 175   jobject o = (jobject)Universe::non_oop_word();
 176   int index = __ oop_recorder()->allocate_oop_index(o);
 177 
 178   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), index);
 179 
 180   __ patchable_mov_oop(reg, o, index);
 181   patching_epilog(patch, lir_patch_normal, reg, info);
 182 }
 183 
 184 
 185 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
 186   Metadata* o = (Metadata*)Universe::non_oop_word();
 187   int index = __ oop_recorder()->allocate_metadata_index(o);
 188   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 189 
 190   __ patchable_mov_metadata(reg, o, index);
 191   patching_epilog(patch, lir_patch_normal, reg, info);
 192 }
 193 
 194 
 195 int LIR_Assembler::initial_frame_size_in_bytes() const {
 196   // Subtracts two words to account for return address and link
 197   return frame_map()->framesize()*VMRegImpl::stack_slot_size - 2*wordSize;
 198 }
 199 
 200 
 201 int LIR_Assembler::emit_exception_handler() {
 202   address handler_base = __ start_a_stub(exception_handler_size());
 203   if (handler_base == nullptr) {
 204     bailout("exception handler overflow");
 205     return -1;
 206   }
 207 
 208   int offset = code_offset();
 209 
 210   // check that there is really an exception
 211   __ verify_not_null_oop(Rexception_obj);
 212 
 213   __ call(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id), relocInfo::runtime_call_type);
 214   __ should_not_reach_here();
 215 
 216   assert(code_offset() - offset <= exception_handler_size(), "overflow");
 217   __ end_a_stub();
 218 
 219   return offset;
 220 }
 221 
 222 // Emit the code to remove the frame from the stack in the exception
 223 // unwind path.
 224 int LIR_Assembler::emit_unwind_handler() {
 225 #ifndef PRODUCT
 226   if (CommentedAssembly) {
 227     _masm->block_comment("Unwind handler");
 228   }
 229 #endif
 230 
 231   int offset = code_offset();
 232 
 233   // Fetch the exception from TLS and clear out exception related thread state
 234   Register zero = __ zero_register(Rtemp);
 235   __ ldr(Rexception_obj, Address(Rthread, JavaThread::exception_oop_offset()));
 236   __ str(zero, Address(Rthread, JavaThread::exception_oop_offset()));
 237   __ str(zero, Address(Rthread, JavaThread::exception_pc_offset()));
 238 
 239   __ bind(_unwind_handler_entry);
 240   __ verify_not_null_oop(Rexception_obj);
 241 
 242   // Perform needed unlocking
 243   MonitorExitStub* stub = nullptr;
 244   if (method()->is_synchronized()) {
 245     monitor_address(0, FrameMap::R0_opr);
 246     stub = new MonitorExitStub(FrameMap::R0_opr, true, 0);
 247     __ unlock_object(R2, R1, R0, *stub->entry());
 248     __ bind(*stub->continuation());
 249   }
 250 
 251   // remove the activation and dispatch to the unwind handler
 252   __ remove_frame(initial_frame_size_in_bytes()); // restores FP and LR
 253   __ jump(Runtime1::entry_for(Runtime1::unwind_exception_id), relocInfo::runtime_call_type, Rtemp);
 254 
 255   // Emit the slow path assembly
 256   if (stub != nullptr) {
 257     stub->emit_code(this);
 258   }
 259 
 260   return offset;
 261 }
 262 
 263 
 264 int LIR_Assembler::emit_deopt_handler() {
 265   address handler_base = __ start_a_stub(deopt_handler_size());
 266   if (handler_base == nullptr) {
 267     bailout("deopt handler overflow");
 268     return -1;
 269   }
 270 
 271   int offset = code_offset();
 272 
 273   __ mov_relative_address(LR, __ pc());
 274   __ push(LR); // stub expects LR to be saved
 275   __ jump(SharedRuntime::deopt_blob()->unpack(), relocInfo::runtime_call_type, noreg);
 276 
 277   assert(code_offset() - offset <= deopt_handler_size(), "overflow");
 278   __ end_a_stub();
 279 
 280   return offset;
 281 }
 282 
 283 
 284 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
 285   // Pop the frame before safepoint polling
 286   __ remove_frame(initial_frame_size_in_bytes());
 287   __ read_polling_page(Rtemp, relocInfo::poll_return_type);
 288   __ ret();
 289 }
 290 
 291 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
 292 
 293   int offset = __ offset();
 294   __ get_polling_page(Rtemp);
 295   __ relocate(relocInfo::poll_type);
 296   add_debug_info_for_branch(info); // help pc_desc_at to find correct scope for current PC
 297   __ ldr(Rtemp, Address(Rtemp));
 298 
 299   return offset;
 300 }
 301 
 302 
 303 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 304   if (from_reg != to_reg) {
 305     __ mov(to_reg, from_reg);
 306   }
 307 }
 308 
 309 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 310   assert(src->is_constant() && dest->is_register(), "must be");
 311   LIR_Const* c = src->as_constant_ptr();
 312 
 313   switch (c->type()) {
 314     case T_ADDRESS:
 315     case T_INT:
 316       assert(patch_code == lir_patch_none, "no patching handled here");
 317       __ mov_slow(dest->as_register(), c->as_jint());
 318       break;
 319 
 320     case T_LONG:
 321       assert(patch_code == lir_patch_none, "no patching handled here");
 322       __ mov_slow(dest->as_register_lo(), c->as_jint_lo());
 323       __ mov_slow(dest->as_register_hi(), c->as_jint_hi());
 324       break;
 325 
 326     case T_OBJECT:
 327       if (patch_code == lir_patch_none) {
 328         __ mov_oop(dest->as_register(), c->as_jobject());
 329       } else {
 330         jobject2reg_with_patching(dest->as_register(), info);
 331       }
 332       break;
 333 
 334     case T_METADATA:
 335       if (patch_code == lir_patch_none) {
 336         __ mov_metadata(dest->as_register(), c->as_metadata());
 337       } else {
 338         klass2reg_with_patching(dest->as_register(), info);
 339       }
 340       break;
 341 
 342     case T_FLOAT:
 343       if (dest->is_single_fpu()) {
 344         __ mov_float(dest->as_float_reg(), c->as_jfloat());
 345       } else {
 346         // Simple getters can return float constant directly into r0
 347         __ mov_slow(dest->as_register(), c->as_jint_bits());
 348       }
 349       break;
 350 
 351     case T_DOUBLE:
 352       if (dest->is_double_fpu()) {
 353         __ mov_double(dest->as_double_reg(), c->as_jdouble());
 354       } else {
 355         // Simple getters can return double constant directly into r1r0
 356         __ mov_slow(dest->as_register_lo(), c->as_jint_lo_bits());
 357         __ mov_slow(dest->as_register_hi(), c->as_jint_hi_bits());
 358       }
 359       break;
 360 
 361     default:
 362       ShouldNotReachHere();
 363   }
 364 }
 365 
 366 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 367   assert(src->is_constant(), "must be");
 368   assert(dest->is_stack(), "must be");
 369   LIR_Const* c = src->as_constant_ptr();
 370 
 371   switch (c->type()) {
 372     case T_INT:  // fall through
 373     case T_FLOAT:
 374       __ mov_slow(Rtemp, c->as_jint_bits());
 375       __ str_32(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 376       break;
 377 
 378     case T_ADDRESS:
 379       __ mov_slow(Rtemp, c->as_jint());
 380       __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 381       break;
 382 
 383     case T_OBJECT:
 384       __ mov_oop(Rtemp, c->as_jobject());
 385       __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 386       break;
 387 
 388     case T_LONG:  // fall through
 389     case T_DOUBLE:
 390       __ mov_slow(Rtemp, c->as_jint_lo_bits());
 391       __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes));
 392       if (c->as_jint_hi_bits() != c->as_jint_lo_bits()) {
 393         __ mov_slow(Rtemp, c->as_jint_hi_bits());
 394       }
 395       __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 396       break;
 397 
 398     default:
 399       ShouldNotReachHere();
 400   }
 401 }
 402 
 403 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type,
 404                               CodeEmitInfo* info, bool wide) {
 405   assert((src->as_constant_ptr()->type() == T_OBJECT && src->as_constant_ptr()->as_jobject() == nullptr),"cannot handle otherwise");
 406   __ mov(Rtemp, 0);
 407 
 408   int null_check_offset = code_offset();
 409   __ str(Rtemp, as_Address(dest->as_address_ptr()));
 410 
 411   if (info != nullptr) {
 412     assert(false, "arm32 didn't support this before, investigate if bug");
 413     add_debug_info_for_null_check(null_check_offset, info);
 414   }
 415 }
 416 
 417 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
 418   assert(src->is_register() && dest->is_register(), "must be");
 419 
 420   if (src->is_single_cpu()) {
 421     if (dest->is_single_cpu()) {
 422       move_regs(src->as_register(), dest->as_register());
 423     } else if (dest->is_single_fpu()) {
 424       __ fmsr(dest->as_float_reg(), src->as_register());
 425     } else {
 426       ShouldNotReachHere();
 427     }
 428   } else if (src->is_double_cpu()) {
 429     if (dest->is_double_cpu()) {
 430       __ long_move(dest->as_register_lo(), dest->as_register_hi(), src->as_register_lo(), src->as_register_hi());
 431     } else {
 432       __ fmdrr(dest->as_double_reg(), src->as_register_lo(), src->as_register_hi());
 433     }
 434   } else if (src->is_single_fpu()) {
 435     if (dest->is_single_fpu()) {
 436       __ mov_float(dest->as_float_reg(), src->as_float_reg());
 437     } else if (dest->is_single_cpu()) {
 438       __ mov_fpr2gpr_float(dest->as_register(), src->as_float_reg());
 439     } else {
 440       ShouldNotReachHere();
 441     }
 442   } else if (src->is_double_fpu()) {
 443     if (dest->is_double_fpu()) {
 444       __ mov_double(dest->as_double_reg(), src->as_double_reg());
 445     } else if (dest->is_double_cpu()) {
 446       __ fmrrd(dest->as_register_lo(), dest->as_register_hi(), src->as_double_reg());
 447     } else {
 448       ShouldNotReachHere();
 449     }
 450   } else {
 451     ShouldNotReachHere();
 452   }
 453 }
 454 
 455 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
 456   assert(src->is_register(), "should not call otherwise");
 457   assert(dest->is_stack(), "should not call otherwise");
 458 
 459   Address addr = dest->is_single_word() ?
 460     frame_map()->address_for_slot(dest->single_stack_ix()) :
 461     frame_map()->address_for_slot(dest->double_stack_ix());
 462 
 463   assert(lo_word_offset_in_bytes == 0 && hi_word_offset_in_bytes == 4, "little ending");
 464   if (src->is_single_fpu() || src->is_double_fpu()) {
 465     if (addr.disp() >= 1024) { BAILOUT("Too exotic case to handle here"); }
 466   }
 467 
 468   if (src->is_single_cpu()) {
 469     switch (type) {
 470       case T_OBJECT:
 471       case T_ARRAY:    __ verify_oop(src->as_register());   // fall through
 472       case T_ADDRESS:
 473       case T_METADATA: __ str(src->as_register(), addr);    break;
 474       case T_FLOAT:    // used in intBitsToFloat intrinsic implementation, fall through
 475       case T_INT:      __ str_32(src->as_register(), addr); break;
 476       default:
 477         ShouldNotReachHere();
 478     }
 479   } else if (src->is_double_cpu()) {
 480     __ str(src->as_register_lo(), addr);
 481     __ str(src->as_register_hi(), frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 482   } else if (src->is_single_fpu()) {
 483     __ str_float(src->as_float_reg(), addr);
 484   } else if (src->is_double_fpu()) {
 485     __ str_double(src->as_double_reg(), addr);
 486   } else {
 487     ShouldNotReachHere();
 488   }
 489 }
 490 
 491 
 492 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type,
 493                             LIR_PatchCode patch_code, CodeEmitInfo* info,
 494                             bool pop_fpu_stack, bool wide) {
 495   LIR_Address* to_addr = dest->as_address_ptr();
 496   Register base_reg = to_addr->base()->as_pointer_register();
 497   const bool needs_patching = (patch_code != lir_patch_none);
 498 
 499   PatchingStub* patch = nullptr;
 500   if (needs_patching) {
 501     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 502   }
 503 
 504   int null_check_offset = code_offset();
 505 
 506   switch (type) {
 507     case T_ARRAY:
 508     case T_OBJECT:
 509       if (UseCompressedOops && !wide) {
 510         ShouldNotReachHere();
 511       } else {
 512         __ str(src->as_register(), as_Address(to_addr));
 513       }
 514       break;
 515 
 516     case T_ADDRESS:
 517       __ str(src->as_pointer_register(), as_Address(to_addr));
 518       break;
 519 
 520     case T_BYTE:
 521     case T_BOOLEAN:
 522       __ strb(src->as_register(), as_Address(to_addr));
 523       break;
 524 
 525     case T_CHAR:
 526     case T_SHORT:
 527       __ strh(src->as_register(), as_Address(to_addr));
 528       break;
 529 
 530     case T_INT:
 531 #ifdef __SOFTFP__
 532     case T_FLOAT:
 533 #endif // __SOFTFP__
 534       __ str_32(src->as_register(), as_Address(to_addr));
 535       break;
 536 
 537 
 538 #ifdef __SOFTFP__
 539     case T_DOUBLE:
 540 #endif // __SOFTFP__
 541     case T_LONG: {
 542       Register from_lo = src->as_register_lo();
 543       Register from_hi = src->as_register_hi();
 544       if (to_addr->index()->is_register()) {
 545         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 546         assert(to_addr->disp() == 0, "Not yet supporting both");
 547         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 548         base_reg = Rtemp;
 549         __ str(from_lo, Address(Rtemp));
 550         if (patch != nullptr) {
 551           __ nop(); // see comment before patching_epilog for 2nd str
 552           patching_epilog(patch, lir_patch_low, base_reg, info);
 553           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 554           patch_code = lir_patch_high;
 555         }
 556         __ str(from_hi, Address(Rtemp, BytesPerWord));
 557       } else if (base_reg == from_lo) {
 558         __ str(from_hi, as_Address_hi(to_addr));
 559         if (patch != nullptr) {
 560           __ nop(); // see comment before patching_epilog for 2nd str
 561           patching_epilog(patch, lir_patch_high, base_reg, info);
 562           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 563           patch_code = lir_patch_low;
 564         }
 565         __ str(from_lo, as_Address_lo(to_addr));
 566       } else {
 567         __ str(from_lo, as_Address_lo(to_addr));
 568         if (patch != nullptr) {
 569           __ nop(); // see comment before patching_epilog for 2nd str
 570           patching_epilog(patch, lir_patch_low, base_reg, info);
 571           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 572           patch_code = lir_patch_high;
 573         }
 574         __ str(from_hi, as_Address_hi(to_addr));
 575       }
 576       break;
 577     }
 578 
 579 #ifndef __SOFTFP__
 580     case T_FLOAT:
 581       if (to_addr->index()->is_register()) {
 582         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 583         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 584         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 585         __ fsts(src->as_float_reg(), Address(Rtemp, to_addr->disp()));
 586       } else {
 587         __ fsts(src->as_float_reg(), as_Address(to_addr));
 588       }
 589       break;
 590 
 591     case T_DOUBLE:
 592       if (to_addr->index()->is_register()) {
 593         assert(to_addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 594         __ add(Rtemp, base_reg, to_addr->index()->as_register());
 595         if ((to_addr->disp() <= -4096) || (to_addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 596         __ fstd(src->as_double_reg(), Address(Rtemp, to_addr->disp()));
 597       } else {
 598         __ fstd(src->as_double_reg(), as_Address(to_addr));
 599       }
 600       break;
 601 #endif // __SOFTFP__
 602 
 603 
 604     default:
 605       ShouldNotReachHere();
 606   }
 607 
 608   if (info != nullptr) {
 609     add_debug_info_for_null_check(null_check_offset, info);
 610   }
 611 
 612   if (patch != nullptr) {
 613     // Offset embedded into LDR/STR instruction may appear not enough
 614     // to address a field. So, provide a space for one more instruction
 615     // that will deal with larger offsets.
