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/vmreg.hpp"
  31 #include "metaprogramming/enableIf.hpp"
  32 #include "oops/compressedOops.hpp"
  33 #include "oops/compressedKlass.hpp"
  34 #include "runtime/vm_version.hpp"
  35 #include "utilities/macros.hpp"
  36 #include "utilities/powerOfTwo.hpp"
  37 #include "runtime/signature.hpp"
  38 
  39 
  40 class ciInlineKlass;
  41 
  42 class OopMap;
  43 
  44 // MacroAssembler extends Assembler by frequently used macros.
  45 //
  46 // Instructions for which a 'better' code sequence exists depending
  47 // on arguments should also go in here.
  48 
  49 class MacroAssembler: public Assembler {
  50   friend class LIR_Assembler;
  51 
  52  public:
  53   using Assembler::mov;
  54   using Assembler::movi;
  55 
  56  protected:
  57 
  58   // Support for VM calls
  59   //
  60   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  61   // may customize this version by overriding it for its purposes (e.g., to save/restore
  62   // additional registers when doing a VM call).
  63   virtual void call_VM_leaf_base(
  64     address entry_point,               // the entry point
  65     int     number_of_arguments,        // the number of arguments to pop after the call
  66     Label *retaddr = nullptr
  67   );
  68 
  69   virtual void call_VM_leaf_base(
  70     address entry_point,               // the entry point
  71     int     number_of_arguments,        // the number of arguments to pop after the call
  72     Label &retaddr) {
  73     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  74   }
  75 
  76   // This is the base routine called by the different versions of call_VM. The interpreter
  77   // may customize this version by overriding it for its purposes (e.g., to save/restore
  78   // additional registers when doing a VM call).
  79   //
  80   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  81   // returns the register which contains the thread upon return. If a thread register has been
  82   // specified, the return value will correspond to that register. If no last_java_sp is specified
  83   // (noreg) than rsp will be used instead.
  84   virtual void call_VM_base(           // returns the register containing the thread upon return
  85     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  86     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  87     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  88     address  entry_point,              // the entry point
  89     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  90     bool     check_exceptions          // whether to check for pending exceptions after return
  91   );
  92 
  93   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  94 
  95   enum KlassDecodeMode {
  96     KlassDecodeNone,
  97     KlassDecodeZero,
  98     KlassDecodeXor,
  99     KlassDecodeMovk
 100   };
 101 
 102   // Calculate decoding mode based on given parameters, used for checking then ultimately setting.
 103   static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range);
 104 
 105  private:
 106   static KlassDecodeMode _klass_decode_mode;
 107 
 108   // Returns above setting with asserts
 109   static KlassDecodeMode klass_decode_mode();
 110 
 111  public:
 112   // Checks the decode mode and returns false if not compatible with preferred decoding mode.
 113   static bool check_klass_decode_mode(address base, int shift, const size_t range);
 114 
 115   // Sets the decode mode and returns false if cannot be set.
 116   static bool set_klass_decode_mode(address base, int shift, const size_t range);
 117 
 118  public:
 119   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 120 
 121  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 122  // The implementation is only non-empty for the InterpreterMacroAssembler,
 123  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 124  virtual void check_and_handle_popframe(Register java_thread);
 125  virtual void check_and_handle_earlyret(Register java_thread);
 126 
 127   void safepoint_poll(Label& slow_path, bool at_return, bool acquire, bool in_nmethod, Register tmp = rscratch1);
 128   void rt_call(address dest, Register tmp = rscratch1);
 129 
 130   // Load Effective Address
 131   void lea(Register r, const Address &a) {
 132     InstructionMark im(this);
 133     a.lea(this, r);
 134   }
 135 
 136   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 137      accesses, and these can exceed the offset range. */
 138   Address legitimize_address(const Address &a, int size, Register scratch) {
 139     if (a.getMode() == Address::base_plus_offset) {
 140       if (! Address::offset_ok_for_immed(a.offset(), exact_log2(size))) {
 141         block_comment("legitimize_address {");
 142         lea(scratch, a);
 143         block_comment("} legitimize_address");
 144         return Address(scratch);
 145       }
 146     }
 147     return a;
 148   }
 149 
 150   void addmw(Address a, Register incr, Register scratch) {
 151     ldrw(scratch, a);
 152     addw(scratch, scratch, incr);
 153     strw(scratch, a);
 154   }
 155 
 156   // Add constant to memory word
 157   void addmw(Address a, int imm, Register scratch) {
 158     ldrw(scratch, a);
 159     if (imm > 0)
 160       addw(scratch, scratch, (unsigned)imm);
 161     else
 162       subw(scratch, scratch, (unsigned)-imm);
 163     strw(scratch, a);
 164   }
 165 
 166   void bind(Label& L) {
 167     Assembler::bind(L);
 168     code()->clear_last_insn();
 169     code()->set_last_label(pc());
 170   }
 171 
 172   void membar(Membar_mask_bits order_constraint);
 173 
 174   using Assembler::ldr;
 175   using Assembler::str;
 176   using Assembler::ldrw;
 177   using Assembler::strw;
 178 
 179   void ldr(Register Rx, const Address &adr);
 180   void ldrw(Register Rw, const Address &adr);
 181   void str(Register Rx, const Address &adr);
 182   void strw(Register Rx, const Address &adr);
 183 
 184   // Frame creation and destruction shared between JITs.
 185   void build_frame(int framesize);
 186   void remove_frame(int framesize);
 187 
 188   virtual void _call_Unimplemented(address call_site) {
 189     mov(rscratch2, call_site);
 190   }
 191 
 192 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 193 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 194 // https://reviews.llvm.org/D3311
 195 
 196 #ifdef _WIN64
 197 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 198 #else
 199 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 200 #endif
 201 
 202   // aliases defined in AARCH64 spec
 203 
 204   template<class T>
 205   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 206 
 207   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 208   inline void cmp(Register Rd, unsigned imm) = delete;
 209 
 210   template<class T>
 211   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 212 
 213   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 214   inline void cmn(Register Rd, unsigned imm) = delete;
 215 
 216   void cset(Register Rd, Assembler::Condition cond) {
 217     csinc(Rd, zr, zr, ~cond);
 218   }
 219   void csetw(Register Rd, Assembler::Condition cond) {
 220     csincw(Rd, zr, zr, ~cond);
 221   }
 222 
 223   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 224     csneg(Rd, Rn, Rn, ~cond);
 225   }
 226   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 227     csnegw(Rd, Rn, Rn, ~cond);
 228   }
 229 
 230   inline void movw(Register Rd, Register Rn) {
 231     if (Rd == sp || Rn == sp) {
 232       Assembler::addw(Rd, Rn, 0U);
 233     } else {
 234       orrw(Rd, zr, Rn);
 235     }
 236   }
 237   inline void mov(Register Rd, Register Rn) {
 238     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 239     if (Rd == Rn) {
 240     } else if (Rd == sp || Rn == sp) {
 241       Assembler::add(Rd, Rn, 0U);
 242     } else {
 243       orr(Rd, zr, Rn);
 244     }
 245   }
 246 
 247   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 248   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 249 
 250   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 251   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 252 
 253   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 254   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 255 
 256   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 257     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 258   }
 259   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 260     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 261   }
 262 
 263   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 264     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 265   }
 266   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 267     bfm(Rd, Rn, lsb , (lsb + width - 1));
 268   }
 269 
 270   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 271     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 272   }
 273   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 274     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 275   }
 276 
 277   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 278     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 279   }
 280   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 281     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 282   }
 283 
 284   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 285     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 286   }
 287   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 288     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 289   }
 290 
 291   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 292     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 293   }
 294   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 295     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 296   }
 297 
 298   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 299     sbfmw(Rd, Rn, imm, 31);
 300   }
 301 
 302   inline void asr(Register Rd, Register Rn, unsigned imm) {
 303     sbfm(Rd, Rn, imm, 63);
 304   }
 305 
 306   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 307     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 308   }
 309 
 310   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 311     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 312   }
 313 
 314   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 315     ubfmw(Rd, Rn, imm, 31);
 316   }
 317 
 318   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 319     ubfm(Rd, Rn, imm, 63);
 320   }
 321 
 322   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 323     extrw(Rd, Rn, Rn, imm);
 324   }
 325 
 326   inline void ror(Register Rd, Register Rn, unsigned imm) {
 327     extr(Rd, Rn, Rn, imm);
 328   }
 329 
 330   inline void sxtbw(Register Rd, Register Rn) {
 331     sbfmw(Rd, Rn, 0, 7);
 332   }
 333   inline void sxthw(Register Rd, Register Rn) {
 334     sbfmw(Rd, Rn, 0, 15);
 335   }
 336   inline void sxtb(Register Rd, Register Rn) {
 337     sbfm(Rd, Rn, 0, 7);
 338   }
 339   inline void sxth(Register Rd, Register Rn) {
 340     sbfm(Rd, Rn, 0, 15);
 341   }
 342   inline void sxtw(Register Rd, Register Rn) {
 343     sbfm(Rd, Rn, 0, 31);
 344   }
 345 
 346   inline void uxtbw(Register Rd, Register Rn) {
 347     ubfmw(Rd, Rn, 0, 7);
 348   }
 349   inline void uxthw(Register Rd, Register Rn) {
 350     ubfmw(Rd, Rn, 0, 15);
 351   }
 352   inline void uxtb(Register Rd, Register Rn) {
 353     ubfm(Rd, Rn, 0, 7);
 354   }
 355   inline void uxth(Register Rd, Register Rn) {
 356     ubfm(Rd, Rn, 0, 15);
 357   }
 358   inline void uxtw(Register Rd, Register Rn) {
 359     ubfm(Rd, Rn, 0, 31);
 360   }
 361 
 362   inline void cmnw(Register Rn, Register Rm) {
 363     addsw(zr, Rn, Rm);
 364   }
 365   inline void cmn(Register Rn, Register Rm) {
 366     adds(zr, Rn, Rm);
 367   }
 368 
 369   inline void cmpw(Register Rn, Register Rm) {
 370     subsw(zr, Rn, Rm);
 371   }
 372   inline void cmp(Register Rn, Register Rm) {
 373     subs(zr, Rn, Rm);
 374   }
 375 
 376   inline void negw(Register Rd, Register Rn) {
 377     subw(Rd, zr, Rn);
 378   }
 379 
 380   inline void neg(Register Rd, Register Rn) {
 381     sub(Rd, zr, Rn);
 382   }
 383 
 384   inline void negsw(Register Rd, Register Rn) {
 385     subsw(Rd, zr, Rn);
 386   }
 387 
 388   inline void negs(Register Rd, Register Rn) {
 389     subs(Rd, zr, Rn);
 390   }
 391 
 392   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 393     addsw(zr, Rn, Rm, kind, shift);
 394   }
 395   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 396     adds(zr, Rn, Rm, kind, shift);
 397   }
 398 
 399   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 400     subsw(zr, Rn, Rm, kind, shift);
 401   }
 402   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 403     subs(zr, Rn, Rm, kind, shift);
 404   }
 405 
 406   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 407     subw(Rd, zr, Rn, kind, shift);
 408   }
 409 
 410   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 411     sub(Rd, zr, Rn, kind, shift);
 412   }
 413 
 414   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 415     subsw(Rd, zr, Rn, kind, shift);
 416   }
 417 
 418   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 419     subs(Rd, zr, Rn, kind, shift);
 420   }
 421 
 422   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 423     msubw(Rd, Rn, Rm, zr);
 424   }
 425   inline void mneg(Register Rd, Register Rn, Register Rm) {
 426     msub(Rd, Rn, Rm, zr);
 427   }
 428 
 429   inline void mulw(Register Rd, Register Rn, Register Rm) {
 430     maddw(Rd, Rn, Rm, zr);
 431   }
 432   inline void mul(Register Rd, Register Rn, Register Rm) {
 433     madd(Rd, Rn, Rm, zr);
 434   }
 435 
 436   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 437     smsubl(Rd, Rn, Rm, zr);
 438   }
 439   inline void smull(Register Rd, Register Rn, Register Rm) {
 440     smaddl(Rd, Rn, Rm, zr);
 441   }
 442 
 443   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 444     umsubl(Rd, Rn, Rm, zr);
 445   }
 446   inline void umull(Register Rd, Register Rn, Register Rm) {
 447     umaddl(Rd, Rn, Rm, zr);
 448   }
 449 
 450 #define WRAP(INSN)                                                            \
 451   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 452     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 453       nop();                                                                  \
 454     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 455   }
 456 
 457   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 458   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 459 #undef WRAP
 460 
 461 
 462   // macro assembly operations needed for aarch64
 463 
 464 public:
 465 
 466   enum FpPushPopMode {
 467     PushPopFull,
 468     PushPopSVE,
 469     PushPopNeon,
 470     PushPopFp
 471   };
 472 
 473   // first two private routines for loading 32 bit or 64 bit constants
 474 private:
 475 
 476   void mov_immediate64(Register dst, uint64_t imm64);
 477   void mov_immediate32(Register dst, uint32_t imm32);
 478 
 479   int push(unsigned int bitset, Register stack);
 480   int pop(unsigned int bitset, Register stack);
 481 
 482   int push_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 483   int pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode);
 484 
 485   int push_p(unsigned int bitset, Register stack);
 486   int pop_p(unsigned int bitset, Register stack);
 487 
 488   void mov(Register dst, Address a);
 489 
 490 public:
 491 
 492   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
 493   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 494 
 495   void push_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) push_fp(regs.bits(), stack, mode); }
 496   void pop_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) pop_fp(regs.bits(), stack, mode); }
 497 
 498   static RegSet call_clobbered_gp_registers();
 499 
 500   void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); }
 501   void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); }
 502 
 503   // Push and pop everything that might be clobbered by a native
 504   // runtime call except rscratch1 and rscratch2.  (They are always
 505   // scratch, so we don't have to protect them.)  Only save the lower
 506   // 64 bits of each vector register. Additional registers can be excluded
 507   // in a passed RegSet.
 508   void push_call_clobbered_registers_except(RegSet exclude);
 509   void pop_call_clobbered_registers_except(RegSet exclude);
 510 
 511   void push_call_clobbered_registers() {
 512     push_call_clobbered_registers_except(RegSet());
 513   }
 514   void pop_call_clobbered_registers() {
 515     pop_call_clobbered_registers_except(RegSet());
 516   }
 517 
 518 
 519   // now mov instructions for loading absolute addresses and 32 or
 520   // 64 bit integers
 521 
 522   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 523 
 524   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 525   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 526 
 527   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 528 
 529   void mov(Register dst, RegisterOrConstant src) {
 530     if (src.is_register())
 531       mov(dst, src.as_register());
 532     else
 533       mov(dst, src.as_constant());
 534   }
 535 
 536   void movptr(Register r, uintptr_t imm64);
 537 
 538   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 539 
 540   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 541     orr(Vd, T, Vn, Vn);
 542   }
 543 
 544   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 545     fcvtsh(tmp, src);
 546     smov(dst, tmp, H, 0);
 547   }
 548 
 549   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 550     mov(tmp, H, 0, src);
 551     fcvths(dst, tmp);
 552   }
 553 
 554   // Generalized Test Bit And Branch, including a "far" variety which
 555   // spans more than 32KiB.
 556   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 557     assert(cond == EQ || cond == NE, "must be");
 558 
 559     if (isfar)
 560       cond = ~cond;
 561 
 562     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 563     if (cond == Assembler::EQ)
 564       branch = &Assembler::tbz;
 565     else
 566       branch = &Assembler::tbnz;
 567 
 568     if (isfar) {
 569       Label L;
 570       (this->*branch)(Rt, bitpos, L);
 571       b(dest);
 572       bind(L);
 573     } else {
 574       (this->*branch)(Rt, bitpos, dest);
 575     }
 576   }
 577 
 578   // macro instructions for accessing and updating floating point
 579   // status register
 580   //
 581   // FPSR : op1 == 011
 582   //        CRn == 0100
 583   //        CRm == 0100
 584   //        op2 == 001
 585 
 586   inline void get_fpsr(Register reg)
 587   {
 588     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 589   }
 590 
 591   inline void set_fpsr(Register reg)
 592   {
 593     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 594   }
 595 
 596   inline void clear_fpsr()
 597   {
 598     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 599   }
 600 
 601   // FPCR : op1 == 011
 602   //        CRn == 0100
 603   //        CRm == 0100
 604   //        op2 == 000
 605 
 606   inline void get_fpcr(Register reg) {
 607     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 608   }
 609 
 610   inline void set_fpcr(Register reg) {
 611     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 612   }
 613 
 614   // DCZID_EL0: op1 == 011
 615   //            CRn == 0000
 616   //            CRm == 0000
 617   //            op2 == 111
 618   inline void get_dczid_el0(Register reg)
 619   {
 620     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 621   }
 622 
 623   // CTR_EL0:   op1 == 011
 624   //            CRn == 0000
 625   //            CRm == 0000
 626   //            op2 == 001
 627   inline void get_ctr_el0(Register reg)
 628   {
 629     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 630   }
 631 
 632   inline void get_nzcv(Register reg) {
 633     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 634   }
 635 
 636   inline void set_nzcv(Register reg) {
 637     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 638   }
 639 
 640   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 641   int corrected_idivl(Register result, Register ra, Register rb,
 642                       bool want_remainder, Register tmp = rscratch1);
 643   int corrected_idivq(Register result, Register ra, Register rb,
 644                       bool want_remainder, Register tmp = rscratch1);
 645 
 646   // Support for null-checks
 647   //
 648   // Generates code that causes a null OS exception if the content of reg is null.
 649   // If the accessed location is M[reg + offset] and the offset is known, provide the
 650   // offset. No explicit code generation is needed if the offset is within a certain
 651   // range (0 <= offset <= page_size).
 652 
 653   virtual void null_check(Register reg, int offset = -1);
 654   static bool needs_explicit_null_check(intptr_t offset);
 655   static bool uses_implicit_null_check(void* address);
 656 
 657   // markWord tests, kills markWord reg
 658   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 659 
 660   // inlineKlass queries, kills temp_reg
 661   void test_klass_is_inline_type(Register klass, Register temp_reg, Label& is_inline_type);
 662   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type);
 663 
 664   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 665   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 666   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 667   void test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker);
 668 
 669   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 670   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 671   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 672   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 673   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 674   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 675 
 676   // Check array klass layout helper for flat or null-free arrays...
 677   void test_flat_array_layout(Register lh, Label& is_flat_array);
 678   void test_non_flat_array_layout(Register lh, Label& is_non_flat_array);
 679 
 680   static address target_addr_for_insn(address insn_addr, unsigned insn);
 681   static address target_addr_for_insn_or_null(address insn_addr, unsigned insn);
 682   static address target_addr_for_insn(address insn_addr) {
 683     unsigned insn = *(unsigned*)insn_addr;
 684     return target_addr_for_insn(insn_addr, insn);
 685   }
 686   static address target_addr_for_insn_or_null(address insn_addr) {
 687     unsigned insn = *(unsigned*)insn_addr;
 688     return target_addr_for_insn_or_null(insn_addr, insn);
 689   }
 690 
 691   // Required platform-specific helpers for Label::patch_instructions.
 692   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 693   static int pd_patch_instruction_size(address branch, address target);
 694   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 695     pd_patch_instruction_size(branch, target);
 696   }
 697   static address pd_call_destination(address branch) {
 698     return target_addr_for_insn(branch);
 699   }
 700 #ifndef PRODUCT
 701   static void pd_print_patched_instruction(address branch);
 702 #endif
 703 
 704   static int patch_oop(address insn_addr, address o);
 705   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 706 
 707   // Return whether code is emitted to a scratch blob.
 708   virtual bool in_scratch_emit_size() {
 709     return false;
 710   }
 711   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 712   static int max_trampoline_stub_size();
 713   void emit_static_call_stub();
 714   static int static_call_stub_size();
 715 
 716   // The following 4 methods return the offset of the appropriate move instruction
 717 
 718   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 719   int load_unsigned_byte(Register dst, Address src);
 720   int load_unsigned_short(Register dst, Address src);
 721 
 722   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 723   int load_signed_byte(Register dst, Address src);
 724   int load_signed_short(Register dst, Address src);
 725 
 726   int load_signed_byte32(Register dst, Address src);
 727   int load_signed_short32(Register dst, Address src);
 728 
 729   // Support for sign-extension (hi:lo = extend_sign(lo))
 730   void extend_sign(Register hi, Register lo);
 731 
 732   // Load and store values by size and signed-ness
 733   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 734   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 735 
 736   // Support for inc/dec with optimal instruction selection depending on value
 737 
 738   // x86_64 aliases an unqualified register/address increment and
 739   // decrement to call incrementq and decrementq but also supports
 740   // explicitly sized calls to incrementq/decrementq or
 741   // incrementl/decrementl
 742 
 743   // for aarch64 the proper convention would be to use
 744   // increment/decrement for 64 bit operations and
 745   // incrementw/decrementw for 32 bit operations. so when porting
 746   // x86_64 code we can leave calls to increment/decrement as is,
 747   // replace incrementq/decrementq with increment/decrement and
 748   // replace incrementl/decrementl with incrementw/decrementw.
 749 
 750   // n.b. increment/decrement calls with an Address destination will
 751   // need to use a scratch register to load the value to be
 752   // incremented. increment/decrement calls which add or subtract a
 753   // constant value greater than 2^12 will need to use a 2nd scratch
 754   // register to hold the constant. so, a register increment/decrement
 755   // may trash rscratch2 and an address increment/decrement trash
 756   // rscratch and rscratch2
 757 
 758   void decrementw(Address dst, int value = 1);
 759   void decrementw(Register reg, int value = 1);
 760 
 761   void decrement(Register reg, int value = 1);
 762   void decrement(Address dst, int value = 1);
 763 
 764   void incrementw(Address dst, int value = 1);
 765   void incrementw(Register reg, int value = 1);
 766 
 767   void increment(Register reg, int value = 1);
 768   void increment(Address dst, int value = 1);
 769 
 770 
 771   // Alignment
 772   void align(int modulus);
 773   void align(int modulus, int target);
 774 
 775   // nop
 776   void post_call_nop();
 777 
 778   // Stack frame creation/removal
 779   void enter(bool strip_ret_addr = false);
 780   void leave();
 781 
 782   // ROP Protection
 783   void protect_return_address();
 784   void protect_return_address(Register return_reg);
 785   void authenticate_return_address();
 786   void authenticate_return_address(Register return_reg);
 787   void strip_return_address();
 788   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 789 
 790   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 791   // The pointer will be loaded into the thread register.
 792   void get_thread(Register thread);
 793 
 794   // support for argument shuffling
 795   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 796   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 797   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 798   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 799   void object_move(
 800                    OopMap* map,
 801                    int oop_handle_offset,
 802                    int framesize_in_slots,
 803                    VMRegPair src,
 804                    VMRegPair dst,
 805                    bool is_receiver,
 806                    int* receiver_offset);
 807 
 808 
 809   // Support for VM calls
 810   //
 811   // It is imperative that all calls into the VM are handled via the call_VM macros.
 812   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 813   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 814 
 815 
 816   void call_VM(Register oop_result,
 817                address entry_point,
 818                bool check_exceptions = true);
 819   void call_VM(Register oop_result,
 820                address entry_point,
 821                Register arg_1,
 822                bool check_exceptions = true);
 823   void call_VM(Register oop_result,
 824                address entry_point,
 825                Register arg_1, Register arg_2,
 826                bool check_exceptions = true);
 827   void call_VM(Register oop_result,
 828                address entry_point,
 829                Register arg_1, Register arg_2, Register arg_3,
 830                bool check_exceptions = true);
 831 
 832   // Overloadings with last_Java_sp
 833   void call_VM(Register oop_result,
 834                Register last_java_sp,
 835                address entry_point,
 836                int number_of_arguments = 0,
 837                bool check_exceptions = true);
 838   void call_VM(Register oop_result,
 839                Register last_java_sp,
 840                address entry_point,
 841                Register arg_1, bool
 842                check_exceptions = true);
 843   void call_VM(Register oop_result,
 844                Register last_java_sp,
 845                address entry_point,
 846                Register arg_1, Register arg_2,
 847                bool check_exceptions = true);
 848   void call_VM(Register oop_result,
 849                Register last_java_sp,
 850                address entry_point,
 851                Register arg_1, Register arg_2, Register arg_3,
 852                bool check_exceptions = true);
 853 
 854   void get_vm_result  (Register oop_result, Register thread);
 855   void get_vm_result_2(Register metadata_result, Register thread);
 856 
 857   // These always tightly bind to MacroAssembler::call_VM_base
 858   // bypassing the virtual implementation
 859   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 860   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 861   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 862   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);
 863   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);
 864 
 865   void call_VM_leaf(address entry_point,
 866                     int number_of_arguments = 0);
 867   void call_VM_leaf(address entry_point,
 868                     Register arg_1);
 869   void call_VM_leaf(address entry_point,
 870                     Register arg_1, Register arg_2);
 871   void call_VM_leaf(address entry_point,
 872                     Register arg_1, Register arg_2, Register arg_3);
 873 
 874   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 875   // bypassing the virtual implementation
 876   void super_call_VM_leaf(address entry_point);
 877   void super_call_VM_leaf(address entry_point, Register arg_1);
 878   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 879   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 880   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 881 
 882   // last Java Frame (fills frame anchor)
 883   void set_last_Java_frame(Register last_java_sp,
 884                            Register last_java_fp,
 885                            address last_java_pc,
 886                            Register scratch);
 887 
 888   void set_last_Java_frame(Register last_java_sp,
 889                            Register last_java_fp,
 890                            Label &last_java_pc,
 891                            Register scratch);
 892 
 893   void set_last_Java_frame(Register last_java_sp,
 894                            Register last_java_fp,
 895                            Register last_java_pc,
 896                            Register scratch);
 897 
 898   void reset_last_Java_frame(Register thread);
 899 
 900   // thread in the default location (rthread)
 901   void reset_last_Java_frame(bool clear_fp);
 902 
 903   // Stores
 904   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 905   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 906 
 907   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 908   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 909 
 910   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 911   void c2bool(Register x);
 912 
 913   void load_method_holder_cld(Register rresult, Register rmethod);
 914   void load_method_holder(Register holder, Register method);
 915 
 916   // oop manipulations
 917   void load_metadata(Register dst, Register src);
 918 
 919   void load_narrow_klass_compact(Register dst, Register src);
 920   void load_klass(Register dst, Register src);
 921   void store_klass(Register dst, Register src);
 922   void cmp_klass(Register obj, Register klass, Register tmp);
 923   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 924 
 925   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 926   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 927   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 928 
 929   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 930                       Register tmp1, Register tmp2);
 931 
 932   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 933                        Register tmp1, Register tmp2, Register tmp3);
 934 
 935   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 936 
 937   // inline type data payload offsets...
 938   void payload_offset(Register inline_klass, Register offset);
 939   void payload_address(Register oop, Register data, Register inline_klass);
 940   // get data payload ptr a flat value array at index, kills rcx and index
 941   void data_for_value_array_index(Register array, Register array_klass,
 942                                   Register index, Register data);
 943 
 944   void load_heap_oop(Register dst, Address src, Register tmp1,
 945                      Register tmp2, DecoratorSet decorators = 0);
 946 
 947   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 948                               Register tmp2, DecoratorSet decorators = 0);
 949   void store_heap_oop(Address dst, Register val, Register tmp1,
 950                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 951 
 952   // currently unimplemented
 953   // Used for storing null. All other oop constants should be
 954   // stored using routines that take a jobject.
 955   void store_heap_oop_null(Address dst);
 956 
 957   void load_prototype_header(Register dst, Register src);
 958 
 959   void store_klass_gap(Register dst, Register src);
 960 
 961   // This dummy is to prevent a call to store_heap_oop from
 962   // converting a zero (like null) into a Register by giving
 963   // the compiler two choices it can't resolve
 964 
 965   void store_heap_oop(Address dst, void* dummy);
 966 
 967   void encode_heap_oop(Register d, Register s);
 968   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 969   void decode_heap_oop(Register d, Register s);
 970   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 971   void encode_heap_oop_not_null(Register r);
 972   void decode_heap_oop_not_null(Register r);
 973   void encode_heap_oop_not_null(Register dst, Register src);
 974   void decode_heap_oop_not_null(Register dst, Register src);
 975 
 976   void set_narrow_oop(Register dst, jobject obj);
 977 
 978   void encode_klass_not_null(Register r);
 979   void decode_klass_not_null(Register r);
 980   void encode_klass_not_null(Register dst, Register src);
 981   void decode_klass_not_null(Register dst, Register src);
 982 
 983   void set_narrow_klass(Register dst, Klass* k);
 984 
 985   // if heap base register is used - reinit it with the correct value
 986   void reinit_heapbase();
 987 
 988   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 989 
 990   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
 991                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 992   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
 993                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 994 
 995   void push_cont_fastpath(Register java_thread = rthread);
 996   void pop_cont_fastpath(Register java_thread = rthread);
 997 
 998   void inc_held_monitor_count(Register tmp);
 999   void dec_held_monitor_count(Register tmp);
1000 
1001   // Round up to a power of two
1002   void round_to(Register reg, int modulus);
1003 
1004   // java.lang.Math::round intrinsics
1005   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1006   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1007 
1008   // allocation
1009 
1010   // Object / value buffer allocation...
1011   // Allocate instance of klass, assumes klass initialized by caller
1012   // new_obj prefers to be rax
1013   // Kills t1 and t2, perserves klass, return allocation in new_obj (rsi on LP64)
1014   void allocate_instance(Register klass, Register new_obj,
1015                          Register t1, Register t2,
1016                          bool clear_fields, Label& alloc_failed);
1017 
1018   void tlab_allocate(
1019     Register obj,                      // result: pointer to object after successful allocation
1020     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1021     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1022     Register t1,                       // temp register
1023     Register t2,                       // temp register
1024     Label&   slow_case                 // continuation point if fast allocation fails
1025   );
1026   void verify_tlab();
1027 
1028   // For field "index" within "klass", return inline_klass ...
1029   void get_inline_type_field_klass(Register klass, Register index, Register inline_klass);
1030   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1031 
1032 
1033   // interface method calling
1034   void lookup_interface_method(Register recv_klass,
1035                                Register intf_klass,
1036                                RegisterOrConstant itable_index,
1037                                Register method_result,
1038                                Register scan_temp,
1039                                Label& no_such_interface,
1040                    bool return_method = true);
1041 
1042   void lookup_interface_method_stub(Register recv_klass,
1043                                     Register holder_klass,
1044                                     Register resolved_klass,
1045                                     Register method_result,
1046                                     Register temp_reg,
1047                                     Register temp_reg2,
1048                                     int itable_index,
1049                                     Label& L_no_such_interface);
1050 
1051   // virtual method calling
1052   // n.b. x86 allows RegisterOrConstant for vtable_index
1053   void lookup_virtual_method(Register recv_klass,
1054                              RegisterOrConstant vtable_index,
1055                              Register method_result);
1056 
1057   // Test sub_klass against super_klass, with fast and slow paths.
1058 
1059   // The fast path produces a tri-state answer: yes / no / maybe-slow.
1060   // One of the three labels can be null, meaning take the fall-through.
1061   // If super_check_offset is -1, the value is loaded up from super_klass.
1062   // No registers are killed, except temp_reg.
1063   void check_klass_subtype_fast_path(Register sub_klass,
1064                                      Register super_klass,
1065                                      Register temp_reg,
1066                                      Label* L_success,
1067                                      Label* L_failure,
1068                                      Label* L_slow_path,
1069                                      Register super_check_offset = noreg);
1070 
1071   // The rest of the type check; must be wired to a corresponding fast path.
1072   // It does not repeat the fast path logic, so don't use it standalone.
1073   // The temp_reg and temp2_reg can be noreg, if no temps are available.
1074   // Updates the sub's secondary super cache as necessary.
1075   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
1076   void check_klass_subtype_slow_path(Register sub_klass,
1077                                      Register super_klass,
1078                                      Register temp_reg,
1079                                      Register temp2_reg,
1080                                      Label* L_success,
1081                                      Label* L_failure,
1082                                      bool set_cond_codes = false);
1083 
1084   void check_klass_subtype_slow_path_linear(Register sub_klass,
1085                                             Register super_klass,
1086                                             Register temp_reg,
1087                                             Register temp2_reg,
1088                                             Label* L_success,
1089                                             Label* L_failure,
1090                                             bool set_cond_codes = false);
1091 
1092   void check_klass_subtype_slow_path_table(Register sub_klass,
1093                                            Register super_klass,
1094                                            Register temp_reg,
1095                                            Register temp2_reg,
1096                                            Register temp3_reg,
1097                                            Register result_reg,
1098                                            FloatRegister vtemp_reg,
1099                                            Label* L_success,
1100                                            Label* L_failure,
1101                                            bool set_cond_codes = false);
1102 
1103   // If r is valid, return r.
1104   // If r is invalid, remove a register r2 from available_regs, add r2
1105   // to regs_to_push, then return r2.
1106   Register allocate_if_noreg(const Register r,
1107                              RegSetIterator<Register> &available_regs,
1108                              RegSet &regs_to_push);
1109 
1110   // Secondary subtype checking
1111   void lookup_secondary_supers_table_var(Register sub_klass,
1112                                          Register r_super_klass,
1113                                          Register temp1,
1114                                          Register temp2,
1115                                          Register temp3,
1116                                          FloatRegister vtemp,
1117                                          Register result,
1118                                          Label *L_success);
1119 
1120 
1121   // As above, but with a constant super_klass.
1122   // The result is in Register result, not the condition codes.
1123   bool lookup_secondary_supers_table_const(Register r_sub_klass,
1124                                            Register r_super_klass,
1125                                            Register temp1,
1126                                            Register temp2,
1127                                            Register temp3,
1128                                            FloatRegister vtemp,
1129                                            Register result,
1130                                            u1 super_klass_slot,
1131                                            bool stub_is_near = false);
1132 
1133   void verify_secondary_supers_table(Register r_sub_klass,
1134                                      Register r_super_klass,
1135                                      Register temp1,
1136                                      Register temp2,
1137                                      Register result);
1138 
1139   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
1140                                                Register r_array_base,
1141                                                Register r_array_index,
1142                                                Register r_bitmap,
1143                                                Register temp1,
1144                                                Register result,
1145                                                bool is_stub = true);
1146 
1147   // Simplified, combined version, good for typical uses.
1148   // Falls through on failure.
1149   void check_klass_subtype(Register sub_klass,
1150                            Register super_klass,
1151                            Register temp_reg,
1152                            Label& L_success);
1153 
1154   void clinit_barrier(Register klass,
1155                       Register thread,
1156                       Label* L_fast_path = nullptr,
1157                       Label* L_slow_path = nullptr);
1158 
1159   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1160 
1161   void verify_sve_vector_length(Register tmp = rscratch1);
1162   void reinitialize_ptrue() {
1163     if (UseSVE > 0) {
1164       sve_ptrue(ptrue, B);
1165     }
1166   }
1167   void verify_ptrue();
1168 
1169   // Debugging
1170 
1171   // only if +VerifyOops
1172   void _verify_oop(Register reg, const char* s, const char* file, int line);
1173   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1174 
1175   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1176     if (VerifyOops) {
1177       _verify_oop(reg, s, file, line);
1178     }
1179   }
1180   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1181     if (VerifyOops) {
1182       _verify_oop_addr(reg, s, file, line);
1183     }
1184   }
1185 
1186 // TODO: verify method and klass metadata (compare against vptr?)
1187   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1188   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1189 
1190 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1191 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1192 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1193 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1194 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1195 
1196   // Restore cpu control state after JNI call
1197   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1198 
1199   // prints msg, dumps registers and stops execution
1200   void stop(const char* msg);
1201 
1202   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1203 
1204   void untested()                                { stop("untested"); }
1205 
1206   void unimplemented(const char* what = "");
1207 
1208   void should_not_reach_here()                   { stop("should not reach here"); }
1209 
1210   void _assert_asm(Condition cc, const char* msg);
1211 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1212 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1213 
1214   // Stack overflow checking
1215   void bang_stack_with_offset(int offset) {
1216     // stack grows down, caller passes positive offset
1217     assert(offset > 0, "must bang with negative offset");
1218     sub(rscratch2, sp, offset);
1219     str(zr, Address(rscratch2));
1220   }
1221 
1222   // Writes to stack successive pages until offset reached to check for
1223   // stack overflow + shadow pages.  Also, clobbers tmp
1224   void bang_stack_size(Register size, Register tmp);
1225 
1226   // Check for reserved stack access in method being exited (for JIT)
1227   void reserved_stack_check();
1228 
1229   // Arithmetics
1230 
1231   // Clobber: rscratch1, rscratch2
1232   void addptr(const Address &dst, int32_t src);
1233 
1234   // Clobber: rscratch1
1235   void cmpptr(Register src1, Address src2);
1236 
1237   void cmpoop(Register obj1, Register obj2);
1238 
1239   // Various forms of CAS
1240 
1241   void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
1242                           Label &succeed, Label *fail);
1243   void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
1244                   Label &succeed, Label *fail);
1245 
1246   void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
1247                   Label &succeed, Label *fail);
1248 
1249   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1250   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1251   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1252   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1253 
1254   void atomic_xchg(Register prev, Register newv, Register addr);
1255   void atomic_xchgw(Register prev, Register newv, Register addr);
1256   void atomic_xchgl(Register prev, Register newv, Register addr);
1257   void atomic_xchglw(Register prev, Register newv, Register addr);
1258   void atomic_xchgal(Register prev, Register newv, Register addr);
1259   void atomic_xchgalw(Register prev, Register newv, Register addr);
1260 
1261   void orptr(Address adr, RegisterOrConstant src) {
1262     ldr(rscratch1, adr);
1263     if (src.is_register())
1264       orr(rscratch1, rscratch1, src.as_register());
1265     else
1266       orr(rscratch1, rscratch1, src.as_constant());
1267     str(rscratch1, adr);
1268   }
1269 
1270   // A generic CAS; success or failure is in the EQ flag.
1271   // Clobbers rscratch1
1272   void cmpxchg(Register addr, Register expected, Register new_val,
1273                enum operand_size size,
1274                bool acquire, bool release, bool weak,
1275                Register result);
1276 
1277 #ifdef ASSERT
1278   // Template short-hand support to clean-up after a failed call to trampoline
1279   // call generation (see trampoline_call() below),  when a set of Labels must
1280   // be reset (before returning).
1281   template<typename Label, typename... More>
1282   void reset_labels(Label &lbl, More&... more) {
1283     lbl.reset(); reset_labels(more...);
1284   }
1285   template<typename Label>
1286   void reset_labels(Label &lbl) {
1287     lbl.reset();
1288   }
1289 #endif
1290 
1291 private:
1292   void compare_eq(Register rn, Register rm, enum operand_size size);
1293 
1294 public:
1295   // AArch64 OpenJDK uses four different types of calls:
1296   //   - direct call: bl pc_relative_offset
1297   //     This is the shortest and the fastest, but the offset has the range:
1298   //     +/-128MB for the release build, +/-2MB for the debug build.
1299   //
1300   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1301   //     This is longer than a direct call. The offset has
1302   //     the range +/-4GB. As the code cache size is limited to 4GB,
1303   //     far calls can reach anywhere in the code cache. If a jump is
1304   //     needed rather than a call, a far jump 'b reg' can be used instead.
1305   //     All instructions are embedded at a call site.
1306   //
1307   //   - trampoline call:
1308   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1309   //     of a direct call, which is used if the destination of a call is in range,
1310   //     and a register-indirect call. It has the advantages of reaching anywhere in
1311   //     the AArch64 address space and being patchable at runtime when the generated
1312   //     code is being executed by other threads.
1313   //
1314   //     [Main code section]
1315   //       bl trampoline
1316   //     [Stub code section]
1317   //     trampoline:
1318   //       ldr reg, pc + 8
1319   //       br reg
1320   //       <64-bit destination address>
1321   //
1322   //     If the destination is in range when the generated code is moved to the code
1323   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1324   //     is not used.
1325   //     The optimization does not remove the trampoline from the stub section.
1326   //     This is necessary because the trampoline may well be redirected later when
1327   //     code is patched, and the new destination may not be reachable by a simple BR
1328   //     instruction.
1329   //
1330   //   - indirect call: move reg, address; blr reg
1331   //     This too can reach anywhere in the address space, but it cannot be
1332   //     patched while code is running, so it must only be modified at a safepoint.
1333   //     This form of call is most suitable for targets at fixed addresses, which
1334   //     will never be patched.
1335   //
1336   // The patching we do conforms to the "Concurrent modification and
1337   // execution of instructions" section of the Arm Architectural
1338   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1339   // or SVC instructions to be modified while another thread is
1340   // executing them.
1341   //
1342   // To patch a trampoline call when the BL can't reach, we first modify
1343   // the 64-bit destination address in the trampoline, then modify the
1344   // BL to point to the trampoline, then flush the instruction cache to
1345   // broadcast the change to all executing threads. See
1346   // NativeCall::set_destination_mt_safe for the details.
1347   //
1348   // There is a benign race in that the other thread might observe the
1349   // modified BL before it observes the modified 64-bit destination
1350   // address. That does not matter because the destination method has been
1351   // invalidated, so there will be a trap at its start.
1352   // For this to work, the destination address in the trampoline is
1353   // always updated, even if we're not using the trampoline.
1354 
1355   // Emit a direct call if the entry address will always be in range,
1356   // otherwise a trampoline call.
1357   // Supported entry.rspec():
1358   // - relocInfo::runtime_call_type
1359   // - relocInfo::opt_virtual_call_type
1360   // - relocInfo::static_call_type
1361   // - relocInfo::virtual_call_type
1362   //
1363   // Return: the call PC or null if CodeCache is full.
1364   // Clobbers: rscratch1
1365   address trampoline_call(Address entry);
1366 
1367   static bool far_branches() {
1368     return ReservedCodeCacheSize > branch_range;
1369   }
1370 
1371   // Check if branches to the non nmethod section require a far jump
1372   static bool codestub_branch_needs_far_jump() {
1373     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1374   }
1375 
1376   // Emit a direct call/jump if the entry address will always be in range,
1377   // otherwise a far call/jump.
1378   // The address must be inside the code cache.
1379   // Supported entry.rspec():
1380   // - relocInfo::external_word_type
1381   // - relocInfo::runtime_call_type
1382   // - relocInfo::none
1383   // In the case of a far call/jump, the entry address is put in the tmp register.
1384   // The tmp register is invalidated.
1385   //
1386   // Far_jump returns the amount of the emitted code.
1387   void far_call(Address entry, Register tmp = rscratch1);
1388   int far_jump(Address entry, Register tmp = rscratch1);
1389 
1390   static int far_codestub_branch_size() {
1391     if (codestub_branch_needs_far_jump()) {
1392       return 3 * 4;  // adrp, add, br
1393     } else {
1394       return 4;
1395     }
1396   }
1397 
1398   // Emit the CompiledIC call idiom
1399   address ic_call(address entry, jint method_index = 0);
1400   static int ic_check_size();
1401   int ic_check(int end_alignment);
1402 
1403 public:
1404 
1405   // Data
1406 
1407   void mov_metadata(Register dst, Metadata* obj);
1408   Address allocate_metadata_address(Metadata* obj);
1409   Address constant_oop_address(jobject obj);
1410 
1411   void movoop(Register dst, jobject obj);
1412 
1413   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1414   void kernel_crc32(Register crc, Register buf, Register len,
1415         Register table0, Register table1, Register table2, Register table3,
1416         Register tmp, Register tmp2, Register tmp3);
1417   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1418   void kernel_crc32c(Register crc, Register buf, Register len,
1419         Register table0, Register table1, Register table2, Register table3,
1420         Register tmp, Register tmp2, Register tmp3);
1421 
1422   // Stack push and pop individual 64 bit registers
1423   void push(Register src);
1424   void pop(Register dst);
1425 
1426   void repne_scan(Register addr, Register value, Register count,
1427                   Register scratch);
1428   void repne_scanw(Register addr, Register value, Register count,
1429                    Register scratch);
1430 
1431   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1432   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1433 
1434   // If a constant does not fit in an immediate field, generate some
1435   // number of MOV instructions and then perform the operation
1436   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1437                              add_sub_imm_insn insn1,
1438                              add_sub_reg_insn insn2, bool is32);
1439   // Separate vsn which sets the flags
1440   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1441                                add_sub_imm_insn insn1,
1442                                add_sub_reg_insn insn2, bool is32);
1443 
1444 #define WRAP(INSN, is32)                                                \
1445   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1446     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1447   }                                                                     \
1448                                                                         \
1449   void INSN(Register Rd, Register Rn, Register Rm,                      \
1450              enum shift_kind kind, unsigned shift = 0) {                \
1451     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1452   }                                                                     \
1453                                                                         \
1454   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1455     Assembler::INSN(Rd, Rn, Rm);                                        \
1456   }                                                                     \
1457                                                                         \
1458   void INSN(Register Rd, Register Rn, Register Rm,                      \
1459            ext::operation option, int amount = 0) {                     \
1460     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1461   }
1462 
1463   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1464 
1465 #undef WRAP
1466 #define WRAP(INSN, is32)                                                \
1467   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1468     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1469   }                                                                     \
1470                                                                         \
1471   void INSN(Register Rd, Register Rn, Register Rm,                      \
1472              enum shift_kind kind, unsigned shift = 0) {                \
1473     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1474   }                                                                     \
1475                                                                         \
1476   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1477     Assembler::INSN(Rd, Rn, Rm);                                        \
1478   }                                                                     \
1479                                                                         \
1480   void INSN(Register Rd, Register Rn, Register Rm,                      \
1481            ext::operation option, int amount = 0) {                     \
1482     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1483   }
1484 
1485   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1486 
1487   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1488   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1489   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1490   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1491 
1492   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1493 
1494   void verified_entry(Compile* C, int sp_inc);
1495 
1496   // Inline type specific methods
1497   #include "asm/macroAssembler_common.hpp"
1498 
1499   int store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter = true);
1500   bool move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]);
1501   bool unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
1502                             VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
1503                             RegState reg_state[]);
1504   bool pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
1505                           VMRegPair* from, int from_count, int& from_index, VMReg to,
1506                           RegState reg_state[], Register val_array);
1507   int extend_stack_for_inline_args(int args_on_stack);
1508   void remove_frame(int initial_framesize, bool needs_stack_repair);
1509   VMReg spill_reg_for(VMReg reg);
1510   void save_stack_increment(int sp_inc, int frame_size);
1511 
1512   void tableswitch(Register index, jint lowbound, jint highbound,
1513                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1514     adr(rscratch1, jumptable);
1515     subsw(rscratch2, index, lowbound);
1516     subsw(zr, rscratch2, highbound - lowbound);
1517     br(Assembler::HS, jumptable_end);
1518     add(rscratch1, rscratch1, rscratch2,
1519         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1520     br(rscratch1);
1521   }
1522 
1523   // Form an address from base + offset in Rd.  Rd may or may not
1524   // actually be used: you must use the Address that is returned.  It
1525   // is up to you to ensure that the shift provided matches the size
1526   // of your data.
1527   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1528 
1529   // Return true iff an address is within the 48-bit AArch64 address
1530   // space.
1531   bool is_valid_AArch64_address(address a) {
1532     return ((uint64_t)a >> 48) == 0;
1533   }
1534 
1535   // Load the base of the cardtable byte map into reg.
1536   void load_byte_map_base(Register reg);
1537 
1538   // Prolog generator routines to support switch between x86 code and
1539   // generated ARM code
1540 
1541   // routine to generate an x86 prolog for a stub function which
1542   // bootstraps into the generated ARM code which directly follows the
1543   // stub
1544   //
1545 
1546   public:
1547 
1548   void ldr_constant(Register dest, const Address &const_addr) {
1549     if (NearCpool) {
1550       ldr(dest, const_addr);
1551     } else {
1552       uint64_t offset;
1553       adrp(dest, InternalAddress(const_addr.target()), offset);
1554       ldr(dest, Address(dest, offset));
1555     }
1556   }
1557 
1558   address read_polling_page(Register r, relocInfo::relocType rtype);
1559   void get_polling_page(Register dest, relocInfo::relocType rtype);
1560 
1561   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1562   void update_byte_crc32(Register crc, Register val, Register table);
1563   void update_word_crc32(Register crc, Register v, Register tmp,
1564         Register table0, Register table1, Register table2, Register table3,
1565         bool upper = false);
1566 
1567   address count_positives(Register ary1, Register len, Register result);
1568 
1569   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1570                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1571 
1572 // Ensure that the inline code and the stub use the same registers.
1573 #define ARRAYS_HASHCODE_REGISTERS \
1574   do {                      \
1575     assert(result == r0  && \
1576            ary    == r1  && \
1577            cnt    == r2  && \
1578            vdata0 == v3  && \
1579            vdata1 == v2  && \
1580            vdata2 == v1  && \
1581            vdata3 == v0  && \
1582            vmul0  == v4  && \
1583            vmul1  == v5  && \
1584            vmul2  == v6  && \
1585            vmul3  == v7  && \
1586            vpow   == v12 && \
1587            vpowm  == v13, "registers must match aarch64.ad"); \
1588   } while (0)
1589 
1590   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1591 
1592   void fill_words(Register base, Register cnt, Register value);
1593   void fill_words(Register base, uint64_t cnt, Register value);
1594 
1595   address zero_words(Register base, uint64_t cnt);
1596   address zero_words(Register ptr, Register cnt);
1597   void zero_dcache_blocks(Register base, Register cnt);
1598 
1599   static const int zero_words_block_size;
1600 
1601   address byte_array_inflate(Register src, Register dst, Register len,
1602                              FloatRegister vtmp1, FloatRegister vtmp2,
1603                              FloatRegister vtmp3, Register tmp4);
1604 
1605   void char_array_compress(Register src, Register dst, Register len,
1606                            Register res,
1607                            FloatRegister vtmp0, FloatRegister vtmp1,
1608                            FloatRegister vtmp2, FloatRegister vtmp3,
1609                            FloatRegister vtmp4, FloatRegister vtmp5);
1610 
1611   void encode_iso_array(Register src, Register dst,
1612                         Register len, Register res, bool ascii,
1613                         FloatRegister vtmp0, FloatRegister vtmp1,
1614                         FloatRegister vtmp2, FloatRegister vtmp3,
1615                         FloatRegister vtmp4, FloatRegister vtmp5);
1616 
1617   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1618       address pio2, address dsin_coef, address dcos_coef);
1619  private:
1620   // begin trigonometric functions support block
1621   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1622   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1623   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1624   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1625   // end trigonometric functions support block
1626   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1627                        Register src1, Register src2);
1628   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1629     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1630   }
1631   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1632                              Register y, Register y_idx, Register z,
1633                              Register carry, Register product,
1634                              Register idx, Register kdx);
1635   void multiply_128_x_128_loop(Register y, Register z,
1636                                Register carry, Register carry2,
1637                                Register idx, Register jdx,
1638                                Register yz_idx1, Register yz_idx2,
1639                                Register tmp, Register tmp3, Register tmp4,
1640                                Register tmp7, Register product_hi);
1641   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1642         Register len, Register tmp0, Register tmp1, Register tmp2,
1643         Register tmp3);
1644   void kernel_crc32_using_crc32(Register crc, Register buf,
1645         Register len, Register tmp0, Register tmp1, Register tmp2,
1646         Register tmp3);
1647   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1648         Register len, Register tmp0, Register tmp1, Register tmp2,
1649         Register tmp3);
1650   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1651         Register len, Register tmp0, Register tmp1, Register tmp2,
1652         Register tmp3);
1653   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1654         Register len, Register tmp0, Register tmp1, Register tmp2,
1655         size_t table_offset);
1656 
1657   void ghash_modmul (FloatRegister result,
1658                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1659                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1660                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1661   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1662 public:
1663   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1664                        Register tmp0, Register tmp1, Register tmp2, Register tmp3,
1665                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1666   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1667   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1668                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1669                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1670   void ghash_multiply_wide(int index,
1671                            FloatRegister result_lo, FloatRegister result_hi,
1672                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1673                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1674   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1675                     FloatRegister p, FloatRegister z, FloatRegister t1);
1676   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1677                     FloatRegister p, FloatRegister z, FloatRegister t1);
1678   void ghash_processBlocks_wide(address p, Register state, Register subkeyH,
1679                                 Register data, Register blocks, int unrolls);
1680 
1681 
1682   void aesenc_loadkeys(Register key, Register keylen);
1683   void aesecb_encrypt(Register from, Register to, Register keylen,
1684                       FloatRegister data = v0, int unrolls = 1);
1685   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1686   void aes_round(FloatRegister input, FloatRegister subkey);
1687 
1688   // ChaCha20 functions support block
1689   void cc20_quarter_round(FloatRegister aVec, FloatRegister bVec,
1690           FloatRegister cVec, FloatRegister dVec, FloatRegister scratch,
1691           FloatRegister tbl);
1692   void cc20_shift_lane_org(FloatRegister bVec, FloatRegister cVec,
1693           FloatRegister dVec, bool colToDiag);
1694 
1695   // Place an ISB after code may have been modified due to a safepoint.
1696   void safepoint_isb();
1697 
1698 private:
1699   // Return the effective address r + (r1 << ext) + offset.
1700   // Uses rscratch2.
1701   Address offsetted_address(Register r, Register r1, Address::extend ext,
1702                             int offset, int size);
1703 
1704 private:
1705   // Returns an address on the stack which is reachable with a ldr/str of size
1706   // Uses rscratch2 if the address is not directly reachable
1707   Address spill_address(int size, int offset, Register tmp=rscratch2);
1708   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1709 
1710   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1711 
1712   // Check whether two loads/stores can be merged into ldp/stp.
1713   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1714 
1715   // Merge current load/store with previous load/store into ldp/stp.
1716   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1717 
1718   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1719   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1720 
1721 public:
1722   void spill(Register Rx, bool is64, int offset) {
1723     if (is64) {
1724       str(Rx, spill_address(8, offset));
1725     } else {
1726       strw(Rx, spill_address(4, offset));
1727     }
1728   }
1729   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1730     str(Vx, T, spill_address(1 << (int)T, offset));
1731   }
1732 
1733   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1734     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1735   }
1736   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1737     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1738   }
1739 
1740   void unspill(Register Rx, bool is64, int offset) {
1741     if (is64) {
1742       ldr(Rx, spill_address(8, offset));
1743     } else {
1744       ldrw(Rx, spill_address(4, offset));
1745     }
1746   }
1747   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1748     ldr(Vx, T, spill_address(1 << (int)T, offset));
1749   }
1750 
1751   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1752     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1753   }
1754   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1755     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1756   }
1757 
1758   void spill_copy128(int src_offset, int dst_offset,
1759                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1760     if (src_offset < 512 && (src_offset & 7) == 0 &&
1761         dst_offset < 512 && (dst_offset & 7) == 0) {
1762       ldp(tmp1, tmp2, Address(sp, src_offset));
1763       stp(tmp1, tmp2, Address(sp, dst_offset));
1764     } else {
1765       unspill(tmp1, true, src_offset);
1766       spill(tmp1, true, dst_offset);
1767       unspill(tmp1, true, src_offset+8);
1768       spill(tmp1, true, dst_offset+8);
1769     }
1770   }
1771   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1772                                             int sve_vec_reg_size_in_bytes) {
1773     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1774     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1775       spill_copy128(src_offset, dst_offset);
1776       src_offset += 16;
1777       dst_offset += 16;
1778     }
1779   }
1780   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1781                                                int sve_predicate_reg_size_in_bytes) {
1782     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1783     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1784     reinitialize_ptrue();
1785   }
1786   void cache_wb(Address line);
1787   void cache_wbsync(bool is_pre);
1788 
1789   // Code for java.lang.Thread::onSpinWait() intrinsic.
1790   void spin_wait();
1791 
1792   void lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow);
1793   void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1794 
1795 private:
1796   // Check the current thread doesn't need a cross modify fence.
1797   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1798 
1799 };
1800 
1801 #ifdef ASSERT
1802 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1803 #endif
1804 
1805 struct tableswitch {
1806   Register _reg;
1807   int _insn_index; jint _first_key; jint _last_key;
1808   Label _after;
1809   Label _branches;
1810 };
1811 
1812 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP