1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2024, 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 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  28 
  29 #include "asm/assembler.inline.hpp"
  30 #include "code/aotCodeCache.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "oops/compressedKlass.hpp"
  35 #include "runtime/vm_version.hpp"
  36 #include "utilities/macros.hpp"
  37 #include "utilities/powerOfTwo.hpp"
  38 #include "runtime/signature.hpp"
  39 
  40 
  41 class ciInlineKlass;
  42 
  43 class OopMap;
  44 
  45 // MacroAssembler extends Assembler by frequently used macros.
  46 //
  47 // Instructions for which a 'better' code sequence exists depending
  48 // on arguments should also go in here.
  49 
  50 class MacroAssembler: public Assembler {
  51   friend class LIR_Assembler;
  52 
  53  public:
  54   using Assembler::mov;
  55   using Assembler::movi;
  56 
  57  protected:
  58 
  59   // Support for VM calls
  60   //
  61   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  62   // may customize this version by overriding it for its purposes (e.g., to save/restore
  63   // additional registers when doing a VM call).
  64   virtual void call_VM_leaf_base(
  65     address entry_point,               // the entry point
  66     int     number_of_arguments,        // the number of arguments to pop after the call
  67     Label *retaddr = nullptr
  68   );
  69 
  70   virtual void call_VM_leaf_base(
  71     address entry_point,               // the entry point
  72     int     number_of_arguments,        // the number of arguments to pop after the call
  73     Label &retaddr) {
  74     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  75   }
  76 
  77   // This is the base routine called by the different versions of call_VM. The interpreter
  78   // may customize this version by overriding it for its purposes (e.g., to save/restore
  79   // additional registers when doing a VM call).
  80   //
  81   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  82   // returns the register which contains the thread upon return. If a thread register has been
  83   // specified, the return value will correspond to that register. If no last_java_sp is specified
  84   // (noreg) than rsp will be used instead.
  85   virtual void call_VM_base(           // returns the register containing the thread upon return
  86     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  87     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  88     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  89     Label*   return_pc,                // to set up last_Java_frame; use nullptr otherwise
  90     address  entry_point,              // the entry point
  91     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  92     bool     check_exceptions          // whether to check for pending exceptions after return
  93   );
  94 
  95   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  96 
  97   enum KlassDecodeMode {
  98     KlassDecodeNone,
  99     KlassDecodeZero,
 100     KlassDecodeXor,
 101     KlassDecodeMovk
 102   };
 103 
 104   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 105   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 106 
 107  private:
 108   static KlassDecodeMode _klass_decode_mode;
 109 
 110   // Returns above setting with asserts
 111   static KlassDecodeMode klass_decode_mode();
 112 
 113  public:
 114   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 115   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 116 
 117   // Sets the decode mode and returns false if cannot be set.
 118   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 119 
 120  public:
 121   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 122 
 123  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 124  // The implementation is only non-empty for the InterpreterMacroAssembler,
 125  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 126  virtual void check_and_handle_popframe(Register java_thread);
 127  virtual void check_and_handle_earlyret(Register java_thread);
 128 
 129   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1);
 130   void rt_call(address dest, Register tmp = rscratch1);
 131 
 132   // Load Effective Address
 133   void lea(Register r, const Address &a) {
 134     InstructionMark im(this);
 135     a.lea(this, r);
 136   }
 137 
 138   // Whether materializing the given address for a LDR/STR requires an
 139   // additional lea instruction.
 140   static bool legitimize_address_requires_lea(const Address &a, int size) {
 141     return a.getMode() == Address::base_plus_offset &&
 142            !Address::offset_ok_for_immed(a.offset(), exact_log2(size));
 143   }
 144 
 145   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 146      accesses, and these can exceed the offset range. */
 147   Address legitimize_address(const Address &a, int size, Register scratch) {
 148     if (legitimize_address_requires_lea(a, size)) {
 149       block_comment("legitimize_address {");
 150       lea(scratch, a);
 151       block_comment("} legitimize_address");
 152       return Address(scratch);
 153     }
 154     return a;
 155   }
 156 
 157   void addmw(Address a, Register incr, Register scratch) {
 158     ldrw(scratch, a);
 159     addw(scratch, scratch, incr);
 160     strw(scratch, a);
 161   }
 162 
 163   // Add constant to memory word
 164   void addmw(Address a, int imm, Register scratch) {
 165     ldrw(scratch, a);
 166     if (imm > 0)
 167       addw(scratch, scratch, (unsigned)imm);
 168     else
 169       subw(scratch, scratch, (unsigned)-imm);
 170     strw(scratch, a);
 171   }
 172 
 173   void bind(Label& L) {
 174     Assembler::bind(L);
 175     code()->clear_last_insn();
 176     code()->set_last_label(pc());
 177   }
 178 
 179   void membar(Membar_mask_bits order_constraint);
 180 
 181   using Assembler::ldr;
 182   using Assembler::str;
 183   using Assembler::ldrw;
 184   using Assembler::strw;
 185 
 186   void ldr(Register Rx, const Address &adr);
 187   void ldrw(Register Rw, const Address &adr);
 188   void str(Register Rx, const Address &adr);
 189   void strw(Register Rx, const Address &adr);
 190 
 191   // Frame creation and destruction shared between JITs.