 616     __ nop();
 617     patching_epilog(patch, patch_code, base_reg, info);
 618   }
 619 }
 620 
 621 
 622 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 623   assert(src->is_stack(), "should not call otherwise");
 624   assert(dest->is_register(), "should not call otherwise");
 625 
 626   Address addr = src->is_single_word() ?
 627     frame_map()->address_for_slot(src->single_stack_ix()) :
 628     frame_map()->address_for_slot(src->double_stack_ix());
 629 
 630   assert(lo_word_offset_in_bytes == 0 && hi_word_offset_in_bytes == 4, "little ending");
 631   if (dest->is_single_fpu() || dest->is_double_fpu()) {
 632     if (addr.disp() >= 1024) { BAILOUT("Too exotic case to handle here"); }
 633   }
 634 
 635   if (dest->is_single_cpu()) {
 636     switch (type) {
 637       case T_OBJECT:
 638       case T_ARRAY:
 639       case T_ADDRESS:
 640       case T_METADATA: __ ldr(dest->as_register(), addr); break;
 641       case T_FLOAT:    // used in floatToRawIntBits intrinsic implementation
 642       case T_INT:      __ ldr_u32(dest->as_register(), addr); break;
 643       default:
 644         ShouldNotReachHere();
 645     }
 646     if ((type == T_OBJECT) || (type == T_ARRAY)) {
 647       __ verify_oop(dest->as_register());
 648     }
 649   } else if (dest->is_double_cpu()) {
 650     __ ldr(dest->as_register_lo(), addr);
 651     __ ldr(dest->as_register_hi(), frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 652   } else if (dest->is_single_fpu()) {
 653     __ ldr_float(dest->as_float_reg(), addr);
 654   } else if (dest->is_double_fpu()) {
 655     __ ldr_double(dest->as_double_reg(), addr);
 656   } else {
 657     ShouldNotReachHere();
 658   }
 659 }
 660 
 661 
 662 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 663   if (src->is_single_stack()) {
 664     switch (src->type()) {
 665       case T_OBJECT:
 666       case T_ARRAY:
 667       case T_ADDRESS:
 668       case T_METADATA:
 669         __ ldr(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 670         __ str(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 671         break;
 672 
 673       case T_INT:
 674       case T_FLOAT:
 675         __ ldr_u32(Rtemp, frame_map()->address_for_slot(src->single_stack_ix()));
 676         __ str_32(Rtemp, frame_map()->address_for_slot(dest->single_stack_ix()));
 677         break;
 678 
 679       default:
 680         ShouldNotReachHere();
 681     }
 682   } else {
 683     assert(src->is_double_stack(), "must be");
 684     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes));
 685     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes));
 686     __ ldr(Rtemp, frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes));
 687     __ str(Rtemp, frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes));
 688   }
 689 }
 690 
 691 
 692 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type,
 693                             LIR_PatchCode patch_code, CodeEmitInfo* info,
 694                             bool wide) {
 695   assert(src->is_address(), "should not call otherwise");
 696   assert(dest->is_register(), "should not call otherwise");
 697   LIR_Address* addr = src->as_address_ptr();
 698 
 699   Register base_reg = addr->base()->as_pointer_register();
 700 
 701   PatchingStub* patch = nullptr;
 702   if (patch_code != lir_patch_none) {
 703     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 704   }
 705   if (info != nullptr) {
 706     add_debug_info_for_null_check_here(info);
 707   }
 708 
 709   switch (type) {
 710     case T_OBJECT:  // fall through
 711     case T_ARRAY:
 712       if (UseCompressedOops && !wide) {
 713         __ ldr_u32(dest->as_register(), as_Address(addr));
 714       } else {
 715         __ ldr(dest->as_register(), as_Address(addr));
 716       }
 717       break;
 718 
 719     case T_ADDRESS:
 720       __ ldr(dest->as_pointer_register(), as_Address(addr));
 721       break;
 722 
 723     case T_INT:
 724 #ifdef __SOFTFP__
 725     case T_FLOAT:
 726 #endif // __SOFTFP__
 727       __ ldr(dest->as_pointer_register(), as_Address(addr));
 728       break;
 729 
 730     case T_BOOLEAN:
 731       __ ldrb(dest->as_register(), as_Address(addr));
 732       break;
 733 
 734     case T_BYTE:
 735       __ ldrsb(dest->as_register(), as_Address(addr));
 736       break;
 737 
 738     case T_CHAR:
 739       __ ldrh(dest->as_register(), as_Address(addr));
 740       break;
 741 
 742     case T_SHORT:
 743       __ ldrsh(dest->as_register(), as_Address(addr));
 744       break;
 745 
 746 
 747 #ifdef __SOFTFP__
 748     case T_DOUBLE:
 749 #endif // __SOFTFP__
 750     case T_LONG: {
 751       Register to_lo = dest->as_register_lo();
 752       Register to_hi = dest->as_register_hi();
 753       if (addr->index()->is_register()) {
 754         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 755         assert(addr->disp() == 0, "Not yet supporting both");
 756         __ add(Rtemp, base_reg, addr->index()->as_register());
 757         base_reg = Rtemp;
 758         __ ldr(to_lo, Address(Rtemp));
 759         if (patch != nullptr) {
 760           __ nop(); // see comment before patching_epilog for 2nd ldr
 761           patching_epilog(patch, lir_patch_low, base_reg, info);
 762           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 763           patch_code = lir_patch_high;
 764         }
 765         __ ldr(to_hi, Address(Rtemp, BytesPerWord));
 766       } else if (base_reg == to_lo) {
 767         __ ldr(to_hi, as_Address_hi(addr));
 768         if (patch != nullptr) {
 769           __ nop(); // see comment before patching_epilog for 2nd ldr
 770           patching_epilog(patch, lir_patch_high, base_reg, info);
 771           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 772           patch_code = lir_patch_low;
 773         }
 774         __ ldr(to_lo, as_Address_lo(addr));
 775       } else {
 776         __ ldr(to_lo, as_Address_lo(addr));
 777         if (patch != nullptr) {
 778           __ nop(); // see comment before patching_epilog for 2nd ldr
 779           patching_epilog(patch, lir_patch_low, base_reg, info);
 780           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 781           patch_code = lir_patch_high;
 782         }
 783         __ ldr(to_hi, as_Address_hi(addr));
 784       }
 785       break;
 786     }
 787 
 788 #ifndef __SOFTFP__
 789     case T_FLOAT:
 790       if (addr->index()->is_register()) {
 791         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 792         __ add(Rtemp, base_reg, addr->index()->as_register());
 793         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 794         __ flds(dest->as_float_reg(), Address(Rtemp, addr->disp()));
 795       } else {
 796         __ flds(dest->as_float_reg(), as_Address(addr));
 797       }
 798       break;
 799 
 800     case T_DOUBLE:
 801       if (addr->index()->is_register()) {
 802         assert(addr->scale() == LIR_Address::times_1,"Unexpected scaled register");
 803         __ add(Rtemp, base_reg, addr->index()->as_register());
 804         if ((addr->disp() <= -4096) || (addr->disp() >= 4096)) { BAILOUT("offset not in range"); }
 805         __ fldd(dest->as_double_reg(), Address(Rtemp, addr->disp()));
 806       } else {
 807         __ fldd(dest->as_double_reg(), as_Address(addr));
 808       }
 809       break;
 810 #endif // __SOFTFP__
 811 
 812 
 813     default:
 814       ShouldNotReachHere();
 815   }
 816 
 817   if (patch != nullptr) {
 818     // Offset embedded into LDR/STR instruction may appear not enough
 819     // to address a field. So, provide a space for one more instruction
 820     // that will deal with larger offsets.
 821     __ nop();
 822     patching_epilog(patch, patch_code, base_reg, info);
 823   }
 824 
 825 }
 826 
 827 
 828 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 829   bool is_32 = op->result_opr()->is_single_cpu();
 830 
 831   if (op->code() == lir_idiv && op->in_opr2()->is_constant() && is_32) {
 832     int c = op->in_opr2()->as_constant_ptr()->as_jint();
 833     assert(is_power_of_2(c), "non power-of-2 constant should be put in a register");
 834 
 835     Register left = op->in_opr1()->as_register();
 836     Register dest = op->result_opr()->as_register();
 837     if (c == 1) {
 838       __ mov(dest, left);
 839     } else if (c == 2) {
 840       __ add_32(dest, left, AsmOperand(left, lsr, 31));
 841       __ asr_32(dest, dest, 1);
 842     } else if (c != (int) 0x80000000) {
 843       int power = log2i_exact(c);
 844       __ asr_32(Rtemp, left, 31);
 845       __ add_32(dest, left, AsmOperand(Rtemp, lsr, 32-power)); // dest = left + (left < 0 ? 2^power - 1 : 0);
 846       __ asr_32(dest, dest, power);                            // dest = dest >>> power;
 847     } else {
 848       // x/0x80000000 is a special case, since dividend is a power of two, but is negative.
 849       // The only possible result values are 0 and 1, with 1 only for dividend == divisor == 0x80000000.
 850       __ cmp_32(left, c);
 851       __ mov(dest, 0, ne);
 852       __ mov(dest, 1, eq);
 853     }
 854   } else {
 855     assert(op->code() == lir_idiv || op->code() == lir_irem, "unexpected op3");
 856     __ call(StubRoutines::Arm::idiv_irem_entry(), relocInfo::runtime_call_type);
 857     add_debug_info_for_div0_here(op->info());
 858   }
 859 }
 860 
 861 
 862 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 863 #ifdef ASSERT
 864   assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
 865   if (op->block() != nullptr)  _branch_target_blocks.append(op->block());
 866   if (op->ublock() != nullptr) _branch_target_blocks.append(op->ublock());
 867   assert(op->info() == nullptr, "CodeEmitInfo?");
 868 #endif // ASSERT
 869 
 870 #ifdef __SOFTFP__
 871   assert (op->code() != lir_cond_float_branch, "this should be impossible");
 872 #else
 873   if (op->code() == lir_cond_float_branch) {
 874     __ fmstat();
 875     __ b(*(op->ublock()->label()), vs);
 876   }
 877 #endif // __SOFTFP__
 878 
 879   AsmCondition acond = al;
 880   switch (op->cond()) {
 881     case lir_cond_equal:        acond = eq; break;
 882     case lir_cond_notEqual:     acond = ne; break;
 883     case lir_cond_less:         acond = lt; break;
 884     case lir_cond_lessEqual:    acond = le; break;
 885     case lir_cond_greaterEqual: acond = ge; break;
 886     case lir_cond_greater:      acond = gt; break;
 887     case lir_cond_aboveEqual:   acond = hs; break;
 888     case lir_cond_belowEqual:   acond = ls; break;
 889     default: assert(op->cond() == lir_cond_always, "must be");
 890   }
 891   __ b(*(op->label()), acond);
 892 }
 893 
 894 
 895 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 896   LIR_Opr src  = op->in_opr();
 897   LIR_Opr dest = op->result_opr();
 898 
 899   switch (op->bytecode()) {
 900     case Bytecodes::_i2l:
 901       move_regs(src->as_register(), dest->as_register_lo());
 902       __ mov(dest->as_register_hi(), AsmOperand(src->as_register(), asr, 31));
 903       break;
 904     case Bytecodes::_l2i:
 905       move_regs(src->as_register_lo(), dest->as_register());
 906       break;
 907     case Bytecodes::_i2b:
 908       __ sign_extend(dest->as_register(), src->as_register(), 8);
 909       break;
 910     case Bytecodes::_i2s:
 911       __ sign_extend(dest->as_register(), src->as_register(), 16);
 912       break;
 913     case Bytecodes::_i2c:
 914       __ zero_extend(dest->as_register(), src->as_register(), 16);
 915       break;
 916     case Bytecodes::_f2d:
 917       __ convert_f2d(dest->as_double_reg(), src->as_float_reg());
 918       break;
 919     case Bytecodes::_d2f:
 920       __ convert_d2f(dest->as_float_reg(), src->as_double_reg());
 921       break;
 922     case Bytecodes::_i2f:
 923       __ fmsr(Stemp, src->as_register());
 924       __ fsitos(dest->as_float_reg(), Stemp);
 925       break;
 926     case Bytecodes::_i2d:
 927       __ fmsr(Stemp, src->as_register());
 928       __ fsitod(dest->as_double_reg(), Stemp);
 929       break;
 930     case Bytecodes::_f2i:
 931       __ ftosizs(Stemp, src->as_float_reg());
 932       __ fmrs(dest->as_register(), Stemp);
 933       break;
 934     case Bytecodes::_d2i:
 935       __ ftosizd(Stemp, src->as_double_reg());
 936       __ fmrs(dest->as_register(), Stemp);
 937       break;
 938     default:
 939       ShouldNotReachHere();
 940   }
 941 }
 942 
 943 
 944 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
 945   if (op->init_check()) {
 946     Register tmp = op->tmp1()->as_register();
 947     __ ldrb(tmp, Address(op->klass()->as_register(), InstanceKlass::init_state_offset()));
 948     add_debug_info_for_null_check_here(op->stub()->info());
 949     __ cmp(tmp, InstanceKlass::fully_initialized);
 950     __ b(*op->stub()->entry(), ne);
 951   }
 952   __ allocate_object(op->obj()->as_register(),
 953                      op->tmp1()->as_register(),
 954                      op->tmp2()->as_register(),
 955                      op->tmp3()->as_register(),
 956                      op->header_size(),
 957                      op->object_size(),
 958                      op->klass()->as_register(),
 959                      *op->stub()->entry());
 960   __ bind(*op->stub()->continuation());
 961 }
 962 
 963 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
 964   if (UseSlowPath ||
 965       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
 966       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
 967     __ b(*op->stub()->entry());
 968   } else {
 969     __ allocate_array(op->obj()->as_register(),
 970                       op->len()->as_register(),
 971                       op->tmp1()->as_register(),
 972                       op->tmp2()->as_register(),
 973                       op->tmp3()->as_register(),
 974                       arrayOopDesc::header_size(op->type()),
 975                       type2aelembytes(op->type()),
 976                       op->klass()->as_register(),
 977                       *op->stub()->entry());
 978   }
 979   __ bind(*op->stub()->continuation());
 980 }
 981 
 982 void LIR_Assembler::type_profile_helper(Register mdo, int mdo_offset_bias,
 983                                         ciMethodData *md, ciProfileData *data,
 984                                         Register recv, Register tmp1, Label* update_done) {
 985   assert_different_registers(mdo, recv, tmp1);
 986   uint i;
 987   for (i = 0; i < VirtualCallData::row_limit(); i++) {
 988     Label next_test;
 989     // See if the receiver is receiver[n].
 990     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
 991                           mdo_offset_bias);
 992     __ ldr(tmp1, receiver_addr);
 993     __ verify_klass_ptr(tmp1);
 994     __ cmp(recv, tmp1);
 995     __ b(next_test, ne);
 996     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
 997                       mdo_offset_bias);
 998     __ ldr(tmp1, data_addr);
 999     __ add(tmp1, tmp1, DataLayout::counter_increment);
1000     __ str(tmp1, data_addr);
1001     __ b(*update_done);
1002     __ bind(next_test);
1003   }
1004 
1005   // Didn't find receiver; find next empty slot and fill it in
1006   for (i = 0; i < VirtualCallData::row_limit(); i++) {
1007     Label next_test;
1008     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)) -
1009                       mdo_offset_bias);
1010     __ ldr(tmp1, recv_addr);
1011     __ cbnz(tmp1, next_test);
1012     __ str(recv, recv_addr);
1013     __ mov(tmp1, DataLayout::counter_increment);
1014     __ str(tmp1, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)) -
1015                          mdo_offset_bias));
1016     __ b(*update_done);
1017     __ bind(next_test);
1018   }
1019 }
1020 
1021 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
1022                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
1023   md = method->method_data_or_null();
1024   assert(md != nullptr, "Sanity");
1025   data = md->bci_to_data(bci);
1026   assert(data != nullptr,       "need data for checkcast");
1027   assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1028   if (md->byte_offset_of_slot(data, DataLayout::header_offset()) + data->size_in_bytes() >= 4096) {
1029     // The offset is large so bias the mdo by the base of the slot so
1030     // that the ldr can use an immediate offset to reference the slots of the data
1031     mdo_offset_bias = md->byte_offset_of_slot(data, DataLayout::header_offset());
1032   }
1033 }
1034 
1035 // On 32-bit ARM, code before this helper should test obj for null (ZF should be set if obj is null).
1036 void LIR_Assembler::typecheck_profile_helper1(ciMethod* method, int bci,
1037                                               ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias,
1038                                               Register obj, Register mdo, Register data_val, Label* obj_is_null) {
1039   assert(method != nullptr, "Should have method");
1040   assert_different_registers(obj, mdo, data_val);
1041   setup_md_access(method, bci, md, data, mdo_offset_bias);
1042   Label not_null;
1043   __ b(not_null, ne);
1044   __ mov_metadata(mdo, md->constant_encoding());
1045   if (mdo_offset_bias > 0) {
1046     __ mov_slow(data_val, mdo_offset_bias);
1047     __ add(mdo, mdo, data_val);
1048   }
1049   Address flags_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()) - mdo_offset_bias);
1050   __ ldrb(data_val, flags_addr);
1051   __ orr(data_val, data_val, (uint)BitData::null_seen_byte_constant());
1052   __ strb(data_val, flags_addr);
1053   __ b(*obj_is_null);
1054   __ bind(not_null);
1055 }
1056 
1057 void LIR_Assembler::typecheck_profile_helper2(ciMethodData* md, ciProfileData* data, int mdo_offset_bias,
1058                                               Register mdo, Register recv, Register value, Register tmp1,
1059                                               Label* profile_cast_success, Label* profile_cast_failure,
1060                                               Label* success, Label* failure) {
1061   assert_different_registers(mdo, value, tmp1);
1062   __ bind(*profile_cast_success);
1063   __ mov_metadata(mdo, md->constant_encoding());
1064   if (mdo_offset_bias > 0) {
1065     __ mov_slow(tmp1, mdo_offset_bias);
1066     __ add(mdo, mdo, tmp1);
1067   }
1068   __ load_klass(recv, value);
1069   type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, success);
1070   __ b(*success);
1071   // Cast failure case
1072   __ bind(*profile_cast_failure);
1073   __ mov_metadata(mdo, md->constant_encoding());
1074   if (mdo_offset_bias > 0) {
1075     __ mov_slow(tmp1, mdo_offset_bias);
1076     __ add(mdo, mdo, tmp1);
1077   }
1078   Address data_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
1079   __ ldr(tmp1, data_addr);
1080   __ sub(tmp1, tmp1, DataLayout::counter_increment);
1081   __ str(tmp1, data_addr);
1082   __ b(*failure);
1083 }
1084 
1085 // Sets `res` to true, if `cond` holds.
1086 static void set_instanceof_result(MacroAssembler* _masm, Register res, AsmCondition cond) {
1087   __ mov(res, 1, cond);
1088 }
1089 
1090 
1091 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1092   // TODO: ARM - can be more effective with one more register
1093   switch (op->code()) {
1094     case lir_store_check: {
1095       CodeStub* stub = op->stub();
1096       Register value = op->object()->as_register();
1097       Register array = op->array()->as_register();
1098       Register klass_RInfo = op->tmp1()->as_register();
1099       Register k_RInfo = op->tmp2()->as_register();
1100       assert_different_registers(klass_RInfo, k_RInfo, Rtemp);
1101       if (op->should_profile()) {
1102         assert_different_registers(value, klass_RInfo, k_RInfo, Rtemp);
1103       }
1104 
1105       // check if it needs to be profiled
1106       ciMethodData* md;
1107       ciProfileData* data;
1108       int mdo_offset_bias = 0;
1109       Label profile_cast_success, profile_cast_failure, done;
1110       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1111       Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1112 
1113       if (op->should_profile()) {
1114         __ cmp(value, 0);
1115         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, value, k_RInfo, Rtemp, &done);
1116       } else {
1117         __ cbz(value, done);
1118       }
1119       assert_different_registers(k_RInfo, value);
1120       add_debug_info_for_null_check_here(op->info_for_exception());
1121       __ load_klass(k_RInfo, array);
1122       __ load_klass(klass_RInfo, value);
1123       __ ldr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
1124       __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1125       // check for immediate positive hit
1126       __ ldr(Rtemp, Address(klass_RInfo, Rtemp));
1127       __ cmp(klass_RInfo, k_RInfo);
1128       __ cond_cmp(Rtemp, k_RInfo, ne);
1129       __ b(*success_target, eq);
1130       // check for immediate negative hit
1131       __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1132       __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1133       __ b(*failure_target, ne);
1134       // slow case
1135       assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1136       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1137       __ cbz(R0, *failure_target);
1138       if (op->should_profile()) {
1139         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1140         if (mdo == value) {
1141           mdo = k_RInfo;
1142           recv = klass_RInfo;
1143         }
1144         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, value, tmp1,
1145                                   &profile_cast_success, &profile_cast_failure,
1146                                   &done, stub->entry());
1147       }
1148       __ bind(done);
1149       break;
1150     }
1151 
1152     case lir_checkcast: {
1153       CodeStub* stub = op->stub();
1154       Register obj = op->object()->as_register();
1155       Register res = op->result_opr()->as_register();
1156       Register klass_RInfo = op->tmp1()->as_register();
1157       Register k_RInfo = op->tmp2()->as_register();
1158       ciKlass* k = op->klass();
1159       assert_different_registers(res, k_RInfo, klass_RInfo, Rtemp);
1160 
1161       if (stub->is_simple_exception_stub()) {
1162       // TODO: ARM - Late binding is used to prevent confusion of register allocator
1163       assert(stub->is_exception_throw_stub(), "must be");
1164       ((SimpleExceptionStub*)stub)->set_obj(op->result_opr());
1165       }
1166       ciMethodData* md;
1167       ciProfileData* data;
1168       int mdo_offset_bias = 0;
1169 
1170       Label done;
1171 
1172       Label profile_cast_failure, profile_cast_success;
1173       Label *failure_target = op->should_profile() ? &profile_cast_failure : op->stub()->entry();
1174       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1175 
1176 
1177       __ movs(res, obj);
1178       if (op->should_profile()) {
1179         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, res, klass_RInfo, Rtemp, &done);
1180       } else {
1181         __ b(done, eq);
1182       }
1183       if (k->is_loaded()) {
1184         __ mov_metadata(k_RInfo, k->constant_encoding());
1185       } else if (k_RInfo != obj) {
1186         klass2reg_with_patching(k_RInfo, op->info_for_patch());
1187         __ movs(res, obj);
1188       } else {
1189         // Patching doesn't update "res" register after GC, so do patching first
1190         klass2reg_with_patching(Rtemp, op->info_for_patch());
1191         __ movs(res, obj);
1192         __ mov(k_RInfo, Rtemp);
1193       }
1194       __ load_klass(klass_RInfo, res, ne);
1195 
1196       if (op->fast_check()) {
1197         __ cmp(klass_RInfo, k_RInfo, ne);
1198         __ b(*failure_target, ne);
1199       } else if (k->is_loaded()) {
1200         __ b(*success_target, eq);
1201         __ ldr(Rtemp, Address(klass_RInfo, k->super_check_offset()));
1202         if (in_bytes(Klass::secondary_super_cache_offset()) != (int) k->super_check_offset()) {
1203           __ cmp(Rtemp, k_RInfo);
1204           __ b(*failure_target, ne);
1205         } else {
1206           __ cmp(klass_RInfo, k_RInfo);
1207           __ cmp(Rtemp, k_RInfo, ne);
1208           __ b(*success_target, eq);
1209           assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1210           __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1211           __ cbz(R0, *failure_target);
1212         }
1213       } else {
1214         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1215         __ b(*success_target, eq);
1216         // check for immediate positive hit
1217         __ ldr(Rtemp, Address(klass_RInfo, Rtemp));
1218         __ cmp(klass_RInfo, k_RInfo);
1219         __ cmp(Rtemp, k_RInfo, ne);
1220         __ b(*success_target, eq);
1221         // check for immediate negative hit
1222         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1223         __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1224         __ b(*failure_target, ne);
1225         // slow case
1226         assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1227         __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1228         __ cbz(R0, *failure_target);
1229       }
1230 
1231       if (op->should_profile()) {
1232         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1233         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, res, tmp1,
1234                                   &profile_cast_success, &profile_cast_failure,
1235                                   &done, stub->entry());
1236       }
1237       __ bind(done);
1238       break;
1239     }
1240 
1241     case lir_instanceof: {
1242       Register obj = op->object()->as_register();
1243       Register res = op->result_opr()->as_register();
1244       Register klass_RInfo = op->tmp1()->as_register();
1245       Register k_RInfo = op->tmp2()->as_register();
1246       ciKlass* k = op->klass();
1247       assert_different_registers(res, klass_RInfo, k_RInfo, Rtemp);
1248 
1249       ciMethodData* md;
1250       ciProfileData* data;
1251       int mdo_offset_bias = 0;
1252 
1253       Label done;
1254 
1255       Label profile_cast_failure, profile_cast_success;
1256       Label *failure_target = op->should_profile() ? &profile_cast_failure : &done;
1257       Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1258 
1259       __ movs(res, obj);
1260 
1261       if (op->should_profile()) {
1262         typecheck_profile_helper1(op->profiled_method(), op->profiled_bci(), md, data, mdo_offset_bias, res, klass_RInfo, Rtemp, &done);
1263       } else {
1264         __ b(done, eq);
1265       }
1266 
1267       if (k->is_loaded()) {
1268         __ mov_metadata(k_RInfo, k->constant_encoding());
1269       } else {
1270         op->info_for_patch()->add_register_oop(FrameMap::as_oop_opr(res));
1271         klass2reg_with_patching(k_RInfo, op->info_for_patch());
1272       }
1273       __ load_klass(klass_RInfo, res);
1274 
1275       if (!op->should_profile()) {
1276         __ mov(res, 0);
1277       }
1278 
1279       if (op->fast_check()) {
1280         __ cmp(klass_RInfo, k_RInfo);
1281         if (!op->should_profile()) {
1282           set_instanceof_result(_masm, res, eq);
1283         } else {
1284           __ b(profile_cast_failure, ne);
1285         }
1286       } else if (k->is_loaded()) {
1287         __ ldr(Rtemp, Address(klass_RInfo, k->super_check_offset()));
1288         if (in_bytes(Klass::secondary_super_cache_offset()) != (int) k->super_check_offset()) {
1289           __ cmp(Rtemp, k_RInfo);
1290           if (!op->should_profile()) {
1291             set_instanceof_result(_masm, res, eq);
1292           } else {
1293             __ b(profile_cast_failure, ne);
1294           }
1295         } else {
1296           __ cmp(klass_RInfo, k_RInfo);
1297           __ cond_cmp(Rtemp, k_RInfo, ne);
1298           if (!op->should_profile()) {
1299             set_instanceof_result(_masm, res, eq);
1300           }
1301           __ b(*success_target, eq);
1302           assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1303           __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1304           if (!op->should_profile()) {
1305             move_regs(R0, res);
1306           } else {
1307             __ cbz(R0, *failure_target);
1308           }
1309         }
1310       } else {
1311         __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1312         // check for immediate positive hit
1313         __ cmp(klass_RInfo, k_RInfo);
1314         if (!op->should_profile()) {
1315           __ ldr(res, Address(klass_RInfo, Rtemp), ne);
1316           __ cond_cmp(res, k_RInfo, ne);
1317           set_instanceof_result(_masm, res, eq);
1318         } else {
1319           __ ldr(Rtemp, Address(klass_RInfo, Rtemp), ne);
1320           __ cond_cmp(Rtemp, k_RInfo, ne);
1321         }
1322         __ b(*success_target, eq);
1323         // check for immediate negative hit
1324         if (op->should_profile()) {
1325           __ ldr_u32(Rtemp, Address(k_RInfo, Klass::super_check_offset_offset()));
1326         }
1327         __ cmp(Rtemp, in_bytes(Klass::secondary_super_cache_offset()));
1328         if (!op->should_profile()) {
1329           __ mov(res, 0, ne);
1330         }
1331         __ b(*failure_target, ne);
1332         // slow case
1333         assert(klass_RInfo == R0 && k_RInfo == R1, "runtime call setup");
1334         __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type);
1335         if (!op->should_profile()) {
1336           move_regs(R0, res);
1337         }
1338         if (op->should_profile()) {
1339           __ cbz(R0, *failure_target);
1340         }
1341       }
1342 
1343       if (op->should_profile()) {
1344         Label done_ok, done_failure;
1345         Register mdo  = klass_RInfo, recv = k_RInfo, tmp1 = Rtemp;
1346         typecheck_profile_helper2(md, data, mdo_offset_bias, mdo, recv, res, tmp1,
1347                                   &profile_cast_success, &profile_cast_failure,
1348                                   &done_ok, &done_failure);
1349         __ bind(done_failure);
1350         __ mov(res, 0);
1351         __ b(done);
1352         __ bind(done_ok);
1353         __ mov(res, 1);
1354       }
1355       __ bind(done);
1356       break;
1357     }
1358     default:
1359       ShouldNotReachHere();
1360   }
1361 }
1362 
1363 
1364 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1365   //   if (*addr == cmpval) {
1366   //     *addr = newval;
1367   //     dest = 1;
1368   //   } else {
1369   //     dest = 0;
1370   //   }
1371   // FIXME: membar_release
1372   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
1373   Register addr = op->addr()->is_register() ?
1374     op->addr()->as_pointer_register() :
1375     op->addr()->as_address_ptr()->base()->as_pointer_register();
1376   assert(op->addr()->is_register() || op->addr()->as_address_ptr()->disp() == 0, "unexpected disp");
1377   assert(op->addr()->is_register() || op->addr()->as_address_ptr()->index() == LIR_Opr::illegalOpr(), "unexpected index");
1378   if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
1379     Register cmpval = op->cmp_value()->as_register();
1380     Register newval = op->new_value()->as_register();
1381     Register dest = op->result_opr()->as_register();
1382     assert_different_registers(dest, addr, cmpval, newval, Rtemp);
1383 
1384     __ atomic_cas_bool(cmpval, newval, addr, 0, Rtemp); // Rtemp free by default at C1 LIR layer
1385     __ mov(dest, 1, eq);
1386     __ mov(dest, 0, ne);
1387   } else if (op->code() == lir_cas_long) {
1388     assert(VM_Version::supports_cx8(), "wrong machine");
1389     Register cmp_value_lo = op->cmp_value()->as_register_lo();
1390     Register cmp_value_hi = op->cmp_value()->as_register_hi();
1391     Register new_value_lo = op->new_value()->as_register_lo();
1392     Register new_value_hi = op->new_value()->as_register_hi();
1393     Register dest = op->result_opr()->as_register();
1394     Register tmp_lo = op->tmp1()->as_register_lo();
1395     Register tmp_hi = op->tmp1()->as_register_hi();
1396 
1397     assert_different_registers(tmp_lo, tmp_hi, cmp_value_lo, cmp_value_hi, dest, new_value_lo, new_value_hi, addr);
1398     assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
1399     assert(new_value_hi->encoding() == new_value_lo->encoding() + 1, "non aligned register pair");
1400     assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
1401     assert((new_value_lo->encoding() & 0x1) == 0, "misaligned register pair");
1402     __ atomic_cas64(tmp_lo, tmp_hi, dest, cmp_value_lo, cmp_value_hi,
1403                     new_value_lo, new_value_hi, addr, 0);
1404   } else {
1405     Unimplemented();
1406   }
1407   // FIXME: is full membar really needed instead of just membar_acquire?
1408   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
1409 }
1410 
1411 
1412 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1413                           LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1414   assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on arm");
1415 
1416   AsmCondition acond = al;
1417   AsmCondition ncond = nv;
1418   if (opr1 != opr2) {
1419     switch (condition) {
1420       case lir_cond_equal:        acond = eq; ncond = ne; break;
1421       case lir_cond_notEqual:     acond = ne; ncond = eq; break;
1422       case lir_cond_less:         acond = lt; ncond = ge; break;
1423       case lir_cond_lessEqual:    acond = le; ncond = gt; break;
1424       case lir_cond_greaterEqual: acond = ge; ncond = lt; break;
1425       case lir_cond_greater:      acond = gt; ncond = le; break;
1426       case lir_cond_aboveEqual:   acond = hs; ncond = lo; break;
1427       case lir_cond_belowEqual:   acond = ls; ncond = hi; break;
1428       default: ShouldNotReachHere();
1429     }
1430   }
1431 
1432   for (;;) {                         // two iterations only
1433     if (opr1 == result) {
1434       // do nothing
1435     } else if (opr1->is_single_cpu()) {
1436       __ mov(result->as_register(), opr1->as_register(), acond);
1437     } else if (opr1->is_double_cpu()) {
1438       __ long_move(result->as_register_lo(), result->as_register_hi(),
1439                    opr1->as_register_lo(), opr1->as_register_hi(), acond);
1440     } else if (opr1->is_single_stack()) {
1441       __ ldr(result->as_register(), frame_map()->address_for_slot(opr1->single_stack_ix()), acond);
1442     } else if (opr1->is_double_stack()) {
1443       __ ldr(result->as_register_lo(),
1444              frame_map()->address_for_slot(opr1->double_stack_ix(), lo_word_offset_in_bytes), acond);
1445       __ ldr(result->as_register_hi(),
1446              frame_map()->address_for_slot(opr1->double_stack_ix(), hi_word_offset_in_bytes), acond);
1447     } else if (opr1->is_illegal()) {
1448       // do nothing: this part of the cmove has been optimized away in the peephole optimizer
1449     } else {
1450       assert(opr1->is_constant(), "must be");
1451       LIR_Const* c = opr1->as_constant_ptr();
1452 
1453       switch (c->type()) {
1454         case T_INT:
1455           __ mov_slow(result->as_register(), c->as_jint(), acond);
1456           break;
1457         case T_LONG:
1458           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1459           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1460           break;
1461         case T_OBJECT:
1462           __ mov_oop(result->as_register(), c->as_jobject(), 0, acond);
1463           break;
1464         case T_FLOAT:
1465 #ifdef __SOFTFP__
1466           // not generated now.
1467           __ mov_slow(result->as_register(), c->as_jint(), acond);
1468 #else
1469           __ mov_float(result->as_float_reg(), c->as_jfloat(), acond);
1470 #endif // __SOFTFP__
1471           break;
1472         case T_DOUBLE:
1473 #ifdef __SOFTFP__
1474           // not generated now.
1475           __ mov_slow(result->as_register_lo(), c->as_jint_lo(), acond);
1476           __ mov_slow(result->as_register_hi(), c->as_jint_hi(), acond);
1477 #else
1478           __ mov_double(result->as_double_reg(), c->as_jdouble(), acond);
1479 #endif // __SOFTFP__
1480           break;
1481         case T_METADATA:
1482           __ mov_metadata(result->as_register(), c->as_metadata(), acond);
1483           break;
1484         default:
1485           ShouldNotReachHere();
1486       }
1487     }
1488 
1489     // Negate the condition and repeat the algorithm with the second operand
1490     if (opr1 == opr2) { break; }
1491     opr1 = opr2;
1492     acond = ncond;
1493   }
1494 }
1495 
1496 #ifdef ASSERT
1497 static int reg_size(LIR_Opr op) {
1498   switch (op->type()) {
1499   case T_FLOAT:
1500   case T_INT:      return BytesPerInt;
1501   case T_LONG:
1502   case T_DOUBLE:   return BytesPerLong;
1503   case T_OBJECT:
1504   case T_ARRAY:
1505   case T_METADATA: return BytesPerWord;
1506   case T_ADDRESS:
1507   case T_ILLEGAL:  // fall through
1508   default: ShouldNotReachHere(); return -1;
1509   }
1510 }
1511 #endif
1512 
1513 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1514   assert(info == nullptr, "unused on this code path");
1515   assert(dest->is_register(), "wrong items state");
1516 
1517   if (right->is_address()) {
1518     // special case for adding shifted/extended register
1519     const Register res = dest->as_pointer_register();
1520     const Register lreg = left->as_pointer_register();
1521     const LIR_Address* addr = right->as_address_ptr();
1522 
1523     assert(addr->base()->as_pointer_register() == lreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1524 
1525     int scale = addr->scale();
1526     AsmShift shift = lsl;
1527 
1528 
1529     assert(reg_size(addr->base()) == reg_size(addr->index()), "should be");
1530     assert(reg_size(addr->base()) == reg_size(dest), "should be");
1531     assert(reg_size(dest) == wordSize, "should be");
1532 
1533     AsmOperand operand(addr->index()->as_pointer_register(), shift, scale);
1534     switch (code) {
1535       case lir_add: __ add(res, lreg, operand); break;
1536       case lir_sub: __ sub(res, lreg, operand); break;
1537       default: ShouldNotReachHere();
1538     }
1539 
1540   } else if (left->is_address()) {
1541     assert(code == lir_sub && right->is_single_cpu(), "special case used by strength_reduce_multiply()");
1542     const LIR_Address* addr = left->as_address_ptr();
1543     const Register res = dest->as_register();
1544     const Register rreg = right->as_register();
1545     assert(addr->base()->as_register() == rreg && addr->index()->is_register() && addr->disp() == 0, "must be");
1546     __ rsb(res, rreg, AsmOperand(addr->index()->as_register(), lsl, addr->scale()));
1547 
1548   } else if (dest->is_single_cpu()) {
1549     assert(left->is_single_cpu(), "unexpected left operand");
1550 
1551     const Register res = dest->as_register();
1552     const Register lreg = left->as_register();
1553 
1554     if (right->is_single_cpu()) {
1555       const Register rreg = right->as_register();
1556       switch (code) {
1557         case lir_add: __ add_32(res, lreg, rreg); break;
1558         case lir_sub: __ sub_32(res, lreg, rreg); break;
1559         case lir_mul: __ mul_32(res, lreg, rreg); break;
1560         default: ShouldNotReachHere();
1561       }
1562     } else {
1563       assert(right->is_constant(), "must be");
1564       const jint c = right->as_constant_ptr()->as_jint();
1565       if (!Assembler::is_arith_imm_in_range(c)) {
1566         BAILOUT("illegal arithmetic operand");
1567       }
1568       switch (code) {
1569         case lir_add: __ add_32(res, lreg, c); break;
1570         case lir_sub: __ sub_32(res, lreg, c); break;
1571         default: ShouldNotReachHere();
1572       }
1573     }
1574 
1575   } else if (dest->is_double_cpu()) {
1576     Register res_lo = dest->as_register_lo();
1577     Register res_hi = dest->as_register_hi();
1578     Register lreg_lo = left->as_register_lo();
1579     Register lreg_hi = left->as_register_hi();
1580     if (right->is_double_cpu()) {
1581       Register rreg_lo = right->as_register_lo();
1582       Register rreg_hi = right->as_register_hi();
1583       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1584         res_lo = Rtemp;
1585       }
1586       switch (code) {
1587         case lir_add:
1588           __ adds(res_lo, lreg_lo, rreg_lo);
1589           __ adc(res_hi, lreg_hi, rreg_hi);
1590           break;
1591         case lir_sub:
1592           __ subs(res_lo, lreg_lo, rreg_lo);
1593           __ sbc(res_hi, lreg_hi, rreg_hi);
1594           break;
1595         default:
1596           ShouldNotReachHere();
1597       }
1598     } else {
1599       assert(right->is_constant(), "must be");
1600       assert((right->as_constant_ptr()->as_jlong() >> 32) == 0, "out of range");
1601       const jint c = (jint) right->as_constant_ptr()->as_jlong();
1602       if (res_lo == lreg_hi) {
1603         res_lo = Rtemp;
1604       }
1605       switch (code) {
1606         case lir_add:
1607           __ adds(res_lo, lreg_lo, c);
1608           __ adc(res_hi, lreg_hi, 0);
1609           break;
1610         case lir_sub:
1611           __ subs(res_lo, lreg_lo, c);
1612           __ sbc(res_hi, lreg_hi, 0);
1613           break;
1614         default:
1615           ShouldNotReachHere();
1616       }
1617     }
1618     move_regs(res_lo, dest->as_register_lo());
1619 
1620   } else if (dest->is_single_fpu()) {
1621     assert(left->is_single_fpu(), "must be");
1622     assert(right->is_single_fpu(), "must be");
1623     const FloatRegister res = dest->as_float_reg();
1624     const FloatRegister lreg = left->as_float_reg();
1625     const FloatRegister rreg = right->as_float_reg();
1626     switch (code) {
1627       case lir_add: __ add_float(res, lreg, rreg); break;
1628       case lir_sub: __ sub_float(res, lreg, rreg); break;
1629       case lir_mul: __ mul_float(res, lreg, rreg); break;
1630       case lir_div: __ div_float(res, lreg, rreg); break;
1631       default: ShouldNotReachHere();
1632     }
1633   } else if (dest->is_double_fpu()) {
1634     assert(left->is_double_fpu(), "must be");
1635     assert(right->is_double_fpu(), "must be");
1636     const FloatRegister res = dest->as_double_reg();
1637     const FloatRegister lreg = left->as_double_reg();
1638     const FloatRegister rreg = right->as_double_reg();
1639     switch (code) {
1640       case lir_add: __ add_double(res, lreg, rreg); break;
1641       case lir_sub: __ sub_double(res, lreg, rreg); break;
1642       case lir_mul: __ mul_double(res, lreg, rreg); break;
1643       case lir_div: __ div_double(res, lreg, rreg); break;
1644       default: ShouldNotReachHere();
1645     }
1646   } else {
1647     ShouldNotReachHere();
1648   }
1649 }
1650 
1651 
1652 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
1653   switch (code) {
1654     case lir_abs:
1655       __ abs_double(dest->as_double_reg(), value->as_double_reg());
1656       break;
1657     case lir_sqrt:
1658       __ sqrt_double(dest->as_double_reg(), value->as_double_reg());
1659       break;
1660     default:
1661       ShouldNotReachHere();
1662   }
1663 }
1664 
1665 
1666 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest) {
1667   assert(dest->is_register(), "wrong items state");
1668   assert(left->is_register(), "wrong items state");
1669 
1670   if (dest->is_single_cpu()) {
1671 
1672     const Register res = dest->as_register();
1673     const Register lreg = left->as_register();
1674 
1675     if (right->is_single_cpu()) {
1676       const Register rreg = right->as_register();
1677       switch (code) {
1678         case lir_logic_and: __ and_32(res, lreg, rreg); break;
1679         case lir_logic_or:  __ orr_32(res, lreg, rreg); break;
1680         case lir_logic_xor: __ eor_32(res, lreg, rreg); break;
1681         default: ShouldNotReachHere();
1682       }
1683     } else {
1684       assert(right->is_constant(), "must be");
1685       const uint c = (uint)right->as_constant_ptr()->as_jint();
1686       if (!Assembler::is_arith_imm_in_range(c)) {
1687         BAILOUT("illegal arithmetic operand");
1688       }
1689       switch (code) {
1690         case lir_logic_and: __ and_32(res, lreg, c); break;
1691         case lir_logic_or:  __ orr_32(res, lreg, c); break;
1692         case lir_logic_xor: __ eor_32(res, lreg, c); break;
1693         default: ShouldNotReachHere();
1694       }
1695     }
1696   } else {
1697     assert(dest->is_double_cpu(), "should be");
1698     Register res_lo = dest->as_register_lo();
1699 
1700     assert (dest->type() == T_LONG, "unexpected result type");
1701     assert (left->type() == T_LONG, "unexpected left type");
1702     assert (right->type() == T_LONG, "unexpected right type");
1703 
1704     const Register res_hi = dest->as_register_hi();
1705     const Register lreg_lo = left->as_register_lo();
1706     const Register lreg_hi = left->as_register_hi();
1707 
1708     if (right->is_register()) {
1709       const Register rreg_lo = right->as_register_lo();
1710       const Register rreg_hi = right->as_register_hi();
1711       if (res_lo == lreg_hi || res_lo == rreg_hi) {
1712         res_lo = Rtemp; // Temp register helps to avoid overlap between result and input
1713       }
1714       switch (code) {
1715         case lir_logic_and:
1716           __ andr(res_lo, lreg_lo, rreg_lo);
1717           __ andr(res_hi, lreg_hi, rreg_hi);
1718           break;
1719         case lir_logic_or:
1720           __ orr(res_lo, lreg_lo, rreg_lo);
1721           __ orr(res_hi, lreg_hi, rreg_hi);
1722           break;
1723         case lir_logic_xor:
1724           __ eor(res_lo, lreg_lo, rreg_lo);
1725           __ eor(res_hi, lreg_hi, rreg_hi);
1726           break;
1727         default:
1728           ShouldNotReachHere();
1729       }
1730       move_regs(res_lo, dest->as_register_lo());
1731     } else {
1732       assert(right->is_constant(), "must be");
1733       const jint c_lo = (jint) right->as_constant_ptr()->as_jlong();
1734       const jint c_hi = (jint) (right->as_constant_ptr()->as_jlong() >> 32);
1735       // Case for logic_or from do_ClassIDIntrinsic()
1736       if (c_hi == 0 && AsmOperand::is_rotated_imm(c_lo)) {
1737         switch (code) {
1738           case lir_logic_and:
1739             __ andr(res_lo, lreg_lo, c_lo);
1740             __ mov(res_hi, 0);
1741             break;
1742           case lir_logic_or:
1743             __ orr(res_lo, lreg_lo, c_lo);
1744             break;
1745           case lir_logic_xor:
1746             __ eor(res_lo, lreg_lo, c_lo);
1747             break;
1748         default:
1749           ShouldNotReachHere();
1750         }
1751       } else if (code == lir_logic_and &&
1752                  c_hi == -1 &&
1753                  (AsmOperand::is_rotated_imm(c_lo) ||
1754                   AsmOperand::is_rotated_imm(~c_lo))) {
1755         // Another case which handles logic_and from do_ClassIDIntrinsic()
1756         if (AsmOperand::is_rotated_imm(c_lo)) {
1757           __ andr(res_lo, lreg_lo, c_lo);
1758         } else {
1759           __ bic(res_lo, lreg_lo, ~c_lo);
1760         }
1761         if (res_hi != lreg_hi) {
1762           __ mov(res_hi, lreg_hi);
1763         }
1764       } else {
1765         BAILOUT("64 bit constant cannot be inlined");
1766       }
1767     }
1768   }
1769 }
1770 
1771 
1772 
1773 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1774   if (opr1->is_single_cpu()) {
1775     if (opr2->is_constant()) {
1776       switch (opr2->as_constant_ptr()->type()) {
1777         case T_INT: {
1778           const jint c = opr2->as_constant_ptr()->as_jint();
1779           if (Assembler::is_arith_imm_in_range(c)) {
1780             __ cmp_32(opr1->as_register(), c);
1781           } else if (Assembler::is_arith_imm_in_range(-c)) {
1782             __ cmn_32(opr1->as_register(), -c);
1783           } else {
1784             // This can happen when compiling lookupswitch
1785             __ mov_slow(Rtemp, c);
1786             __ cmp_32(opr1->as_register(), Rtemp);
1787           }
1788           break;
1789         }
1790         case T_OBJECT:
1791           assert(opr2->as_constant_ptr()->as_jobject() == nullptr, "cannot handle otherwise");
1792           __ cmp(opr1->as_register(), 0);
1793           break;
1794         case T_METADATA:
1795           assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "Only equality tests");
1796           assert(opr2->as_constant_ptr()->as_metadata() == nullptr, "cannot handle otherwise");
1797           __ cmp(opr1->as_register(), 0);
1798           break;
1799         default:
1800           ShouldNotReachHere();
1801       }
1802     } else if (opr2->is_single_cpu()) {
1803       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1804         assert(opr2->type() == T_OBJECT || opr2->type() == T_ARRAY, "incompatibe type");
1805         __ cmpoop(opr1->as_register(), opr2->as_register());
1806       } else if (opr1->type() == T_METADATA || opr1->type() == T_ADDRESS) {
1807         assert(opr2->type() == T_METADATA || opr2->type() == T_ADDRESS, "incompatibe type");
1808         __ cmp(opr1->as_register(), opr2->as_register());
1809       } else {
1810         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY && opr2->type() != T_METADATA && opr2->type() != T_ADDRESS, "incompatibe type");
1811         __ cmp_32(opr1->as_register(), opr2->as_register());
1812       }
1813     } else {
1814       ShouldNotReachHere();
1815     }
1816   } else if (opr1->is_double_cpu()) {
1817     Register xlo = opr1->as_register_lo();
1818     Register xhi = opr1->as_register_hi();
1819     if (opr2->is_constant() && opr2->as_jlong() == 0) {
1820       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "cannot handle otherwise");
1821       __ orrs(Rtemp, xlo, xhi);
1822     } else if (opr2->is_register()) {
1823       Register ylo = opr2->as_register_lo();
1824       Register yhi = opr2->as_register_hi();
1825       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
1826         __ teq(xhi, yhi);
1827         __ teq(xlo, ylo, eq);
1828       } else {
1829         __ subs(Rtemp, xlo, ylo);
1830         __ sbcs(Rtemp, xhi, yhi);
1831       }
1832     } else {
1833       ShouldNotReachHere();
1834     }
1835   } else if (opr1->is_single_fpu()) {
1836     if (opr2->is_constant()) {
1837       assert(opr2->as_jfloat() == 0.0f, "cannot handle otherwise");
1838       __ cmp_zero_float(opr1->as_float_reg());
1839     } else {
1840       __ cmp_float(opr1->as_float_reg(), opr2->as_float_reg());
1841     }
1842   } else if (opr1->is_double_fpu()) {
1843     if (opr2->is_constant()) {
1844       assert(opr2->as_jdouble() == 0.0, "cannot handle otherwise");
1845       __ cmp_zero_double(opr1->as_double_reg());
1846     } else {
1847       __ cmp_double(opr1->as_double_reg(), opr2->as_double_reg());
1848     }
1849   } else {
1850     ShouldNotReachHere();
1851   }
1852 }
1853 
1854 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1855   const Register res = dst->as_register();
1856   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1857     comp_op(lir_cond_unknown, left, right, op);
1858     __ fmstat();
1859     if (code == lir_ucmp_fd2i) {  // unordered is less
1860       __ mvn(res, 0, lt);
1861       __ mov(res, 1, ge);
1862     } else {                      // unordered is greater
1863       __ mov(res, 1, cs);
1864       __ mvn(res, 0, cc);
1865     }
1866     __ mov(res, 0, eq);
1867 
1868   } else {
1869     assert(code == lir_cmp_l2i, "must be");
1870 
1871     Label done;
1872     const Register xlo = left->as_register_lo();
1873     const Register xhi = left->as_register_hi();
1874     const Register ylo = right->as_register_lo();
1875     const Register yhi = right->as_register_hi();
1876     __ cmp(xhi, yhi);
1877     __ mov(res, 1, gt);
1878     __ mvn(res, 0, lt);
1879     __ b(done, ne);
1880     __ subs(res, xlo, ylo);
1881     __ mov(res, 1, hi);
1882     __ mvn(res, 0, lo);
1883     __ bind(done);
1884   }
1885 }
1886 
1887 
1888 void LIR_Assembler::align_call(LIR_Code code) {
1889   // Not needed
1890 }
1891 
1892 
1893 void LIR_Assembler::call(LIR_OpJavaCall *op, relocInfo::relocType rtype) {
1894   int ret_addr_offset = __ patchable_call(op->addr(), rtype);
1895   assert(ret_addr_offset == __ offset(), "embedded return address not allowed");
1896   add_call_info_here(op->info());
1897 }
1898 
1899 
1900 void LIR_Assembler::ic_call(LIR_OpJavaCall *op) {
1901   bool near_range = __ cache_fully_reachable();
1902   address oop_address = pc();
1903 
1904   bool use_movw = VM_Version::supports_movw();
1905 
1906   // Ricklass may contain something that is not a metadata pointer so
1907   // mov_metadata can't be used
1908   InlinedAddress value((address)Universe::non_oop_word());
1909   InlinedAddress addr(op->addr());
1910   if (use_movw) {
1911     __ movw(Ricklass, ((unsigned int)Universe::non_oop_word()) & 0xffff);
1912     __ movt(Ricklass, ((unsigned int)Universe::non_oop_word()) >> 16);
1913   } else {
1914     // No movw/movt, must be load a pc relative value but no
1915     // relocation so no metadata table to load from.
1916     // Use a b instruction rather than a bl, inline constant after the
1917     // branch, use a PC relative ldr to load the constant, arrange for
1918     // the call to return after the constant(s).
1919     __ ldr_literal(Ricklass, value);
1920   }
1921   __ relocate(virtual_call_Relocation::spec(oop_address));
1922   if (near_range && use_movw) {
1923     __ bl(op->addr());
1924   } else {
1925     Label call_return;
1926     __ adr(LR, call_return);
1927     if (near_range) {
1928       __ b(op->addr());
1929     } else {
1930       __ indirect_jump(addr, Rtemp);
1931       __ bind_literal(addr);
1932     }
1933     if (!use_movw) {
1934       __ bind_literal(value);
1935     }
1936     __ bind(call_return);
1937   }
1938   add_call_info(code_offset(), op->info());
1939 }
1940 
1941 void LIR_Assembler::emit_static_call_stub() {
1942   address call_pc = __ pc();
1943   address stub = __ start_a_stub(call_stub_size());
1944   if (stub == nullptr) {
1945     BAILOUT("static call stub overflow");
1946   }
1947 
1948   DEBUG_ONLY(int offset = code_offset();)
1949 
1950   InlinedMetadata metadata_literal(nullptr);
1951   __ relocate(static_stub_Relocation::spec(call_pc));
1952   // If not a single instruction, NativeMovConstReg::next_instruction_address()
1953   // must jump over the whole following ldr_literal.
1954   // (See CompiledStaticCall::set_to_interpreted())
1955 #ifdef ASSERT
1956   address ldr_site = __ pc();
1957 #endif
1958   __ ldr_literal(Rmethod, metadata_literal);
1959   assert(nativeMovConstReg_at(ldr_site)->next_instruction_address() == __ pc(), "Fix ldr_literal or its parsing");
1960   bool near_range = __ cache_fully_reachable();
1961   InlinedAddress dest((address)-1);
1962   if (near_range) {
1963     address branch_site = __ pc();
1964     __ b(branch_site); // b to self maps to special NativeJump -1 destination
1965   } else {
1966     __ indirect_jump(dest, Rtemp);
1967   }
1968   __ bind_literal(metadata_literal); // includes spec_for_immediate reloc
1969   if (!near_range) {
1970     __ bind_literal(dest); // special NativeJump -1 destination
1971   }
1972 
1973   assert(code_offset() - offset <= call_stub_size(), "overflow");
1974   __ end_a_stub();
1975 }
1976 
1977 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1978   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1979   assert(exceptionPC->as_register()  == Rexception_pc, "must match");
1980   info->add_register_oop(exceptionOop);
1981 
1982   Runtime1::StubID handle_id = compilation()->has_fpu_code() ?
1983                                Runtime1::handle_exception_id :
1984                                Runtime1::handle_exception_nofpu_id;
1985   Label return_address;
1986   __ adr(Rexception_pc, return_address);
1987   __ call(Runtime1::entry_for(handle_id), relocInfo::runtime_call_type);
1988   __ bind(return_address);
1989   add_call_info_here(info);  // for exception handler
1990 }
1991 
1992 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1993   assert(exceptionOop->as_register() == Rexception_obj, "must match");
1994   __ b(_unwind_handler_entry);
1995 }
1996 
1997 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
1998   AsmShift shift = lsl;
1999   switch (code) {
2000     case lir_shl:  shift = lsl; break;
2001     case lir_shr:  shift = asr; break;
2002     case lir_ushr: shift = lsr; break;
2003     default: ShouldNotReachHere();
2004   }
2005 
2006   if (dest->is_single_cpu()) {
2007     __ andr(Rtemp, count->as_register(), 31);
2008     __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, Rtemp));
2009   } else if (dest->is_double_cpu()) {
2010     Register dest_lo = dest->as_register_lo();
2011     Register dest_hi = dest->as_register_hi();
2012     Register src_lo  = left->as_register_lo();
2013     Register src_hi  = left->as_register_hi();
2014     Register Rcount  = count->as_register();
2015     // Resolve possible register conflicts
2016     if (shift == lsl && dest_hi == src_lo) {
2017       dest_hi = Rtemp;
2018     } else if (shift != lsl && dest_lo == src_hi) {
2019       dest_lo = Rtemp;
2020     } else if (dest_lo == src_lo && dest_hi == src_hi) {
2021       dest_lo = Rtemp;
2022     } else if (dest_lo == Rcount || dest_hi == Rcount) {
2023       Rcount = Rtemp;
2024     }
2025     __ andr(Rcount, count->as_register(), 63);
2026     __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, Rcount);
2027     move_regs(dest_lo, dest->as_register_lo());
2028     move_regs(dest_hi, dest->as_register_hi());
2029   } else {
2030     ShouldNotReachHere();
2031   }
2032 }
2033 
2034 
2035 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2036   AsmShift shift = lsl;
2037   switch (code) {
2038     case lir_shl:  shift = lsl; break;
2039     case lir_shr:  shift = asr; break;
2040     case lir_ushr: shift = lsr; break;
2041     default: ShouldNotReachHere();
2042   }
2043 
2044   if (dest->is_single_cpu()) {
2045     count &= 31;
2046     if (count != 0) {
2047       __ mov(dest->as_register(), AsmOperand(left->as_register(), shift, count));
2048     } else {
2049       move_regs(left->as_register(), dest->as_register());
2050     }
2051   } else if (dest->is_double_cpu()) {
2052     count &= 63;
2053     if (count != 0) {
2054       Register dest_lo = dest->as_register_lo();
2055       Register dest_hi = dest->as_register_hi();
2056       Register src_lo  = left->as_register_lo();
2057       Register src_hi  = left->as_register_hi();
2058       // Resolve possible register conflicts
2059       if (shift == lsl && dest_hi == src_lo) {
2060         dest_hi = Rtemp;
2061       } else if (shift != lsl && dest_lo == src_hi) {
2062         dest_lo = Rtemp;
2063       }
2064       __ long_shift(dest_lo, dest_hi, src_lo, src_hi, shift, count);
2065       move_regs(dest_lo, dest->as_register_lo());
2066       move_regs(dest_hi, dest->as_register_hi());
2067     } else {
2068       __ long_move(dest->as_register_lo(), dest->as_register_hi(),
2069                    left->as_register_lo(), left->as_register_hi());
2070     }
2071   } else {
2072     ShouldNotReachHere();
2073   }
2074 }
2075 
2076 
2077 // Saves 4 given registers in reserved argument area.
2078 void LIR_Assembler::save_in_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2079   verify_reserved_argument_area_size(4);
2080   __ stmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4));
2081 }
2082 
2083 // Restores 4 given registers from reserved argument area.
2084 void LIR_Assembler::restore_from_reserved_area(Register r1, Register r2, Register r3, Register r4) {
2085   __ ldmia(SP, RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3) | RegisterSet(r4), no_writeback);
2086 }
2087 
2088 
2089 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2090   ciArrayKlass* default_type = op->expected_type();
2091   Register src = op->src()->as_register();
2092   Register src_pos = op->src_pos()->as_register();
2093   Register dst = op->dst()->as_register();
2094   Register dst_pos = op->dst_pos()->as_register();
2095   Register length  = op->length()->as_register();
2096   Register tmp = op->tmp()->as_register();
2097   Register tmp2 = Rtemp;
2098 
2099   assert(src == R0 && src_pos == R1 && dst == R2 && dst_pos == R3, "code assumption");
2100 
2101   CodeStub* stub = op->stub();
2102 
2103   int flags = op->flags();
2104   BasicType basic_type = default_type != nullptr ? default_type->element_type()->basic_type() : T_ILLEGAL;
2105   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2106 
2107   // If we don't know anything or it's an object array, just go through the generic arraycopy
2108   if (default_type == nullptr) {
2109 
2110     // save arguments, because they will be killed by a runtime call
2111     save_in_reserved_area(R0, R1, R2, R3);
2112 
2113     // pass length argument on SP[0]
2114     __ str(length, Address(SP, -2*wordSize, pre_indexed));  // 2 words for a proper stack alignment
2115 
2116     address copyfunc_addr = StubRoutines::generic_arraycopy();
2117     assert(copyfunc_addr != nullptr, "generic arraycopy stub required");
2118 #ifndef PRODUCT
2119     if (PrintC1Statistics) {
2120       __ inc_counter((address)&Runtime1::_generic_arraycopystub_cnt, tmp, tmp2);
2121     }
2122 #endif // !PRODUCT
2123     // the stub is in the code cache so close enough
2124     __ call(copyfunc_addr, relocInfo::runtime_call_type);
2125 
2126     __ add(SP, SP, 2*wordSize);
2127 
2128     __ cbz_32(R0, *stub->continuation());
2129 
2130     __ mvn_32(tmp, R0);
2131     restore_from_reserved_area(R0, R1, R2, R3);  // load saved arguments in slow case only
2132     __ sub_32(length, length, tmp);
2133     __ add_32(src_pos, src_pos, tmp);
2134     __ add_32(dst_pos, dst_pos, tmp);
2135 
2136     __ b(*stub->entry());
2137 
2138     __ bind(*stub->continuation());
2139     return;
2140   }
2141 
2142   assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(),
2143          "must be true at this point");
2144   int elem_size = type2aelembytes(basic_type);
2145   int shift = exact_log2(elem_size);
2146 
2147   // Check for null
2148   if (flags & LIR_OpArrayCopy::src_null_check) {
2149     if (flags & LIR_OpArrayCopy::dst_null_check) {
2150       __ cmp(src, 0);
2151       __ cond_cmp(dst, 0, ne);  // make one instruction shorter if both checks are needed
2152       __ b(*stub->entry(), eq);
2153     } else {
2154       __ cbz(src, *stub->entry());
2155     }
2156   } else if (flags & LIR_OpArrayCopy::dst_null_check) {
2157     __ cbz(dst, *stub->entry());
2158   }
2159 
2160   // If the compiler was not able to prove that exact type of the source or the destination
2161   // of the arraycopy is an array type, check at runtime if the source or the destination is
2162   // an instance type.
2163   if (flags & LIR_OpArrayCopy::type_check) {
2164     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::dst_objarray)) {
2165       __ load_klass(tmp, dst);
2166       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2167       __ mov_slow(tmp, Klass::_lh_neutral_value);
2168       __ cmp_32(tmp2, tmp);
2169       __ b(*stub->entry(), ge);
2170     }
2171 
2172     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::src_objarray)) {
2173       __ load_klass(tmp, src);
2174       __ ldr_u32(tmp2, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2175       __ mov_slow(tmp, Klass::_lh_neutral_value);
2176       __ cmp_32(tmp2, tmp);
2177       __ b(*stub->entry(), ge);
2178     }
2179   }
2180 
2181   // Check if negative
2182   const int all_positive_checks = LIR_OpArrayCopy::src_pos_positive_check |
2183                                   LIR_OpArrayCopy::dst_pos_positive_check |
2184                                   LIR_OpArrayCopy::length_positive_check;
2185   switch (flags & all_positive_checks) {
2186     case LIR_OpArrayCopy::src_pos_positive_check:
2187       __ branch_if_negative_32(src_pos, *stub->entry());
2188       break;
2189     case LIR_OpArrayCopy::dst_pos_positive_check:
2190       __ branch_if_negative_32(dst_pos, *stub->entry());
2191       break;
2192     case LIR_OpArrayCopy::length_positive_check:
2193       __ branch_if_negative_32(length, *stub->entry());
2194       break;
2195     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::dst_pos_positive_check:
2196       __ branch_if_any_negative_32(src_pos, dst_pos, tmp, *stub->entry());
2197       break;
2198     case LIR_OpArrayCopy::src_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2199       __ branch_if_any_negative_32(src_pos, length, tmp, *stub->entry());
2200       break;
2201     case LIR_OpArrayCopy::dst_pos_positive_check | LIR_OpArrayCopy::length_positive_check:
2202       __ branch_if_any_negative_32(dst_pos, length, tmp, *stub->entry());
2203       break;
2204     case all_positive_checks:
2205       __ branch_if_any_negative_32(src_pos, dst_pos, length, tmp, *stub->entry());
2206       break;
2207     default:
2208       assert((flags & all_positive_checks) == 0, "the last option");
2209   }
2210 
2211   // Range checks
2212   if (flags & LIR_OpArrayCopy::src_range_check) {
2213     __ ldr_s32(tmp2, Address(src, arrayOopDesc::length_offset_in_bytes()));
2214     __ add_32(tmp, src_pos, length);
2215     __ cmp_32(tmp, tmp2);
2216     __ b(*stub->entry(), hi);
2217   }
2218   if (flags & LIR_OpArrayCopy::dst_range_check) {
2219     __ ldr_s32(tmp2, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2220     __ add_32(tmp, dst_pos, length);
2221     __ cmp_32(tmp, tmp2);
2222     __ b(*stub->entry(), hi);
2223   }
2224 
2225   // Check if src and dst are of the same type
2226   if (flags & LIR_OpArrayCopy::type_check) {
2227     // We don't know the array types are compatible
2228     if (basic_type != T_OBJECT) {
2229       // Simple test for basic type arrays
2230       if (UseCompressedClassPointers) {
2231         // We don't need decode because we just need to compare
2232         __ ldr_u32(tmp, Address(src, oopDesc::klass_offset_in_bytes()));
2233         __ ldr_u32(tmp2, Address(dst, oopDesc::klass_offset_in_bytes()));
2234         __ cmp_32(tmp, tmp2);
2235       } else {
2236         __ load_klass(tmp, src);
2237         __ load_klass(tmp2, dst);
2238         __ cmp(tmp, tmp2);
2239       }
2240       __ b(*stub->entry(), ne);
2241     } else {
2242       // For object arrays, if src is a sub class of dst then we can
2243       // safely do the copy.
2244       Label cont, slow;
2245 
2246       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2247 
2248       __ load_klass(tmp, src);
2249       __ load_klass(tmp2, dst);
2250 
2251       // We are at a call so all live registers are saved before we
2252       // get here
2253       assert_different_registers(tmp, tmp2, R6, altFP_7_11);
2254 
2255       __ check_klass_subtype_fast_path(tmp, tmp2, R6, altFP_7_11, &cont, copyfunc_addr == nullptr ? stub->entry() : &slow, nullptr);
2256 
2257       __ mov(R6, R0);
2258       __ mov(altFP_7_11, R1);
2259       __ mov(R0, tmp);
2260       __ mov(R1, tmp2);
2261       __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); // does not blow any registers except R0, LR and Rtemp
2262       __ cmp_32(R0, 0);
2263       __ mov(R0, R6);
2264       __ mov(R1, altFP_7_11);
2265 
2266       if (copyfunc_addr != nullptr) { // use stub if available
2267         // src is not a sub class of dst so we have to do a
2268         // per-element check.
2269 
2270         __ b(cont, ne);
2271 
2272         __ bind(slow);
2273 
2274         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2275         if ((flags & mask) != mask) {
2276           // Check that at least both of them object arrays.
2277           assert(flags & mask, "one of the two should be known to be an object array");
2278 
2279           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2280             __ load_klass(tmp, src);
2281           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2282             __ load_klass(tmp, dst);
2283           }
2284           int lh_offset = in_bytes(Klass::layout_helper_offset());
2285 
2286           __ ldr_u32(tmp2, Address(tmp, lh_offset));
2287 
2288           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2289           __ mov_slow(tmp, objArray_lh);
2290           __ cmp_32(tmp, tmp2);
2291           __ b(*stub->entry(), ne);
2292         }
2293 
2294         save_in_reserved_area(R0, R1, R2, R3);
2295 
2296         Register src_ptr = R0;
2297         Register dst_ptr = R1;
2298         Register len     = R2;
2299         Register chk_off = R3;
2300         Register super_k = tmp;
2301 
2302         __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2303         __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2304 
2305         __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2306         __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2307         __ load_klass(tmp, dst);
2308 
2309         int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2310         int sco_offset = in_bytes(Klass::super_check_offset_offset());
2311 
2312         __ ldr(super_k, Address(tmp, ek_offset));
2313 
2314         __ mov(len, length);
2315         __ ldr_u32(chk_off, Address(super_k, sco_offset));
2316         __ push(super_k);
2317 
2318         __ call(copyfunc_addr, relocInfo::runtime_call_type);
2319 
2320 #ifndef PRODUCT
2321         if (PrintC1Statistics) {
2322           Label failed;
2323           __ cbnz_32(R0, failed);
2324           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_cnt, tmp, tmp2);
2325           __ bind(failed);
2326         }
2327 #endif // PRODUCT
2328 
2329         __ add(SP, SP, wordSize);  // Drop super_k argument
2330 
2331         __ cbz_32(R0, *stub->continuation());
2332         __ mvn_32(tmp, R0);
2333 
2334         // load saved arguments in slow case only
2335         restore_from_reserved_area(R0, R1, R2, R3);
2336 
2337         __ sub_32(length, length, tmp);
2338         __ add_32(src_pos, src_pos, tmp);
2339         __ add_32(dst_pos, dst_pos, tmp);
2340 
2341 #ifndef PRODUCT
2342         if (PrintC1Statistics) {
2343           __ inc_counter((address)&Runtime1::_arraycopy_checkcast_attempt_cnt, tmp, tmp2);
2344         }
2345 #endif
2346 
2347         __ b(*stub->entry());
2348 
2349         __ bind(cont);
2350       } else {
2351         __ b(*stub->entry(), eq);
2352         __ bind(cont);
2353       }
2354     }
2355   }
2356 
2357 #ifndef PRODUCT
2358   if (PrintC1Statistics) {
2359     address counter = Runtime1::arraycopy_count_address(basic_type);
2360     __ inc_counter(counter, tmp, tmp2);
2361   }
2362 #endif // !PRODUCT
2363 
2364   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2365   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2366   const char *name;
2367   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2368 
2369   Register src_ptr = R0;
2370   Register dst_ptr = R1;
2371   Register len     = R2;
2372 
2373   __ add(src_ptr, src, arrayOopDesc::base_offset_in_bytes(basic_type));
2374   __ add_ptr_scaled_int32(src_ptr, src_ptr, src_pos, shift);
2375 
2376   __ add(dst_ptr, dst, arrayOopDesc::base_offset_in_bytes(basic_type));
2377   __ add_ptr_scaled_int32(dst_ptr, dst_ptr, dst_pos, shift);
2378 
2379   __ mov(len, length);
2380 
2381   __ call(entry, relocInfo::runtime_call_type);
2382 
2383   __ bind(*stub->continuation());
2384 }
2385 
2386 #ifdef ASSERT
2387  // emit run-time assertion
2388 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2389   assert(op->code() == lir_assert, "must be");
2390 
2391   if (op->in_opr1()->is_valid()) {
2392     assert(op->in_opr2()->is_valid(), "both operands must be valid");
2393     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
2394   } else {
2395     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
2396     assert(op->condition() == lir_cond_always, "no other conditions allowed");
2397   }
2398 
2399   Label ok;
2400   if (op->condition() != lir_cond_always) {
2401     AsmCondition acond = al;
2402     switch (op->condition()) {
2403       case lir_cond_equal:        acond = eq; break;
2404       case lir_cond_notEqual:     acond = ne; break;
2405       case lir_cond_less:         acond = lt; break;
2406       case lir_cond_lessEqual:    acond = le; break;
2407       case lir_cond_greaterEqual: acond = ge; break;
2408       case lir_cond_greater:      acond = gt; break;
2409       case lir_cond_aboveEqual:   acond = hs; break;
2410       case lir_cond_belowEqual:   acond = ls; break;
2411       default:                    ShouldNotReachHere();
2412     }
2413     __ b(ok, acond);
2414   }
2415   if (op->halt()) {
2416     const char* str = __ code_string(op->msg());
2417     __ stop(str);
2418   } else {
2419     breakpoint();
2420   }
2421   __ bind(ok);
2422 }
2423 #endif // ASSERT
2424 
2425 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
2426   fatal("CRC32 intrinsic is not implemented on this platform");
2427 }
2428 
2429 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2430   Register obj = op->obj_opr()->as_pointer_register();
2431   Register hdr = op->hdr_opr()->as_pointer_register();
2432   Register lock = op->lock_opr()->as_pointer_register();
2433 
2434   if (LockingMode == LM_MONITOR) {
2435     if (op->info() != nullptr) {
2436       add_debug_info_for_null_check_here(op->info());
2437       __ null_check(obj);
2438     }
2439     __ b(*op->stub()->entry());
2440   } else if (op->code() == lir_lock) {
2441     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2442     int null_check_offset = __ lock_object(hdr, obj, lock, *op->stub()->entry());
2443     if (op->info() != nullptr) {
2444       add_debug_info_for_null_check(null_check_offset, op->info());
2445     }
2446   } else if (op->code() == lir_unlock) {
2447     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2448   } else {
2449     ShouldNotReachHere();
2450   }
2451   __ bind(*op->stub()->continuation());
2452 }
2453 
2454 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2455   Register obj = op->obj()->as_pointer_register();
2456   Register result = op->result_opr()->as_pointer_register();
2457 
2458   CodeEmitInfo* info = op->info();
2459   if (info != nullptr) {
2460     add_debug_info_for_null_check_here(info);
2461   }
2462 
2463   if (UseCompressedClassPointers) { // On 32 bit arm??
2464     __ ldr_u32(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2465   } else {
2466     __ ldr(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2467   }
2468 }
2469 
2470 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2471   ciMethod* method = op->profiled_method();
2472   int bci          = op->profiled_bci();
2473   ciMethod* callee = op->profiled_callee();
2474 
2475   // Update counter for all call types
2476   ciMethodData* md = method->method_data_or_null();
2477   assert(md != nullptr, "Sanity");
2478   ciProfileData* data = md->bci_to_data(bci);
2479   assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
2480   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2481   Register mdo  = op->mdo()->as_register();
2482   assert(op->tmp1()->is_register(), "tmp1 must be allocated");
2483   Register tmp1 = op->tmp1()->as_pointer_register();
2484   assert_different_registers(mdo, tmp1);
2485   __ mov_metadata(mdo, md->constant_encoding());
2486   int mdo_offset_bias = 0;
2487   int max_offset = 4096;
2488   if (md->byte_offset_of_slot(data, CounterData::count_offset()) + data->size_in_bytes() >= max_offset) {
2489     // The offset is large so bias the mdo by the base of the slot so
2490     // that the ldr can use an immediate offset to reference the slots of the data
2491     mdo_offset_bias = md->byte_offset_of_slot(data, CounterData::count_offset());
2492     __ mov_slow(tmp1, mdo_offset_bias);
2493     __ add(mdo, mdo, tmp1);
2494   }
2495 
2496   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()) - mdo_offset_bias);
2497   // Perform additional virtual call profiling for invokevirtual and
2498   // invokeinterface bytecodes
2499   if (op->should_profile_receiver_type()) {
2500     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2501     Register recv = op->recv()->as_register();
2502     assert_different_registers(mdo, tmp1, recv);
2503     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2504     ciKlass* known_klass = op->known_holder();
2505     if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
2506       // We know the type that will be seen at this call site; we can
2507       // statically update the MethodData* rather than needing to do
2508       // dynamic tests on the receiver type
2509 
2510       // NOTE: we should probably put a lock around this search to
2511       // avoid collisions by concurrent compilations
2512       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2513       uint i;
2514       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2515         ciKlass* receiver = vc_data->receiver(i);
2516         if (known_klass->equals(receiver)) {
2517           Address data_addr(mdo, md->byte_offset_of_slot(data,
2518                                                          VirtualCallData::receiver_count_offset(i)) -
2519                             mdo_offset_bias);
2520           __ ldr(tmp1, data_addr);
2521           __ add(tmp1, tmp1, DataLayout::counter_increment);
2522           __ str(tmp1, data_addr);
2523           return;
2524         }
2525       }
2526 
2527       // Receiver type not found in profile data; select an empty slot
2528 
2529       // Note that this is less efficient than it should be because it
2530       // always does a write to the receiver part of the
2531       // VirtualCallData rather than just the first time
2532       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2533         ciKlass* receiver = vc_data->receiver(i);
2534         if (receiver == nullptr) {
2535           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)) -
2536                             mdo_offset_bias);
2537           __ mov_metadata(tmp1, known_klass->constant_encoding());
2538           __ str(tmp1, recv_addr);
2539           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)) -
2540                             mdo_offset_bias);
2541           __ ldr(tmp1, data_addr);
2542           __ add(tmp1, tmp1, DataLayout::counter_increment);
2543           __ str(tmp1, data_addr);
2544           return;
2545         }
2546       }
2547     } else {
2548       __ load_klass(recv, recv);
2549       Label update_done;
2550       type_profile_helper(mdo, mdo_offset_bias, md, data, recv, tmp1, &update_done);
2551       // Receiver did not match any saved receiver and there is no empty row for it.
2552       // Increment total counter to indicate polymorphic case.
2553       __ ldr(tmp1, counter_addr);
2554       __ add(tmp1, tmp1, DataLayout::counter_increment);
2555       __ str(tmp1, counter_addr);
2556 
2557       __ bind(update_done);
2558     }
2559   } else {
2560     // Static call
2561     __ ldr(tmp1, counter_addr);
2562     __ add(tmp1, tmp1, DataLayout::counter_increment);
2563     __ str(tmp1, counter_addr);
2564   }
2565 }
2566 
2567 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2568   fatal("Type profiling not implemented on this platform");
2569 }
2570 
2571 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
2572   Unimplemented();
2573 }
2574 
2575 
2576 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
2577   Address mon_addr = frame_map()->address_for_monitor_lock(monitor_no);
2578   __ add_slow(dst->as_pointer_register(), mon_addr.base(), mon_addr.disp());
2579 }
2580 
2581 
2582 void LIR_Assembler::align_backward_branch_target() {
2583   // Some ARM processors do better with 8-byte branch target alignment
2584   __ align(8);
2585 }
2586 
2587 
2588 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2589   // tmp must be unused
2590   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2591 
2592   if (left->is_single_cpu()) {
2593     assert (dest->type() == T_INT, "unexpected result type");
2594     assert (left->type() == T_INT, "unexpected left type");
2595     __ neg_32(dest->as_register(), left->as_register());
2596   } else if (left->is_double_cpu()) {
2597     Register dest_lo = dest->as_register_lo();
2598     Register dest_hi = dest->as_register_hi();
2599     Register src_lo = left->as_register_lo();
2600     Register src_hi = left->as_register_hi();
2601     if (dest_lo == src_hi) {
2602       dest_lo = Rtemp;
2603     }
2604     __ rsbs(dest_lo, src_lo, 0);
2605     __ rsc(dest_hi, src_hi, 0);
2606     move_regs(dest_lo, dest->as_register_lo());
2607   } else if (left->is_single_fpu()) {
2608     __ neg_float(dest->as_float_reg(), left->as_float_reg());
2609   } else if (left->is_double_fpu()) {
2610     __ neg_double(dest->as_double_reg(), left->as_double_reg());
2611   } else {
2612     ShouldNotReachHere();
2613   }
2614 }
2615 
2616 
2617 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2618   assert(patch_code == lir_patch_none, "Patch code not supported");
2619   LIR_Address* addr = addr_opr->as_address_ptr();
2620   if (addr->index()->is_illegal()) {
2621     jint c = addr->disp();
2622     if (!Assembler::is_arith_imm_in_range(c)) {
2623       BAILOUT("illegal arithmetic operand");
2624     }
2625     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(), c);
2626   } else {
2627     assert(addr->disp() == 0, "cannot handle otherwise");
2628     __ add(dest->as_pointer_register(), addr->base()->as_pointer_register(),
2629            AsmOperand(addr->index()->as_pointer_register(), lsl, addr->scale()));
2630   }
2631 }
2632 
2633 
2634 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2635   assert(!tmp->is_valid(), "don't need temporary");
2636   __ call(dest);
2637   if (info != nullptr) {
2638     add_call_info_here(info);
2639   }
2640 }
2641 
2642 
2643 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2644   assert(src->is_double_cpu() && dest->is_address() ||
2645          src->is_address() && dest->is_double_cpu(),
2646          "Simple move_op is called for all other cases");
2647 
2648   int null_check_offset;
2649   if (dest->is_address()) {
2650     // Store
2651     const LIR_Address* addr = dest->as_address_ptr();
2652     const Register src_lo = src->as_register_lo();
2653     const Register src_hi = src->as_register_hi();
2654     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2655 
2656     if (src_lo < src_hi) {
2657       null_check_offset = __ offset();
2658       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(src_hi));
2659     } else {
2660       assert(src_lo < Rtemp, "Rtemp is higher than any allocatable register");
2661       __ mov(Rtemp, src_hi);
2662       null_check_offset = __ offset();
2663       __ stmia(addr->base()->as_register(), RegisterSet(src_lo) | RegisterSet(Rtemp));
2664     }
2665   } else {
2666     // Load
2667     const LIR_Address* addr = src->as_address_ptr();
2668     const Register dest_lo = dest->as_register_lo();
2669     const Register dest_hi = dest->as_register_hi();
2670     assert(addr->index()->is_illegal() && addr->disp() == 0, "The address is simple already");
2671 
2672     null_check_offset = __ offset();
2673     if (dest_lo < dest_hi) {
2674       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(dest_hi));
2675     } else {
2676       assert(dest_lo < Rtemp, "Rtemp is higher than any allocatable register");
2677       __ ldmia(addr->base()->as_register(), RegisterSet(dest_lo) | RegisterSet(Rtemp));
2678       __ mov(dest_hi, Rtemp);
2679     }
2680   }
2681 
2682   if (info != nullptr) {
2683     add_debug_info_for_null_check(null_check_offset, info);
2684   }
2685 }
2686 
2687 
2688 void LIR_Assembler::membar() {
2689   __ membar(MacroAssembler::StoreLoad, Rtemp);
2690 }
2691 
2692 void LIR_Assembler::membar_acquire() {
2693   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::LoadLoad | MacroAssembler::LoadStore), Rtemp);
2694 }
2695 
2696 void LIR_Assembler::membar_release() {
2697   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2698 }
2699 
2700 void LIR_Assembler::membar_loadload() {
2701   __ membar(MacroAssembler::LoadLoad, Rtemp);
2702 }
2703 
2704 void LIR_Assembler::membar_storestore() {
2705   __ membar(MacroAssembler::StoreStore, Rtemp);
2706 }
2707 
2708 void LIR_Assembler::membar_loadstore() {
2709   __ membar(MacroAssembler::LoadStore, Rtemp);
2710 }
2711 
2712 void LIR_Assembler::membar_storeload() {
2713   __ membar(MacroAssembler::StoreLoad, Rtemp);
2714 }
2715 
2716 void LIR_Assembler::on_spin_wait() {
2717   Unimplemented();
2718 }
2719 
2720 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2721   // Not used on ARM
2722   Unimplemented();
2723 }
2724 
2725 void LIR_Assembler::peephole(LIR_List* lir) {
2726   LIR_OpList* inst = lir->instructions_list();
2727   const int inst_length = inst->length();
2728   for (int i = 0; i < inst_length; i++) {
2729     LIR_Op* op = inst->at(i);
2730     switch (op->code()) {
2731       case lir_cmp: {
2732         // Replace:
2733         //   cmp rX, y
2734         //   cmove [EQ] y, z, rX
2735         // with
2736         //   cmp rX, y
2737         //   cmove [EQ] illegalOpr, z, rX
2738         //
2739         // or
2740         //   cmp rX, y
2741         //   cmove [NE] z, y, rX
2742         // with
2743         //   cmp rX, y
2744         //   cmove [NE] z, illegalOpr, rX
2745         //
2746         // moves from illegalOpr should be removed when converting LIR to native assembly
2747 
2748         LIR_Op2* cmp = op->as_Op2();
2749         assert(cmp != nullptr, "cmp LIR instruction is not an op2");
2750 
2751         if (i + 1 < inst_length) {
2752           LIR_Op2* cmove = inst->at(i + 1)->as_Op2();
2753           if (cmove != nullptr && cmove->code() == lir_cmove) {
2754             LIR_Opr cmove_res = cmove->result_opr();
2755             bool res_is_op1 = cmove_res == cmp->in_opr1();
2756             bool res_is_op2 = cmove_res == cmp->in_opr2();
2757             LIR_Opr cmp_res, cmp_arg;
2758             if (res_is_op1) {
2759               cmp_res = cmp->in_opr1();
2760               cmp_arg = cmp->in_opr2();
2761             } else if (res_is_op2) {
2762               cmp_res = cmp->in_opr2();
2763               cmp_arg = cmp->in_opr1();
2764             } else {
2765               cmp_res = LIR_OprFact::illegalOpr;
2766               cmp_arg = LIR_OprFact::illegalOpr;
2767             }
2768 
2769             if (cmp_res != LIR_OprFact::illegalOpr) {
2770               LIR_Condition cond = cmove->condition();
2771               if (cond == lir_cond_equal && cmove->in_opr1() == cmp_arg) {
2772                 cmove->set_in_opr1(LIR_OprFact::illegalOpr);
2773               } else if (cond == lir_cond_notEqual && cmove->in_opr2() == cmp_arg) {
2774                 cmove->set_in_opr2(LIR_OprFact::illegalOpr);
2775               }
2776             }
2777           }
2778         }
2779         break;
2780       }
2781 
2782       default:
2783         break;
2784     }
2785   }
2786 }
2787 
2788 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2789   assert(src->is_address(), "sanity");
2790   Address addr = as_Address(src->as_address_ptr());
2791 
2792   if (code == lir_xchg) {
2793   } else {
2794     assert (!data->is_oop(), "xadd for oops");
2795   }
2796 
2797   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreStore | MacroAssembler::LoadStore), Rtemp);
2798 
2799   Label retry;
2800   __ bind(retry);
2801 
2802   if (data->type() == T_INT || data->is_oop()) {
2803     Register dst = dest->as_register();
2804     Register new_val = noreg;
2805     __ ldrex(dst, addr);
2806     if (code == lir_xadd) {
2807       Register tmp_reg = tmp->as_register();
2808       if (data->is_constant()) {
2809         assert_different_registers(dst, tmp_reg);
2810         __ add_32(tmp_reg, dst, data->as_constant_ptr()->as_jint());
2811       } else {
2812         assert_different_registers(dst, tmp_reg, data->as_register());
2813         __ add_32(tmp_reg, dst, data->as_register());
2814       }
2815       new_val = tmp_reg;
2816     } else {
2817       if (UseCompressedOops && data->is_oop()) {
2818         new_val = tmp->as_pointer_register();
2819       } else {
2820         new_val = data->as_register();
2821       }
2822       assert_different_registers(dst, new_val);
2823     }
2824     __ strex(Rtemp, new_val, addr);
2825 
2826   } else if (data->type() == T_LONG) {
2827     Register dst_lo = dest->as_register_lo();
2828     Register new_val_lo = noreg;
2829     Register dst_hi = dest->as_register_hi();
2830 
2831     assert(dst_hi->encoding() == dst_lo->encoding() + 1, "non aligned register pair");
2832     assert((dst_lo->encoding() & 0x1) == 0, "misaligned register pair");
2833 
2834     __ bind(retry);
2835     __ ldrexd(dst_lo, addr);
2836     if (code == lir_xadd) {
2837       Register tmp_lo = tmp->as_register_lo();
2838       Register tmp_hi = tmp->as_register_hi();
2839 
2840       assert(tmp_hi->encoding() == tmp_lo->encoding() + 1, "non aligned register pair");
2841       assert((tmp_lo->encoding() & 0x1) == 0, "misaligned register pair");
2842 
2843       if (data->is_constant()) {
2844         jlong c = data->as_constant_ptr()->as_jlong();
2845         assert((jlong)((jint)c) == c, "overflow");
2846         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi);
2847         __ adds(tmp_lo, dst_lo, (jint)c);
2848         __ adc(tmp_hi, dst_hi, 0);
2849       } else {
2850         Register new_val_lo = data->as_register_lo();
2851         Register new_val_hi = data->as_register_hi();
2852         __ adds(tmp_lo, dst_lo, new_val_lo);
2853         __ adc(tmp_hi, dst_hi, new_val_hi);
2854         assert_different_registers(dst_lo, dst_hi, tmp_lo, tmp_hi, new_val_lo, new_val_hi);
2855       }
2856       new_val_lo = tmp_lo;
2857     } else {
2858       new_val_lo = data->as_register_lo();
2859       Register new_val_hi = data->as_register_hi();
2860 
2861       assert_different_registers(dst_lo, dst_hi, new_val_lo, new_val_hi);
2862       assert(new_val_hi->encoding() == new_val_lo->encoding() + 1, "non aligned register pair");
2863       assert((new_val_lo->encoding() & 0x1) == 0, "misaligned register pair");
2864     }
2865     __ strexd(Rtemp, new_val_lo, addr);
2866   } else {
2867     ShouldNotReachHere();
2868   }
2869 
2870   __ cbnz_32(Rtemp, retry);
2871   __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
2872 
2873 }
2874 
2875 #undef __