1 /*
   2  * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "asm/macroAssembler.inline.hpp"
  27 #include "asm/assembler.hpp"
  28 #include "c1/c1_CodeStubs.hpp"
  29 #include "c1/c1_Compilation.hpp"
  30 #include "c1/c1_LIRAssembler.hpp"
  31 #include "c1/c1_MacroAssembler.hpp"
  32 #include "c1/c1_Runtime1.hpp"
  33 #include "c1/c1_ValueStack.hpp"
  34 #include "ci/ciArrayKlass.hpp"
  35 #include "ci/ciInstance.hpp"
  36 #include "code/aotCodeCache.hpp"
  37 #include "code/compiledIC.hpp"
  38 #include "gc/shared/collectedHeap.hpp"
  39 #include "gc/shared/gc_globals.hpp"
  40 #include "nativeInst_aarch64.hpp"
  41 #include "oops/objArrayKlass.hpp"
  42 #include "runtime/frame.inline.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "runtime/stubRoutines.hpp"
  45 #include "runtime/threadIdentifier.hpp"
  46 #include "utilities/powerOfTwo.hpp"
  47 #include "vmreg_aarch64.inline.hpp"
  48 
  49 
  50 #ifndef PRODUCT
  51 #define COMMENT(x)   do { __ block_comment(x); } while (0)
  52 #else
  53 #define COMMENT(x)
  54 #endif
  55 
  56 NEEDS_CLEANUP // remove this definitions ?
  57 const Register SYNC_header = r0;   // synchronization header
  58 const Register SHIFT_count = r0;   // where count for shift operations must be
  59 
  60 #define __ _masm->
  61 
  62 
  63 static void select_different_registers(Register preserve,
  64                                        Register extra,
  65                                        Register &tmp1,
  66                                        Register &tmp2) {
  67   if (tmp1 == preserve) {
  68     assert_different_registers(tmp1, tmp2, extra);
  69     tmp1 = extra;
  70   } else if (tmp2 == preserve) {
  71     assert_different_registers(tmp1, tmp2, extra);
  72     tmp2 = extra;
  73   }
  74   assert_different_registers(preserve, tmp1, tmp2);
  75 }
  76 
  77 
  78 
  79 static void select_different_registers(Register preserve,
  80                                        Register extra,
  81                                        Register &tmp1,
  82                                        Register &tmp2,
  83                                        Register &tmp3) {
  84   if (tmp1 == preserve) {
  85     assert_different_registers(tmp1, tmp2, tmp3, extra);
  86     tmp1 = extra;
  87   } else if (tmp2 == preserve) {
  88     assert_different_registers(tmp1, tmp2, tmp3, extra);
  89     tmp2 = extra;
  90   } else if (tmp3 == preserve) {
  91     assert_different_registers(tmp1, tmp2, tmp3, extra);
  92     tmp3 = extra;
  93   }
  94   assert_different_registers(preserve, tmp1, tmp2, tmp3);
  95 }
  96 
  97 
  98 bool LIR_Assembler::is_small_constant(LIR_Opr opr) { Unimplemented(); return false; }
  99 
 100 
 101 LIR_Opr LIR_Assembler::receiverOpr() {
 102   return FrameMap::receiver_opr;
 103 }
 104 
 105 LIR_Opr LIR_Assembler::osrBufferPointer() {
 106   return FrameMap::as_pointer_opr(receiverOpr()->as_register());
 107 }
 108 
 109 //--------------fpu register translations-----------------------
 110 
 111 
 112 address LIR_Assembler::float_constant(float f) {
 113   address const_addr = __ float_constant(f);
 114   if (const_addr == nullptr) {
 115     bailout("const section overflow");
 116     return __ code()->consts()->start();
 117   } else {
 118     return const_addr;
 119   }
 120 }
 121 
 122 
 123 address LIR_Assembler::double_constant(double d) {
 124   address const_addr = __ double_constant(d);
 125   if (const_addr == nullptr) {
 126     bailout("const section overflow");
 127     return __ code()->consts()->start();
 128   } else {
 129     return const_addr;
 130   }
 131 }
 132 
 133 address LIR_Assembler::int_constant(jlong n) {
 134   address const_addr = __ long_constant(n);
 135   if (const_addr == nullptr) {
 136     bailout("const section overflow");
 137     return __ code()->consts()->start();
 138   } else {
 139     return const_addr;
 140   }
 141 }
 142 
 143 void LIR_Assembler::breakpoint() { Unimplemented(); }
 144 
 145 void LIR_Assembler::push(LIR_Opr opr) { Unimplemented(); }
 146 
 147 void LIR_Assembler::pop(LIR_Opr opr) { Unimplemented(); }
 148 
 149 bool LIR_Assembler::is_literal_address(LIR_Address* addr) { Unimplemented(); return false; }
 150 //-------------------------------------------
 151 
 152 static Register as_reg(LIR_Opr op) {
 153   return op->is_double_cpu() ? op->as_register_lo() : op->as_register();
 154 }
 155 
 156 static jlong as_long(LIR_Opr data) {
 157   jlong result;
 158   switch (data->type()) {
 159   case T_INT:
 160     result = (data->as_jint());
 161     break;
 162   case T_LONG:
 163     result = (data->as_jlong());
 164     break;
 165   default:
 166     ShouldNotReachHere();
 167     result = 0;  // unreachable
 168   }
 169   return result;
 170 }
 171 
 172 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
 173   Register base = addr->base()->as_pointer_register();
 174   LIR_Opr opr = addr->index();
 175   if (opr->is_cpu_register()) {
 176     Register index;
 177     if (opr->is_single_cpu())
 178       index = opr->as_register();
 179     else
 180       index = opr->as_register_lo();
 181     assert(addr->disp() == 0, "must be");
 182     switch(opr->type()) {
 183       case T_INT:
 184         return Address(base, index, Address::sxtw(addr->scale()));
 185       case T_LONG:
 186         return Address(base, index, Address::lsl(addr->scale()));
 187       default:
 188         ShouldNotReachHere();
 189       }
 190   } else {
 191     assert(addr->scale() == 0,
 192            "expected for immediate operand, was: %d", addr->scale());
 193     ptrdiff_t offset = ptrdiff_t(addr->disp());
 194     // NOTE: Does not handle any 16 byte vector access.
 195     const uint type_size = type2aelembytes(addr->type(), true);
 196     return __ legitimize_address(Address(base, offset), type_size, tmp);
 197   }
 198   return Address();
 199 }
 200 
 201 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 202   ShouldNotReachHere();
 203   return Address();
 204 }
 205 
 206 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 207   return as_Address(addr, rscratch1);
 208 }
 209 
 210 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 211   return as_Address(addr, rscratch1);  // Ouch
 212   // FIXME: This needs to be much more clever.  See x86.
 213 }
 214 
 215 // Ensure a valid Address (base + offset) to a stack-slot. If stack access is
 216 // not encodable as a base + (immediate) offset, generate an explicit address
 217 // calculation to hold the address in a temporary register.
 218 Address LIR_Assembler::stack_slot_address(int index, uint size, Register tmp, int adjust) {
 219   precond(size == 4 || size == 8);
 220   Address addr = frame_map()->address_for_slot(index, adjust);
 221   precond(addr.getMode() == Address::base_plus_offset);
 222   precond(addr.base() == sp);
 223   precond(addr.offset() > 0);
 224   uint mask = size - 1;
 225   assert((addr.offset() & mask) == 0, "scaled offsets only");
 226   return __ legitimize_address(addr, size, tmp);
 227 }
 228 
 229 void LIR_Assembler::osr_entry() {
 230   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 231   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 232   ValueStack* entry_state = osr_entry->state();
 233   int number_of_locks = entry_state->locks_size();
 234 
 235   // we jump here if osr happens with the interpreter
 236   // state set up to continue at the beginning of the
 237   // loop that triggered osr - in particular, we have
 238   // the following registers setup:
 239   //
 240   // r2: osr buffer
 241   //
 242 
 243   // build frame
 244   ciMethod* m = compilation()->method();
 245   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 246 
 247   // OSR buffer is
 248   //
 249   // locals[nlocals-1..0]
 250   // monitors[0..number_of_locks]
 251   //
 252   // locals is a direct copy of the interpreter frame so in the osr buffer
 253   // so first slot in the local array is the last local from the interpreter
 254   // and last slot is local[0] (receiver) from the interpreter
 255   //
 256   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 257   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 258   // in the interpreter frame (the method lock if a sync method)
 259 
 260   // Initialize monitors in the compiled activation.
 261   //   r2: pointer to osr buffer
 262   //
 263   // All other registers are dead at this point and the locals will be
 264   // copied into place by code emitted in the IR.
 265 
 266   Register OSR_buf = osrBufferPointer()->as_pointer_register();
 267   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 268     int monitor_offset = BytesPerWord * method()->max_locals() +
 269       (2 * BytesPerWord) * (number_of_locks - 1);
 270     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 271     // the OSR buffer using 2 word entries: first the lock and then
 272     // the oop.
 273     for (int i = 0; i < number_of_locks; i++) {
 274       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 275 #ifdef ASSERT
 276       // verify the interpreter's monitor has a non-null object
 277       {
 278         Label L;
 279         __ ldr(rscratch1, __ form_address(rscratch1, OSR_buf, slot_offset + 1*BytesPerWord, 0));
 280         __ cbnz(rscratch1, L);
 281         __ stop("locked object is null");
 282         __ bind(L);
 283       }
 284 #endif
 285       __ ldr(r19, __ form_address(rscratch1, OSR_buf, slot_offset, 0));
 286       __ ldr(r20, __ form_address(rscratch1, OSR_buf, slot_offset + BytesPerWord, 0));
 287       __ str(r19, frame_map()->address_for_monitor_lock(i));
 288       __ str(r20, frame_map()->address_for_monitor_object(i));
 289     }
 290   }
 291 }
 292 
 293 
 294 // inline cache check; done before the frame is built.
 295 int LIR_Assembler::check_icache() {
 296   return __ ic_check(CodeEntryAlignment);
 297 }
 298 
 299 void LIR_Assembler::clinit_barrier(ciMethod* method) {
 300   assert(VM_Version::supports_fast_class_init_checks(), "sanity");
 301   assert(!method->holder()->is_not_initialized(), "initialization should have been started");
 302 
 303   Label L_skip_barrier;
 304 
 305   __ mov_metadata(rscratch2, method->holder()->constant_encoding());
 306   __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier /*L_fast_path*/);
 307   __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
 308   __ bind(L_skip_barrier);
 309 }
 310 
 311 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 312   if (o == nullptr) {
 313     __ mov(reg, zr);
 314   } else {
 315     __ movoop(reg, o);
 316   }
 317 }
 318 
 319 void LIR_Assembler::deoptimize_trap(CodeEmitInfo *info) {
 320   address target = nullptr;
 321   relocInfo::relocType reloc_type = relocInfo::none;
 322 
 323   switch (patching_id(info)) {
 324   case PatchingStub::access_field_id:
 325     target = Runtime1::entry_for(StubId::c1_access_field_patching_id);
 326     reloc_type = relocInfo::section_word_type;
 327     break;
 328   case PatchingStub::load_klass_id:
 329     target = Runtime1::entry_for(StubId::c1_load_klass_patching_id);
 330     reloc_type = relocInfo::metadata_type;
 331     break;
 332   case PatchingStub::load_mirror_id:
 333     target = Runtime1::entry_for(StubId::c1_load_mirror_patching_id);
 334     reloc_type = relocInfo::oop_type;
 335     break;
 336   case PatchingStub::load_appendix_id:
 337     target = Runtime1::entry_for(StubId::c1_load_appendix_patching_id);
 338     reloc_type = relocInfo::oop_type;
 339     break;
 340   default: ShouldNotReachHere();
 341   }
 342 
 343   __ far_call(RuntimeAddress(target));
 344   add_call_info_here(info);
 345 }
 346 
 347 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 348   deoptimize_trap(info);
 349 }
 350 
 351 
 352 // This specifies the rsp decrement needed to build the frame
 353 int LIR_Assembler::initial_frame_size_in_bytes() const {
 354   // if rounding, must let FrameMap know!
 355 
 356   return in_bytes(frame_map()->framesize_in_bytes());
 357 }
 358 
 359 
 360 int LIR_Assembler::emit_exception_handler() {
 361   // generate code for exception handler
 362   address handler_base = __ start_a_stub(exception_handler_size());
 363   if (handler_base == nullptr) {
 364     // not enough space left for the handler
 365     bailout("exception handler overflow");
 366     return -1;
 367   }
 368 
 369   int offset = code_offset();
 370 
 371   // the exception oop and pc are in r0, and r3
 372   // no other registers need to be preserved, so invalidate them
 373   __ invalidate_registers(false, true, true, false, true, true);
 374 
 375   // check that there is really an exception
 376   __ verify_not_null_oop(r0);
 377 
 378   // search an exception handler (r0: exception oop, r3: throwing pc)
 379   __ far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_handle_exception_from_callee_id)));
 380   __ should_not_reach_here();
 381   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 382   __ end_a_stub();
 383 
 384   return offset;
 385 }
 386 
 387 
 388 // Emit the code to remove the frame from the stack in the exception
 389 // unwind path.
 390 int LIR_Assembler::emit_unwind_handler() {
 391 #ifndef PRODUCT
 392   if (CommentedAssembly) {
 393     _masm->block_comment("Unwind handler");
 394   }
 395 #endif
 396 
 397   int offset = code_offset();
 398 
 399   // Fetch the exception from TLS and clear out exception related thread state
 400   __ ldr(r0, Address(rthread, JavaThread::exception_oop_offset()));
 401   __ str(zr, Address(rthread, JavaThread::exception_oop_offset()));
 402   __ str(zr, Address(rthread, JavaThread::exception_pc_offset()));
 403 
 404   __ bind(_unwind_handler_entry);
 405   __ verify_not_null_oop(r0);
 406   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 407     __ mov(r19, r0);  // Preserve the exception
 408   }
 409 
 410   // Perform needed unlocking
 411   MonitorExitStub* stub = nullptr;
 412   if (method()->is_synchronized()) {
 413     monitor_address(0, FrameMap::r0_opr);
 414     stub = new MonitorExitStub(FrameMap::r0_opr, 0);
 415     __ unlock_object(r5, r4, r0, r6, *stub->entry());
 416     __ bind(*stub->continuation());
 417   }
 418 
 419   if (compilation()->env()->dtrace_method_probes()) {
 420     __ mov(c_rarg0, rthread);
 421     __ mov_metadata(c_rarg1, method()->constant_encoding());
 422     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), c_rarg0, c_rarg1);
 423   }
 424 
 425   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 426     __ mov(r0, r19);  // Restore the exception
 427   }
 428 
 429   // remove the activation and dispatch to the unwind handler
 430   __ block_comment("remove_frame and dispatch to the unwind handler");
 431   __ remove_frame(initial_frame_size_in_bytes());
 432   __ far_jump(RuntimeAddress(Runtime1::entry_for(StubId::c1_unwind_exception_id)));
 433 
 434   // Emit the slow path assembly
 435   if (stub != nullptr) {
 436     stub->emit_code(this);
 437   }
 438 
 439   return offset;
 440 }
 441 
 442 
 443 int LIR_Assembler::emit_deopt_handler() {
 444   // generate code for exception handler
 445   address handler_base = __ start_a_stub(deopt_handler_size());
 446   if (handler_base == nullptr) {
 447     // not enough space left for the handler
 448     bailout("deopt handler overflow");
 449     return -1;
 450   }
 451 
 452   int offset = code_offset();
 453 
 454   Label start;
 455   __ bind(start);
 456 
 457   __ far_call(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
 458 
 459   int entry_offset = __ offset();
 460   __ b(start);
 461 
 462   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 463   assert(code_offset() - entry_offset >= NativePostCallNop::first_check_size,
 464          "out of bounds read in post-call NOP check");
 465   __ end_a_stub();
 466 
 467   return entry_offset;
 468 }
 469 
 470 void LIR_Assembler::add_debug_info_for_branch(address adr, CodeEmitInfo* info) {
 471   _masm->code_section()->relocate(adr, relocInfo::poll_type);
 472   int pc_offset = code_offset();
 473   flush_debug_info(pc_offset);
 474   info->record_debug_info(compilation()->debug_info_recorder(), pc_offset);
 475   if (info->exception_handlers() != nullptr) {
 476     compilation()->add_exception_handlers_for_pco(pc_offset, info->exception_handlers());
 477   }
 478 }
 479 
 480 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
 481   assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == r0, "word returns are in r0,");
 482 
 483   // Pop the stack before the safepoint code
 484   __ remove_frame(initial_frame_size_in_bytes());
 485 
 486   if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
 487     __ reserved_stack_check();
 488   }
 489 
 490   code_stub->set_safepoint_offset(__ offset());
 491   __ relocate(relocInfo::poll_return_type);
 492   __ safepoint_poll(*code_stub->entry(), true /* at_return */, true /* in_nmethod */);
 493   __ ret(lr);
 494 }
 495 
 496 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
 497   guarantee(info != nullptr, "Shouldn't be null");
 498   __ get_polling_page(rscratch1, relocInfo::poll_type);
 499   add_debug_info_for_branch(info);  // This isn't just debug info:
 500                                     // it's the oop map
 501   __ read_polling_page(rscratch1, relocInfo::poll_type);
 502   return __ offset();
 503 }
 504 
 505 
 506 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 507   if (from_reg == r31_sp)
 508     from_reg = sp;
 509   if (to_reg == r31_sp)
 510     to_reg = sp;
 511   __ mov(to_reg, from_reg);
 512 }
 513 
 514 void LIR_Assembler::swap_reg(Register a, Register b) { Unimplemented(); }
 515 
 516 
 517 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 518   assert(src->is_constant(), "should not call otherwise");
 519   assert(dest->is_register(), "should not call otherwise");
 520   LIR_Const* c = src->as_constant_ptr();
 521 
 522   switch (c->type()) {
 523     case T_INT: {
 524       assert(patch_code == lir_patch_none, "no patching handled here");
 525       __ movw(dest->as_register(), c->as_jint());
 526       break;
 527     }
 528 
 529     case T_ADDRESS: {
 530       assert(patch_code == lir_patch_none, "no patching handled here");
 531       __ mov(dest->as_register(), c->as_jint());
 532       break;
 533     }
 534 
 535     case T_LONG: {
 536       assert(patch_code == lir_patch_none, "no patching handled here");
 537 #if INCLUDE_CDS
 538       if (AOTCodeCache::is_on_for_dump()) {
 539         address b = c->as_pointer();
 540         if (b == (address)ThreadIdentifier::unsafe_offset()) {
 541           __ lea(dest->as_register_lo(), ExternalAddress(b));
 542           break;
 543         }
 544         if (AOTRuntimeConstants::contains(b)) {
 545           __ load_aotrc_address(dest->as_register_lo(), b);
 546           break;
 547         }
 548       }
 549 #endif
 550       __ mov(dest->as_register_lo(), (intptr_t)c->as_jlong());
 551       break;
 552     }
 553 
 554     case T_OBJECT: {
 555         if (patch_code == lir_patch_none) {
 556           jobject2reg(c->as_jobject(), dest->as_register());
 557         } else {
 558           jobject2reg_with_patching(dest->as_register(), info);
 559         }
 560       break;
 561     }
 562 
 563     case T_METADATA: {
 564       if (patch_code != lir_patch_none) {
 565         klass2reg_with_patching(dest->as_register(), info);
 566       } else {
 567         __ mov_metadata(dest->as_register(), c->as_metadata());
 568       }
 569       break;
 570     }
 571 
 572     case T_FLOAT: {
 573       if (__ operand_valid_for_float_immediate(c->as_jfloat())) {
 574         __ fmovs(dest->as_float_reg(), (c->as_jfloat()));
 575       } else {
 576         __ adr(rscratch1, InternalAddress(float_constant(c->as_jfloat())));
 577         __ ldrs(dest->as_float_reg(), Address(rscratch1));
 578       }
 579       break;
 580     }
 581 
 582     case T_DOUBLE: {
 583       if (__ operand_valid_for_float_immediate(c->as_jdouble())) {
 584         __ fmovd(dest->as_double_reg(), (c->as_jdouble()));
 585       } else {
 586         __ adr(rscratch1, InternalAddress(double_constant(c->as_jdouble())));
 587         __ ldrd(dest->as_double_reg(), Address(rscratch1));
 588       }
 589       break;
 590     }
 591 
 592     default:
 593       ShouldNotReachHere();
 594   }
 595 }
 596 
 597 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 598   LIR_Const* c = src->as_constant_ptr();
 599   switch (c->type()) {
 600   case T_OBJECT:
 601     {
 602       if (! c->as_jobject())
 603         __ str(zr, frame_map()->address_for_slot(dest->single_stack_ix()));
 604       else {
 605         const2reg(src, FrameMap::rscratch1_opr, lir_patch_none, nullptr);
 606         reg2stack(FrameMap::rscratch1_opr, dest, c->type());
 607       }
 608     }
 609     break;
 610   case T_ADDRESS:
 611     {
 612       const2reg(src, FrameMap::rscratch1_opr, lir_patch_none, nullptr);
 613       reg2stack(FrameMap::rscratch1_opr, dest, c->type());
 614     }
 615   case T_INT:
 616   case T_FLOAT:
 617     {
 618       Register reg = zr;
 619       if (c->as_jint_bits() == 0)
 620         __ strw(zr, frame_map()->address_for_slot(dest->single_stack_ix()));
 621       else {
 622         __ movw(rscratch1, c->as_jint_bits());
 623         __ strw(rscratch1, frame_map()->address_for_slot(dest->single_stack_ix()));
 624       }
 625     }
 626     break;
 627   case T_LONG:
 628   case T_DOUBLE:
 629     {
 630       Register reg = zr;
 631       if (c->as_jlong_bits() == 0)
 632         __ str(zr, frame_map()->address_for_slot(dest->double_stack_ix(),
 633                                                  lo_word_offset_in_bytes));
 634       else {
 635         __ mov(rscratch1, (intptr_t)c->as_jlong_bits());
 636         __ str(rscratch1, frame_map()->address_for_slot(dest->double_stack_ix(),
 637                                                         lo_word_offset_in_bytes));
 638       }
 639     }
 640     break;
 641   default:
 642     ShouldNotReachHere();
 643   }
 644 }
 645 
 646 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 647   assert(src->is_constant(), "should not call otherwise");
 648   LIR_Const* c = src->as_constant_ptr();
 649   LIR_Address* to_addr = dest->as_address_ptr();
 650 
 651   void (Assembler::* insn)(Register Rt, const Address &adr);
 652 
 653   switch (type) {
 654   case T_ADDRESS:
 655     assert(c->as_jint() == 0, "should be");
 656     insn = &Assembler::str;
 657     break;
 658   case T_LONG:
 659     assert(c->as_jlong() == 0, "should be");
 660     insn = &Assembler::str;
 661     break;
 662   case T_INT:
 663     assert(c->as_jint() == 0, "should be");
 664     insn = &Assembler::strw;
 665     break;
 666   case T_OBJECT:
 667   case T_ARRAY:
 668     assert(c->as_jobject() == nullptr, "should be");
 669     if (UseCompressedOops && !wide) {
 670       insn = &Assembler::strw;
 671     } else {
 672       insn = &Assembler::str;
 673     }
 674     break;
 675   case T_CHAR:
 676   case T_SHORT:
 677     assert(c->as_jint() == 0, "should be");
 678     insn = &Assembler::strh;
 679     break;
 680   case T_BOOLEAN:
 681   case T_BYTE:
 682     assert(c->as_jint() == 0, "should be");
 683     insn = &Assembler::strb;
 684     break;
 685   default:
 686     ShouldNotReachHere();
 687     insn = &Assembler::str;  // unreachable
 688   }
 689 
 690   if (info) add_debug_info_for_null_check_here(info);
 691   (_masm->*insn)(zr, as_Address(to_addr, rscratch1));
 692 }
 693 
 694 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
 695   assert(src->is_register(), "should not call otherwise");
 696   assert(dest->is_register(), "should not call otherwise");
 697 
 698   // move between cpu-registers
 699   if (dest->is_single_cpu()) {
 700     if (src->type() == T_LONG) {
 701       // Can do LONG -> OBJECT
 702       move_regs(src->as_register_lo(), dest->as_register());
 703       return;
 704     }
 705     assert(src->is_single_cpu(), "must match");
 706     if (src->type() == T_OBJECT) {
 707       __ verify_oop(src->as_register());
 708     }
 709     move_regs(src->as_register(), dest->as_register());
 710 
 711   } else if (dest->is_double_cpu()) {
 712     if (is_reference_type(src->type())) {
 713       // Surprising to me but we can see move of a long to t_object
 714       __ verify_oop(src->as_register());
 715       move_regs(src->as_register(), dest->as_register_lo());
 716       return;
 717     }
 718     assert(src->is_double_cpu(), "must match");
 719     Register f_lo = src->as_register_lo();
 720     Register f_hi = src->as_register_hi();
 721     Register t_lo = dest->as_register_lo();
 722     Register t_hi = dest->as_register_hi();
 723     assert(f_hi == f_lo, "must be same");
 724     assert(t_hi == t_lo, "must be same");
 725     move_regs(f_lo, t_lo);
 726 
 727   } else if (dest->is_single_fpu()) {
 728     __ fmovs(dest->as_float_reg(), src->as_float_reg());
 729 
 730   } else if (dest->is_double_fpu()) {
 731     __ fmovd(dest->as_double_reg(), src->as_double_reg());
 732 
 733   } else {
 734     ShouldNotReachHere();
 735   }
 736 }
 737 
 738 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 739   precond(src->is_register() && dest->is_stack());
 740 
 741   uint const c_sz32 = sizeof(uint32_t);
 742   uint const c_sz64 = sizeof(uint64_t);
 743 
 744   if (src->is_single_cpu()) {
 745     int index = dest->single_stack_ix();
 746     if (is_reference_type(type)) {
 747       __ str(src->as_register(), stack_slot_address(index, c_sz64, rscratch1));
 748       __ verify_oop(src->as_register());
 749     } else if (type == T_METADATA || type == T_DOUBLE || type == T_ADDRESS) {
 750       __ str(src->as_register(), stack_slot_address(index, c_sz64, rscratch1));
 751     } else {
 752       __ strw(src->as_register(), stack_slot_address(index, c_sz32, rscratch1));
 753     }
 754 
 755   } else if (src->is_double_cpu()) {
 756     int index = dest->double_stack_ix();
 757     Address dest_addr_LO = stack_slot_address(index, c_sz64, rscratch1, lo_word_offset_in_bytes);
 758     __ str(src->as_register_lo(), dest_addr_LO);
 759 
 760   } else if (src->is_single_fpu()) {
 761     int index = dest->single_stack_ix();
 762     __ strs(src->as_float_reg(), stack_slot_address(index, c_sz32, rscratch1));
 763 
 764   } else if (src->is_double_fpu()) {
 765     int index = dest->double_stack_ix();
 766     __ strd(src->as_double_reg(), stack_slot_address(index, c_sz64, rscratch1));
 767 
 768   } else {
 769     ShouldNotReachHere();
 770   }
 771 }
 772 
 773 
 774 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide) {
 775   LIR_Address* to_addr = dest->as_address_ptr();
 776   PatchingStub* patch = nullptr;
 777   Register compressed_src = rscratch1;
 778 
 779   if (patch_code != lir_patch_none) {
 780     deoptimize_trap(info);
 781     return;
 782   }
 783 
 784   if (is_reference_type(type)) {
 785     __ verify_oop(src->as_register());
 786 
 787     if (UseCompressedOops && !wide) {
 788       __ encode_heap_oop(compressed_src, src->as_register());
 789     } else {
 790       compressed_src = src->as_register();
 791     }
 792   }
 793 
 794   int null_check_here = code_offset();
 795   switch (type) {
 796     case T_FLOAT: {
 797       __ strs(src->as_float_reg(), as_Address(to_addr));
 798       break;
 799     }
 800 
 801     case T_DOUBLE: {
 802       __ strd(src->as_double_reg(), as_Address(to_addr));
 803       break;
 804     }
 805 
 806     case T_ARRAY:   // fall through
 807     case T_OBJECT:  // fall through
 808       if (UseCompressedOops && !wide) {
 809         __ strw(compressed_src, as_Address(to_addr, rscratch2));
 810       } else {
 811          __ str(compressed_src, as_Address(to_addr));
 812       }
 813       break;
 814     case T_METADATA:
 815       // We get here to store a method pointer to the stack to pass to
 816       // a dtrace runtime call. This can't work on 64 bit with
 817       // compressed klass ptrs: T_METADATA can be a compressed klass
 818       // ptr or a 64 bit method pointer.
 819       ShouldNotReachHere();
 820       __ str(src->as_register(), as_Address(to_addr));
 821       break;
 822     case T_ADDRESS:
 823       __ str(src->as_register(), as_Address(to_addr));
 824       break;
 825     case T_INT:
 826       __ strw(src->as_register(), as_Address(to_addr));
 827       break;
 828 
 829     case T_LONG: {
 830       __ str(src->as_register_lo(), as_Address_lo(to_addr));
 831       break;
 832     }
 833 
 834     case T_BYTE:    // fall through
 835     case T_BOOLEAN: {
 836       __ strb(src->as_register(), as_Address(to_addr));
 837       break;
 838     }
 839 
 840     case T_CHAR:    // fall through
 841     case T_SHORT:
 842       __ strh(src->as_register(), as_Address(to_addr));
 843       break;
 844 
 845     default:
 846       ShouldNotReachHere();
 847   }
 848   if (info != nullptr) {
 849     add_debug_info_for_null_check(null_check_here, info);
 850   }
 851 }
 852 
 853 
 854 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 855   precond(src->is_stack() && dest->is_register());
 856 
 857   uint const c_sz32 = sizeof(uint32_t);
 858   uint const c_sz64 = sizeof(uint64_t);
 859 
 860   if (dest->is_single_cpu()) {
 861     int index = src->single_stack_ix();
 862     if (is_reference_type(type)) {
 863       __ ldr(dest->as_register(), stack_slot_address(index, c_sz64, rscratch1));
 864       __ verify_oop(dest->as_register());
 865     } else if (type == T_METADATA || type == T_ADDRESS) {
 866       __ ldr(dest->as_register(), stack_slot_address(index, c_sz64, rscratch1));
 867     } else {
 868       __ ldrw(dest->as_register(), stack_slot_address(index, c_sz32, rscratch1));
 869     }
 870 
 871   } else if (dest->is_double_cpu()) {
 872     int index = src->double_stack_ix();
 873     Address src_addr_LO = stack_slot_address(index, c_sz64, rscratch1, lo_word_offset_in_bytes);
 874     __ ldr(dest->as_register_lo(), src_addr_LO);
 875 
 876   } else if (dest->is_single_fpu()) {
 877     int index = src->single_stack_ix();
 878     __ ldrs(dest->as_float_reg(), stack_slot_address(index, c_sz32, rscratch1));
 879 
 880   } else if (dest->is_double_fpu()) {
 881     int index = src->double_stack_ix();
 882     __ ldrd(dest->as_double_reg(), stack_slot_address(index, c_sz64, rscratch1));
 883 
 884   } else {
 885     ShouldNotReachHere();
 886   }
 887 }
 888 
 889 
 890 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
 891   address target = nullptr;
 892   relocInfo::relocType reloc_type = relocInfo::none;
 893 
 894   switch (patching_id(info)) {
 895   case PatchingStub::access_field_id:
 896     target = Runtime1::entry_for(StubId::c1_access_field_patching_id);
 897     reloc_type = relocInfo::section_word_type;
 898     break;
 899   case PatchingStub::load_klass_id:
 900     target = Runtime1::entry_for(StubId::c1_load_klass_patching_id);
 901     reloc_type = relocInfo::metadata_type;
 902     break;
 903   case PatchingStub::load_mirror_id:
 904     target = Runtime1::entry_for(StubId::c1_load_mirror_patching_id);
 905     reloc_type = relocInfo::oop_type;
 906     break;
 907   case PatchingStub::load_appendix_id:
 908     target = Runtime1::entry_for(StubId::c1_load_appendix_patching_id);
 909     reloc_type = relocInfo::oop_type;
 910     break;
 911   default: ShouldNotReachHere();
 912   }
 913 
 914   __ far_call(RuntimeAddress(target));
 915   add_call_info_here(info);
 916 }
 917 
 918 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 919 
 920   LIR_Opr temp;
 921   if (type == T_LONG || type == T_DOUBLE)
 922     temp = FrameMap::rscratch1_long_opr;
 923   else
 924     temp = FrameMap::rscratch1_opr;
 925 
 926   stack2reg(src, temp, src->type());
 927   reg2stack(temp, dest, dest->type());
 928 }
 929 
 930 
 931 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide) {
 932   LIR_Address* addr = src->as_address_ptr();
 933   LIR_Address* from_addr = src->as_address_ptr();
 934 
 935   if (addr->base()->type() == T_OBJECT) {
 936     __ verify_oop(addr->base()->as_pointer_register());
 937   }
 938 
 939   if (patch_code != lir_patch_none) {
 940     deoptimize_trap(info);
 941     return;
 942   }
 943 
 944   if (info != nullptr) {
 945     add_debug_info_for_null_check_here(info);
 946   }
 947   int null_check_here = code_offset();
 948   switch (type) {
 949     case T_FLOAT: {
 950       __ ldrs(dest->as_float_reg(), as_Address(from_addr));
 951       break;
 952     }
 953 
 954     case T_DOUBLE: {
 955       __ ldrd(dest->as_double_reg(), as_Address(from_addr));
 956       break;
 957     }
 958 
 959     case T_ARRAY:   // fall through
 960     case T_OBJECT:  // fall through
 961       if (UseCompressedOops && !wide) {
 962         __ ldrw(dest->as_register(), as_Address(from_addr));
 963       } else {
 964         __ ldr(dest->as_register(), as_Address(from_addr));
 965       }
 966       break;
 967     case T_METADATA:
 968       // We get here to store a method pointer to the stack to pass to
 969       // a dtrace runtime call. This can't work on 64 bit with
 970       // compressed klass ptrs: T_METADATA can be a compressed klass
 971       // ptr or a 64 bit method pointer.
 972       ShouldNotReachHere();
 973       __ ldr(dest->as_register(), as_Address(from_addr));
 974       break;
 975     case T_ADDRESS:
 976       __ ldr(dest->as_register(), as_Address(from_addr));
 977       break;
 978     case T_INT:
 979       __ ldrw(dest->as_register(), as_Address(from_addr));
 980       break;
 981 
 982     case T_LONG: {
 983       __ ldr(dest->as_register_lo(), as_Address_lo(from_addr));
 984       break;
 985     }
 986 
 987     case T_BYTE:
 988       __ ldrsb(dest->as_register(), as_Address(from_addr));
 989       break;
 990     case T_BOOLEAN: {
 991       __ ldrb(dest->as_register(), as_Address(from_addr));
 992       break;
 993     }
 994 
 995     case T_CHAR:
 996       __ ldrh(dest->as_register(), as_Address(from_addr));
 997       break;
 998     case T_SHORT:
 999       __ ldrsh(dest->as_register(), as_Address(from_addr));
1000       break;
1001 
1002     default:
1003       ShouldNotReachHere();
1004   }
1005 
1006   if (is_reference_type(type)) {
1007     if (UseCompressedOops && !wide) {
1008       __ decode_heap_oop(dest->as_register());
1009     }
1010 
1011     __ verify_oop(dest->as_register());
1012   }
1013 }
1014 
1015 
1016 int LIR_Assembler::array_element_size(BasicType type) const {
1017   int elem_size = type2aelembytes(type);
1018   return exact_log2(elem_size);
1019 }
1020 
1021 
1022 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1023   switch (op->code()) {
1024   case lir_idiv:
1025   case lir_irem:
1026     arithmetic_idiv(op->code(),
1027                     op->in_opr1(),
1028                     op->in_opr2(),
1029                     op->in_opr3(),
1030                     op->result_opr(),
1031                     op->info());
1032     break;
1033   case lir_fmad:
1034     __ fmaddd(op->result_opr()->as_double_reg(),
1035               op->in_opr1()->as_double_reg(),
1036               op->in_opr2()->as_double_reg(),
1037               op->in_opr3()->as_double_reg());
1038     break;
1039   case lir_fmaf:
1040     __ fmadds(op->result_opr()->as_float_reg(),
1041               op->in_opr1()->as_float_reg(),
1042               op->in_opr2()->as_float_reg(),
1043               op->in_opr3()->as_float_reg());
1044     break;
1045   default:      ShouldNotReachHere(); break;
1046   }
1047 }
1048 
1049 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1050 #ifdef ASSERT
1051   assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
1052   if (op->block() != nullptr)  _branch_target_blocks.append(op->block());
1053   if (op->ublock() != nullptr) _branch_target_blocks.append(op->ublock());
1054 #endif
1055 
1056   if (op->cond() == lir_cond_always) {
1057     if (op->info() != nullptr) add_debug_info_for_branch(op->info());
1058     __ b(*(op->label()));
1059   } else {
1060     Assembler::Condition acond;
1061     if (op->code() == lir_cond_float_branch) {
1062       bool is_unordered = (op->ublock() == op->block());
1063       // Assembler::EQ does not permit unordered branches, so we add
1064       // another branch here.  Likewise, Assembler::NE does not permit
1065       // ordered branches.
1066       if ((is_unordered && op->cond() == lir_cond_equal)
1067           || (!is_unordered && op->cond() == lir_cond_notEqual))
1068         __ br(Assembler::VS, *(op->ublock()->label()));
1069       switch(op->cond()) {
1070       case lir_cond_equal:        acond = Assembler::EQ; break;
1071       case lir_cond_notEqual:     acond = Assembler::NE; break;
1072       case lir_cond_less:         acond = (is_unordered ? Assembler::LT : Assembler::LO); break;
1073       case lir_cond_lessEqual:    acond = (is_unordered ? Assembler::LE : Assembler::LS); break;
1074       case lir_cond_greaterEqual: acond = (is_unordered ? Assembler::HS : Assembler::GE); break;
1075       case lir_cond_greater:      acond = (is_unordered ? Assembler::HI : Assembler::GT); break;
1076       default:                    ShouldNotReachHere();
1077         acond = Assembler::EQ;  // unreachable
1078       }
1079     } else {
1080       switch (op->cond()) {
1081         case lir_cond_equal:        acond = Assembler::EQ; break;
1082         case lir_cond_notEqual:     acond = Assembler::NE; break;
1083         case lir_cond_less:         acond = Assembler::LT; break;
1084         case lir_cond_lessEqual:    acond = Assembler::LE; break;
1085         case lir_cond_greaterEqual: acond = Assembler::GE; break;
1086         case lir_cond_greater:      acond = Assembler::GT; break;
1087         case lir_cond_belowEqual:   acond = Assembler::LS; break;
1088         case lir_cond_aboveEqual:   acond = Assembler::HS; break;
1089         default:                    ShouldNotReachHere();
1090           acond = Assembler::EQ;  // unreachable
1091       }
1092     }
1093     __ br(acond,*(op->label()));
1094   }
1095 }
1096 
1097 
1098 
1099 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1100   LIR_Opr src  = op->in_opr();
1101   LIR_Opr dest = op->result_opr();
1102 
1103   switch (op->bytecode()) {
1104     case Bytecodes::_i2f:
1105       {
1106         __ scvtfws(dest->as_float_reg(), src->as_register());
1107         break;
1108       }
1109     case Bytecodes::_i2d:
1110       {
1111         __ scvtfwd(dest->as_double_reg(), src->as_register());
1112         break;
1113       }
1114     case Bytecodes::_l2d:
1115       {
1116         __ scvtfd(dest->as_double_reg(), src->as_register_lo());
1117         break;
1118       }
1119     case Bytecodes::_l2f:
1120       {
1121         __ scvtfs(dest->as_float_reg(), src->as_register_lo());
1122         break;
1123       }
1124     case Bytecodes::_f2d:
1125       {
1126         __ fcvts(dest->as_double_reg(), src->as_float_reg());
1127         break;
1128       }
1129     case Bytecodes::_d2f:
1130       {
1131         __ fcvtd(dest->as_float_reg(), src->as_double_reg());
1132         break;
1133       }
1134     case Bytecodes::_i2c:
1135       {
1136         __ ubfx(dest->as_register(), src->as_register(), 0, 16);
1137         break;
1138       }
1139     case Bytecodes::_i2l:
1140       {
1141         __ sxtw(dest->as_register_lo(), src->as_register());
1142         break;
1143       }
1144     case Bytecodes::_i2s:
1145       {
1146         __ sxth(dest->as_register(), src->as_register());
1147         break;
1148       }
1149     case Bytecodes::_i2b:
1150       {
1151         __ sxtb(dest->as_register(), src->as_register());
1152         break;
1153       }
1154     case Bytecodes::_l2i:
1155       {
1156         _masm->block_comment("FIXME: This could be a no-op");
1157         __ uxtw(dest->as_register(), src->as_register_lo());
1158         break;
1159       }
1160     case Bytecodes::_d2l:
1161       {
1162         __ fcvtzd(dest->as_register_lo(), src->as_double_reg());
1163         break;
1164       }
1165     case Bytecodes::_f2i:
1166       {
1167         __ fcvtzsw(dest->as_register(), src->as_float_reg());
1168         break;
1169       }
1170     case Bytecodes::_f2l:
1171       {
1172         __ fcvtzs(dest->as_register_lo(), src->as_float_reg());
1173         break;
1174       }
1175     case Bytecodes::_d2i:
1176       {
1177         __ fcvtzdw(dest->as_register(), src->as_double_reg());
1178         break;
1179       }
1180     default: ShouldNotReachHere();
1181   }
1182 }
1183 
1184 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1185   if (op->init_check()) {
1186     __ lea(rscratch1, Address(op->klass()->as_register(), InstanceKlass::init_state_offset()));
1187     __ ldarb(rscratch1, rscratch1);
1188     __ cmpw(rscratch1, InstanceKlass::fully_initialized);
1189     add_debug_info_for_null_check_here(op->stub()->info());
1190     __ br(Assembler::NE, *op->stub()->entry());
1191   }
1192   __ allocate_object(op->obj()->as_register(),
1193                      op->tmp1()->as_register(),
1194                      op->tmp2()->as_register(),
1195                      op->header_size(),
1196                      op->object_size(),
1197                      op->klass()->as_register(),
1198                      *op->stub()->entry());
1199   __ bind(*op->stub()->continuation());
1200 }
1201 
1202 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1203   Register len =  op->len()->as_register();
1204   __ uxtw(len, len);
1205 
1206   if (UseSlowPath ||
1207       (!UseFastNewObjectArray && is_reference_type(op->type())) ||
1208       (!UseFastNewTypeArray   && !is_reference_type(op->type()))) {
1209     __ b(*op->stub()->entry());
1210   } else {
1211     Register tmp1 = op->tmp1()->as_register();
1212     Register tmp2 = op->tmp2()->as_register();
1213     Register tmp3 = op->tmp3()->as_register();
1214     if (len == tmp1) {
1215       tmp1 = tmp3;
1216     } else if (len == tmp2) {
1217       tmp2 = tmp3;
1218     } else if (len == tmp3) {
1219       // everything is ok
1220     } else {
1221       __ mov(tmp3, len);
1222     }
1223     __ allocate_array(op->obj()->as_register(),
1224                       len,
1225                       tmp1,
1226                       tmp2,
1227                       arrayOopDesc::base_offset_in_bytes(op->type()),
1228                       array_element_size(op->type()),
1229                       op->klass()->as_register(),
1230                       *op->stub()->entry(),
1231                       op->zero_array());
1232   }
1233   __ bind(*op->stub()->continuation());
1234 }
1235 
1236 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md,
1237                                         ciProfileData *data, Register recv) {
1238 
1239   int mdp_offset = md->byte_offset_of_slot(data, in_ByteSize(0));
1240   __ profile_receiver_type(recv, mdo, mdp_offset);
1241 }
1242 
1243 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1244   // we always need a stub for the failure case.
1245   CodeStub* stub = op->stub();
1246   Register obj = op->object()->as_register();
1247   Register k_RInfo = op->tmp1()->as_register();
1248   Register klass_RInfo = op->tmp2()->as_register();
1249   Register dst = op->result_opr()->as_register();
1250   ciKlass* k = op->klass();
1251   Register Rtmp1 = noreg;
1252 
1253   // check if it needs to be profiled
1254   ciMethodData* md;
1255   ciProfileData* data;
1256 
1257   const bool should_profile = op->should_profile();
1258 
1259   if (should_profile) {
1260     ciMethod* method = op->profiled_method();
1261     assert(method != nullptr, "Should have method");
1262     int bci = op->profiled_bci();
1263     md = method->method_data_or_null();
1264     assert(md != nullptr, "Sanity");
1265     data = md->bci_to_data(bci);
1266     assert(data != nullptr,                "need data for type check");
1267     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1268   }
1269   Label* success_target = success;
1270   Label* failure_target = failure;
1271 
1272   if (obj == k_RInfo) {
1273     k_RInfo = dst;
1274   } else if (obj == klass_RInfo) {
1275     klass_RInfo = dst;
1276   }
1277   if (k->is_loaded() && !UseCompressedClassPointers) {
1278     select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1279   } else {
1280     Rtmp1 = op->tmp3()->as_register();
1281     select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1282   }
1283 
1284   assert_different_registers(obj, k_RInfo, klass_RInfo);
1285 
1286   if (should_profile) {
1287     Register mdo  = klass_RInfo;
1288     __ mov_metadata(mdo, md->constant_encoding());
1289     Label not_null;
1290     __ cbnz(obj, not_null);
1291     // Object is null; update MDO and exit
1292     Address data_addr
1293       = __ form_address(rscratch2, mdo,
1294                         md->byte_offset_of_slot(data, DataLayout::flags_offset()),
1295                         0);
1296     __ ldrb(rscratch1, data_addr);
1297     __ orr(rscratch1, rscratch1, BitData::null_seen_byte_constant());
1298     __ strb(rscratch1, data_addr);
1299     __ b(*obj_is_null);
1300     __ bind(not_null);
1301 
1302     Register recv = k_RInfo;
1303     __ load_klass(recv, obj);
1304     type_profile_helper(mdo, md, data, recv);
1305   } else {
1306     __ cbz(obj, *obj_is_null);
1307   }
1308 
1309   if (!k->is_loaded()) {
1310     klass2reg_with_patching(k_RInfo, op->info_for_patch());
1311   } else {
1312     __ mov_metadata(k_RInfo, k->constant_encoding());
1313   }
1314   __ verify_oop(obj);
1315 
1316   if (op->fast_check()) {
1317     // get object class
1318     // not a safepoint as obj null check happens earlier
1319     __ load_klass(rscratch1, obj);
1320     __ cmp( rscratch1, k_RInfo);
1321 
1322     __ br(Assembler::NE, *failure_target);
1323     // successful cast, fall through to profile or jump
1324   } else {
1325     // get object class
1326     // not a safepoint as obj null check happens earlier
1327     __ load_klass(klass_RInfo, obj);
1328     if (k->is_loaded()) {
1329       // See if we get an immediate positive hit
1330       __ ldr(rscratch1, Address(klass_RInfo, int64_t(k->super_check_offset())));
1331       __ cmp(k_RInfo, rscratch1);
1332       if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
1333         __ br(Assembler::NE, *failure_target);
1334         // successful cast, fall through to profile or jump
1335       } else {
1336         // See if we get an immediate positive hit
1337         __ br(Assembler::EQ, *success_target);
1338         // check for self
1339         __ cmp(klass_RInfo, k_RInfo);
1340         __ br(Assembler::EQ, *success_target);
1341 
1342         __ stp(klass_RInfo, k_RInfo, Address(__ pre(sp, -2 * wordSize)));
1343         __ far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_slow_subtype_check_id)));
1344         __ ldr(klass_RInfo, Address(__ post(sp, 2 * wordSize)));
1345         // result is a boolean
1346         __ cbzw(klass_RInfo, *failure_target);
1347         // successful cast, fall through to profile or jump
1348       }
1349     } else {
1350       // perform the fast part of the checking logic
1351       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, nullptr);
1352       // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1353       __ stp(klass_RInfo, k_RInfo, Address(__ pre(sp, -2 * wordSize)));
1354       __ far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_slow_subtype_check_id)));
1355       __ ldp(k_RInfo, klass_RInfo, Address(__ post(sp, 2 * wordSize)));
1356       // result is a boolean
1357       __ cbz(k_RInfo, *failure_target);
1358       // successful cast, fall through to profile or jump
1359     }
1360   }
1361   __ b(*success);
1362 }
1363 
1364 
1365 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1366   const bool should_profile = op->should_profile();
1367 
1368   LIR_Code code = op->code();
1369   if (code == lir_store_check) {
1370     Register value = op->object()->as_register();
1371     Register array = op->array()->as_register();
1372     Register k_RInfo = op->tmp1()->as_register();
1373     Register klass_RInfo = op->tmp2()->as_register();
1374     Register Rtmp1 = op->tmp3()->as_register();
1375 
1376     CodeStub* stub = op->stub();
1377 
1378     // check if it needs to be profiled
1379     ciMethodData* md;
1380     ciProfileData* data;
1381 
1382     if (should_profile) {
1383       ciMethod* method = op->profiled_method();
1384       assert(method != nullptr, "Should have method");
1385       int bci = op->profiled_bci();
1386       md = method->method_data_or_null();
1387       assert(md != nullptr, "Sanity");
1388       data = md->bci_to_data(bci);
1389       assert(data != nullptr,                "need data for type check");
1390       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1391     }
1392     Label done;
1393     Label* success_target = &done;
1394     Label* failure_target = stub->entry();
1395 
1396     if (should_profile) {
1397       Label not_null;
1398       Register mdo  = klass_RInfo;
1399       __ mov_metadata(mdo, md->constant_encoding());
1400       __ cbnz(value, not_null);
1401       // Object is null; update MDO and exit
1402       Address data_addr
1403         = __ form_address(rscratch2, mdo,
1404                           md->byte_offset_of_slot(data, DataLayout::flags_offset()), 0);
1405       __ ldrb(rscratch1, data_addr);
1406       __ orr(rscratch1, rscratch1, BitData::null_seen_byte_constant());
1407       __ strb(rscratch1, data_addr);
1408       __ b(done);
1409       __ bind(not_null);
1410 
1411       Register recv = k_RInfo;
1412       __ load_klass(recv, value);
1413       type_profile_helper(mdo, md, data, recv);
1414     } else {
1415       __ cbz(value, done);
1416     }
1417 
1418     add_debug_info_for_null_check_here(op->info_for_exception());
1419     __ load_klass(k_RInfo, array);
1420     __ load_klass(klass_RInfo, value);
1421 
1422     // get instance klass (it's already uncompressed)
1423     __ ldr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
1424     // perform the fast part of the checking logic
1425     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, nullptr);
1426     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1427     __ stp(klass_RInfo, k_RInfo, Address(__ pre(sp, -2 * wordSize)));
1428     __ far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_slow_subtype_check_id)));
1429     __ ldp(k_RInfo, klass_RInfo, Address(__ post(sp, 2 * wordSize)));
1430     // result is a boolean
1431     __ cbzw(k_RInfo, *failure_target);
1432     // fall through to the success case
1433 
1434     __ bind(done);
1435   } else if (code == lir_checkcast) {
1436     Register obj = op->object()->as_register();
1437     Register dst = op->result_opr()->as_register();
1438     Label success;
1439     emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1440     __ bind(success);
1441     if (dst != obj) {
1442       __ mov(dst, obj);
1443     }
1444   } else if (code == lir_instanceof) {
1445     Register obj = op->object()->as_register();
1446     Register dst = op->result_opr()->as_register();
1447     Label success, failure, done;
1448     emit_typecheck_helper(op, &success, &failure, &failure);
1449     __ bind(failure);
1450     __ mov(dst, zr);
1451     __ b(done);
1452     __ bind(success);
1453     __ mov(dst, 1);
1454     __ bind(done);
1455   } else {
1456     ShouldNotReachHere();
1457   }
1458 }
1459 
1460 void LIR_Assembler::casw(Register addr, Register newval, Register cmpval) {
1461   __ cmpxchg(addr, cmpval, newval, Assembler::word, /* acquire*/ true, /* release*/ true, /* weak*/ false, rscratch1);
1462   __ cset(rscratch1, Assembler::NE);
1463   __ membar(__ AnyAny);
1464 }
1465 
1466 void LIR_Assembler::casl(Register addr, Register newval, Register cmpval) {
1467   __ cmpxchg(addr, cmpval, newval, Assembler::xword, /* acquire*/ true, /* release*/ true, /* weak*/ false, rscratch1);
1468   __ cset(rscratch1, Assembler::NE);
1469   __ membar(__ AnyAny);
1470 }
1471 
1472 
1473 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1474   Register addr;
1475   if (op->addr()->is_register()) {
1476     addr = as_reg(op->addr());
1477   } else {
1478     assert(op->addr()->is_address(), "what else?");
1479     LIR_Address* addr_ptr = op->addr()->as_address_ptr();
1480     assert(addr_ptr->disp() == 0, "need 0 disp");
1481     assert(addr_ptr->index() == LIR_Opr::illegalOpr(), "need 0 index");
1482     addr = as_reg(addr_ptr->base());
1483   }
1484   Register newval = as_reg(op->new_value());
1485   Register cmpval = as_reg(op->cmp_value());
1486 
1487   if (op->code() == lir_cas_obj) {
1488     if (UseCompressedOops) {
1489       Register t1 = op->tmp1()->as_register();
1490       assert(op->tmp1()->is_valid(), "must be");
1491       __ encode_heap_oop(t1, cmpval);
1492       cmpval = t1;
1493       __ encode_heap_oop(rscratch2, newval);
1494       newval = rscratch2;
1495       casw(addr, newval, cmpval);
1496     } else {
1497       casl(addr, newval, cmpval);
1498     }
1499   } else if (op->code() == lir_cas_int) {
1500     casw(addr, newval, cmpval);
1501   } else {
1502     casl(addr, newval, cmpval);
1503   }
1504 }
1505 
1506 
1507 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1508                           LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1509   assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on aarch64");
1510 
1511   Assembler::Condition acond, ncond;
1512   switch (condition) {
1513   case lir_cond_equal:        acond = Assembler::EQ; ncond = Assembler::NE; break;
1514   case lir_cond_notEqual:     acond = Assembler::NE; ncond = Assembler::EQ; break;
1515   case lir_cond_less:         acond = Assembler::LT; ncond = Assembler::GE; break;
1516   case lir_cond_lessEqual:    acond = Assembler::LE; ncond = Assembler::GT; break;
1517   case lir_cond_greaterEqual: acond = Assembler::GE; ncond = Assembler::LT; break;
1518   case lir_cond_greater:      acond = Assembler::GT; ncond = Assembler::LE; break;
1519   case lir_cond_belowEqual:
1520   case lir_cond_aboveEqual:
1521   default:                    ShouldNotReachHere();
1522     acond = Assembler::EQ; ncond = Assembler::NE;  // unreachable
1523   }
1524 
1525   assert(result->is_single_cpu() || result->is_double_cpu(),
1526          "expect single register for result");
1527   if (opr1->is_constant() && opr2->is_constant()
1528       && opr1->type() == T_INT && opr2->type() == T_INT) {
1529     jint val1 = opr1->as_jint();
1530     jint val2 = opr2->as_jint();
1531     if (val1 == 0 && val2 == 1) {
1532       __ cset(result->as_register(), ncond);
1533       return;
1534     } else if (val1 == 1 && val2 == 0) {
1535       __ cset(result->as_register(), acond);
1536       return;
1537     }
1538   }
1539 
1540   if (opr1->is_constant() && opr2->is_constant()
1541       && opr1->type() == T_LONG && opr2->type() == T_LONG) {
1542     jlong val1 = opr1->as_jlong();
1543     jlong val2 = opr2->as_jlong();
1544     if (val1 == 0 && val2 == 1) {
1545       __ cset(result->as_register_lo(), ncond);
1546       return;
1547     } else if (val1 == 1 && val2 == 0) {
1548       __ cset(result->as_register_lo(), acond);
1549       return;
1550     }
1551   }
1552 
1553   if (opr1->is_stack()) {
1554     stack2reg(opr1, FrameMap::rscratch1_opr, result->type());
1555     opr1 = FrameMap::rscratch1_opr;
1556   } else if (opr1->is_constant()) {
1557     LIR_Opr tmp
1558       = opr1->type() == T_LONG ? FrameMap::rscratch1_long_opr : FrameMap::rscratch1_opr;
1559     const2reg(opr1, tmp, lir_patch_none, nullptr);
1560     opr1 = tmp;
1561   }
1562 
1563   if (opr2->is_stack()) {
1564     stack2reg(opr2, FrameMap::rscratch2_opr, result->type());
1565     opr2 = FrameMap::rscratch2_opr;
1566   } else if (opr2->is_constant()) {
1567     LIR_Opr tmp
1568       = opr2->type() == T_LONG ? FrameMap::rscratch2_long_opr : FrameMap::rscratch2_opr;
1569     const2reg(opr2, tmp, lir_patch_none, nullptr);
1570     opr2 = tmp;
1571   }
1572 
1573   if (result->type() == T_LONG)
1574     __ csel(result->as_register_lo(), opr1->as_register_lo(), opr2->as_register_lo(), acond);
1575   else
1576     __ csel(result->as_register(), opr1->as_register(), opr2->as_register(), acond);
1577 }
1578 
1579 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info) {
1580   assert(info == nullptr, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1581 
1582   if (left->is_single_cpu()) {
1583     Register lreg = left->as_register();
1584     Register dreg = as_reg(dest);
1585 
1586     if (right->is_single_cpu()) {
1587       // cpu register - cpu register
1588 
1589       assert(left->type() == T_INT && right->type() == T_INT && dest->type() == T_INT,
1590              "should be");
1591       Register rreg = right->as_register();
1592       switch (code) {
1593       case lir_add: __ addw (dest->as_register(), lreg, rreg); break;
1594       case lir_sub: __ subw (dest->as_register(), lreg, rreg); break;
1595       case lir_mul: __ mulw (dest->as_register(), lreg, rreg); break;
1596       default:      ShouldNotReachHere();
1597       }
1598 
1599     } else if (right->is_double_cpu()) {
1600       Register rreg = right->as_register_lo();
1601       // single_cpu + double_cpu: can happen with obj+long
1602       assert(code == lir_add || code == lir_sub, "mismatched arithmetic op");
1603       switch (code) {
1604       case lir_add: __ add(dreg, lreg, rreg); break;
1605       case lir_sub: __ sub(dreg, lreg, rreg); break;
1606       default: ShouldNotReachHere();
1607       }
1608     } else if (right->is_constant()) {
1609       // cpu register - constant
1610       jlong c;
1611 
1612       // FIXME.  This is fugly: we really need to factor all this logic.
1613       switch(right->type()) {
1614       case T_LONG:
1615         c = right->as_constant_ptr()->as_jlong();
1616         break;
1617       case T_INT:
1618       case T_ADDRESS:
1619         c = right->as_constant_ptr()->as_jint();
1620         break;
1621       default:
1622         ShouldNotReachHere();
1623         c = 0;  // unreachable
1624         break;
1625       }
1626 
1627       assert(code == lir_add || code == lir_sub, "mismatched arithmetic op");
1628       if (c == 0 && dreg == lreg) {
1629         COMMENT("effective nop elided");
1630         return;
1631       }
1632       switch(left->type()) {
1633       case T_INT:
1634         switch (code) {
1635         case lir_add: __ addw(dreg, lreg, c); break;
1636         case lir_sub: __ subw(dreg, lreg, c); break;
1637         default: ShouldNotReachHere();
1638         }
1639         break;
1640       case T_OBJECT:
1641       case T_ADDRESS:
1642         switch (code) {
1643         case lir_add: __ add(dreg, lreg, c); break;
1644         case lir_sub: __ sub(dreg, lreg, c); break;
1645         default: ShouldNotReachHere();
1646         }
1647         break;
1648       default:
1649         ShouldNotReachHere();
1650       }
1651     } else {
1652       ShouldNotReachHere();
1653     }
1654 
1655   } else if (left->is_double_cpu()) {
1656     Register lreg_lo = left->as_register_lo();
1657 
1658     if (right->is_double_cpu()) {
1659       // cpu register - cpu register
1660       Register rreg_lo = right->as_register_lo();
1661       switch (code) {
1662       case lir_add: __ add (dest->as_register_lo(), lreg_lo, rreg_lo); break;
1663       case lir_sub: __ sub (dest->as_register_lo(), lreg_lo, rreg_lo); break;
1664       case lir_mul: __ mul (dest->as_register_lo(), lreg_lo, rreg_lo); break;
1665       case lir_div: __ corrected_idivq(dest->as_register_lo(), lreg_lo, rreg_lo, false, rscratch1); break;
1666       case lir_rem: __ corrected_idivq(dest->as_register_lo(), lreg_lo, rreg_lo, true, rscratch1); break;
1667       default:
1668         ShouldNotReachHere();
1669       }
1670 
1671     } else if (right->is_constant()) {
1672       jlong c = right->as_constant_ptr()->as_jlong();
1673       Register dreg = as_reg(dest);
1674       switch (code) {
1675         case lir_add:
1676         case lir_sub:
1677           if (c == 0 && dreg == lreg_lo) {
1678             COMMENT("effective nop elided");
1679             return;
1680           }
1681           code == lir_add ? __ add(dreg, lreg_lo, c) : __ sub(dreg, lreg_lo, c);
1682           break;
1683         case lir_div:
1684           assert(c > 0 && is_power_of_2(c), "divisor must be power-of-2 constant");
1685           if (c == 1) {
1686             // move lreg_lo to dreg if divisor is 1
1687             __ mov(dreg, lreg_lo);
1688           } else {
1689             unsigned int shift = log2i_exact(c);
1690             // use rscratch1 as intermediate result register
1691             __ asr(rscratch1, lreg_lo, 63);
1692             __ add(rscratch1, lreg_lo, rscratch1, Assembler::LSR, 64 - shift);
1693             __ asr(dreg, rscratch1, shift);
1694           }
1695           break;
1696         case lir_rem:
1697           assert(c > 0 && is_power_of_2(c), "divisor must be power-of-2 constant");
1698           if (c == 1) {
1699             // move 0 to dreg if divisor is 1
1700             __ mov(dreg, zr);
1701           } else {
1702             // use rscratch1 as intermediate result register
1703             __ negs(rscratch1, lreg_lo);
1704             __ andr(dreg, lreg_lo, c - 1);
1705             __ andr(rscratch1, rscratch1, c - 1);
1706             __ csneg(dreg, dreg, rscratch1, Assembler::MI);
1707           }
1708           break;
1709         default:
1710           ShouldNotReachHere();
1711       }
1712     } else {
1713       ShouldNotReachHere();
1714     }
1715   } else if (left->is_single_fpu()) {
1716     assert(right->is_single_fpu(), "right hand side of float arithmetics needs to be float register");
1717     switch (code) {
1718     case lir_add: __ fadds (dest->as_float_reg(), left->as_float_reg(), right->as_float_reg()); break;
1719     case lir_sub: __ fsubs (dest->as_float_reg(), left->as_float_reg(), right->as_float_reg()); break;
1720     case lir_mul: __ fmuls (dest->as_float_reg(), left->as_float_reg(), right->as_float_reg()); break;
1721     case lir_div: __ fdivs (dest->as_float_reg(), left->as_float_reg(), right->as_float_reg()); break;
1722     default:
1723       ShouldNotReachHere();
1724     }
1725   } else if (left->is_double_fpu()) {
1726     if (right->is_double_fpu()) {
1727       // fpu register - fpu register
1728       switch (code) {
1729       case lir_add: __ faddd (dest->as_double_reg(), left->as_double_reg(), right->as_double_reg()); break;
1730       case lir_sub: __ fsubd (dest->as_double_reg(), left->as_double_reg(), right->as_double_reg()); break;
1731       case lir_mul: __ fmuld (dest->as_double_reg(), left->as_double_reg(), right->as_double_reg()); break;
1732       case lir_div: __ fdivd (dest->as_double_reg(), left->as_double_reg(), right->as_double_reg()); break;
1733       default:
1734         ShouldNotReachHere();
1735       }
1736     } else {
1737       if (right->is_constant()) {
1738         ShouldNotReachHere();
1739       }
1740       ShouldNotReachHere();
1741     }
1742   } else if (left->is_single_stack() || left->is_address()) {
1743     assert(left == dest, "left and dest must be equal");
1744     ShouldNotReachHere();
1745   } else {
1746     ShouldNotReachHere();
1747   }
1748 }
1749 
1750 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr tmp, LIR_Opr dest, LIR_Op* op) {
1751   switch(code) {
1752   case lir_abs : __ fabsd(dest->as_double_reg(), value->as_double_reg()); break;
1753   case lir_sqrt: __ fsqrtd(dest->as_double_reg(), value->as_double_reg()); break;
1754   case lir_f2hf: __ flt_to_flt16(dest->as_register(), value->as_float_reg(), tmp->as_float_reg()); break;
1755   case lir_hf2f: __ flt16_to_flt(dest->as_float_reg(), value->as_register(), tmp->as_float_reg()); break;
1756   default      : ShouldNotReachHere();
1757   }
1758 }
1759 
1760 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1761 
1762   assert(left->is_single_cpu() || left->is_double_cpu(), "expect single or double register");
1763   Register Rleft = left->is_single_cpu() ? left->as_register() :
1764                                            left->as_register_lo();
1765    if (dst->is_single_cpu()) {
1766      Register Rdst = dst->as_register();
1767      if (right->is_constant()) {
1768        switch (code) {
1769          case lir_logic_and: __ andw (Rdst, Rleft, right->as_jint()); break;
1770          case lir_logic_or:  __ orrw (Rdst, Rleft, right->as_jint()); break;
1771          case lir_logic_xor: __ eorw (Rdst, Rleft, right->as_jint()); break;
1772          default: ShouldNotReachHere(); break;
1773        }
1774      } else {
1775        Register Rright = right->is_single_cpu() ? right->as_register() :
1776                                                   right->as_register_lo();
1777        switch (code) {
1778          case lir_logic_and: __ andw (Rdst, Rleft, Rright); break;
1779          case lir_logic_or:  __ orrw (Rdst, Rleft, Rright); break;
1780          case lir_logic_xor: __ eorw (Rdst, Rleft, Rright); break;
1781          default: ShouldNotReachHere(); break;
1782        }
1783      }
1784    } else {
1785      Register Rdst = dst->as_register_lo();
1786      if (right->is_constant()) {
1787        switch (code) {
1788          case lir_logic_and: __ andr (Rdst, Rleft, right->as_jlong()); break;
1789          case lir_logic_or:  __ orr (Rdst, Rleft, right->as_jlong()); break;
1790          case lir_logic_xor: __ eor (Rdst, Rleft, right->as_jlong()); break;
1791          default: ShouldNotReachHere(); break;
1792        }
1793      } else {
1794        Register Rright = right->is_single_cpu() ? right->as_register() :
1795                                                   right->as_register_lo();
1796        switch (code) {
1797          case lir_logic_and: __ andr (Rdst, Rleft, Rright); break;
1798          case lir_logic_or:  __ orr (Rdst, Rleft, Rright); break;
1799          case lir_logic_xor: __ eor (Rdst, Rleft, Rright); break;
1800          default: ShouldNotReachHere(); break;
1801        }
1802      }
1803    }
1804 }
1805 
1806 
1807 
1808 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr illegal, LIR_Opr result, CodeEmitInfo* info) {
1809 
1810   // opcode check
1811   assert((code == lir_idiv) || (code == lir_irem), "opcode must be idiv or irem");
1812   bool is_irem = (code == lir_irem);
1813 
1814   // operand check
1815   assert(left->is_single_cpu(),   "left must be register");
1816   assert(right->is_single_cpu() || right->is_constant(),  "right must be register or constant");
1817   assert(result->is_single_cpu(), "result must be register");
1818   Register lreg = left->as_register();
1819   Register dreg = result->as_register();
1820 
1821   // power-of-2 constant check and codegen
1822   if (right->is_constant()) {
1823     int c = right->as_constant_ptr()->as_jint();
1824     assert(c > 0 && is_power_of_2(c), "divisor must be power-of-2 constant");
1825     if (is_irem) {
1826       if (c == 1) {
1827         // move 0 to dreg if divisor is 1
1828         __ movw(dreg, zr);
1829       } else {
1830         // use rscratch1 as intermediate result register
1831         __ negsw(rscratch1, lreg);
1832         __ andw(dreg, lreg, c - 1);
1833         __ andw(rscratch1, rscratch1, c - 1);
1834         __ csnegw(dreg, dreg, rscratch1, Assembler::MI);
1835       }
1836     } else {
1837       if (c == 1) {
1838         // move lreg to dreg if divisor is 1
1839         __ movw(dreg, lreg);
1840       } else {
1841         unsigned int shift = exact_log2(c);
1842         // use rscratch1 as intermediate result register
1843         __ asrw(rscratch1, lreg, 31);
1844         __ addw(rscratch1, lreg, rscratch1, Assembler::LSR, 32 - shift);
1845         __ asrw(dreg, rscratch1, shift);
1846       }
1847     }
1848   } else {
1849     Register rreg = right->as_register();
1850     __ corrected_idivl(dreg, lreg, rreg, is_irem, rscratch1);
1851   }
1852 }
1853 
1854 
1855 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1856   if (opr1->is_constant() && opr2->is_single_cpu()) {
1857     // tableswitch
1858     Register reg = as_reg(opr2);
1859     struct tableswitch &table = switches[opr1->as_constant_ptr()->as_jint()];
1860     __ tableswitch(reg, table._first_key, table._last_key, table._branches, table._after);
1861   } else if (opr1->is_single_cpu() || opr1->is_double_cpu()) {
1862     Register reg1 = as_reg(opr1);
1863     if (opr2->is_single_cpu()) {
1864       // cpu register - cpu register
1865       Register reg2 = opr2->as_register();
1866       if (is_reference_type(opr1->type())) {
1867         __ cmpoop(reg1, reg2);
1868       } else {
1869         assert(!is_reference_type(opr2->type()), "cmp int, oop?");
1870         __ cmpw(reg1, reg2);
1871       }
1872       return;
1873     }
1874     if (opr2->is_double_cpu()) {
1875       // cpu register - cpu register
1876       Register reg2 = opr2->as_register_lo();
1877       __ cmp(reg1, reg2);
1878       return;
1879     }
1880 
1881     if (opr2->is_constant()) {
1882       bool is_32bit = false; // width of register operand
1883       jlong imm;
1884 
1885       switch(opr2->type()) {
1886       case T_INT:
1887         imm = opr2->as_constant_ptr()->as_jint();
1888         is_32bit = true;
1889         break;
1890       case T_LONG:
1891         imm = opr2->as_constant_ptr()->as_jlong();
1892         break;
1893       case T_ADDRESS:
1894         imm = opr2->as_constant_ptr()->as_jint();
1895         break;
1896       case T_METADATA:
1897         imm = (intptr_t)(opr2->as_constant_ptr()->as_metadata());
1898         break;
1899       case T_OBJECT:
1900       case T_ARRAY:
1901         jobject2reg(opr2->as_constant_ptr()->as_jobject(), rscratch1);
1902         __ cmpoop(reg1, rscratch1);
1903         return;
1904       default:
1905         ShouldNotReachHere();
1906         imm = 0;  // unreachable
1907         break;
1908       }
1909 
1910       if (Assembler::operand_valid_for_add_sub_immediate(imm)) {
1911         if (is_32bit)
1912           __ cmpw(reg1, imm);
1913         else
1914           __ subs(zr, reg1, imm);
1915         return;
1916       } else {
1917         __ mov(rscratch1, imm);
1918         if (is_32bit)
1919           __ cmpw(reg1, rscratch1);
1920         else
1921           __ cmp(reg1, rscratch1);
1922         return;
1923       }
1924     } else
1925       ShouldNotReachHere();
1926   } else if (opr1->is_single_fpu()) {
1927     FloatRegister reg1 = opr1->as_float_reg();
1928     assert(opr2->is_single_fpu(), "expect single float register");
1929     FloatRegister reg2 = opr2->as_float_reg();
1930     __ fcmps(reg1, reg2);
1931   } else if (opr1->is_double_fpu()) {
1932     FloatRegister reg1 = opr1->as_double_reg();
1933     assert(opr2->is_double_fpu(), "expect double float register");
1934     FloatRegister reg2 = opr2->as_double_reg();
1935     __ fcmpd(reg1, reg2);
1936   } else {
1937     ShouldNotReachHere();
1938   }
1939 }
1940 
1941 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op){
1942   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1943     bool is_unordered_less = (code == lir_ucmp_fd2i);
1944     if (left->is_single_fpu()) {
1945       __ float_cmp(true, is_unordered_less ? -1 : 1, left->as_float_reg(), right->as_float_reg(), dst->as_register());
1946     } else if (left->is_double_fpu()) {
1947       __ float_cmp(false, is_unordered_less ? -1 : 1, left->as_double_reg(), right->as_double_reg(), dst->as_register());
1948     } else {
1949       ShouldNotReachHere();
1950     }
1951   } else if (code == lir_cmp_l2i) {
1952     Label done;
1953     __ cmp(left->as_register_lo(), right->as_register_lo());
1954     __ mov(dst->as_register(), (uint64_t)-1L);
1955     __ br(Assembler::LT, done);
1956     __ csinc(dst->as_register(), zr, zr, Assembler::EQ);
1957     __ bind(done);
1958   } else {
1959     ShouldNotReachHere();
1960   }
1961 }
1962 
1963 
1964 void LIR_Assembler::align_call(LIR_Code code) {  }
1965 
1966 
1967 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
1968   address call = __ trampoline_call(Address(op->addr(), rtype));
1969   if (call == nullptr) {
1970     bailout("trampoline stub overflow");
1971     return;
1972   }
1973   add_call_info(code_offset(), op->info());
1974   __ post_call_nop();
1975 }
1976 
1977 
1978 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
1979   address call = __ ic_call(op->addr());
1980   if (call == nullptr) {
1981     bailout("trampoline stub overflow");
1982     return;
1983   }
1984   add_call_info(code_offset(), op->info());
1985   __ post_call_nop();
1986 }
1987 
1988 void LIR_Assembler::emit_static_call_stub() {
1989   address call_pc = __ pc();
1990   address stub = __ start_a_stub(call_stub_size());
1991   if (stub == nullptr) {
1992     bailout("static call stub overflow");
1993     return;
1994   }
1995 
1996   int start = __ offset();
1997 
1998   __ relocate(static_stub_Relocation::spec(call_pc));
1999   __ emit_static_call_stub();
2000 
2001   assert(__ offset() - start + CompiledDirectCall::to_trampoline_stub_size()
2002         <= call_stub_size(), "stub too big");
2003   __ end_a_stub();
2004 }
2005 
2006 
2007 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2008   assert(exceptionOop->as_register() == r0, "must match");
2009   assert(exceptionPC->as_register() == r3, "must match");
2010 
2011   // exception object is not added to oop map by LinearScan
2012   // (LinearScan assumes that no oops are in fixed registers)
2013   info->add_register_oop(exceptionOop);
2014   StubId unwind_id;
2015 
2016   // get current pc information
2017   // pc is only needed if the method has an exception handler, the unwind code does not need it.
2018   if (compilation()->debug_info_recorder()->last_pc_offset() == __ offset()) {
2019     // As no instructions have been generated yet for this LIR node it's
2020     // possible that an oop map already exists for the current offset.
2021     // In that case insert an dummy NOP here to ensure all oop map PCs
2022     // are unique. See JDK-8237483.
2023     __ nop();
2024   }
2025   int pc_for_athrow_offset = __ offset();
2026   InternalAddress pc_for_athrow(__ pc());
2027   __ adr(exceptionPC->as_register(), pc_for_athrow);
2028   add_call_info(pc_for_athrow_offset, info); // for exception handler
2029 
2030   __ verify_not_null_oop(r0);
2031   // search an exception handler (r0: exception oop, r3: throwing pc)
2032   if (compilation()->has_fpu_code()) {
2033     unwind_id = StubId::c1_handle_exception_id;
2034   } else {
2035     unwind_id = StubId::c1_handle_exception_nofpu_id;
2036   }
2037   __ far_call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2038 
2039   // FIXME: enough room for two byte trap   ????
2040   __ nop();
2041 }
2042 
2043 
2044 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2045   assert(exceptionOop->as_register() == r0, "must match");
2046 
2047   __ b(_unwind_handler_entry);
2048 }
2049 
2050 
2051 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2052   Register lreg = left->is_single_cpu() ? left->as_register() : left->as_register_lo();
2053   Register dreg = dest->is_single_cpu() ? dest->as_register() : dest->as_register_lo();
2054 
2055   switch (left->type()) {
2056     case T_INT: {
2057       switch (code) {
2058       case lir_shl:  __ lslvw (dreg, lreg, count->as_register()); break;
2059       case lir_shr:  __ asrvw (dreg, lreg, count->as_register()); break;
2060       case lir_ushr: __ lsrvw (dreg, lreg, count->as_register()); break;
2061       default:
2062         ShouldNotReachHere();
2063         break;
2064       }
2065       break;
2066     case T_LONG:
2067     case T_ADDRESS:
2068     case T_OBJECT:
2069       switch (code) {
2070       case lir_shl:  __ lslv (dreg, lreg, count->as_register()); break;
2071       case lir_shr:  __ asrv (dreg, lreg, count->as_register()); break;
2072       case lir_ushr: __ lsrv (dreg, lreg, count->as_register()); break;
2073       default:
2074         ShouldNotReachHere();
2075         break;
2076       }
2077       break;
2078     default:
2079       ShouldNotReachHere();
2080       break;
2081     }
2082   }
2083 }
2084 
2085 
2086 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2087   Register dreg = dest->is_single_cpu() ? dest->as_register() : dest->as_register_lo();
2088   Register lreg = left->is_single_cpu() ? left->as_register() : left->as_register_lo();
2089 
2090   switch (left->type()) {
2091     case T_INT: {
2092       switch (code) {
2093       case lir_shl:  __ lslw (dreg, lreg, count); break;
2094       case lir_shr:  __ asrw (dreg, lreg, count); break;
2095       case lir_ushr: __ lsrw (dreg, lreg, count); break;
2096       default:
2097         ShouldNotReachHere();
2098         break;
2099       }
2100       break;
2101     case T_LONG:
2102     case T_ADDRESS:
2103     case T_OBJECT:
2104       switch (code) {
2105       case lir_shl:  __ lsl (dreg, lreg, count); break;
2106       case lir_shr:  __ asr (dreg, lreg, count); break;
2107       case lir_ushr: __ lsr (dreg, lreg, count); break;
2108       default:
2109         ShouldNotReachHere();
2110         break;
2111       }
2112       break;
2113     default:
2114       ShouldNotReachHere();
2115       break;
2116     }
2117   }
2118 }
2119 
2120 
2121 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
2122   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2123   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2124   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2125   __ str (r, Address(sp, offset_from_rsp_in_bytes));
2126 }
2127 
2128 
2129 void LIR_Assembler::store_parameter(jint c,     int offset_from_rsp_in_words) {
2130   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2131   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2132   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2133   __ mov (rscratch1, c);
2134   __ str (rscratch1, Address(sp, offset_from_rsp_in_bytes));
2135 }
2136 
2137 
2138 void LIR_Assembler::store_parameter(jobject o,  int offset_from_rsp_in_words) {
2139   ShouldNotReachHere();
2140   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2141   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2142   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2143   __ lea(rscratch1, __ constant_oop_address(o));
2144   __ str(rscratch1, Address(sp, offset_from_rsp_in_bytes));
2145 }
2146 
2147 
2148 // This code replaces a call to arraycopy; no exception may
2149 // be thrown in this code, they must be thrown in the System.arraycopy
2150 // activation frame; we could save some checks if this would not be the case
2151 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2152   ciArrayKlass* default_type = op->expected_type();
2153   Register src = op->src()->as_register();
2154   Register dst = op->dst()->as_register();
2155   Register src_pos = op->src_pos()->as_register();
2156   Register dst_pos = op->dst_pos()->as_register();
2157   Register length  = op->length()->as_register();
2158   Register tmp = op->tmp()->as_register();
2159 
2160   CodeStub* stub = op->stub();
2161   int flags = op->flags();
2162   BasicType basic_type = default_type != nullptr ? default_type->element_type()->basic_type() : T_ILLEGAL;
2163   if (is_reference_type(basic_type)) basic_type = T_OBJECT;
2164 
2165   // if we don't know anything, just go through the generic arraycopy
2166   if (default_type == nullptr // || basic_type == T_OBJECT
2167       ) {
2168     Label done;
2169     assert(src == r1 && src_pos == r2, "mismatch in calling convention");
2170 
2171     // Save the arguments in case the generic arraycopy fails and we
2172     // have to fall back to the JNI stub
2173     __ stp(dst,     dst_pos, Address(sp, 0*BytesPerWord));
2174     __ stp(length,  src_pos, Address(sp, 2*BytesPerWord));
2175     __ str(src,              Address(sp, 4*BytesPerWord));
2176 
2177     address copyfunc_addr = StubRoutines::generic_arraycopy();
2178     assert(copyfunc_addr != nullptr, "generic arraycopy stub required");
2179 
2180     // The arguments are in java calling convention so we shift them
2181     // to C convention
2182     assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
2183     __ mov(c_rarg0, j_rarg0);
2184     assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
2185     __ mov(c_rarg1, j_rarg1);
2186     assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
2187     __ mov(c_rarg2, j_rarg2);
2188     assert_different_registers(c_rarg3, j_rarg4);
2189     __ mov(c_rarg3, j_rarg3);
2190     __ mov(c_rarg4, j_rarg4);
2191 #ifndef PRODUCT
2192     if (PrintC1Statistics) {
2193       __ incrementw(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
2194     }
2195 #endif
2196     __ far_call(RuntimeAddress(copyfunc_addr));
2197 
2198     __ cbz(r0, *stub->continuation());
2199 
2200     // Reload values from the stack so they are where the stub
2201     // expects them.
2202     __ ldp(dst,     dst_pos, Address(sp, 0*BytesPerWord));
2203     __ ldp(length,  src_pos, Address(sp, 2*BytesPerWord));
2204     __ ldr(src,              Address(sp, 4*BytesPerWord));
2205 
2206     // r0 is -1^K where K == partial copied count
2207     __ eonw(rscratch1, r0, zr);
2208     // adjust length down and src/end pos up by partial copied count
2209     __ subw(length, length, rscratch1);
2210     __ addw(src_pos, src_pos, rscratch1);
2211     __ addw(dst_pos, dst_pos, rscratch1);
2212     __ b(*stub->entry());
2213 
2214     __ bind(*stub->continuation());
2215     return;
2216   }
2217 
2218   assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
2219 
2220   int elem_size = type2aelembytes(basic_type);
2221   int scale = exact_log2(elem_size);
2222 
2223   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
2224   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
2225 
2226   // test for null
2227   if (flags & LIR_OpArrayCopy::src_null_check) {
2228     __ cbz(src, *stub->entry());
2229   }
2230   if (flags & LIR_OpArrayCopy::dst_null_check) {
2231     __ cbz(dst, *stub->entry());
2232   }
2233 
2234   // If the compiler was not able to prove that exact type of the source or the destination
2235   // of the arraycopy is an array type, check at runtime if the source or the destination is
2236   // an instance type.
2237   if (flags & LIR_OpArrayCopy::type_check) {
2238     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::dst_objarray)) {
2239       __ load_klass(tmp, dst);
2240       __ ldrw(rscratch1, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2241       __ cmpw(rscratch1, Klass::_lh_neutral_value);
2242       __ br(Assembler::GE, *stub->entry());
2243     }
2244 
2245     if (!(flags & LIR_OpArrayCopy::LIR_OpArrayCopy::src_objarray)) {
2246       __ load_klass(tmp, src);
2247       __ ldrw(rscratch1, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2248       __ cmpw(rscratch1, Klass::_lh_neutral_value);
2249       __ br(Assembler::GE, *stub->entry());
2250     }
2251   }
2252 
2253   // check if negative
2254   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2255     __ cmpw(src_pos, 0);
2256     __ br(Assembler::LT, *stub->entry());
2257   }
2258   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2259     __ cmpw(dst_pos, 0);
2260     __ br(Assembler::LT, *stub->entry());
2261   }
2262 
2263   if (flags & LIR_OpArrayCopy::length_positive_check) {
2264     __ cmpw(length, 0);
2265     __ br(Assembler::LT, *stub->entry());
2266   }
2267 
2268   if (flags & LIR_OpArrayCopy::src_range_check) {
2269     __ addw(tmp, src_pos, length);
2270     __ ldrw(rscratch1, src_length_addr);
2271     __ cmpw(tmp, rscratch1);
2272     __ br(Assembler::HI, *stub->entry());
2273   }
2274   if (flags & LIR_OpArrayCopy::dst_range_check) {
2275     __ addw(tmp, dst_pos, length);
2276     __ ldrw(rscratch1, dst_length_addr);
2277     __ cmpw(tmp, rscratch1);
2278     __ br(Assembler::HI, *stub->entry());
2279   }
2280 
2281   if (flags & LIR_OpArrayCopy::type_check) {
2282     // We don't know the array types are compatible
2283     if (basic_type != T_OBJECT) {
2284       // Simple test for basic type arrays
2285       __ cmp_klasses_from_objects(src, dst, tmp, rscratch1);
2286       __ br(Assembler::NE, *stub->entry());
2287     } else {
2288       // For object arrays, if src is a sub class of dst then we can
2289       // safely do the copy.
2290       Label cont, slow;
2291 
2292 #define PUSH(r1, r2)                                    \
2293       stp(r1, r2, __ pre(sp, -2 * wordSize));
2294 
2295 #define POP(r1, r2)                                     \
2296       ldp(r1, r2, __ post(sp, 2 * wordSize));
2297 
2298       __ PUSH(src, dst);
2299 
2300       __ load_klass(src, src);
2301       __ load_klass(dst, dst);
2302 
2303       __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, nullptr);
2304 
2305       __ PUSH(src, dst);
2306       __ far_call(RuntimeAddress(Runtime1::entry_for(StubId::c1_slow_subtype_check_id)));
2307       __ POP(src, dst);
2308 
2309       __ cbnz(src, cont);
2310 
2311       __ bind(slow);
2312       __ POP(src, dst);
2313 
2314       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2315       if (copyfunc_addr != nullptr) { // use stub if available
2316         // src is not a sub class of dst so we have to do a
2317         // per-element check.
2318 
2319         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2320         if ((flags & mask) != mask) {
2321           // Check that at least both of them object arrays.
2322           assert(flags & mask, "one of the two should be known to be an object array");
2323 
2324           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2325             __ load_klass(tmp, src);
2326           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2327             __ load_klass(tmp, dst);
2328           }
2329           int lh_offset = in_bytes(Klass::layout_helper_offset());
2330           Address klass_lh_addr(tmp, lh_offset);
2331           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2332           __ ldrw(rscratch1, klass_lh_addr);
2333           __ mov(rscratch2, objArray_lh);
2334           __ eorw(rscratch1, rscratch1, rscratch2);
2335           __ cbnzw(rscratch1, *stub->entry());
2336         }
2337 
2338        // Spill because stubs can use any register they like and it's
2339        // easier to restore just those that we care about.
2340         __ stp(dst,     dst_pos, Address(sp, 0*BytesPerWord));
2341         __ stp(length,  src_pos, Address(sp, 2*BytesPerWord));
2342         __ str(src,              Address(sp, 4*BytesPerWord));
2343 
2344         __ lea(c_rarg0, Address(src, src_pos, Address::uxtw(scale)));
2345         __ add(c_rarg0, c_rarg0, arrayOopDesc::base_offset_in_bytes(basic_type));
2346         assert_different_registers(c_rarg0, dst, dst_pos, length);
2347         __ lea(c_rarg1, Address(dst, dst_pos, Address::uxtw(scale)));
2348         __ add(c_rarg1, c_rarg1, arrayOopDesc::base_offset_in_bytes(basic_type));
2349         assert_different_registers(c_rarg1, dst, length);
2350         __ uxtw(c_rarg2, length);
2351         assert_different_registers(c_rarg2, dst);
2352 
2353         __ load_klass(c_rarg4, dst);
2354         __ ldr(c_rarg4, Address(c_rarg4, ObjArrayKlass::element_klass_offset()));
2355         __ ldrw(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset()));
2356         __ far_call(RuntimeAddress(copyfunc_addr));
2357 
2358 #ifndef PRODUCT
2359         if (PrintC1Statistics) {
2360           Label failed;
2361           __ cbnz(r0, failed);
2362           __ incrementw(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt));
2363           __ bind(failed);
2364         }
2365 #endif
2366 
2367         __ cbz(r0, *stub->continuation());
2368 
2369 #ifndef PRODUCT
2370         if (PrintC1Statistics) {
2371           __ incrementw(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt));
2372         }
2373 #endif
2374         assert_different_registers(dst, dst_pos, length, src_pos, src, r0, rscratch1);
2375 
2376         // Restore previously spilled arguments
2377         __ ldp(dst,     dst_pos, Address(sp, 0*BytesPerWord));
2378         __ ldp(length,  src_pos, Address(sp, 2*BytesPerWord));
2379         __ ldr(src,              Address(sp, 4*BytesPerWord));
2380 
2381         // return value is -1^K where K is partial copied count
2382         __ eonw(rscratch1, r0, zr);
2383         // adjust length down and src/end pos up by partial copied count
2384         __ subw(length, length, rscratch1);
2385         __ addw(src_pos, src_pos, rscratch1);
2386         __ addw(dst_pos, dst_pos, rscratch1);
2387       }
2388 
2389       __ b(*stub->entry());
2390 
2391       __ bind(cont);
2392       __ POP(src, dst);
2393     }
2394   }
2395 
2396 #ifdef ASSERT
2397   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2398     // Sanity check the known type with the incoming class.  For the
2399     // primitive case the types must match exactly with src.klass and
2400     // dst.klass each exactly matching the default type.  For the
2401     // object array case, if no type check is needed then either the
2402     // dst type is exactly the expected type and the src type is a
2403     // subtype which we can't check or src is the same array as dst
2404     // but not necessarily exactly of type default_type.
2405     Label known_ok, halt;
2406     __ mov_metadata(tmp, default_type->constant_encoding());
2407 
2408     if (basic_type != T_OBJECT) {
2409       __ cmp_klass(dst, tmp, rscratch1);
2410       __ br(Assembler::NE, halt);
2411       __ cmp_klass(src, tmp, rscratch1);
2412       __ br(Assembler::EQ, known_ok);
2413     } else {
2414       __ cmp_klass(dst, tmp, rscratch1);
2415       __ br(Assembler::EQ, known_ok);
2416       __ cmp(src, dst);
2417       __ br(Assembler::EQ, known_ok);
2418     }
2419     __ bind(halt);
2420     __ stop("incorrect type information in arraycopy");
2421     __ bind(known_ok);
2422   }
2423 #endif
2424 
2425 #ifndef PRODUCT
2426   if (PrintC1Statistics) {
2427     __ incrementw(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)));
2428   }
2429 #endif
2430 
2431   __ lea(c_rarg0, Address(src, src_pos, Address::uxtw(scale)));
2432   __ add(c_rarg0, c_rarg0, arrayOopDesc::base_offset_in_bytes(basic_type));
2433   assert_different_registers(c_rarg0, dst, dst_pos, length);
2434   __ lea(c_rarg1, Address(dst, dst_pos, Address::uxtw(scale)));
2435   __ add(c_rarg1, c_rarg1, arrayOopDesc::base_offset_in_bytes(basic_type));
2436   assert_different_registers(c_rarg1, dst, length);
2437   __ uxtw(c_rarg2, length);
2438   assert_different_registers(c_rarg2, dst);
2439 
2440   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2441   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2442   const char *name;
2443   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2444 
2445  CodeBlob *cb = CodeCache::find_blob(entry);
2446  if (cb) {
2447    __ far_call(RuntimeAddress(entry));
2448  } else {
2449    __ call_VM_leaf(entry, 3);
2450  }
2451 
2452   if (stub != nullptr) {
2453     __ bind(*stub->continuation());
2454   }
2455 }
2456 
2457 
2458 
2459 
2460 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2461   Register obj = op->obj_opr()->as_register();  // may not be an oop
2462   Register hdr = op->hdr_opr()->as_register();
2463   Register lock = op->lock_opr()->as_register();
2464   Register temp = op->scratch_opr()->as_register();
2465   if (op->code() == lir_lock) {
2466     // add debug info for NullPointerException only if one is possible
2467     int null_check_offset = __ lock_object(hdr, obj, lock, temp, *op->stub()->entry());
2468     if (op->info() != nullptr) {
2469       add_debug_info_for_null_check(null_check_offset, op->info());
2470     }
2471     // done
2472   } else if (op->code() == lir_unlock) {
2473     __ unlock_object(hdr, obj, lock, temp, *op->stub()->entry());
2474   } else {
2475     Unimplemented();
2476   }
2477   __ bind(*op->stub()->continuation());
2478 }
2479 
2480 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2481   Register obj = op->obj()->as_pointer_register();
2482   Register result = op->result_opr()->as_pointer_register();
2483 
2484   CodeEmitInfo* info = op->info();
2485   if (info != nullptr) {
2486     add_debug_info_for_null_check_here(info);
2487   }
2488 
2489   __ load_klass(result, obj);
2490 }
2491 
2492 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2493   ciMethod* method = op->profiled_method();
2494   int bci          = op->profiled_bci();
2495   ciMethod* callee = op->profiled_callee();
2496 
2497   // Update counter for all call types
2498   ciMethodData* md = method->method_data_or_null();
2499   assert(md != nullptr, "Sanity");
2500   ciProfileData* data = md->bci_to_data(bci);
2501   assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
2502   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2503   Register mdo  = op->mdo()->as_register();
2504   __ mov_metadata(mdo, md->constant_encoding());
2505   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2506   // Perform additional virtual call profiling for invokevirtual and
2507   // invokeinterface bytecodes
2508   if (op->should_profile_receiver_type()) {
2509     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2510     Register recv = op->recv()->as_register();
2511     assert_different_registers(mdo, recv);
2512     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2513     ciKlass* known_klass = op->known_holder();
2514     if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
2515       // We know the type that will be seen at this call site; we can
2516       // statically update the MethodData* rather than needing to do
2517       // dynamic tests on the receiver type.
2518       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2519       for (uint i = 0; i < VirtualCallData::row_limit(); i++) {
2520         ciKlass* receiver = vc_data->receiver(i);
2521         if (known_klass->equals(receiver)) {
2522           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2523           __ addptr(data_addr, DataLayout::counter_increment);
2524           return;
2525         }
2526       }
2527       // Receiver type is not found in profile data.
2528       // Fall back to runtime helper to handle the rest at runtime.
2529       __ mov_metadata(recv, known_klass->constant_encoding());
2530     } else {
2531       __ load_klass(recv, recv);
2532     }
2533     type_profile_helper(mdo, md, data, recv);
2534   } else {
2535     // Static call
2536     __ addptr(counter_addr, DataLayout::counter_increment);
2537   }
2538 }
2539 
2540 
2541 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
2542   __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
2543 }
2544 
2545 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
2546   assert(op->crc()->is_single_cpu(),  "crc must be register");
2547   assert(op->val()->is_single_cpu(),  "byte value must be register");
2548   assert(op->result_opr()->is_single_cpu(), "result must be register");
2549   Register crc = op->crc()->as_register();
2550   Register val = op->val()->as_register();
2551   Register res = op->result_opr()->as_register();
2552 
2553   assert_different_registers(val, crc, res);
2554   uint64_t offset;
2555   __ adrp(res, ExternalAddress(StubRoutines::crc_table_addr()), offset);
2556   __ add(res, res, offset);
2557 
2558   __ mvnw(crc, crc); // ~crc
2559   __ update_byte_crc32(crc, val, res);
2560   __ mvnw(res, crc); // ~crc
2561 }
2562 
2563 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2564   COMMENT("emit_profile_type {");
2565   Register obj = op->obj()->as_register();
2566   Register tmp = op->tmp()->as_pointer_register();
2567   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2568   ciKlass* exact_klass = op->exact_klass();
2569   intptr_t current_klass = op->current_klass();
2570   bool not_null = op->not_null();
2571   bool no_conflict = op->no_conflict();
2572 
2573   Label update, next, none;
2574 
2575   bool do_null = !not_null;
2576   bool exact_klass_set = exact_klass != nullptr && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2577   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2578 
2579   assert(do_null || do_update, "why are we here?");
2580   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2581   assert(mdo_addr.base() != rscratch1, "wrong register");
2582 
2583   __ verify_oop(obj);
2584 
2585   if (tmp != obj) {
2586     assert_different_registers(obj, tmp, rscratch1, rscratch2, mdo_addr.base(), mdo_addr.index());
2587     __ mov(tmp, obj);
2588   } else {
2589     assert_different_registers(obj, rscratch1, rscratch2, mdo_addr.base(), mdo_addr.index());
2590   }
2591   if (do_null) {
2592     __ cbnz(tmp, update);
2593     if (!TypeEntries::was_null_seen(current_klass)) {
2594       __ ldr(rscratch2, mdo_addr);
2595       __ orr(rscratch2, rscratch2, TypeEntries::null_seen);
2596       __ str(rscratch2, mdo_addr);
2597     }
2598     if (do_update) {
2599 #ifndef ASSERT
2600       __ b(next);
2601     }
2602 #else
2603       __ b(next);
2604     }
2605   } else {
2606     __ cbnz(tmp, update);
2607     __ stop("unexpected null obj");
2608 #endif
2609   }
2610 
2611   __ bind(update);
2612 
2613   if (do_update) {
2614 #ifdef ASSERT
2615     if (exact_klass != nullptr) {
2616       Label ok;
2617       __ load_klass(tmp, tmp);
2618       __ mov_metadata(rscratch1, exact_klass->constant_encoding());
2619       __ eor(rscratch1, tmp, rscratch1);
2620       __ cbz(rscratch1, ok);
2621       __ stop("exact klass and actual klass differ");
2622       __ bind(ok);
2623     }
2624 #endif
2625     if (!no_conflict) {
2626       if (exact_klass == nullptr || TypeEntries::is_type_none(current_klass)) {
2627         if (exact_klass != nullptr) {
2628           __ mov_metadata(tmp, exact_klass->constant_encoding());
2629         } else {
2630           __ load_klass(tmp, tmp);
2631         }
2632 
2633         __ ldr(rscratch2, mdo_addr);
2634         __ eor(tmp, tmp, rscratch2);
2635         __ andr(rscratch1, tmp, TypeEntries::type_klass_mask);
2636         // klass seen before, nothing to do. The unknown bit may have been
2637         // set already but no need to check.
2638         __ cbz(rscratch1, next);
2639 
2640         __ tbnz(tmp, exact_log2(TypeEntries::type_unknown), next); // already unknown. Nothing to do anymore.
2641 
2642         if (TypeEntries::is_type_none(current_klass)) {
2643           __ cbz(rscratch2, none);
2644           __ cmp(rscratch2, (u1)TypeEntries::null_seen);
2645           __ br(Assembler::EQ, none);
2646           // There is a chance that the checks above
2647           // fail if another thread has just set the
2648           // profiling to this obj's klass
2649           __ dmb(Assembler::ISHLD);
2650           __ eor(tmp, tmp, rscratch2); // get back original value before XOR
2651           __ ldr(rscratch2, mdo_addr);
2652           __ eor(tmp, tmp, rscratch2);
2653           __ andr(rscratch1, tmp, TypeEntries::type_klass_mask);
2654           __ cbz(rscratch1, next);
2655         }
2656       } else {
2657         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
2658                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
2659 
2660         __ ldr(tmp, mdo_addr);
2661         __ tbnz(tmp, exact_log2(TypeEntries::type_unknown), next); // already unknown. Nothing to do anymore.
2662       }
2663 
2664       // different than before. Cannot keep accurate profile.
2665       __ ldr(rscratch2, mdo_addr);
2666       __ orr(rscratch2, rscratch2, TypeEntries::type_unknown);
2667       __ str(rscratch2, mdo_addr);
2668 
2669       if (TypeEntries::is_type_none(current_klass)) {
2670         __ b(next);
2671 
2672         __ bind(none);
2673         // first time here. Set profile type.
2674         __ str(tmp, mdo_addr);
2675 #ifdef ASSERT
2676         __ andr(tmp, tmp, TypeEntries::type_mask);
2677         __ verify_klass_ptr(tmp);
2678 #endif
2679       }
2680     } else {
2681       // There's a single possible klass at this profile point
2682       assert(exact_klass != nullptr, "should be");
2683       if (TypeEntries::is_type_none(current_klass)) {
2684         __ mov_metadata(tmp, exact_klass->constant_encoding());
2685         __ ldr(rscratch2, mdo_addr);
2686         __ eor(tmp, tmp, rscratch2);
2687         __ andr(rscratch1, tmp, TypeEntries::type_klass_mask);
2688         __ cbz(rscratch1, next);
2689 #ifdef ASSERT
2690         {
2691           Label ok;
2692           __ ldr(rscratch1, mdo_addr);
2693           __ cbz(rscratch1, ok);
2694           __ cmp(rscratch1, (u1)TypeEntries::null_seen);
2695           __ br(Assembler::EQ, ok);
2696           // may have been set by another thread
2697           __ dmb(Assembler::ISHLD);
2698           __ mov_metadata(rscratch1, exact_klass->constant_encoding());
2699           __ ldr(rscratch2, mdo_addr);
2700           __ eor(rscratch2, rscratch1, rscratch2);
2701           __ andr(rscratch2, rscratch2, TypeEntries::type_mask);
2702           __ cbz(rscratch2, ok);
2703 
2704           __ stop("unexpected profiling mismatch");
2705           __ bind(ok);
2706         }
2707 #endif
2708         // first time here. Set profile type.
2709         __ str(tmp, mdo_addr);
2710 #ifdef ASSERT
2711         __ andr(tmp, tmp, TypeEntries::type_mask);
2712         __ verify_klass_ptr(tmp);
2713 #endif
2714       } else {
2715         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
2716                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
2717 
2718         __ ldr(tmp, mdo_addr);
2719         __ tbnz(tmp, exact_log2(TypeEntries::type_unknown), next); // already unknown. Nothing to do anymore.
2720 
2721         __ orr(tmp, tmp, TypeEntries::type_unknown);
2722         __ str(tmp, mdo_addr);
2723         // FIXME: Write barrier needed here?
2724       }
2725     }
2726 
2727     __ bind(next);
2728   }
2729   COMMENT("} emit_profile_type");
2730 }
2731 
2732 
2733 void LIR_Assembler::align_backward_branch_target() {
2734 }
2735 
2736 
2737 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2738   // tmp must be unused
2739   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2740 
2741   if (left->is_single_cpu()) {
2742     assert(dest->is_single_cpu(), "expect single result reg");
2743     __ negw(dest->as_register(), left->as_register());
2744   } else if (left->is_double_cpu()) {
2745     assert(dest->is_double_cpu(), "expect double result reg");
2746     __ neg(dest->as_register_lo(), left->as_register_lo());
2747   } else if (left->is_single_fpu()) {
2748     assert(dest->is_single_fpu(), "expect single float result reg");
2749     __ fnegs(dest->as_float_reg(), left->as_float_reg());
2750   } else {
2751     assert(left->is_double_fpu(), "expect double float operand reg");
2752     assert(dest->is_double_fpu(), "expect double float result reg");
2753     __ fnegd(dest->as_double_reg(), left->as_double_reg());
2754   }
2755 }
2756 
2757 
2758 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2759   if (patch_code != lir_patch_none) {
2760     deoptimize_trap(info);
2761     return;
2762   }
2763 
2764   __ lea(dest->as_pointer_register(), as_Address(addr->as_address_ptr()));
2765 }
2766 
2767 
2768 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2769   assert(!tmp->is_valid(), "don't need temporary");
2770 
2771   CodeBlob *cb = CodeCache::find_blob(dest);
2772   if (cb) {
2773     __ far_call(RuntimeAddress(dest));
2774   } else {
2775     __ mov(rscratch1, RuntimeAddress(dest));
2776     __ blr(rscratch1);
2777   }
2778 
2779   if (info != nullptr) {
2780     add_call_info_here(info);
2781   }
2782   __ post_call_nop();
2783 }
2784 
2785 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2786   if (dest->is_address() || src->is_address()) {
2787     move_op(src, dest, type, lir_patch_none, info, /*wide*/false);
2788   } else {
2789     ShouldNotReachHere();
2790   }
2791 }
2792 
2793 #ifdef ASSERT
2794 // emit run-time assertion
2795 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2796   assert(op->code() == lir_assert, "must be");
2797 
2798   if (op->in_opr1()->is_valid()) {
2799     assert(op->in_opr2()->is_valid(), "both operands must be valid");
2800     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
2801   } else {
2802     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
2803     assert(op->condition() == lir_cond_always, "no other conditions allowed");
2804   }
2805 
2806   Label ok;
2807   if (op->condition() != lir_cond_always) {
2808     Assembler::Condition acond = Assembler::AL;
2809     switch (op->condition()) {
2810       case lir_cond_equal:        acond = Assembler::EQ;  break;
2811       case lir_cond_notEqual:     acond = Assembler::NE;  break;
2812       case lir_cond_less:         acond = Assembler::LT;  break;
2813       case lir_cond_lessEqual:    acond = Assembler::LE;  break;
2814       case lir_cond_greaterEqual: acond = Assembler::GE;  break;
2815       case lir_cond_greater:      acond = Assembler::GT;  break;
2816       case lir_cond_belowEqual:   acond = Assembler::LS;  break;
2817       case lir_cond_aboveEqual:   acond = Assembler::HS;  break;
2818       default:                    ShouldNotReachHere();
2819     }
2820     __ br(acond, ok);
2821   }
2822   if (op->halt()) {
2823     const char* str = __ code_string(op->msg());
2824     __ stop(str);
2825   } else {
2826     breakpoint();
2827   }
2828   __ bind(ok);
2829 }
2830 #endif
2831 
2832 #ifndef PRODUCT
2833 #define COMMENT(x)   do { __ block_comment(x); } while (0)
2834 #else
2835 #define COMMENT(x)
2836 #endif
2837 
2838 void LIR_Assembler::membar() {
2839   COMMENT("membar");
2840   __ membar(MacroAssembler::AnyAny);
2841 }
2842 
2843 void LIR_Assembler::membar_acquire() {
2844   __ membar(Assembler::LoadLoad|Assembler::LoadStore);
2845 }
2846 
2847 void LIR_Assembler::membar_release() {
2848   __ membar(Assembler::LoadStore|Assembler::StoreStore);
2849 }
2850 
2851 void LIR_Assembler::membar_loadload() {
2852   __ membar(Assembler::LoadLoad);
2853 }
2854 
2855 void LIR_Assembler::membar_storestore() {
2856   __ membar(MacroAssembler::StoreStore);
2857 }
2858 
2859 void LIR_Assembler::membar_loadstore() { __ membar(MacroAssembler::LoadStore); }
2860 
2861 void LIR_Assembler::membar_storeload() { __ membar(MacroAssembler::StoreLoad); }
2862 
2863 void LIR_Assembler::on_spin_wait() {
2864   __ spin_wait();
2865 }
2866 
2867 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2868   __ mov(result_reg->as_register(), rthread);
2869 }
2870 
2871 
2872 void LIR_Assembler::peephole(LIR_List *lir) {
2873 #if 0
2874   if (tableswitch_count >= max_tableswitches)
2875     return;
2876 
2877   /*
2878     This finite-state automaton recognizes sequences of compare-and-
2879     branch instructions.  We will turn them into a tableswitch.  You
2880     could argue that C1 really shouldn't be doing this sort of
2881     optimization, but without it the code is really horrible.
2882   */
2883 
2884   enum { start_s, cmp1_s, beq_s, cmp_s } state;
2885   int first_key, last_key = -2147483648;
2886   int next_key = 0;
2887   int start_insn = -1;
2888   int last_insn = -1;
2889   Register reg = noreg;
2890   LIR_Opr reg_opr;
2891   state = start_s;
2892 
2893   LIR_OpList* inst = lir->instructions_list();
2894   for (int i = 0; i < inst->length(); i++) {
2895     LIR_Op* op = inst->at(i);
2896     switch (state) {
2897     case start_s:
2898       first_key = -1;
2899       start_insn = i;
2900       switch (op->code()) {
2901       case lir_cmp:
2902         LIR_Opr opr1 = op->as_Op2()->in_opr1();
2903         LIR_Opr opr2 = op->as_Op2()->in_opr2();
2904         if (opr1->is_cpu_register() && opr1->is_single_cpu()
2905             && opr2->is_constant()
2906             && opr2->type() == T_INT) {
2907           reg_opr = opr1;
2908           reg = opr1->as_register();
2909           first_key = opr2->as_constant_ptr()->as_jint();
2910           next_key = first_key + 1;
2911           state = cmp_s;
2912           goto next_state;
2913         }
2914         break;
2915       }
2916       break;
2917     case cmp_s:
2918       switch (op->code()) {
2919       case lir_branch:
2920         if (op->as_OpBranch()->cond() == lir_cond_equal) {
2921           state = beq_s;
2922           last_insn = i;
2923           goto next_state;
2924         }
2925       }
2926       state = start_s;
2927       break;
2928     case beq_s:
2929       switch (op->code()) {
2930       case lir_cmp: {
2931         LIR_Opr opr1 = op->as_Op2()->in_opr1();
2932         LIR_Opr opr2 = op->as_Op2()->in_opr2();
2933         if (opr1->is_cpu_register() && opr1->is_single_cpu()
2934             && opr1->as_register() == reg
2935             && opr2->is_constant()
2936             && opr2->type() == T_INT
2937             && opr2->as_constant_ptr()->as_jint() == next_key) {
2938           last_key = next_key;
2939           next_key++;
2940           state = cmp_s;
2941           goto next_state;
2942         }
2943       }
2944       }
2945       last_key = next_key;
2946       state = start_s;
2947       break;
2948     default:
2949       assert(false, "impossible state");
2950     }
2951     if (state == start_s) {
2952       if (first_key < last_key - 5L && reg != noreg) {
2953         {
2954           // printf("found run register %d starting at insn %d low value %d high value %d\n",
2955           //        reg->encoding(),
2956           //        start_insn, first_key, last_key);
2957           //   for (int i = 0; i < inst->length(); i++) {
2958           //     inst->at(i)->print();
2959           //     tty->print("\n");
2960           //   }
2961           //   tty->print("\n");
2962         }
2963 
2964         struct tableswitch *sw = &switches[tableswitch_count];
2965         sw->_insn_index = start_insn, sw->_first_key = first_key,
2966           sw->_last_key = last_key, sw->_reg = reg;
2967         inst->insert_before(last_insn + 1, new LIR_OpLabel(&sw->_after));
2968         {
2969           // Insert the new table of branches
2970           int offset = last_insn;
2971           for (int n = first_key; n < last_key; n++) {
2972             inst->insert_before
2973               (last_insn + 1,
2974                new LIR_OpBranch(lir_cond_always, T_ILLEGAL,
2975                                 inst->at(offset)->as_OpBranch()->label()));
2976             offset -= 2, i++;
2977           }
2978         }
2979         // Delete all the old compare-and-branch instructions
2980         for (int n = first_key; n < last_key; n++) {
2981           inst->remove_at(start_insn);
2982           inst->remove_at(start_insn);
2983         }
2984         // Insert the tableswitch instruction
2985         inst->insert_before(start_insn,
2986                             new LIR_Op2(lir_cmp, lir_cond_always,
2987                                         LIR_OprFact::intConst(tableswitch_count),
2988                                         reg_opr));
2989         inst->insert_before(start_insn + 1, new LIR_OpLabel(&sw->_branches));
2990         tableswitch_count++;
2991       }
2992       reg = noreg;
2993       last_key = -2147483648;
2994     }
2995   next_state:
2996     ;
2997   }
2998 #endif
2999 }
3000 
3001 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp_op) {
3002   Address addr = as_Address(src->as_address_ptr());
3003   BasicType type = src->type();
3004   bool is_oop = is_reference_type(type);
3005 
3006   void (MacroAssembler::* add)(Register prev, RegisterOrConstant incr, Register addr);
3007   void (MacroAssembler::* xchg)(Register prev, Register newv, Register addr);
3008 
3009   switch(type) {
3010   case T_INT:
3011     xchg = &MacroAssembler::atomic_xchgalw;
3012     add = &MacroAssembler::atomic_addalw;
3013     break;
3014   case T_LONG:
3015     xchg = &MacroAssembler::atomic_xchgal;
3016     add = &MacroAssembler::atomic_addal;
3017     break;
3018   case T_OBJECT:
3019   case T_ARRAY:
3020     if (UseCompressedOops) {
3021       xchg = &MacroAssembler::atomic_xchgalw;
3022       add = &MacroAssembler::atomic_addalw;
3023     } else {
3024       xchg = &MacroAssembler::atomic_xchgal;
3025       add = &MacroAssembler::atomic_addal;
3026     }
3027     break;
3028   default:
3029     ShouldNotReachHere();
3030     xchg = &MacroAssembler::atomic_xchgal;
3031     add = &MacroAssembler::atomic_addal; // unreachable
3032   }
3033 
3034   switch (code) {
3035   case lir_xadd:
3036     {
3037       RegisterOrConstant inc;
3038       Register tmp = as_reg(tmp_op);
3039       Register dst = as_reg(dest);
3040       if (data->is_constant()) {
3041         inc = RegisterOrConstant(as_long(data));
3042         assert_different_registers(dst, addr.base(), tmp,
3043                                    rscratch1, rscratch2);
3044       } else {
3045         inc = RegisterOrConstant(as_reg(data));
3046         assert_different_registers(inc.as_register(), dst, addr.base(), tmp,
3047                                    rscratch1, rscratch2);
3048       }
3049       __ lea(tmp, addr);
3050       (_masm->*add)(dst, inc, tmp);
3051       break;
3052     }
3053   case lir_xchg:
3054     {
3055       Register tmp = tmp_op->as_register();
3056       Register obj = as_reg(data);
3057       Register dst = as_reg(dest);
3058       if (is_oop && UseCompressedOops) {
3059         __ encode_heap_oop(rscratch2, obj);
3060         obj = rscratch2;
3061       }
3062       assert_different_registers(obj, addr.base(), tmp, rscratch1);
3063       assert_different_registers(dst, addr.base(), tmp, rscratch1);
3064       __ lea(tmp, addr);
3065       (_masm->*xchg)(dst, obj, tmp);
3066       if (is_oop && UseCompressedOops) {
3067         __ decode_heap_oop(dst);
3068       }
3069     }
3070     break;
3071   default:
3072     ShouldNotReachHere();
3073   }
3074   if(!UseLSE) {
3075     __ membar(__ AnyAny);
3076   }
3077 }
3078 
3079 #undef __