 192   DEBUG_ONLY(void build_frame(int framesize);)
 193   void build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills));
 194   void remove_frame(int framesize);
 195 
 196   virtual void _call_Unimplemented(address call_site) {
 197     mov(rscratch2, call_site);
 198   }
 199 
 200 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 201 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 202 // https://reviews.llvm.org/D3311
 203 
 204 #ifdef _WIN64
 205 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 206 #else
 207 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 208 #endif
 209 
 210   // aliases defined in AARCH64 spec
 211 
 212   template<class T>
 213   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 214 
 215   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 216   inline void cmp(Register Rd, unsigned imm) = delete;
 217 
 218   template<class T>
 219   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 220 
 221   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 222   inline void cmn(Register Rd, unsigned imm) = delete;
 223 
 224   void cset(Register Rd, Assembler::Condition cond) {
 225     csinc(Rd, zr, zr, ~cond);
 226   }
 227   void csetw(Register Rd, Assembler::Condition cond) {
 228     csincw(Rd, zr, zr, ~cond);
 229   }
 230 
 231   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 232     csneg(Rd, Rn, Rn, ~cond);
 233   }
 234   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 235     csnegw(Rd, Rn, Rn, ~cond);
 236   }
 237 
 238   inline void movw(Register Rd, Register Rn) {
 239     if (Rd == sp || Rn == sp) {
 240       Assembler::addw(Rd, Rn, 0U);
 241     } else {
 242       orrw(Rd, zr, Rn);
 243     }
 244   }
 245   inline void mov(Register Rd, Register Rn) {
 246     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 247     if (Rd == Rn) {
 248     } else if (Rd == sp || Rn == sp) {
 249       Assembler::add(Rd, Rn, 0U);
 250     } else {
 251       orr(Rd, zr, Rn);
 252     }
 253   }
 254 
 255   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 256   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 257 
 258   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 259   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 260 
 261   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 262   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 263 
 264   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 265     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 266   }
 267   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 268     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 269   }
 270 
 271   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 272     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 273   }
 274   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 275     bfm(Rd, Rn, lsb , (lsb + width - 1));
 276   }
 277 
 278   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 279     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 280   }
 281   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 282     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 283   }
 284 
 285   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 286     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 287   }
 288   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 289     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 290   }
 291 
 292   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 293     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 294   }
 295   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 296     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 297   }
 298 
 299   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 300     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 301   }
 302   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 303     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 304   }
 305 
 306   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 307     sbfmw(Rd, Rn, imm, 31);
 308   }
 309 
 310   inline void asr(Register Rd, Register Rn, unsigned imm) {
 311     sbfm(Rd, Rn, imm, 63);
 312   }
 313 
 314   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 315     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 316   }
 317 
 318   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 319     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 320   }
 321 
 322   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 323     ubfmw(Rd, Rn, imm, 31);
 324   }
 325 
 326   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 327     ubfm(Rd, Rn, imm, 63);
 328   }
 329 
 330   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 331     extrw(Rd, Rn, Rn, imm);
 332   }
 333 
 334   inline void ror(Register Rd, Register Rn, unsigned imm) {
 335     extr(Rd, Rn, Rn, imm);
 336   }
 337 
 338   inline void rolw(Register Rd, Register Rn, unsigned imm) {
 339     extrw(Rd, Rn, Rn, (32 - imm));
 340   }
 341 
 342   inline void rol(Register Rd, Register Rn, unsigned imm) {
 343     extr(Rd, Rn, Rn, (64 - imm));
 344   }
 345 
 346   using Assembler::rax1;
 347   using Assembler::eor3;
 348 
 349   inline void rax1(Register Rd, Register Rn, Register Rm) {
 350     eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1)
 351   }
 352 
 353   inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) {
 354     assert(Rd != Rn, "Use tmp register");
 355     eor(Rd, Rm, Rk);
 356     eor(Rd, Rd, Rn);
 357   }
 358 
 359   inline void sxtbw(Register Rd, Register Rn) {
 360     sbfmw(Rd, Rn, 0, 7);
 361   }
 362   inline void sxthw(Register Rd, Register Rn) {
 363     sbfmw(Rd, Rn, 0, 15);
 364   }
 365   inline void sxtb(Register Rd, Register Rn) {
 366     sbfm(Rd, Rn, 0, 7);
 367   }
 368   inline void sxth(Register Rd, Register Rn) {
 369     sbfm(Rd, Rn, 0, 15);
 370   }
 371   inline void sxtw(Register Rd, Register Rn) {
 372     sbfm(Rd, Rn, 0, 31);
 373   }
 374 
 375   inline void uxtbw(Register Rd, Register Rn) {
 376     ubfmw(Rd, Rn, 0, 7);
 377   }
 378   inline void uxthw(Register Rd, Register Rn) {
 379     ubfmw(Rd, Rn, 0, 15);
 380   }
 381   inline void uxtb(Register Rd, Register Rn) {
 382     ubfm(Rd, Rn, 0, 7);
 383   }
 384   inline void uxth(Register Rd, Register Rn) {
 385     ubfm(Rd, Rn, 0, 15);
 386   }
 387   inline void uxtw(Register Rd, Register Rn) {
 388     ubfm(Rd, Rn, 0, 31);
 389   }
 390 
 391   inline void cmnw(Register Rn, Register Rm) {
 392     addsw(zr, Rn, Rm);
 393   }
 394   inline void cmn(Register Rn, Register Rm) {
 395     adds(zr, Rn, Rm);
 396   }
 397 
 398   inline void cmpw(Register Rn, Register Rm) {
 399     subsw(zr, Rn, Rm);
 400   }
 401   inline void cmp(Register Rn, Register Rm) {
 402     subs(zr, Rn, Rm);
 403   }
 404 
 405   inline void negw(Register Rd, Register Rn) {
 406     subw(Rd, zr, Rn);
 407   }
 408 
 409   inline void neg(Register Rd, Register Rn) {
 410     sub(Rd, zr, Rn);
 411   }
 412 
 413   inline void negsw(Register Rd, Register Rn) {
 414     subsw(Rd, zr, Rn);
 415   }
 416 
 417   inline void negs(Register Rd, Register Rn) {
 418     subs(Rd, zr, Rn);
 419   }
 420 
 421   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 422     addsw(zr, Rn, Rm, kind, shift);
 423   }
 424   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 425     adds(zr, Rn, Rm, kind, shift);
 426   }
 427 
 428   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 429     subsw(zr, Rn, Rm, kind, shift);
 430   }
 431   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 432     subs(zr, Rn, Rm, kind, shift);
 433   }
 434 
 435   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 436     subw(Rd, zr, Rn, kind, shift);
 437   }
 438 
 439   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 440     sub(Rd, zr, Rn, kind, shift);
 441   }
 442 
 443   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 444     subsw(Rd, zr, Rn, kind, shift);
 445   }
 446 
 447   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 448     subs(Rd, zr, Rn, kind, shift);
 449   }
 450 
 451   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 452     msubw(Rd, Rn, Rm, zr);
 453   }
 454   inline void mneg(Register Rd, Register Rn, Register Rm) {
 455     msub(Rd, Rn, Rm, zr);
 456   }
 457 
 458   inline void mulw(Register Rd, Register Rn, Register Rm) {
 459     maddw(Rd, Rn, Rm, zr);
 460   }
 461   inline void mul(Register Rd, Register Rn, Register Rm) {
 462     madd(Rd, Rn, Rm, zr);
 463   }
 464 
 465   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 466     smsubl(Rd, Rn, Rm, zr);
 467   }
 468   inline void smull(Register Rd, Register Rn, Register Rm) {
 469     smaddl(Rd, Rn, Rm, zr);
 470   }
 471 
 472   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 473     umsubl(Rd, Rn, Rm, zr);
 474   }
 475   inline void umull(Register Rd, Register Rn, Register Rm) {
 476     umaddl(Rd, Rn, Rm, zr);
 477   }
 478 
 479 #define WRAP(INSN)                                                            \
 480   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 481     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 482       nop();                                                                  \
 483     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 484   }
 485 
 486   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 487   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 488 #undef WRAP
 489 
 490 
 491   // macro assembly operations needed for aarch64
 492 
 493 public:
 494 
 495   enum FpPushPopMode {
 496     PushPopFull,
 497     PushPopSVE,
 498     PushPopNeon,
 499     PushPopFp
 500   };
 501 
 502   // first two private routines for loading 32 bit or 64 bit constants
 503 private:
 504 
 505   void mov_immediate64(Register dst, uint64_t imm64);
 506   void mov_immediate32(Register dst, uint32_t imm32);
 507 
 508   int push(unsigned int bitset, Register stack);
 509   int pop(unsigned int bitset, Register stack);
 510 
 511   int push_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 512   int pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 513 
 514   int push_p(unsigned int bitset, Register stack);
 515   int pop_p(unsigned int bitset, Register stack);
 516 
 517   void mov(Register dst, Address a);
 518 
 519 public:
 520 
 521   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
 522   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 523 
 524   void push_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) push_fp(regs.bits(), stack, mode); }
 525   void pop_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) pop_fp(regs.bits(), stack, mode); }
 526 
 527   static RegSet call_clobbered_gp_registers();
 528 
 529   void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); }
 530   void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); }
 531 
 532   // Push and pop everything that might be clobbered by a native
 533   // runtime call except rscratch1 and rscratch2.  (They are always
 534   // scratch, so we don't have to protect them.)  Only save the lower
 535   // 64 bits of each vector register. Additional registers can be excluded
 536   // in a passed RegSet.
 537   void push_call_clobbered_registers_except(RegSet exclude);
 538   void pop_call_clobbered_registers_except(RegSet exclude);
 539 
 540   void push_call_clobbered_registers() {
 541     push_call_clobbered_registers_except(RegSet());
 542   }
 543   void pop_call_clobbered_registers() {
 544     pop_call_clobbered_registers_except(RegSet());
 545   }
 546 
 547 
 548   // now mov instructions for loading absolute addresses and 32 or
 549   // 64 bit integers
 550 
 551   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 552 
 553   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 554   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 555 
 556   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 557 
 558   void mov(Register dst, RegisterOrConstant src) {
 559     if (src.is_register())
 560       mov(dst, src.as_register());
 561     else
 562       mov(dst, src.as_constant());
 563   }
 564 
 565   void movptr(Register r, uintptr_t imm64);
 566 
 567   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 568 
 569   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 570     orr(Vd, T, Vn, Vn);
 571   }
 572 
 573   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 574     fcvtsh(tmp, src);
 575     smov(dst, tmp, H, 0);
 576   }
 577 
 578   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 579     mov(tmp, H, 0, src);
 580     fcvths(dst, tmp);
 581   }
 582 
 583   // Generalized Test Bit And Branch, including a "far" variety which
 584   // spans more than 32KiB.
 585   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 586     assert(cond == EQ || cond == NE, "must be");
 587 
 588     if (isfar)
 589       cond = ~cond;
 590 
 591     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 592     if (cond == Assembler::EQ)
 593       branch = &Assembler::tbz;
 594     else
 595       branch = &Assembler::tbnz;
 596 
 597     if (isfar) {
 598       Label L;
 599       (this->*branch)(Rt, bitpos, L);
 600       b(dest);
 601       bind(L);
 602     } else {
 603       (this->*branch)(Rt, bitpos, dest);
 604     }
 605   }
 606 
 607   // macro instructions for accessing and updating floating point
 608   // status register
 609   //
 610   // FPSR : op1 == 011
 611   //        CRn == 0100
 612   //        CRm == 0100
 613   //        op2 == 001
 614 
 615   inline void get_fpsr(Register reg)
 616   {
 617     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 618   }
 619 
 620   inline void set_fpsr(Register reg)
 621   {
 622     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 623   }
 624 
 625   inline void clear_fpsr()
 626   {
 627     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 628   }
 629 
 630   // FPCR : op1 == 011
 631   //        CRn == 0100
 632   //        CRm == 0100
 633   //        op2 == 000
 634 
 635   inline void get_fpcr(Register reg) {
 636     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 637   }
 638 
 639   inline void set_fpcr(Register reg) {
 640     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 641   }
 642 
 643   // DCZID_EL0: op1 == 011
 644   //            CRn == 0000
 645   //            CRm == 0000
 646   //            op2 == 111
 647   inline void get_dczid_el0(Register reg)
 648   {
 649     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 650   }
 651 
 652   // CTR_EL0:   op1 == 011
 653   //            CRn == 0000
 654   //            CRm == 0000
 655   //            op2 == 001
 656   inline void get_ctr_el0(Register reg)
 657   {
 658     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 659   }
 660 
 661   inline void get_nzcv(Register reg) {
 662     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 663   }
 664 
 665   inline void set_nzcv(Register reg) {
 666     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 667   }
 668 
 669   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 670   int corrected_idivl(Register result, Register ra, Register rb,
 671                       bool want_remainder, Register tmp = rscratch1);
 672   int corrected_idivq(Register result, Register ra, Register rb,
 673                       bool want_remainder, Register tmp = rscratch1);
 674 
 675   // Support for null-checks
 676   //
 677   // Generates code that causes a null OS exception if the content of reg is null.
 678   // If the accessed location is M[reg + offset] and the offset is known, provide the
 679   // offset. No explicit code generation is needed if the offset is within a certain
 680   // range (0 <= offset <= page_size).
 681 
 682   virtual void null_check(Register reg, int offset = -1);
 683   static bool needs_explicit_null_check(intptr_t offset);
 684   static bool uses_implicit_null_check(void* address);
 685 
 686   // markWord tests, kills markWord reg
 687   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 688 
 689   // inlineKlass queries, kills temp_reg
 690   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 691 
 692   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 693   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 694   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 695   void test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker);
 696 
 697   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 698   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 699   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 700   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 701   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 702   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 703 
 704   // Check array klass layout helper for flat or null-free arrays...
 705   void test_flat_array_layout(Register lh, Label& is_flat_array);
 706   void test_non_flat_array_layout(Register lh, Label& is_non_flat_array);
 707 
 708   static address target_addr_for_insn(address insn_addr);
 709   static address target_addr_for_insn_or_null(address insn_addr);
 710 
 711   // Required platform-specific helpers for Label::patch_instructions.
 712   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 713   static int pd_patch_instruction_size(address branch, address target);
 714   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 715     pd_patch_instruction_size(branch, target);
 716   }
 717   static address pd_call_destination(address branch) {
 718     return target_addr_for_insn(branch);
 719   }
 720 #ifndef PRODUCT
 721   static void pd_print_patched_instruction(address branch);
 722 #endif
 723 
 724   static int patch_oop(address insn_addr, address o);
 725   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 726 
 727   // Return whether code is emitted to a scratch blob.
 728   virtual bool in_scratch_emit_size() {
 729     return false;
 730   }
 731   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 732   static int max_trampoline_stub_size();
 733   void emit_static_call_stub();
 734   static int static_call_stub_size();
 735 
 736   // The following 4 methods return the offset of the appropriate move instruction
 737 
 738   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 739   int load_unsigned_byte(Register dst, Address src);
 740   int load_unsigned_short(Register dst, Address src);
 741 
 742   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 743   int load_signed_byte(Register dst, Address src);
 744   int load_signed_short(Register dst, Address src);
 745 
 746   int load_signed_byte32(Register dst, Address src);
 747   int load_signed_short32(Register dst, Address src);
 748 
 749   // Support for sign-extension (hi:lo = extend_sign(lo))
 750   void extend_sign(Register hi, Register lo);
 751 
 752   // Load and store values by size and signed-ness
 753   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 754   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 755 
 756   // Support for inc/dec with optimal instruction selection depending on value
 757 
 758   // x86_64 aliases an unqualified register/address increment and
 759   // decrement to call incrementq and decrementq but also supports
 760   // explicitly sized calls to incrementq/decrementq or
 761   // incrementl/decrementl
 762 
 763   // for aarch64 the proper convention would be to use
 764   // increment/decrement for 64 bit operations and
 765   // incrementw/decrementw for 32 bit operations. so when porting
 766   // x86_64 code we can leave calls to increment/decrement as is,
 767   // replace incrementq/decrementq with increment/decrement and
 768   // replace incrementl/decrementl with incrementw/decrementw.
 769 
 770   // n.b. increment/decrement calls with an Address destination will
 771   // need to use a scratch register to load the value to be
 772   // incremented. increment/decrement calls which add or subtract a
 773   // constant value greater than 2^12 will need to use a 2nd scratch
 774   // register to hold the constant. so, a register increment/decrement
 775   // may trash rscratch2 and an address increment/decrement trash
 776   // rscratch and rscratch2
 777 
 778   void decrementw(Address dst, int value = 1);
 779   void decrementw(Register reg, int value = 1);
 780 
 781   void decrement(Register reg, int value = 1);
 782   void decrement(Address dst, int value = 1);
 783 
 784   void incrementw(Address dst, int value = 1);
 785   void incrementw(Register reg, int value = 1);
 786 
 787   void increment(Register reg, int value = 1);
 788   void increment(Address dst, int value = 1);
 789 
 790 
 791   // Alignment
 792   void align(int modulus);
 793   void align(int modulus, int target);
 794 
 795   // nop
 796   void post_call_nop();
 797 
 798   // Stack frame creation/removal
 799   void enter(bool strip_ret_addr = false);
 800   void leave();
 801 
 802   // ROP Protection
 803   void protect_return_address();
 804   void protect_return_address(Register return_reg);
 805   void authenticate_return_address();
 806   void authenticate_return_address(Register return_reg);
 807   void strip_return_address();
 808   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 809 
 810   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 811   // The pointer will be loaded into the thread register.
 812   void get_thread(Register thread);
 813 
 814   // support for argument shuffling
 815   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 816   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 817   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 818   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 819   void object_move(
 820                    OopMap* map,
 821                    int oop_handle_offset,
 822                    int framesize_in_slots,
 823                    VMRegPair src,
 824                    VMRegPair dst,
 825                    bool is_receiver,
 826                    int* receiver_offset);
 827 
 828 
 829   // Support for VM calls
 830   //
 831   // It is imperative that all calls into the VM are handled via the call_VM macros.
 832   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 833   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 834 
 835 
 836   void call_VM(Register oop_result,
 837                address entry_point,
 838                bool check_exceptions = true);
 839   void call_VM(Register oop_result,
 840                address entry_point,
 841                Register arg_1,
 842                bool check_exceptions = true);
 843   void call_VM(Register oop_result,
 844                address entry_point,
 845                Register arg_1, Register arg_2,
 846                bool check_exceptions = true);
 847   void call_VM(Register oop_result,
 848                address entry_point,
 849                Register arg_1, Register arg_2, Register arg_3,
 850                bool check_exceptions = true);
 851 
 852   // Overloadings with last_Java_sp
 853   void call_VM(Register oop_result,
 854                Register last_java_sp,
 855                address entry_point,
 856                int number_of_arguments = 0,
 857                bool check_exceptions = true);
 858   void call_VM(Register oop_result,
 859                Register last_java_sp,
 860                address entry_point,
 861                Register arg_1, bool
 862                check_exceptions = true);
 863   void call_VM(Register oop_result,
 864                Register last_java_sp,
 865                address entry_point,
 866                Register arg_1, Register arg_2,
 867                bool check_exceptions = true);
 868   void call_VM(Register oop_result,
 869                Register last_java_sp,
 870                address entry_point,
 871                Register arg_1, Register arg_2, Register arg_3,
 872                bool check_exceptions = true);
 873 
 874   void get_vm_result_oop(Register oop_result, Register thread);
 875   void get_vm_result_metadata(Register metadata_result, Register thread);
 876 
 877   // These always tightly bind to MacroAssembler::call_VM_base
 878   // bypassing the virtual implementation
 879   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 880   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 881   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 882   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 883   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
 884 
 885   void call_VM_leaf(address entry_point,
 886                     int number_of_arguments = 0);
 887   void call_VM_leaf(address entry_point,
 888                     Register arg_1);
 889   void call_VM_leaf(address entry_point,
 890                     Register arg_1, Register arg_2);
 891   void call_VM_leaf(address entry_point,
 892                     Register arg_1, Register arg_2, Register arg_3);
 893 
 894   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 895   // bypassing the virtual implementation
 896   void super_call_VM_leaf(address entry_point);
 897   void super_call_VM_leaf(address entry_point, Register arg_1);
 898   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 899   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 900   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 901 
 902   // last Java Frame (fills frame anchor)
 903   void set_last_Java_frame(Register last_java_sp,
 904                            Register last_java_fp,
 905                            address last_java_pc,
 906                            Register scratch);
 907 
 908   void set_last_Java_frame(Register last_java_sp,
 909                            Register last_java_fp,
 910                            Label &last_java_pc,
 911                            Register scratch);
 912 
 913   void set_last_Java_frame(Register last_java_sp,
 914                            Register last_java_fp,
 915                            Register last_java_pc,
 916                            Register scratch);
 917 
 918   void reset_last_Java_frame(Register thread);
 919 
 920   // thread in the default location (rthread)
 921   void reset_last_Java_frame(bool clear_fp);
 922 
 923   // Stores
 924   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 925   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 926 
 927   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 928   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 929 
 930   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 931   void c2bool(Register x);
 932 
 933   void load_method_holder_cld(Register rresult, Register rmethod);
 934   void load_method_holder(Register holder, Register method);
 935 
 936   // oop manipulations
 937   void load_metadata(Register dst, Register src);
 938 
 939   void load_narrow_klass_compact(Register dst, Register src);
 940   void load_klass(Register dst, Register src);
 941   void store_klass(Register dst, Register src);
 942   void cmp_klass(Register obj, Register klass, Register tmp);
 943   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 944 
 945   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 946   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 947   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 948 
 949   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 950                       Register tmp1, Register tmp2);
 951 
 952   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 953                        Register tmp1, Register tmp2, Register tmp3);
 954 
 955   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 956 
 957   // inline type data payload offsets...
 958   void payload_offset(Register inline_klass, Register offset);
 959   void payload_address(Register oop, Register data, Register inline_klass);
 960   // get data payload ptr a flat value array at index, kills rcx and index
 961   void data_for_value_array_index(Register array, Register array_klass,
 962                                   Register index, Register data);
 963 
 964   void load_heap_oop(Register dst, Address src, Register tmp1,
 965                      Register tmp2, DecoratorSet decorators = 0);
 966 
 967   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 968                               Register tmp2, DecoratorSet decorators = 0);
 969   void store_heap_oop(Address dst, Register val, Register tmp1,
 970                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 971 
 972   // currently unimplemented
 973   // Used for storing null. All other oop constants should be
 974   // stored using routines that take a jobject.
 975   void store_heap_oop_null(Address dst);
 976 
 977   void load_prototype_header(Register dst, Register src);
 978 
 979   void store_klass_gap(Register dst, Register src);
 980 
 981   // This dummy is to prevent a call to store_heap_oop from
 982   // converting a zero (like null) into a Register by giving
 983   // the compiler two choices it can't resolve
 984 
 985   void store_heap_oop(Address dst, void* dummy);
 986 
 987   void encode_heap_oop(Register d, Register s);
 988   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 989   void decode_heap_oop(Register d, Register s);
 990   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 991   void encode_heap_oop_not_null(Register r);
 992   void decode_heap_oop_not_null(Register r);
 993   void encode_heap_oop_not_null(Register dst, Register src);
 994   void decode_heap_oop_not_null(Register dst, Register src);
 995 
 996   void set_narrow_oop(Register dst, jobject obj);
 997 
 998   void decode_klass_not_null_for_aot(Register dst, Register src);
 999   void encode_klass_not_null_for_aot(Register dst, Register src);
1000   void encode_klass_not_null(Register r);
1001   void decode_klass_not_null(Register r);
1002   void encode_klass_not_null(Register dst, Register src);
1003   void decode_klass_not_null(Register dst, Register src);
1004 
1005   void set_narrow_klass(Register dst, Klass* k);
1006 
1007   // if heap base register is used - reinit it with the correct value
1008   void reinit_heapbase();
1009 
1010   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1011 
1012   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1013                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1014   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1015                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1016 
1017   void push_cont_fastpath(Register java_thread = rthread);
1018   void pop_cont_fastpath(Register java_thread = rthread);
1019 
1020   // Round up to a power of two
1021   void round_to(Register reg, int modulus);
1022 
1023   // java.lang.Math::round intrinsics
1024   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1025   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1026 
1027   // allocation
1028 
1029   // Object / value buffer allocation...
1030   // Allocate instance of klass, assumes klass initialized by caller
1031   // new_obj prefers to be rax
1032   // Kills t1 and t2, perserves klass, return allocation in new_obj (rsi on LP64)
1033   void allocate_instance(Register klass, Register new_obj,
1034                          Register t1, Register t2,
1035                          bool clear_fields, Label& alloc_failed);
1036 
1037   void tlab_allocate(
1038     Register obj,                      // result: pointer to object after successful allocation
1039     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1040     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1041     Register t1,                       // temp register
1042     Register t2,                       // temp register
1043     Label&   slow_case                 // continuation point if fast allocation fails
1044   );
1045   void verify_tlab();
1046 
1047   // For field "index" within "klass", return inline_klass ...
1048   void get_inline_type_field_klass(Register klass, Register index, Register inline_klass);
1049   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1050 
1051 
1052   // interface method calling
1053   void lookup_interface_method(Register recv_klass,
1054                                Register intf_klass,
1055                                RegisterOrConstant itable_index,
1056                                Register method_result,
1057                                Register scan_temp,
1058                                Label& no_such_interface,
1059                    bool return_method = true);
1060 
1061   void lookup_interface_method_stub(Register recv_klass,
1062                                     Register holder_klass,
1063                                     Register resolved_klass,
1064                                     Register method_result,
1065                                     Register temp_reg,
1066                                     Register temp_reg2,
1067                                     int itable_index,
1068                                     Label& L_no_such_interface);
1069 
1070   // virtual method calling
1071   // n.b. x86 allows RegisterOrConstant for vtable_index
1072   void lookup_virtual_method(Register recv_klass,
1073                              RegisterOrConstant vtable_index,
1074                              Register method_result);
1075 
1076   // Test sub_klass against super_klass, with fast and slow paths.
1077 
1078   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1079   // One of the three labels can be null, meaning take the fall-through.
1080   // If super_check_offset is -1, the value is loaded up from super_klass.
1081   // No registers are killed, except temp_reg.
1082   void check_klass_subtype_fast_path(Register sub_klass,
1083                                      Register super_klass,
1084                                      Register temp_reg,
1085                                      Label* L_success,
1086                                      Label* L_failure,
1087                                      Label* L_slow_path,
1088                                      Register super_check_offset = noreg);
1089 
1090   // The rest of the type check; must be wired to a corresponding fast path.
1091   // It does not repeat the fast path logic, so don't use it standalone.
1092   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1093   // Updates the sub's secondary super cache as necessary.
1094   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1095   void check_klass_subtype_slow_path(Register sub_klass,
1096                                      Register super_klass,
1097                                      Register temp_reg,
1098                                      Register temp2_reg,
1099                                      Label* L_success,
1100                                      Label* L_failure,
1101                                      bool set_cond_codes = false);
1102 
1103   void check_klass_subtype_slow_path_linear(Register sub_klass,
1104                                             Register super_klass,
1105                                             Register temp_reg,
1106                                             Register temp2_reg,
1107                                             Label* L_success,
1108                                             Label* L_failure,
1109                                             bool set_cond_codes = false);
1110 
1111   void check_klass_subtype_slow_path_table(Register sub_klass,
1112                                            Register super_klass,
1113                                            Register temp_reg,
1114                                            Register temp2_reg,
1115                                            Register temp3_reg,
1116                                            Register result_reg,
1117                                            FloatRegister vtemp_reg,
1118                                            Label* L_success,
1119                                            Label* L_failure,
1120                                            bool set_cond_codes = false);
1121 
1122   // If r is valid, return r.
1123   // If r is invalid, remove a register r2 from available_regs, add r2
1124   // to regs_to_push, then return r2.
1125   Register allocate_if_noreg(const Register r,
1126                              RegSetIterator<Register> &available_regs,
1127                              RegSet &regs_to_push);
1128 
1129   // Secondary subtype checking
1130   void lookup_secondary_supers_table_var(Register sub_klass,
1131                                          Register r_super_klass,
1132                                          Register temp1,
1133                                          Register temp2,
1134                                          Register temp3,
1135                                          FloatRegister vtemp,
1136                                          Register result,
1137                                          Label *L_success);
1138 
1139 
1140   // As above, but with a constant super_klass.
1141   // The result is in Register result, not the condition codes.
1142   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1143                                            Register r_super_klass,
1144                                            Register temp1,
1145                                            Register temp2,
1146                                            Register temp3,
1147                                            FloatRegister vtemp,
1148                                            Register result,
1149                                            u1 super_klass_slot,
1150                                            bool stub_is_near = false);
1151 
1152   void verify_secondary_supers_table(Register r_sub_klass,
1153                                      Register r_super_klass,
1154                                      Register temp1,
1155                                      Register temp2,
1156                                      Register result);
1157 
1158   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1159                                                Register r_array_base,
1160                                                Register r_array_index,
1161                                                Register r_bitmap,
1162                                                Register temp1,
1163                                                Register result,
1164                                                bool is_stub = true);
1165 
1166   // Simplified, combined version, good for typical uses.
1167   // Falls through on failure.
1168   void check_klass_subtype(Register sub_klass,
1169                            Register super_klass,
1170                            Register temp_reg,
1171                            Label& L_success);
1172 
1173   void clinit_barrier(Register klass,
1174                       Register thread,
1175                       Label* L_fast_path = nullptr,
1176                       Label* L_slow_path = nullptr);
1177 
1178   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1179 
1180   void verify_sve_vector_length(Register tmp = rscratch1);
1181   void reinitialize_ptrue() {
1182     if (UseSVE > 0) {
1183       sve_ptrue(ptrue, B);
1184     }
1185   }
1186   void verify_ptrue();
1187 
1188   // Debugging
1189 
1190   // only if +VerifyOops
1191   void _verify_oop(Register reg, const char* s, const char* file, int line);
1192   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1193 
1194   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1195     if (VerifyOops) {
1196       _verify_oop(reg, s, file, line);
1197     }
1198   }
1199   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1200     if (VerifyOops) {
1201       _verify_oop_addr(reg, s, file, line);
1202     }
1203   }
1204 
1205 // TODO: verify method and klass metadata (compare against vptr?)
1206   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1207   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1208 
1209 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1210 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1211 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1212 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1213 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1214 
1215   // Restore cpu control state after JNI call
1216   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1217 
1218   // prints msg, dumps registers and stops execution
1219   void stop(const char* msg);
1220 
1221   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1222 
1223   void untested()                                { stop("untested"); }
1224 
1225   void unimplemented(const char* what = "");
1226 
1227   void should_not_reach_here()                   { stop("should not reach here"); }
1228 
1229   void _assert_asm(Condition cc, const char* msg);
1230 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1231 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1232 
1233   // Stack overflow checking
1234   void bang_stack_with_offset(int offset) {
1235     // stack grows down, caller passes positive offset
1236     assert(offset > 0, "must bang with negative offset");
1237     sub(rscratch2, sp, offset);
1238     str(zr, Address(rscratch2));
1239   }
1240 
1241   // Writes to stack successive pages until offset reached to check for
1242   // stack overflow + shadow pages.  Also, clobbers tmp
1243   void bang_stack_size(Register size, Register tmp);
1244 
1245   // Check for reserved stack access in method being exited (for JIT)
1246   void reserved_stack_check();
1247 
1248   // Arithmetics
1249 
1250   // Clobber: rscratch1, rscratch2
1251   void addptr(const Address &dst, int32_t src);
1252 
1253   // Clobber: rscratch1
1254   void cmpptr(Register src1, Address src2);
1255 
1256   void cmpoop(Register obj1, Register obj2);
1257 
1258   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1259   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1260   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1261   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1262 
1263   void atomic_xchg(Register prev, Register newv, Register addr);
1264   void atomic_xchgw(Register prev, Register newv, Register addr);
1265   void atomic_xchgl(Register prev, Register newv, Register addr);
1266   void atomic_xchglw(Register prev, Register newv, Register addr);
1267   void atomic_xchgal(Register prev, Register newv, Register addr);
1268   void atomic_xchgalw(Register prev, Register newv, Register addr);
1269 
1270   void orptr(Address adr, RegisterOrConstant src) {
1271     ldr(rscratch1, adr);
1272     if (src.is_register())
1273       orr(rscratch1, rscratch1, src.as_register());
1274     else
1275       orr(rscratch1, rscratch1, src.as_constant());
1276     str(rscratch1, adr);
1277   }
1278 
1279   // A generic CAS; success or failure is in the EQ flag.
1280   // Clobbers rscratch1
1281   void cmpxchg(Register addr, Register expected, Register new_val,
1282                enum operand_size size,
1283                bool acquire, bool release, bool weak,
1284                Register result);
1285 
1286 #ifdef ASSERT
1287   // Template short-hand support to clean-up after a failed call to trampoline
1288   // call generation (see trampoline_call() below),  when a set of Labels must
1289   // be reset (before returning).
1290   template<typename Label, typename... More>
1291   void reset_labels(Label &lbl, More&... more) {
1292     lbl.reset(); reset_labels(more...);
1293   }
1294   template<typename Label>
1295   void reset_labels(Label &lbl) {
1296     lbl.reset();
1297   }
1298 #endif
1299 
1300 private:
1301   void compare_eq(Register rn, Register rm, enum operand_size size);
1302 
1303 public:
1304   // AArch64 OpenJDK uses four different types of calls:
1305   //   - direct call: bl pc_relative_offset
1306   //     This is the shortest and the fastest, but the offset has the range:
1307   //     +/-128MB for the release build, +/-2MB for the debug build.
1308   //
1309   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1310   //     This is longer than a direct call. The offset has
1311   //     the range +/-4GB. As the code cache size is limited to 4GB,
1312   //     far calls can reach anywhere in the code cache. If a jump is
1313   //     needed rather than a call, a far jump 'b reg' can be used instead.
1314   //     All instructions are embedded at a call site.
1315   //
1316   //   - trampoline call:
1317   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1318   //     of a direct call, which is used if the destination of a call is in range,
1319   //     and a register-indirect call. It has the advantages of reaching anywhere in
1320   //     the AArch64 address space and being patchable at runtime when the generated
1321   //     code is being executed by other threads.
1322   //
1323   //     [Main code section]
1324   //       bl trampoline
1325   //     [Stub code section]
1326   //     trampoline:
1327   //       ldr reg, pc + 8
1328   //       br reg
1329   //       <64-bit destination address>
1330   //
1331   //     If the destination is in range when the generated code is moved to the code
1332   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1333   //     is not used.
1334   //     The optimization does not remove the trampoline from the stub section.
1335   //     This is necessary because the trampoline may well be redirected later when
1336   //     code is patched, and the new destination may not be reachable by a simple BR
1337   //     instruction.
1338   //
1339   //   - indirect call: move reg, address; blr reg
1340   //     This too can reach anywhere in the address space, but it cannot be
1341   //     patched while code is running, so it must only be modified at a safepoint.
1342   //     This form of call is most suitable for targets at fixed addresses, which
1343   //     will never be patched.
1344   //
1345   // The patching we do conforms to the "Concurrent modification and
1346   // execution of instructions" section of the Arm Architectural
1347   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1348   // or SVC instructions to be modified while another thread is
1349   // executing them.
1350   //
1351   // To patch a trampoline call when the BL can't reach, we first modify
1352   // the 64-bit destination address in the trampoline, then modify the
1353   // BL to point to the trampoline, then flush the instruction cache to
1354   // broadcast the change to all executing threads. See
1355   // NativeCall::set_destination_mt_safe for the details.
1356   //
1357   // There is a benign race in that the other thread might observe the
1358   // modified BL before it observes the modified 64-bit destination
1359   // address. That does not matter because the destination method has been
1360   // invalidated, so there will be a trap at its start.
1361   // For this to work, the destination address in the trampoline is
1362   // always updated, even if we're not using the trampoline.
1363 
1364   // Emit a direct call if the entry address will always be in range,
1365   // otherwise a trampoline call.
1366   // Supported entry.rspec():
1367   // - relocInfo::runtime_call_type
1368   // - relocInfo::opt_virtual_call_type
1369   // - relocInfo::static_call_type
1370   // - relocInfo::virtual_call_type
1371   //
1372   // Return: the call PC or null if CodeCache is full.
1373   // Clobbers: rscratch1
1374   address trampoline_call(Address entry);
1375 
1376   static bool far_branches() {
1377     return ReservedCodeCacheSize > branch_range;
1378   }
1379 
1380   // Check if branches to the non nmethod section require a far jump
1381   static bool codestub_branch_needs_far_jump() {
1382     if (AOTCodeCache::is_on_for_dump()) {
1383       // To calculate far_codestub_branch_size correctly.
1384       return true;
1385     }
1386     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1387   }
1388 
1389   // Emit a direct call/jump if the entry address will always be in range,
1390   // otherwise a far call/jump.
1391   // The address must be inside the code cache.
1392   // Supported entry.rspec():
1393   // - relocInfo::external_word_type
1394   // - relocInfo::runtime_call_type
1395   // - relocInfo::none
1396   // In the case of a far call/jump, the entry address is put in the tmp register.
1397   // The tmp register is invalidated.
1398   //
1399   // Far_jump returns the amount of the emitted code.
1400   void far_call(Address entry, Register tmp = rscratch1);
1401   int far_jump(Address entry, Register tmp = rscratch1);
1402 
1403   static int far_codestub_branch_size() {
1404     if (codestub_branch_needs_far_jump()) {
1405       return 3 * 4;  // adrp, add, br
1406     } else {
1407       return 4;
1408     }
1409   }
1410 
1411   // Emit the CompiledIC call idiom
1412   address ic_call(address entry, jint method_index = 0);
1413   static int ic_check_size();
1414   int ic_check(int end_alignment);
1415 
1416 public:
1417 
1418   // Data
1419 
1420   void mov_metadata(Register dst, Metadata* obj);
1421   Address allocate_metadata_address(Metadata* obj);
1422   Address constant_oop_address(jobject obj);
1423 
1424   void movoop(Register dst, jobject obj);
1425 
1426   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1427   void kernel_crc32(Register crc, Register buf, Register len,
1428         Register table0, Register table1, Register table2, Register table3,
1429         Register tmp, Register tmp2, Register tmp3);
1430   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1431   void kernel_crc32c(Register crc, Register buf, Register len,
1432         Register table0, Register table1, Register table2, Register table3,
1433         Register tmp, Register tmp2, Register tmp3);
1434 
1435   // Stack push and pop individual 64 bit registers
1436   void push(Register src);
1437   void pop(Register dst);
1438 
1439   void repne_scan(Register addr, Register value, Register count,
1440                   Register scratch);
1441   void repne_scanw(Register addr, Register value, Register count,
1442                    Register scratch);
1443 
1444   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1445   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1446 
1447   // If a constant does not fit in an immediate field, generate some
1448   // number of MOV instructions and then perform the operation
1449   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1450                              add_sub_imm_insn insn1,
1451                              add_sub_reg_insn insn2, bool is32);
1452   // Separate vsn which sets the flags
1453   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1454                                add_sub_imm_insn insn1,
1455                                add_sub_reg_insn insn2, bool is32);
1456 
1457 #define WRAP(INSN, is32)                                                \
1458   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1459     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1460   }                                                                     \
1461                                                                         \
1462   void INSN(Register Rd, Register Rn, Register Rm,                      \
1463              enum shift_kind kind, unsigned shift = 0) {                \
1464     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1465   }                                                                     \
1466                                                                         \
1467   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1468     Assembler::INSN(Rd, Rn, Rm);                                        \
1469   }                                                                     \
1470                                                                         \
1471   void INSN(Register Rd, Register Rn, Register Rm,                      \
1472            ext::operation option, int amount = 0) {                     \
1473     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1474   }
1475 
1476   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1477 
1478 #undef WRAP
1479 #define WRAP(INSN, is32)                                                \
1480   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1481     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1482   }                                                                     \
1483                                                                         \
1484   void INSN(Register Rd, Register Rn, Register Rm,                      \
1485              enum shift_kind kind, unsigned shift = 0) {                \
1486     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1487   }                                                                     \
1488                                                                         \
1489   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1490     Assembler::INSN(Rd, Rn, Rm);                                        \
1491   }                                                                     \
1492                                                                         \
1493   void INSN(Register Rd, Register Rn, Register Rm,                      \
1494            ext::operation option, int amount = 0) {                     \
1495     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1496   }
1497 
1498   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1499 
1500   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1501   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1502   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1503   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1504 
1505   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1506 
1507   void verified_entry(Compile* C, int sp_inc);
1508 
1509   // Inline type specific methods
1510   #include "asm/macroAssembler_common.hpp"
1511 
1512   int store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter = true);
1513   bool move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]);
1514   bool unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
1515                             VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
1516                             RegState reg_state[]);
1517   bool pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
1518                           VMRegPair* from, int from_count, int& from_index, VMReg to,
1519                           RegState reg_state[], Register val_array);
1520   int extend_stack_for_inline_args(int args_on_stack);
1521   void remove_frame(int initial_framesize, bool needs_stack_repair);
1522   VMReg spill_reg_for(VMReg reg);
1523   void save_stack_increment(int sp_inc, int frame_size);
1524 
1525   void tableswitch(Register index, jint lowbound, jint highbound,
1526                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1527     adr(rscratch1, jumptable);
1528     subsw(rscratch2, index, lowbound);
1529     subsw(zr, rscratch2, highbound - lowbound);
1530     br(Assembler::HS, jumptable_end);
1531     add(rscratch1, rscratch1, rscratch2,
1532         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1533     br(rscratch1);
1534   }
1535 
1536   // Form an address from base + offset in Rd.  Rd may or may not
1537   // actually be used: you must use the Address that is returned.  It
1538   // is up to you to ensure that the shift provided matches the size
1539   // of your data.
1540   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1541 
1542   // Return true iff an address is within the 48-bit AArch64 address
1543   // space.
1544   bool is_valid_AArch64_address(address a) {
1545     return ((uint64_t)a >> 48) == 0;
1546   }
1547 
1548   // Load the base of the cardtable byte map into reg.
1549   void load_byte_map_base(Register reg);
1550 
1551   // Prolog generator routines to support switch between x86 code and
1552   // generated ARM code
1553 
1554   // routine to generate an x86 prolog for a stub function which
1555   // bootstraps into the generated ARM code which directly follows the
1556   // stub
1557   //
1558 
1559   public:
1560 
1561   address read_polling_page(Register r, relocInfo::relocType rtype);
1562   void get_polling_page(Register dest, relocInfo::relocType rtype);
1563 
1564   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1565   void update_byte_crc32(Register crc, Register val, Register table);
1566   void update_word_crc32(Register crc, Register v, Register tmp,
1567         Register table0, Register table1, Register table2, Register table3,
1568         bool upper = false);
1569 
1570   address count_positives(Register ary1, Register len, Register result);
1571 
1572   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1573                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1574 
1575 // Ensure that the inline code and the stub use the same registers.
1576 #define ARRAYS_HASHCODE_REGISTERS \
1577   do {                      \
1578     assert(result == r0  && \
1579            ary    == r1  && \
1580            cnt    == r2  && \
1581            vdata0 == v3  && \
1582            vdata1 == v2  && \
1583            vdata2 == v1  && \
1584            vdata3 == v0  && \
1585            vmul0  == v4  && \
1586            vmul1  == v5  && \
1587            vmul2  == v6  && \
1588            vmul3  == v7  && \
1589            vpow   == v12 && \
1590            vpowm  == v13, "registers must match aarch64.ad"); \
1591   } while (0)
1592 
1593   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1594 
1595   void fill_words(Register base, Register cnt, Register value);
1596   void fill_words(Register base, uint64_t cnt, Register value);
1597 
1598   address zero_words(Register base, uint64_t cnt);
1599   address zero_words(Register ptr, Register cnt);
1600   void zero_dcache_blocks(Register base, Register cnt);
1601 
1602   static const int zero_words_block_size;
1603 
1604   address byte_array_inflate(Register src, Register dst, Register len,
1605                              FloatRegister vtmp1, FloatRegister vtmp2,
1606                              FloatRegister vtmp3, Register tmp4);
1607 
1608   void char_array_compress(Register src, Register dst, Register len,
1609                            Register res,
1610                            FloatRegister vtmp0, FloatRegister vtmp1,
1611                            FloatRegister vtmp2, FloatRegister vtmp3,
1612                            FloatRegister vtmp4, FloatRegister vtmp5);
1613 
1614   void encode_iso_array(Register src, Register dst,
1615                         Register len, Register res, bool ascii,
1616                         FloatRegister vtmp0, FloatRegister vtmp1,
1617                         FloatRegister vtmp2, FloatRegister vtmp3,
1618                         FloatRegister vtmp4, FloatRegister vtmp5);
1619 
1620   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1621       address pio2, address dsin_coef, address dcos_coef);
1622  private:
1623   // begin trigonometric functions support block
1624   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1625   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1626   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1627   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1628   // end trigonometric functions support block
1629   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1630                        Register src1, Register src2);
1631   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1632     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1633   }
1634   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1635                              Register y, Register y_idx, Register z,
1636                              Register carry, Register product,
1637                              Register idx, Register kdx);
1638   void multiply_128_x_128_loop(Register y, Register z,
1639                                Register carry, Register carry2,
1640                                Register idx, Register jdx,
1641                                Register yz_idx1, Register yz_idx2,
1642                                Register tmp, Register tmp3, Register tmp4,
1643                                Register tmp7, Register product_hi);
1644   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1645         Register len, Register tmp0, Register tmp1, Register tmp2,
1646         Register tmp3);
1647   void kernel_crc32_using_crc32(Register crc, Register buf,
1648         Register len, Register tmp0, Register tmp1, Register tmp2,
1649         Register tmp3);
1650   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1651         Register len, Register tmp0, Register tmp1, Register tmp2,
1652         Register tmp3);
1653   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1654         Register len, Register tmp0, Register tmp1, Register tmp2,
1655         Register tmp3);
1656   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1657         Register len, Register tmp0, Register tmp1, Register tmp2,
1658         size_t table_offset);
1659 
1660   void ghash_modmul (FloatRegister result,
1661                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1662                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1663                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1664   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1665 public:
1666   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1667                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1668                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1669   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1670   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1671                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1672                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1673   void ghash_multiply_wide(int index,
1674                            FloatRegister result_lo, FloatRegister result_hi,
1675                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1676                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1677   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1678                     FloatRegister p, FloatRegister z, FloatRegister t1);
1679   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1680                     FloatRegister p, FloatRegister z, FloatRegister t1);
1681   void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH,
1682                                 Register data, Register blocks, int unrolls);
1683 
1684 
1685   void aesenc_loadkeys(Register key, Register keylen);
1686   void aesecb_encrypt(Register from, Register to, Register keylen,
1687                       FloatRegister data = v0, int unrolls = 1);
1688   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1689   void aes_round(FloatRegister input, FloatRegister subkey);
1690 
1691   // ChaCha20 functions support block
1692   void cc20_qr_add4(FloatRegister (&addFirst)[4],
1693           FloatRegister (&addSecond)[4]);
1694   void cc20_qr_xor4(FloatRegister (&firstElem)[4],
1695           FloatRegister (&secondElem)[4], FloatRegister (&result)[4]);
1696   void cc20_qr_lrot4(FloatRegister (&sourceReg)[4],
1697           FloatRegister (&destReg)[4], int bits, FloatRegister table);
1698   void cc20_set_qr_registers(FloatRegister (&vectorSet)[4],
1699           const FloatRegister (&stateVectors)[16], int idx1, int idx2,
1700           int idx3, int idx4);
1701 
1702   // Place an ISB after code may have been modified due to a safepoint.
1703   void safepoint_isb();
1704 
1705 private:
1706   // Return the effective address r + (r1 << ext) + offset.
1707   // Uses rscratch2.
1708   Address offsetted_address(Register r, Register r1, Address::extend ext,
1709                             int offset, int size);
1710 
1711 private:
1712   // Returns an address on the stack which is reachable with a ldr/str of size
1713   // Uses rscratch2 if the address is not directly reachable
1714   Address spill_address(int size, int offset, Register tmp=rscratch2);
1715   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1716 
1717   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1718 
1719   // Check whether two loads/stores can be merged into ldp/stp.
1720   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1721 
1722   // Merge current load/store with previous load/store into ldp/stp.
1723   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1724 
1725   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1726   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1727 
1728 public:
1729   void spill(Register Rx, bool is64, int offset) {
1730     if (is64) {
1731       str(Rx, spill_address(8, offset));
1732     } else {
1733       strw(Rx, spill_address(4, offset));
1734     }
1735   }
1736   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1737     str(Vx, T, spill_address(1 << (int)T, offset));
1738   }
1739 
1740   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1741     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1742   }
1743   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1744     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1745   }
1746 
1747   void unspill(Register Rx, bool is64, int offset) {
1748     if (is64) {
1749       ldr(Rx, spill_address(8, offset));
1750     } else {
1751       ldrw(Rx, spill_address(4, offset));
1752     }
1753   }
1754   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1755     ldr(Vx, T, spill_address(1 << (int)T, offset));
1756   }
1757 
1758   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1759     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1760   }
1761   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1762     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1763   }
1764 
1765   void spill_copy128(int src_offset, int dst_offset,
1766                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1767     if (src_offset < 512 && (src_offset & 7) == 0 &&
1768         dst_offset < 512 && (dst_offset & 7) == 0) {
1769       ldp(tmp1, tmp2, Address(sp, src_offset));
1770       stp(tmp1, tmp2, Address(sp, dst_offset));
1771     } else {
1772       unspill(tmp1, true, src_offset);
1773       spill(tmp1, true, dst_offset);
1774       unspill(tmp1, true, src_offset+8);
1775       spill(tmp1, true, dst_offset+8);
1776     }
1777   }
1778   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1779                                             int sve_vec_reg_size_in_bytes) {
1780     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1781     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1782       spill_copy128(src_offset, dst_offset);
1783       src_offset += 16;
1784       dst_offset += 16;
1785     }
1786   }
1787   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1788                                                int sve_predicate_reg_size_in_bytes) {
1789     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1790     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1791     reinitialize_ptrue();
1792   }
1793   void cache_wb(Address line);
1794   void cache_wbsync(bool is_pre);
1795 
1796   // Code for java.lang.Thread::onSpinWait() intrinsic.
1797   void spin_wait();
1798 
1799   void lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1800   void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1801 
1802 private:
1803   // Check the current thread doesn't need a cross modify fence.
1804   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1805 
1806 };
1807 
1808 #ifdef ASSERT
1809 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1810 #endif
1811 
1812 struct tableswitch {
1813   Register _reg;
1814   int _insn_index; jint _first_key; jint _last_key;
1815   Label _after;
1816   Label _branches;
1817 };
1818 
1819 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP