1 /*
   2  * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2021, 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/powerOfTwo.hpp"
  36 
  37 class OopMap;
  38 
  39 // MacroAssembler extends Assembler by frequently used macros.
  40 //
  41 // Instructions for which a 'better' code sequence exists depending
  42 // on arguments should also go in here.
  43 
  44 class MacroAssembler: public Assembler {
  45   friend class LIR_Assembler;
  46 
  47  public:
  48   using Assembler::mov;
  49   using Assembler::movi;
  50 
  51  protected:
  52 
  53   // Support for VM calls
  54   //
  55   // This is the base routine called by the different versions of call_VM_leaf. The interpreter
  56   // may customize this version by overriding it for its purposes (e.g., to save/restore
  57   // additional registers when doing a VM call).
  58   virtual void call_VM_leaf_base(
  59     address entry_point,               // the entry point
  60     int     number_of_arguments,        // the number of arguments to pop after the call
  61     Label *retaddr = nullptr
  62   );
  63 
  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) {
  68     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
  69   }
  70 
  71   // This is the base routine called by the different versions of call_VM. The interpreter
  72   // may customize this version by overriding it for its purposes (e.g., to save/restore
  73   // additional registers when doing a VM call).
  74   //
  75   // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base
  76   // returns the register which contains the thread upon return. If a thread register has been
  77   // specified, the return value will correspond to that register. If no last_java_sp is specified
  78   // (noreg) than rsp will be used instead.
  79   virtual void call_VM_base(           // returns the register containing the thread upon return
  80     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
  81     Register java_thread,              // the thread if computed before     ; use noreg otherwise
  82     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
  83     address  entry_point,              // the entry point
  84     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
  85     bool     check_exceptions          // whether to check for pending exceptions after return
  86   );
  87 
  88   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
  89 
  90   enum KlassDecodeMode {
  91     KlassDecodeNone,
  92     KlassDecodeZero,
  93     KlassDecodeXor,
  94     KlassDecodeMovk
  95   };
  96 
  97   KlassDecodeMode klass_decode_mode();
  98 
  99  private:
 100   static KlassDecodeMode _klass_decode_mode;
 101 
 102  public:
 103   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 104 
 105  // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 106  // The implementation is only non-empty for the InterpreterMacroAssembler,
 107  // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 108  virtual void check_and_handle_popframe(Register java_thread);
 109  virtual void check_and_handle_earlyret(Register java_thread);
 110 
 111   void safepoint_poll(Label& slow_path, bool at_return, bool acquire, bool in_nmethod, Register tmp = rscratch1);
 112   void rt_call(address dest, Register tmp = rscratch1);
 113 
 114   // Load Effective Address
 115   void lea(Register r, const Address &a) {
 116     InstructionMark im(this);
 117     a.lea(this, r);
 118   }
 119 
 120   /* Sometimes we get misaligned loads and stores, usually from Unsafe
 121      accesses, and these can exceed the offset range. */
 122   Address legitimize_address(const Address &a, int size, Register scratch) {
 123     if (a.getMode() == Address::base_plus_offset) {
 124       if (! Address::offset_ok_for_immed(a.offset(), exact_log2(size))) {
 125         block_comment("legitimize_address {");
 126         lea(scratch, a);
 127         block_comment("} legitimize_address");
 128         return Address(scratch);
 129       }
 130     }
 131     return a;
 132   }
 133 
 134   void addmw(Address a, Register incr, Register scratch) {
 135     ldrw(scratch, a);
 136     addw(scratch, scratch, incr);
 137     strw(scratch, a);
 138   }
 139 
 140   // Add constant to memory word
 141   void addmw(Address a, int imm, Register scratch) {
 142     ldrw(scratch, a);
 143     if (imm > 0)
 144       addw(scratch, scratch, (unsigned)imm);
 145     else
 146       subw(scratch, scratch, (unsigned)-imm);
 147     strw(scratch, a);
 148   }
 149 
 150   void bind(Label& L) {
 151     Assembler::bind(L);
 152     code()->clear_last_insn();
 153   }
 154 
 155   void membar(Membar_mask_bits order_constraint);
 156 
 157   using Assembler::ldr;
 158   using Assembler::str;
 159   using Assembler::ldrw;
 160   using Assembler::strw;
 161 
 162   void ldr(Register Rx, const Address &adr);
 163   void ldrw(Register Rw, const Address &adr);
 164   void str(Register Rx, const Address &adr);
 165   void strw(Register Rx, const Address &adr);
 166 
 167   // Frame creation and destruction shared between JITs.
 168   void build_frame(int framesize);
 169   void remove_frame(int framesize);
 170 
 171   virtual void _call_Unimplemented(address call_site) {
 172     mov(rscratch2, call_site);
 173   }
 174 
 175 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 176 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 177 // https://reviews.llvm.org/D3311
 178 
 179 #ifdef _WIN64
 180 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 181 #else
 182 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 183 #endif
 184 
 185   // aliases defined in AARCH64 spec
 186 
 187   template<class T>
 188   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
 189 
 190   inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
 191   inline void cmp(Register Rd, unsigned imm) = delete;
 192 
 193   template<class T>
 194   inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); }
 195 
 196   inline void cmn(Register Rd, unsigned char imm8)  { adds(zr, Rd, imm8); }
 197   inline void cmn(Register Rd, unsigned imm) = delete;
 198 
 199   void cset(Register Rd, Assembler::Condition cond) {
 200     csinc(Rd, zr, zr, ~cond);
 201   }
 202   void csetw(Register Rd, Assembler::Condition cond) {
 203     csincw(Rd, zr, zr, ~cond);
 204   }
 205 
 206   void cneg(Register Rd, Register Rn, Assembler::Condition cond) {
 207     csneg(Rd, Rn, Rn, ~cond);
 208   }
 209   void cnegw(Register Rd, Register Rn, Assembler::Condition cond) {
 210     csnegw(Rd, Rn, Rn, ~cond);
 211   }
 212 
 213   inline void movw(Register Rd, Register Rn) {
 214     if (Rd == sp || Rn == sp) {
 215       Assembler::addw(Rd, Rn, 0U);
 216     } else {
 217       orrw(Rd, zr, Rn);
 218     }
 219   }
 220   inline void mov(Register Rd, Register Rn) {
 221     assert(Rd != r31_sp && Rn != r31_sp, "should be");
 222     if (Rd == Rn) {
 223     } else if (Rd == sp || Rn == sp) {
 224       Assembler::add(Rd, Rn, 0U);
 225     } else {
 226       orr(Rd, zr, Rn);
 227     }
 228   }
 229 
 230   inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); }
 231   inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); }
 232 
 233   inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); }
 234   inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); }
 235 
 236   inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); }
 237   inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); }
 238 
 239   inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 240     bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 241   }
 242   inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 243     bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 244   }
 245 
 246   inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 247     bfmw(Rd, Rn, lsb, (lsb + width - 1));
 248   }
 249   inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 250     bfm(Rd, Rn, lsb , (lsb + width - 1));
 251   }
 252 
 253   inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 254     sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 255   }
 256   inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 257     sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 258   }
 259 
 260   inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 261     sbfmw(Rd, Rn, lsb, (lsb + width - 1));
 262   }
 263   inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 264     sbfm(Rd, Rn, lsb , (lsb + width - 1));
 265   }
 266 
 267   inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 268     ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1));
 269   }
 270   inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 271     ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1));
 272   }
 273 
 274   inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 275     ubfmw(Rd, Rn, lsb, (lsb + width - 1));
 276   }
 277   inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) {
 278     ubfm(Rd, Rn, lsb , (lsb + width - 1));
 279   }
 280 
 281   inline void asrw(Register Rd, Register Rn, unsigned imm) {
 282     sbfmw(Rd, Rn, imm, 31);
 283   }
 284 
 285   inline void asr(Register Rd, Register Rn, unsigned imm) {
 286     sbfm(Rd, Rn, imm, 63);
 287   }
 288 
 289   inline void lslw(Register Rd, Register Rn, unsigned imm) {
 290     ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm));
 291   }
 292 
 293   inline void lsl(Register Rd, Register Rn, unsigned imm) {
 294     ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm));
 295   }
 296 
 297   inline void lsrw(Register Rd, Register Rn, unsigned imm) {
 298     ubfmw(Rd, Rn, imm, 31);
 299   }
 300 
 301   inline void lsr(Register Rd, Register Rn, unsigned imm) {
 302     ubfm(Rd, Rn, imm, 63);
 303   }
 304 
 305   inline void rorw(Register Rd, Register Rn, unsigned imm) {
 306     extrw(Rd, Rn, Rn, imm);
 307   }
 308 
 309   inline void ror(Register Rd, Register Rn, unsigned imm) {
 310     extr(Rd, Rn, Rn, imm);
 311   }
 312 
 313   inline void sxtbw(Register Rd, Register Rn) {
 314     sbfmw(Rd, Rn, 0, 7);
 315   }
 316   inline void sxthw(Register Rd, Register Rn) {
 317     sbfmw(Rd, Rn, 0, 15);
 318   }
 319   inline void sxtb(Register Rd, Register Rn) {
 320     sbfm(Rd, Rn, 0, 7);
 321   }
 322   inline void sxth(Register Rd, Register Rn) {
 323     sbfm(Rd, Rn, 0, 15);
 324   }
 325   inline void sxtw(Register Rd, Register Rn) {
 326     sbfm(Rd, Rn, 0, 31);
 327   }
 328 
 329   inline void uxtbw(Register Rd, Register Rn) {
 330     ubfmw(Rd, Rn, 0, 7);
 331   }
 332   inline void uxthw(Register Rd, Register Rn) {
 333     ubfmw(Rd, Rn, 0, 15);
 334   }
 335   inline void uxtb(Register Rd, Register Rn) {
 336     ubfm(Rd, Rn, 0, 7);
 337   }
 338   inline void uxth(Register Rd, Register Rn) {
 339     ubfm(Rd, Rn, 0, 15);
 340   }
 341   inline void uxtw(Register Rd, Register Rn) {
 342     ubfm(Rd, Rn, 0, 31);
 343   }
 344 
 345   inline void cmnw(Register Rn, Register Rm) {
 346     addsw(zr, Rn, Rm);
 347   }
 348   inline void cmn(Register Rn, Register Rm) {
 349     adds(zr, Rn, Rm);
 350   }
 351 
 352   inline void cmpw(Register Rn, Register Rm) {
 353     subsw(zr, Rn, Rm);
 354   }
 355   inline void cmp(Register Rn, Register Rm) {
 356     subs(zr, Rn, Rm);
 357   }
 358 
 359   inline void negw(Register Rd, Register Rn) {
 360     subw(Rd, zr, Rn);
 361   }
 362 
 363   inline void neg(Register Rd, Register Rn) {
 364     sub(Rd, zr, Rn);
 365   }
 366 
 367   inline void negsw(Register Rd, Register Rn) {
 368     subsw(Rd, zr, Rn);
 369   }
 370 
 371   inline void negs(Register Rd, Register Rn) {
 372     subs(Rd, zr, Rn);
 373   }
 374 
 375   inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 376     addsw(zr, Rn, Rm, kind, shift);
 377   }
 378   inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 379     adds(zr, Rn, Rm, kind, shift);
 380   }
 381 
 382   inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 383     subsw(zr, Rn, Rm, kind, shift);
 384   }
 385   inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) {
 386     subs(zr, Rn, Rm, kind, shift);
 387   }
 388 
 389   inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 390     subw(Rd, zr, Rn, kind, shift);
 391   }
 392 
 393   inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 394     sub(Rd, zr, Rn, kind, shift);
 395   }
 396 
 397   inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 398     subsw(Rd, zr, Rn, kind, shift);
 399   }
 400 
 401   inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) {
 402     subs(Rd, zr, Rn, kind, shift);
 403   }
 404 
 405   inline void mnegw(Register Rd, Register Rn, Register Rm) {
 406     msubw(Rd, Rn, Rm, zr);
 407   }
 408   inline void mneg(Register Rd, Register Rn, Register Rm) {
 409     msub(Rd, Rn, Rm, zr);
 410   }
 411 
 412   inline void mulw(Register Rd, Register Rn, Register Rm) {
 413     maddw(Rd, Rn, Rm, zr);
 414   }
 415   inline void mul(Register Rd, Register Rn, Register Rm) {
 416     madd(Rd, Rn, Rm, zr);
 417   }
 418 
 419   inline void smnegl(Register Rd, Register Rn, Register Rm) {
 420     smsubl(Rd, Rn, Rm, zr);
 421   }
 422   inline void smull(Register Rd, Register Rn, Register Rm) {
 423     smaddl(Rd, Rn, Rm, zr);
 424   }
 425 
 426   inline void umnegl(Register Rd, Register Rn, Register Rm) {
 427     umsubl(Rd, Rn, Rm, zr);
 428   }
 429   inline void umull(Register Rd, Register Rn, Register Rm) {
 430     umaddl(Rd, Rn, Rm, zr);
 431   }
 432 
 433 #define WRAP(INSN)                                                            \
 434   void INSN(Register Rd, Register Rn, Register Rm, Register Ra) {             \
 435     if (VM_Version::supports_a53mac() && Ra != zr)                            \
 436       nop();                                                                  \
 437     Assembler::INSN(Rd, Rn, Rm, Ra);                                          \
 438   }
 439 
 440   WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw)
 441   WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl)
 442 #undef WRAP
 443 
 444 
 445   // macro assembly operations needed for aarch64
 446 
 447   // first two private routines for loading 32 bit or 64 bit constants
 448 private:
 449 
 450   void mov_immediate64(Register dst, uint64_t imm64);
 451   void mov_immediate32(Register dst, uint32_t imm32);
 452 
 453   int push(unsigned int bitset, Register stack);
 454   int pop(unsigned int bitset, Register stack);
 455 
 456   int push_fp(unsigned int bitset, Register stack);
 457   int pop_fp(unsigned int bitset, Register stack);
 458 
 459   int push_p(unsigned int bitset, Register stack);
 460   int pop_p(unsigned int bitset, Register stack);
 461 
 462   void mov(Register dst, Address a);
 463 
 464 public:
 465   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
 466   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 467 
 468   void push_fp(FloatRegSet regs, Register stack) { if (regs.bits()) push_fp(regs.bits(), stack); }
 469   void pop_fp(FloatRegSet regs, Register stack) { if (regs.bits()) pop_fp(regs.bits(), stack); }
 470 
 471   static RegSet call_clobbered_gp_registers();
 472 
 473   void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); }
 474   void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); }
 475 
 476   // Push and pop everything that might be clobbered by a native
 477   // runtime call except rscratch1 and rscratch2.  (They are always
 478   // scratch, so we don't have to protect them.)  Only save the lower
 479   // 64 bits of each vector register. Additional registers can be excluded
 480   // in a passed RegSet.
 481   void push_call_clobbered_registers_except(RegSet exclude);
 482   void pop_call_clobbered_registers_except(RegSet exclude);
 483 
 484   void push_call_clobbered_registers() {
 485     push_call_clobbered_registers_except(RegSet());
 486   }
 487   void pop_call_clobbered_registers() {
 488     pop_call_clobbered_registers_except(RegSet());
 489   }
 490 
 491 
 492   // now mov instructions for loading absolute addresses and 32 or
 493   // 64 bit integers
 494 
 495   inline void mov(Register dst, address addr)             { mov_immediate64(dst, (uint64_t)addr); }
 496 
 497   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 498   inline void mov(Register dst, T o)                      { mov_immediate64(dst, (uint64_t)o); }
 499 
 500   inline void movw(Register dst, uint32_t imm32)          { mov_immediate32(dst, imm32); }
 501 
 502   void mov(Register dst, RegisterOrConstant src) {
 503     if (src.is_register())
 504       mov(dst, src.as_register());
 505     else
 506       mov(dst, src.as_constant());
 507   }
 508 
 509   void movptr(Register r, uintptr_t imm64);
 510 
 511   void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64);
 512 
 513   void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) {
 514     orr(Vd, T, Vn, Vn);
 515   }
 516 
 517   void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) {
 518     fcvtsh(tmp, src);
 519     smov(dst, tmp, H, 0);
 520   }
 521 
 522   void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) {
 523     mov(tmp, H, 0, src);
 524     fcvths(dst, tmp);
 525   }
 526 
 527   // Generalized Test Bit And Branch, including a "far" variety which
 528   // spans more than 32KiB.
 529   void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) {
 530     assert(cond == EQ || cond == NE, "must be");
 531 
 532     if (isfar)
 533       cond = ~cond;
 534 
 535     void (Assembler::* branch)(Register Rt, int bitpos, Label &L);
 536     if (cond == Assembler::EQ)
 537       branch = &Assembler::tbz;
 538     else
 539       branch = &Assembler::tbnz;
 540 
 541     if (isfar) {
 542       Label L;
 543       (this->*branch)(Rt, bitpos, L);
 544       b(dest);
 545       bind(L);
 546     } else {
 547       (this->*branch)(Rt, bitpos, dest);
 548     }
 549   }
 550 
 551   // macro instructions for accessing and updating floating point
 552   // status register
 553   //
 554   // FPSR : op1 == 011
 555   //        CRn == 0100
 556   //        CRm == 0100
 557   //        op2 == 001
 558 
 559   inline void get_fpsr(Register reg)
 560   {
 561     mrs(0b11, 0b0100, 0b0100, 0b001, reg);
 562   }
 563 
 564   inline void set_fpsr(Register reg)
 565   {
 566     msr(0b011, 0b0100, 0b0100, 0b001, reg);
 567   }
 568 
 569   inline void clear_fpsr()
 570   {
 571     msr(0b011, 0b0100, 0b0100, 0b001, zr);
 572   }
 573 
 574   // FPCR : op1 == 011
 575   //        CRn == 0100
 576   //        CRm == 0100
 577   //        op2 == 000
 578 
 579   inline void get_fpcr(Register reg) {
 580     mrs(0b11, 0b0100, 0b0100, 0b000, reg);
 581   }
 582 
 583   inline void set_fpcr(Register reg) {
 584     msr(0b011, 0b0100, 0b0100, 0b000, reg);
 585   }
 586 
 587   // DCZID_EL0: op1 == 011
 588   //            CRn == 0000
 589   //            CRm == 0000
 590   //            op2 == 111
 591   inline void get_dczid_el0(Register reg)
 592   {
 593     mrs(0b011, 0b0000, 0b0000, 0b111, reg);
 594   }
 595 
 596   // CTR_EL0:   op1 == 011
 597   //            CRn == 0000
 598   //            CRm == 0000
 599   //            op2 == 001
 600   inline void get_ctr_el0(Register reg)
 601   {
 602     mrs(0b011, 0b0000, 0b0000, 0b001, reg);
 603   }
 604 
 605   inline void get_nzcv(Register reg) {
 606     mrs(0b011, 0b0100, 0b0010, 0b000, reg);
 607   }
 608 
 609   inline void set_nzcv(Register reg) {
 610     msr(0b011, 0b0100, 0b0010, 0b000, reg);
 611   }
 612 
 613   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 614   int corrected_idivl(Register result, Register ra, Register rb,
 615                       bool want_remainder, Register tmp = rscratch1);
 616   int corrected_idivq(Register result, Register ra, Register rb,
 617                       bool want_remainder, Register tmp = rscratch1);
 618 
 619   // Support for null-checks
 620   //
 621   // Generates code that causes a null OS exception if the content of reg is null.
 622   // If the accessed location is M[reg + offset] and the offset is known, provide the
 623   // offset. No explicit code generation is needed if the offset is within a certain
 624   // range (0 <= offset <= page_size).
 625 
 626   virtual void null_check(Register reg, int offset = -1);
 627   static bool needs_explicit_null_check(intptr_t offset);
 628   static bool uses_implicit_null_check(void* address);
 629 
 630   static address target_addr_for_insn(address insn_addr, unsigned insn);
 631   static address target_addr_for_insn_or_null(address insn_addr, unsigned insn);
 632   static address target_addr_for_insn(address insn_addr) {
 633     unsigned insn = *(unsigned*)insn_addr;
 634     return target_addr_for_insn(insn_addr, insn);
 635   }
 636   static address target_addr_for_insn_or_null(address insn_addr) {
 637     unsigned insn = *(unsigned*)insn_addr;
 638     return target_addr_for_insn_or_null(insn_addr, insn);
 639   }
 640 
 641   // Required platform-specific helpers for Label::patch_instructions.
 642   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 643   static int pd_patch_instruction_size(address branch, address target);
 644   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 645     pd_patch_instruction_size(branch, target);
 646   }
 647   static address pd_call_destination(address branch) {
 648     return target_addr_for_insn(branch);
 649   }
 650 #ifndef PRODUCT
 651   static void pd_print_patched_instruction(address branch);
 652 #endif
 653 
 654   static int patch_oop(address insn_addr, address o);
 655   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 656 
 657   // Return whether code is emitted to a scratch blob.
 658   virtual bool in_scratch_emit_size() {
 659     return false;
 660   }
 661   address emit_trampoline_stub(int insts_call_instruction_offset, address target);
 662   static int max_trampoline_stub_size();
 663   void emit_static_call_stub();
 664   static int static_call_stub_size();
 665 
 666   // The following 4 methods return the offset of the appropriate move instruction
 667 
 668   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 669   int load_unsigned_byte(Register dst, Address src);
 670   int load_unsigned_short(Register dst, Address src);
 671 
 672   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 673   int load_signed_byte(Register dst, Address src);
 674   int load_signed_short(Register dst, Address src);
 675 
 676   int load_signed_byte32(Register dst, Address src);
 677   int load_signed_short32(Register dst, Address src);
 678 
 679   // Support for sign-extension (hi:lo = extend_sign(lo))
 680   void extend_sign(Register hi, Register lo);
 681 
 682   // Load and store values by size and signed-ness
 683   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 684   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 685 
 686   // Support for inc/dec with optimal instruction selection depending on value
 687 
 688   // x86_64 aliases an unqualified register/address increment and
 689   // decrement to call incrementq and decrementq but also supports
 690   // explicitly sized calls to incrementq/decrementq or
 691   // incrementl/decrementl
 692 
 693   // for aarch64 the proper convention would be to use
 694   // increment/decrement for 64 bit operations and
 695   // incrementw/decrementw for 32 bit operations. so when porting
 696   // x86_64 code we can leave calls to increment/decrement as is,
 697   // replace incrementq/decrementq with increment/decrement and
 698   // replace incrementl/decrementl with incrementw/decrementw.
 699 
 700   // n.b. increment/decrement calls with an Address destination will
 701   // need to use a scratch register to load the value to be
 702   // incremented. increment/decrement calls which add or subtract a
 703   // constant value greater than 2^12 will need to use a 2nd scratch
 704   // register to hold the constant. so, a register increment/decrement
 705   // may trash rscratch2 and an address increment/decrement trash
 706   // rscratch and rscratch2
 707 
 708   void decrementw(Address dst, int value = 1);
 709   void decrementw(Register reg, int value = 1);
 710 
 711   void decrement(Register reg, int value = 1);
 712   void decrement(Address dst, int value = 1);
 713 
 714   void incrementw(Address dst, int value = 1);
 715   void incrementw(Register reg, int value = 1);
 716 
 717   void increment(Register reg, int value = 1);
 718   void increment(Address dst, int value = 1);
 719 
 720 
 721   // Alignment
 722   void align(int modulus);
 723   void align(int modulus, int target);
 724 
 725   // nop
 726   void post_call_nop();
 727 
 728   // Stack frame creation/removal
 729   void enter(bool strip_ret_addr = false);
 730   void leave();
 731 
 732   // ROP Protection
 733   void protect_return_address();
 734   void protect_return_address(Register return_reg);
 735   void authenticate_return_address();
 736   void authenticate_return_address(Register return_reg);
 737   void strip_return_address();
 738   void check_return_address(Register return_reg=lr) PRODUCT_RETURN;
 739 
 740   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 741   // The pointer will be loaded into the thread register.
 742   void get_thread(Register thread);
 743 
 744   // support for argument shuffling
 745   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 746   void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 747   void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 748   void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1);
 749   void object_move(
 750                    OopMap* map,
 751                    int oop_handle_offset,
 752                    int framesize_in_slots,
 753                    VMRegPair src,
 754                    VMRegPair dst,
 755                    bool is_receiver,
 756                    int* receiver_offset);
 757 
 758 
 759   // Support for VM calls
 760   //
 761   // It is imperative that all calls into the VM are handled via the call_VM macros.
 762   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 763   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 764 
 765 
 766   void call_VM(Register oop_result,
 767                address entry_point,
 768                bool check_exceptions = true);
 769   void call_VM(Register oop_result,
 770                address entry_point,
 771                Register arg_1,
 772                bool check_exceptions = true);
 773   void call_VM(Register oop_result,
 774                address entry_point,
 775                Register arg_1, Register arg_2,
 776                bool check_exceptions = true);
 777   void call_VM(Register oop_result,
 778                address entry_point,
 779                Register arg_1, Register arg_2, Register arg_3,
 780                bool check_exceptions = true);
 781 
 782   // Overloadings with last_Java_sp
 783   void call_VM(Register oop_result,
 784                Register last_java_sp,
 785                address entry_point,
 786                int number_of_arguments = 0,
 787                bool check_exceptions = true);
 788   void call_VM(Register oop_result,
 789                Register last_java_sp,
 790                address entry_point,
 791                Register arg_1, bool
 792                check_exceptions = true);
 793   void call_VM(Register oop_result,
 794                Register last_java_sp,
 795                address entry_point,
 796                Register arg_1, Register arg_2,
 797                bool check_exceptions = true);
 798   void call_VM(Register oop_result,
 799                Register last_java_sp,
 800                address entry_point,
 801                Register arg_1, Register arg_2, Register arg_3,
 802                bool check_exceptions = true);
 803 
 804   void get_vm_result  (Register oop_result, Register thread);
 805   void get_vm_result_2(Register metadata_result, Register thread);
 806 
 807   // These always tightly bind to MacroAssembler::call_VM_base
 808   // bypassing the virtual implementation
 809   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 810   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 811   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 812   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);
 813   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);
 814 
 815   void call_VM_leaf(address entry_point,
 816                     int number_of_arguments = 0);
 817   void call_VM_leaf(address entry_point,
 818                     Register arg_1);
 819   void call_VM_leaf(address entry_point,
 820                     Register arg_1, Register arg_2);
 821   void call_VM_leaf(address entry_point,
 822                     Register arg_1, Register arg_2, Register arg_3);
 823 
 824   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 825   // bypassing the virtual implementation
 826   void super_call_VM_leaf(address entry_point);
 827   void super_call_VM_leaf(address entry_point, Register arg_1);
 828   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 829   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 830   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 831 
 832   // last Java Frame (fills frame anchor)
 833   void set_last_Java_frame(Register last_java_sp,
 834                            Register last_java_fp,
 835                            address last_java_pc,
 836                            Register scratch);
 837 
 838   void set_last_Java_frame(Register last_java_sp,
 839                            Register last_java_fp,
 840                            Label &last_java_pc,
 841                            Register scratch);
 842 
 843   void set_last_Java_frame(Register last_java_sp,
 844                            Register last_java_fp,
 845                            Register last_java_pc,
 846                            Register scratch);
 847 
 848   void reset_last_Java_frame(Register thread);
 849 
 850   // thread in the default location (rthread)
 851   void reset_last_Java_frame(bool clear_fp);
 852 
 853   // Stores
 854   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 855   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 856 
 857   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 858   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 859 
 860   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 861   void c2bool(Register x);
 862 
 863   void load_method_holder_cld(Register rresult, Register rmethod);
 864   void load_method_holder(Register holder, Register method);
 865 
 866   // oop manipulations
 867   void load_nklass(Register dst, Register src);
 868   void load_klass(Register dst, Register src);
 869   void store_klass(Register dst, Register src);
 870   void cmp_klass(Register oop, Register trial_klass, Register tmp);
 871   void cmp_klass(Register src, Register dst, Register tmp1, Register tmp2);
 872 
 873   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 874   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 875   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 876 
 877   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 878                       Register tmp1, Register tmp2);
 879 
 880   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 881                        Register tmp1, Register tmp2, Register tmp3);
 882 
 883   void load_heap_oop(Register dst, Address src, Register tmp1,
 884                      Register tmp2, DecoratorSet decorators = 0);
 885 
 886   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 887                               Register tmp2, DecoratorSet decorators = 0);
 888   void store_heap_oop(Address dst, Register val, Register tmp1,
 889                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 890 
 891   // currently unimplemented
 892   // Used for storing null. All other oop constants should be
 893   // stored using routines that take a jobject.
 894   void store_heap_oop_null(Address dst);
 895 
 896   void store_klass_gap(Register dst, Register src);
 897 
 898   // This dummy is to prevent a call to store_heap_oop from
 899   // converting a zero (like null) into a Register by giving
 900   // the compiler two choices it can't resolve
 901 
 902   void store_heap_oop(Address dst, void* dummy);
 903 
 904   void encode_heap_oop(Register d, Register s);
 905   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 906   void decode_heap_oop(Register d, Register s);
 907   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 908   void encode_heap_oop_not_null(Register r);
 909   void decode_heap_oop_not_null(Register r);
 910   void encode_heap_oop_not_null(Register dst, Register src);
 911   void decode_heap_oop_not_null(Register dst, Register src);
 912 
 913   void set_narrow_oop(Register dst, jobject obj);
 914 
 915   void encode_klass_not_null(Register r);
 916   void decode_klass_not_null(Register r);
 917   void encode_klass_not_null(Register dst, Register src);
 918   void decode_klass_not_null(Register dst, Register src);
 919 
 920   void set_narrow_klass(Register dst, Klass* k);
 921 
 922   // if heap base register is used - reinit it with the correct value
 923   void reinit_heapbase();
 924 
 925   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 926 
 927   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
 928                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 929   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
 930                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 931 
 932   void push_cont_fastpath(Register java_thread);
 933   void pop_cont_fastpath(Register java_thread);
 934 
 935   // Round up to a power of two
 936   void round_to(Register reg, int modulus);
 937 
 938   // java.lang.Math::round intrinsics
 939   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
 940   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
 941 
 942   // allocation
 943   void tlab_allocate(
 944     Register obj,                      // result: pointer to object after successful allocation
 945     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 946     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 947     Register t1,                       // temp register
 948     Register t2,                       // temp register
 949     Label&   slow_case                 // continuation point if fast allocation fails
 950   );
 951   void verify_tlab();
 952 
 953   // interface method calling
 954   void lookup_interface_method(Register recv_klass,
 955                                Register intf_klass,
 956                                RegisterOrConstant itable_index,
 957                                Register method_result,
 958                                Register scan_temp,
 959                                Label& no_such_interface,
 960                    bool return_method = true);
 961 
 962   void lookup_interface_method_stub(Register recv_klass,
 963                                     Register holder_klass,
 964                                     Register resolved_klass,
 965                                     Register method_result,
 966                                     Register temp_reg,
 967                                     Register temp_reg2,
 968                                     int itable_index,
 969                                     Label& L_no_such_interface);
 970 
 971   // virtual method calling
 972   // n.b. x86 allows RegisterOrConstant for vtable_index
 973   void lookup_virtual_method(Register recv_klass,
 974                              RegisterOrConstant vtable_index,
 975                              Register method_result);
 976 
 977   // Test sub_klass against super_klass, with fast and slow paths.
 978 
 979   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 980   // One of the three labels can be null, meaning take the fall-through.
 981   // If super_check_offset is -1, the value is loaded up from super_klass.
 982   // No registers are killed, except temp_reg.
 983   void check_klass_subtype_fast_path(Register sub_klass,
 984                                      Register super_klass,
 985                                      Register temp_reg,
 986                                      Label* L_success,
 987                                      Label* L_failure,
 988                                      Label* L_slow_path,
 989                 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
 990 
 991   // The rest of the type check; must be wired to a corresponding fast path.
 992   // It does not repeat the fast path logic, so don't use it standalone.
 993   // The temp_reg and temp2_reg can be noreg, if no temps are available.
 994   // Updates the sub's secondary super cache as necessary.
 995   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
 996   void check_klass_subtype_slow_path(Register sub_klass,
 997                                      Register super_klass,
 998                                      Register temp_reg,
 999                                      Register temp2_reg,
1000                                      Label* L_success,
1001                                      Label* L_failure,
1002                                      bool set_cond_codes = false);
1003 
1004   // Simplified, combined version, good for typical uses.
1005   // Falls through on failure.
1006   void check_klass_subtype(Register sub_klass,
1007                            Register super_klass,
1008                            Register temp_reg,
1009                            Label& L_success);
1010 
1011   void clinit_barrier(Register klass,
1012                       Register thread,
1013                       Label* L_fast_path = nullptr,
1014                       Label* L_slow_path = nullptr);
1015 
1016   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
1017 
1018   void verify_sve_vector_length(Register tmp = rscratch1);
1019   void reinitialize_ptrue() {
1020     if (UseSVE > 0) {
1021       sve_ptrue(ptrue, B);
1022     }
1023   }
1024   void verify_ptrue();
1025 
1026   // Debugging
1027 
1028   // only if +VerifyOops
1029   void _verify_oop(Register reg, const char* s, const char* file, int line);
1030   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
1031 
1032   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
1033     if (VerifyOops) {
1034       _verify_oop(reg, s, file, line);
1035     }
1036   }
1037   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
1038     if (VerifyOops) {
1039       _verify_oop_addr(reg, s, file, line);
1040     }
1041   }
1042 
1043 // TODO: verify method and klass metadata (compare against vptr?)
1044   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
1045   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
1046 
1047 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
1048 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
1049 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
1050 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
1051 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
1052 
1053   // Restore cpu control state after JNI call
1054   void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2);
1055 
1056   // prints msg, dumps registers and stops execution
1057   void stop(const char* msg);
1058 
1059   static void debug64(char* msg, int64_t pc, int64_t regs[]);
1060 
1061   void untested()                                { stop("untested"); }
1062 
1063   void unimplemented(const char* what = "");
1064 
1065   void should_not_reach_here()                   { stop("should not reach here"); }
1066 
1067   void _assert_asm(Condition cc, const char* msg);
1068 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg)
1069 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg))
1070 
1071   // Stack overflow checking
1072   void bang_stack_with_offset(int offset) {
1073     // stack grows down, caller passes positive offset
1074     assert(offset > 0, "must bang with negative offset");
1075     sub(rscratch2, sp, offset);
1076     str(zr, Address(rscratch2));
1077   }
1078 
1079   // Writes to stack successive pages until offset reached to check for
1080   // stack overflow + shadow pages.  Also, clobbers tmp
1081   void bang_stack_size(Register size, Register tmp);
1082 
1083   // Check for reserved stack access in method being exited (for JIT)
1084   void reserved_stack_check();
1085 
1086   // Arithmetics
1087 
1088   void addptr(const Address &dst, int32_t src);
1089   void cmpptr(Register src1, Address src2);
1090 
1091   void cmpoop(Register obj1, Register obj2);
1092 
1093   // Various forms of CAS
1094 
1095   void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp,
1096                           Label &succeed, Label *fail);
1097   void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp,
1098                   Label &succeed, Label *fail);
1099 
1100   void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp,
1101                   Label &succeed, Label *fail);
1102 
1103   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1104   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1105   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1106   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1107 
1108   void atomic_xchg(Register prev, Register newv, Register addr);
1109   void atomic_xchgw(Register prev, Register newv, Register addr);
1110   void atomic_xchgl(Register prev, Register newv, Register addr);
1111   void atomic_xchglw(Register prev, Register newv, Register addr);
1112   void atomic_xchgal(Register prev, Register newv, Register addr);
1113   void atomic_xchgalw(Register prev, Register newv, Register addr);
1114 
1115   void orptr(Address adr, RegisterOrConstant src) {
1116     ldr(rscratch1, adr);
1117     if (src.is_register())
1118       orr(rscratch1, rscratch1, src.as_register());
1119     else
1120       orr(rscratch1, rscratch1, src.as_constant());
1121     str(rscratch1, adr);
1122   }
1123 
1124   // A generic CAS; success or failure is in the EQ flag.
1125   // Clobbers rscratch1
1126   void cmpxchg(Register addr, Register expected, Register new_val,
1127                enum operand_size size,
1128                bool acquire, bool release, bool weak,
1129                Register result);
1130 
1131 #ifdef ASSERT
1132   // Template short-hand support to clean-up after a failed call to trampoline
1133   // call generation (see trampoline_call() below),  when a set of Labels must
1134   // be reset (before returning).
1135   template<typename Label, typename... More>
1136   void reset_labels(Label &lbl, More&... more) {
1137     lbl.reset(); reset_labels(more...);
1138   }
1139   template<typename Label>
1140   void reset_labels(Label &lbl) {
1141     lbl.reset();
1142   }
1143 #endif
1144 
1145 private:
1146   void compare_eq(Register rn, Register rm, enum operand_size size);
1147 
1148 public:
1149   // AArch64 OpenJDK uses four different types of calls:
1150   //   - direct call: bl pc_relative_offset
1151   //     This is the shortest and the fastest, but the offset has the range:
1152   //     +/-128MB for the release build, +/-2MB for the debug build.
1153   //
1154   //   - far call: adrp reg, pc_relative_offset; add; bl reg
1155   //     This is longer than a direct call. The offset has
1156   //     the range +/-4GB. As the code cache size is limited to 4GB,
1157   //     far calls can reach anywhere in the code cache. If a jump is
1158   //     needed rather than a call, a far jump 'b reg' can be used instead.
1159   //     All instructions are embedded at a call site.
1160   //
1161   //   - trampoline call:
1162   //     This is only available in C1/C2-generated code (nmethod). It is a combination
1163   //     of a direct call, which is used if the destination of a call is in range,
1164   //     and a register-indirect call. It has the advantages of reaching anywhere in
1165   //     the AArch64 address space and being patchable at runtime when the generated
1166   //     code is being executed by other threads.
1167   //
1168   //     [Main code section]
1169   //       bl trampoline
1170   //     [Stub code section]
1171   //     trampoline:
1172   //       ldr reg, pc + 8
1173   //       br reg
1174   //       <64-bit destination address>
1175   //
1176   //     If the destination is in range when the generated code is moved to the code
1177   //     cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline
1178   //     is not used.
1179   //     The optimization does not remove the trampoline from the stub section.
1180   //     This is necessary because the trampoline may well be redirected later when
1181   //     code is patched, and the new destination may not be reachable by a simple BR
1182   //     instruction.
1183   //
1184   //   - indirect call: move reg, address; blr reg
1185   //     This too can reach anywhere in the address space, but it cannot be
1186   //     patched while code is running, so it must only be modified at a safepoint.
1187   //     This form of call is most suitable for targets at fixed addresses, which
1188   //     will never be patched.
1189   //
1190   // The patching we do conforms to the "Concurrent modification and
1191   // execution of instructions" section of the Arm Architectural
1192   // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC,
1193   // or SVC instructions to be modified while another thread is
1194   // executing them.
1195   //
1196   // To patch a trampoline call when the BL can't reach, we first modify
1197   // the 64-bit destination address in the trampoline, then modify the
1198   // BL to point to the trampoline, then flush the instruction cache to
1199   // broadcast the change to all executing threads. See
1200   // NativeCall::set_destination_mt_safe for the details.
1201   //
1202   // There is a benign race in that the other thread might observe the
1203   // modified BL before it observes the modified 64-bit destination
1204   // address. That does not matter because the destination method has been
1205   // invalidated, so there will be a trap at its start.
1206   // For this to work, the destination address in the trampoline is
1207   // always updated, even if we're not using the trampoline.
1208 
1209   // Emit a direct call if the entry address will always be in range,
1210   // otherwise a trampoline call.
1211   // Supported entry.rspec():
1212   // - relocInfo::runtime_call_type
1213   // - relocInfo::opt_virtual_call_type
1214   // - relocInfo::static_call_type
1215   // - relocInfo::virtual_call_type
1216   //
1217   // Return: the call PC or null if CodeCache is full.
1218   address trampoline_call(Address entry);
1219 
1220   static bool far_branches() {
1221     return ReservedCodeCacheSize > branch_range;
1222   }
1223 
1224   // Check if branches to the non nmethod section require a far jump
1225   static bool codestub_branch_needs_far_jump() {
1226     return CodeCache::max_distance_to_non_nmethod() > branch_range;
1227   }
1228 
1229   // Emit a direct call/jump if the entry address will always be in range,
1230   // otherwise a far call/jump.
1231   // The address must be inside the code cache.
1232   // Supported entry.rspec():
1233   // - relocInfo::external_word_type
1234   // - relocInfo::runtime_call_type
1235   // - relocInfo::none
1236   // In the case of a far call/jump, the entry address is put in the tmp register.
1237   // The tmp register is invalidated.
1238   //
1239   // Far_jump returns the amount of the emitted code.
1240   void far_call(Address entry, Register tmp = rscratch1);
1241   int far_jump(Address entry, Register tmp = rscratch1);
1242 
1243   static int far_codestub_branch_size() {
1244     if (codestub_branch_needs_far_jump()) {
1245       return 3 * 4;  // adrp, add, br
1246     } else {
1247       return 4;
1248     }
1249   }
1250 
1251   // Emit the CompiledIC call idiom
1252   address ic_call(address entry, jint method_index = 0);
1253   static int ic_check_size();
1254   int ic_check(int end_alignment);
1255 
1256 public:
1257 
1258   // Data
1259 
1260   void mov_metadata(Register dst, Metadata* obj);
1261   Address allocate_metadata_address(Metadata* obj);
1262   Address constant_oop_address(jobject obj);
1263 
1264   void movoop(Register dst, jobject obj);
1265 
1266   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1267   void kernel_crc32(Register crc, Register buf, Register len,
1268         Register table0, Register table1, Register table2, Register table3,
1269         Register tmp, Register tmp2, Register tmp3);
1270   // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic.
1271   void kernel_crc32c(Register crc, Register buf, Register len,
1272         Register table0, Register table1, Register table2, Register table3,
1273         Register tmp, Register tmp2, Register tmp3);
1274 
1275   // Stack push and pop individual 64 bit registers
1276   void push(Register src);
1277   void pop(Register dst);
1278 
1279   void repne_scan(Register addr, Register value, Register count,
1280                   Register scratch);
1281   void repne_scanw(Register addr, Register value, Register count,
1282                    Register scratch);
1283 
1284   typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm);
1285   typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift);
1286 
1287   // If a constant does not fit in an immediate field, generate some
1288   // number of MOV instructions and then perform the operation
1289   void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm,
1290                              add_sub_imm_insn insn1,
1291                              add_sub_reg_insn insn2, bool is32);
1292   // Separate vsn which sets the flags
1293   void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm,
1294                                add_sub_imm_insn insn1,
1295                                add_sub_reg_insn insn2, bool is32);
1296 
1297 #define WRAP(INSN, is32)                                                \
1298   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1299     wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1300   }                                                                     \
1301                                                                         \
1302   void INSN(Register Rd, Register Rn, Register Rm,                      \
1303              enum shift_kind kind, unsigned shift = 0) {                \
1304     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1305   }                                                                     \
1306                                                                         \
1307   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1308     Assembler::INSN(Rd, Rn, Rm);                                        \
1309   }                                                                     \
1310                                                                         \
1311   void INSN(Register Rd, Register Rn, Register Rm,                      \
1312            ext::operation option, int amount = 0) {                     \
1313     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1314   }
1315 
1316   WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true)
1317 
1318 #undef WRAP
1319 #define WRAP(INSN, is32)                                                \
1320   void INSN(Register Rd, Register Rn, uint64_t imm) {                   \
1321     wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \
1322   }                                                                     \
1323                                                                         \
1324   void INSN(Register Rd, Register Rn, Register Rm,                      \
1325              enum shift_kind kind, unsigned shift = 0) {                \
1326     Assembler::INSN(Rd, Rn, Rm, kind, shift);                           \
1327   }                                                                     \
1328                                                                         \
1329   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1330     Assembler::INSN(Rd, Rn, Rm);                                        \
1331   }                                                                     \
1332                                                                         \
1333   void INSN(Register Rd, Register Rn, Register Rm,                      \
1334            ext::operation option, int amount = 0) {                     \
1335     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1336   }
1337 
1338   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1339 
1340   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1341   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1342   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1343   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1344 
1345   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1346 
1347   void tableswitch(Register index, jint lowbound, jint highbound,
1348                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1349     adr(rscratch1, jumptable);
1350     subsw(rscratch2, index, lowbound);
1351     subsw(zr, rscratch2, highbound - lowbound);
1352     br(Assembler::HS, jumptable_end);
1353     add(rscratch1, rscratch1, rscratch2,
1354         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1355     br(rscratch1);
1356   }
1357 
1358   // Form an address from base + offset in Rd.  Rd may or may not
1359   // actually be used: you must use the Address that is returned.  It
1360   // is up to you to ensure that the shift provided matches the size
1361   // of your data.
1362   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1363 
1364   // Return true iff an address is within the 48-bit AArch64 address
1365   // space.
1366   bool is_valid_AArch64_address(address a) {
1367     return ((uint64_t)a >> 48) == 0;
1368   }
1369 
1370   // Load the base of the cardtable byte map into reg.
1371   void load_byte_map_base(Register reg);
1372 
1373   // Prolog generator routines to support switch between x86 code and
1374   // generated ARM code
1375 
1376   // routine to generate an x86 prolog for a stub function which
1377   // bootstraps into the generated ARM code which directly follows the
1378   // stub
1379   //
1380 
1381   public:
1382 
1383   void ldr_constant(Register dest, const Address &const_addr) {
1384     if (NearCpool) {
1385       ldr(dest, const_addr);
1386     } else {
1387       uint64_t offset;
1388       adrp(dest, InternalAddress(const_addr.target()), offset);
1389       ldr(dest, Address(dest, offset));
1390     }
1391   }
1392 
1393   address read_polling_page(Register r, relocInfo::relocType rtype);
1394   void get_polling_page(Register dest, relocInfo::relocType rtype);
1395 
1396   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1397   void update_byte_crc32(Register crc, Register val, Register table);
1398   void update_word_crc32(Register crc, Register v, Register tmp,
1399         Register table0, Register table1, Register table2, Register table3,
1400         bool upper = false);
1401 
1402   address count_positives(Register ary1, Register len, Register result);
1403 
1404   address arrays_equals(Register a1, Register a2, Register result, Register cnt1,
1405                         Register tmp1, Register tmp2, Register tmp3, int elem_size);
1406 
1407   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1408 
1409   void fill_words(Register base, Register cnt, Register value);
1410   address zero_words(Register base, uint64_t cnt);
1411   address zero_words(Register ptr, Register cnt);
1412   void zero_dcache_blocks(Register base, Register cnt);
1413 
1414   static const int zero_words_block_size;
1415 
1416   address byte_array_inflate(Register src, Register dst, Register len,
1417                              FloatRegister vtmp1, FloatRegister vtmp2,
1418                              FloatRegister vtmp3, Register tmp4);
1419 
1420   void char_array_compress(Register src, Register dst, Register len,
1421                            Register res,
1422                            FloatRegister vtmp0, FloatRegister vtmp1,
1423                            FloatRegister vtmp2, FloatRegister vtmp3,
1424                            FloatRegister vtmp4, FloatRegister vtmp5);
1425 
1426   void encode_iso_array(Register src, Register dst,
1427                         Register len, Register res, bool ascii,
1428                         FloatRegister vtmp0, FloatRegister vtmp1,
1429                         FloatRegister vtmp2, FloatRegister vtmp3,
1430                         FloatRegister vtmp4, FloatRegister vtmp5);
1431 
1432   void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi,
1433       address pio2, address dsin_coef, address dcos_coef);
1434  private:
1435   // begin trigonometric functions support block
1436   void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2);
1437   void generate__kernel_rem_pio2(address two_over_pi, address pio2);
1438   void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef);
1439   void generate_kernel_cos(FloatRegister x, address dcos_coef);
1440   // end trigonometric functions support block
1441   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1442                        Register src1, Register src2);
1443   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) {
1444     add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2);
1445   }
1446   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1447                              Register y, Register y_idx, Register z,
1448                              Register carry, Register product,
1449                              Register idx, Register kdx);
1450   void multiply_128_x_128_loop(Register y, Register z,
1451                                Register carry, Register carry2,
1452                                Register idx, Register jdx,
1453                                Register yz_idx1, Register yz_idx2,
1454                                Register tmp, Register tmp3, Register tmp4,
1455                                Register tmp7, Register product_hi);
1456   void kernel_crc32_using_crypto_pmull(Register crc, Register buf,
1457         Register len, Register tmp0, Register tmp1, Register tmp2,
1458         Register tmp3);
1459   void kernel_crc32_using_crc32(Register crc, Register buf,
1460         Register len, Register tmp0, Register tmp1, Register tmp2,
1461         Register tmp3);
1462   void kernel_crc32c_using_crypto_pmull(Register crc, Register buf,
1463         Register len, Register tmp0, Register tmp1, Register tmp2,
1464         Register tmp3);
1465   void kernel_crc32c_using_crc32c(Register crc, Register buf,
1466         Register len, Register tmp0, Register tmp1, Register tmp2,
1467         Register tmp3);
1468   void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf,
1469         Register len, Register tmp0, Register tmp1, Register tmp2,
1470         size_t table_offset);
1471 
1472   void ghash_modmul (FloatRegister result,
1473                      FloatRegister result_lo, FloatRegister result_hi, FloatRegister b,
1474                      FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p,
1475                      FloatRegister t1, FloatRegister t2, FloatRegister t3);
1476   void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state);
1477 public:
1478   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z,
1479                        Register zlen, Register tmp1, Register tmp2, Register tmp3,
1480                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
1481   void mul_add(Register out, Register in, Register offs, Register len, Register k);
1482   void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi,
1483                       FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1484                       FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1485   void ghash_multiply_wide(int index,
1486                            FloatRegister result_lo, FloatRegister result_hi,
1487                            FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0,
1488                            FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3);
1489   void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi,
1490                     FloatRegister p, FloatRegister z, FloatRegister t1);
1491   void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi,
1492                     FloatRegister p, FloatRegister z, FloatRegister t1);
1493   void ghash_processBlocks_wide(address p, Register state, Register subkeyH,
1494                                 Register data, Register blocks, int unrolls);
1495 
1496 
1497   void aesenc_loadkeys(Register key, Register keylen);
1498   void aesecb_encrypt(Register from, Register to, Register keylen,
1499                       FloatRegister data = v0, int unrolls = 1);
1500   void aesecb_decrypt(Register from, Register to, Register key, Register keylen);
1501   void aes_round(FloatRegister input, FloatRegister subkey);
1502 
1503   // ChaCha20 functions support block
1504   void cc20_quarter_round(FloatRegister aVec, FloatRegister bVec,
1505           FloatRegister cVec, FloatRegister dVec, FloatRegister scratch,
1506           FloatRegister tbl);
1507   void cc20_shift_lane_org(FloatRegister bVec, FloatRegister cVec,
1508           FloatRegister dVec, bool colToDiag);
1509 
1510   // Place an ISB after code may have been modified due to a safepoint.
1511   void safepoint_isb();
1512 
1513 private:
1514   // Return the effective address r + (r1 << ext) + offset.
1515   // Uses rscratch2.
1516   Address offsetted_address(Register r, Register r1, Address::extend ext,
1517                             int offset, int size);
1518 
1519 private:
1520   // Returns an address on the stack which is reachable with a ldr/str of size
1521   // Uses rscratch2 if the address is not directly reachable
1522   Address spill_address(int size, int offset, Register tmp=rscratch2);
1523   Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2);
1524 
1525   bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const;
1526 
1527   // Check whether two loads/stores can be merged into ldp/stp.
1528   bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const;
1529 
1530   // Merge current load/store with previous load/store into ldp/stp.
1531   void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1532 
1533   // Try to merge two loads/stores into ldp/stp. If success, returns true else false.
1534   bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store);
1535 
1536 public:
1537   void spill(Register Rx, bool is64, int offset) {
1538     if (is64) {
1539       str(Rx, spill_address(8, offset));
1540     } else {
1541       strw(Rx, spill_address(4, offset));
1542     }
1543   }
1544   void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1545     str(Vx, T, spill_address(1 << (int)T, offset));
1546   }
1547 
1548   void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1549     sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1550   }
1551   void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1552     sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1553   }
1554 
1555   void unspill(Register Rx, bool is64, int offset) {
1556     if (is64) {
1557       ldr(Rx, spill_address(8, offset));
1558     } else {
1559       ldrw(Rx, spill_address(4, offset));
1560     }
1561   }
1562   void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) {
1563     ldr(Vx, T, spill_address(1 << (int)T, offset));
1564   }
1565 
1566   void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) {
1567     sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset));
1568   }
1569   void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) {
1570     sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset));
1571   }
1572 
1573   void spill_copy128(int src_offset, int dst_offset,
1574                      Register tmp1=rscratch1, Register tmp2=rscratch2) {
1575     if (src_offset < 512 && (src_offset & 7) == 0 &&
1576         dst_offset < 512 && (dst_offset & 7) == 0) {
1577       ldp(tmp1, tmp2, Address(sp, src_offset));
1578       stp(tmp1, tmp2, Address(sp, dst_offset));
1579     } else {
1580       unspill(tmp1, true, src_offset);
1581       spill(tmp1, true, dst_offset);
1582       unspill(tmp1, true, src_offset+8);
1583       spill(tmp1, true, dst_offset+8);
1584     }
1585   }
1586   void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset,
1587                                             int sve_vec_reg_size_in_bytes) {
1588     assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size");
1589     for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) {
1590       spill_copy128(src_offset, dst_offset);
1591       src_offset += 16;
1592       dst_offset += 16;
1593     }
1594   }
1595   void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset,
1596                                                int sve_predicate_reg_size_in_bytes) {
1597     sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset));
1598     sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset));
1599     reinitialize_ptrue();
1600   }
1601   void cache_wb(Address line);
1602   void cache_wbsync(bool is_pre);
1603 
1604   // Code for java.lang.Thread::onSpinWait() intrinsic.
1605   void spin_wait();
1606 
1607   void lightweight_lock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1608   void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow);
1609 
1610 private:
1611   // Check the current thread doesn't need a cross modify fence.
1612   void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
1613 
1614 };
1615 
1616 #ifdef ASSERT
1617 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1618 #endif
1619 
1620 /**
1621  * class SkipIfEqual:
1622  *
1623  * Instantiating this class will result in assembly code being output that will
1624  * jump around any code emitted between the creation of the instance and it's
1625  * automatic destruction at the end of a scope block, depending on the value of
1626  * the flag passed to the constructor, which will be checked at run-time.
1627  */
1628 class SkipIfEqual {
1629  private:
1630   MacroAssembler* _masm;
1631   Label _label;
1632 
1633  public:
1634    SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1635    ~SkipIfEqual();
1636 };
1637 
1638 struct tableswitch {
1639   Register _reg;
1640   int _insn_index; jint _first_key; jint _last_key;
1641   Label _after;
1642   Label _branches;
1643 };
1644 
1645 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP