1 /*
   2  * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
   4  * Copyright (c) 2020, 2024, Huawei Technologies Co., Ltd. All rights reserved.
   5  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   6  *
   7  * This code is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License version 2 only, as
   9  * published by the Free Software Foundation.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  *
  25  */
  26 
  27 #ifndef CPU_RISCV_MACROASSEMBLER_RISCV_HPP
  28 #define CPU_RISCV_MACROASSEMBLER_RISCV_HPP
  29 
  30 #include "asm/assembler.inline.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "utilities/powerOfTwo.hpp"
  35 
  36 // MacroAssembler extends Assembler by frequently used macros.
  37 //
  38 // Instructions for which a 'better' code sequence exists depending
  39 // on arguments should also go in here.
  40 
  41 class MacroAssembler: public Assembler {
  42 
  43  public:
  44 
  45   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
  46 
  47   void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp_reg = t0);
  48 
  49   // Alignment
  50   int align(int modulus, int extra_offset = 0);
  51 
  52   static inline void assert_alignment(address pc, int alignment = MacroAssembler::instruction_size) {
  53     assert(is_aligned(pc, alignment), "bad alignment");
  54   }
  55 
  56   // nop
  57   void post_call_nop();
  58 
  59   // Stack frame creation/removal
  60   // Note that SP must be updated to the right place before saving/restoring RA and FP
  61   // because signal based thread suspend/resume could happen asynchronously.
  62   void enter() {
  63     subi(sp, sp, 2 * wordSize);
  64     sd(ra, Address(sp, wordSize));
  65     sd(fp, Address(sp));
  66     addi(fp, sp, 2 * wordSize);
  67   }
  68 
  69   void leave() {
  70     subi(sp, fp, 2 * wordSize);
  71     ld(fp, Address(sp));
  72     ld(ra, Address(sp, wordSize));
  73     addi(sp, sp, 2 * wordSize);
  74   }
  75 
  76 
  77   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
  78   // The pointer will be loaded into the thread register.
  79   void get_thread(Register thread);
  80 
  81   // Support for VM calls
  82   //
  83   // It is imperative that all calls into the VM are handled via the call_VM macros.
  84   // They make sure that the stack linkage is setup correctly. call_VM's correspond
  85   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
  86 
  87   void call_VM(Register oop_result,
  88                address entry_point,
  89                bool check_exceptions = true);
  90   void call_VM(Register oop_result,
  91                address entry_point,
  92                Register arg_1,
  93                bool check_exceptions = true);
  94   void call_VM(Register oop_result,
  95                address entry_point,
  96                Register arg_1, Register arg_2,
  97                bool check_exceptions = true);
  98   void call_VM(Register oop_result,
  99                address entry_point,
 100                Register arg_1, Register arg_2, Register arg_3,
 101                bool check_exceptions = true);
 102 
 103   // Overloadings with last_Java_sp
 104   void call_VM(Register oop_result,
 105                Register last_java_sp,
 106                address entry_point,
 107                int number_of_arguments = 0,
 108                bool check_exceptions = true);
 109   void call_VM(Register oop_result,
 110                Register last_java_sp,
 111                address entry_point,
 112                Register arg_1,
 113                bool check_exceptions = true);
 114   void call_VM(Register oop_result,
 115                Register last_java_sp,
 116                address entry_point,
 117                Register arg_1, Register arg_2,
 118                bool check_exceptions = true);
 119   void call_VM(Register oop_result,
 120                Register last_java_sp,
 121                address entry_point,
 122                Register arg_1, Register arg_2, Register arg_3,
 123                bool check_exceptions = true);
 124 
 125   void get_vm_result_oop(Register oop_result, Register java_thread);
 126   void get_vm_result_metadata(Register metadata_result, Register java_thread);
 127 
 128   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 129   // bypassing the virtual implementation
 130   void call_VM_leaf(address entry_point,
 131                     int number_of_arguments = 0);
 132   void call_VM_leaf(address entry_point,
 133                     Register arg_0);
 134   void call_VM_leaf(address entry_point,
 135                     Register arg_0, Register arg_1);
 136   void call_VM_leaf(address entry_point,
 137                     Register arg_0, Register arg_1, Register arg_2);
 138 
 139   // These always tightly bind to MacroAssembler::call_VM_base
 140   // bypassing the virtual implementation
 141   void super_call_VM_leaf(address entry_point, Register arg_0);
 142   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1);
 143   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2);
 144   void super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3);
 145 
 146   // last Java Frame (fills frame anchor)
 147   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, address last_java_pc, Register tmp);
 148   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, Label &last_java_pc, Register tmp);
 149   void set_last_Java_frame(Register last_java_sp, Register last_java_fp, Register last_java_pc);
 150 
 151   // thread in the default location (xthread)
 152   void reset_last_Java_frame(bool clear_fp);
 153 
 154   virtual void call_VM_leaf_base(
 155     address entry_point,                // the entry point
 156     int     number_of_arguments,        // the number of arguments to pop after the call
 157     Label*  retaddr = nullptr
 158   );
 159 
 160   virtual void call_VM_leaf_base(
 161     address entry_point,                // the entry point
 162     int     number_of_arguments,        // the number of arguments to pop after the call
 163     Label&  retaddr) {
 164     call_VM_leaf_base(entry_point, number_of_arguments, &retaddr);
 165   }
 166 
 167   virtual void call_VM_base(           // returns the register containing the thread upon return
 168     Register oop_result,               // where an oop-result ends up if any; use noreg otherwise
 169     Register java_thread,              // the thread if computed before     ; use noreg otherwise
 170     Register last_java_sp,             // to set up last_Java_frame in stubs; use noreg otherwise
 171     address  entry_point,              // the entry point
 172     int      number_of_arguments,      // the number of arguments (w/o thread) to pop after the call
 173     bool     check_exceptions          // whether to check for pending exceptions after return
 174   );
 175 
 176   void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions);
 177 
 178   virtual void check_and_handle_earlyret(Register java_thread);
 179   virtual void check_and_handle_popframe(Register java_thread);
 180 
 181   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 182   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 183   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 184   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 185 
 186   void movoop(Register dst, jobject obj);
 187   void mov_metadata(Register dst, Metadata* obj);
 188   void bang_stack_size(Register size, Register tmp);
 189   void set_narrow_oop(Register dst, jobject obj);
 190   void set_narrow_klass(Register dst, Klass* k);
 191 
 192   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 193   void access_load_at(BasicType type, DecoratorSet decorators, Register dst,
 194                       Address src, Register tmp1, Register tmp2);
 195   void access_store_at(BasicType type, DecoratorSet decorators, Address dst,
 196                        Register val, Register tmp1, Register tmp2, Register tmp3);
 197   void load_klass(Register dst, Register src, Register tmp = t0);
 198   void load_narrow_klass_compact(Register dst, Register src);
 199   void store_klass(Register dst, Register src, Register tmp = t0);
 200   void cmp_klass_compressed(Register oop, Register trial_klass, Register tmp, Label &L, bool equal);
 201 
 202   void encode_klass_not_null(Register r, Register tmp = t0);
 203   void decode_klass_not_null(Register r, Register tmp = t0);
 204   void encode_klass_not_null(Register dst, Register src, Register tmp);
 205   void decode_klass_not_null(Register dst, Register src, Register tmp);
 206   void decode_heap_oop_not_null(Register r);
 207   void decode_heap_oop_not_null(Register dst, Register src);
 208   void decode_heap_oop(Register d, Register s);
 209   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 210   void encode_heap_oop_not_null(Register r);
 211   void encode_heap_oop_not_null(Register dst, Register src);
 212   void encode_heap_oop(Register d, Register s);
 213   void encode_heap_oop(Register r) { encode_heap_oop(r, r); };
 214   void load_heap_oop(Register dst, Address src, Register tmp1,
 215                      Register tmp2, DecoratorSet decorators = 0);
 216   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 217                               Register tmp2, DecoratorSet decorators = 0);
 218   void store_heap_oop(Address dst, Register val, Register tmp1,
 219                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 220 
 221   void store_klass_gap(Register dst, Register src);
 222 
 223   // currently unimplemented
 224   // Used for storing null. All other oop constants should be
 225   // stored using routines that take a jobject.
 226   void store_heap_oop_null(Address dst);
 227 
 228   // This dummy is to prevent a call to store_heap_oop from
 229   // converting a zero (linked null) into a Register by giving
 230   // the compiler two choices it can't resolve
 231 
 232   void store_heap_oop(Address dst, void* dummy);
 233 
 234   // Support for null-checks
 235   //
 236   // Generates code that causes a null OS exception if the content of reg is null.
 237   // If the accessed location is M[reg + offset] and the offset is known, provide the
 238   // offset. No explicit code generateion is needed if the offset is within a certain
 239   // range (0 <= offset <= page_size).
 240 
 241   virtual void null_check(Register reg, int offset = -1);
 242   static bool needs_explicit_null_check(intptr_t offset);
 243   static bool uses_implicit_null_check(void* address);
 244 
 245   // interface method calling
 246   void lookup_interface_method(Register recv_klass,
 247                                Register intf_klass,
 248                                RegisterOrConstant itable_index,
 249                                Register method_result,
 250                                Register scan_tmp,
 251                                Label& no_such_interface,
 252                                bool return_method = true);
 253 
 254   void lookup_interface_method_stub(Register recv_klass,
 255                                     Register holder_klass,
 256                                     Register resolved_klass,
 257                                     Register method_result,
 258                                     Register temp_reg,
 259                                     Register temp_reg2,
 260                                     int itable_index,
 261                                     Label& L_no_such_interface);
 262 
 263   // virtual method calling
 264   // n.n. x86 allows RegisterOrConstant for vtable_index
 265   void lookup_virtual_method(Register recv_klass,
 266                              RegisterOrConstant vtable_index,
 267                              Register method_result);
 268 
 269   // Form an address from base + offset in Rd. Rd my or may not
 270   // actually be used: you must use the Address that is returned. It
 271   // is up to you to ensure that the shift provided matches the size
 272   // of your data.
 273   Address form_address(Register Rd, Register base, int64_t byte_offset);
 274 
 275   // Sometimes we get misaligned loads and stores, usually from Unsafe
 276   // accesses, and these can exceed the offset range.
 277   Address legitimize_address(Register Rd, const Address &adr) {
 278     if (adr.getMode() == Address::base_plus_offset) {
 279       if (!is_simm12(adr.offset())) {
 280         return form_address(Rd, adr.base(), adr.offset());
 281       }
 282     }
 283     return adr;
 284   }
 285 
 286   // allocation
 287   void tlab_allocate(
 288     Register obj,                   // result: pointer to object after successful allocation
 289     Register var_size_in_bytes,     // object size in bytes if unknown at compile time; invalid otherwise
 290     int      con_size_in_bytes,     // object size in bytes if   known at compile time
 291     Register tmp1,                  // temp register
 292     Register tmp2,                  // temp register
 293     Label&   slow_case,             // continuation point of fast allocation fails
 294     bool     is_far = false
 295   );
 296 
 297   // Test sub_klass against super_klass, with fast and slow paths.
 298 
 299   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 300   // One of the three labels can be null, meaning take the fall-through.
 301   // If super_check_offset is -1, the value is loaded up from super_klass.
 302   // No registers are killed, except tmp_reg
 303   void check_klass_subtype_fast_path(Register sub_klass,
 304                                      Register super_klass,
 305                                      Register tmp_reg,
 306                                      Label* L_success,
 307                                      Label* L_failure,
 308                                      Label* L_slow_path,
 309                                      Register super_check_offset = noreg);
 310 
 311   // The reset of the type check; must be wired to a corresponding fast path.
 312   // It does not repeat the fast path logic, so don't use it standalone.
 313   // The tmp1_reg and tmp2_reg can be noreg, if no temps are available.
 314   // Updates the sub's secondary super cache as necessary.
 315   void check_klass_subtype_slow_path(Register sub_klass,
 316                                      Register super_klass,
 317                                      Register tmp1_reg,
 318                                      Register tmp2_reg,
 319                                      Label* L_success,
 320                                      Label* L_failure,
 321                                      bool set_cond_codes = false);
 322 
 323   void check_klass_subtype_slow_path_linear(Register sub_klass,
 324                                             Register super_klass,
 325                                             Register tmp1_reg,
 326                                             Register tmp2_reg,
 327                                             Label* L_success,
 328                                             Label* L_failure,
 329                                             bool set_cond_codes = false);
 330 
 331   void check_klass_subtype_slow_path_table(Register sub_klass,
 332                                            Register super_klass,
 333                                            Register tmp1_reg,
 334                                            Register tmp2_reg,
 335                                            Label* L_success,
 336                                            Label* L_failure,
 337                                            bool set_cond_codes = false);
 338 
 339   // If r is valid, return r.
 340   // If r is invalid, remove a register r2 from available_regs, add r2
 341   // to regs_to_push, then return r2.
 342   Register allocate_if_noreg(const Register r,
 343                              RegSetIterator<Register> &available_regs,
 344                              RegSet &regs_to_push);
 345 
 346   // Secondary subtype checking
 347   void lookup_secondary_supers_table_var(Register sub_klass,
 348                                          Register r_super_klass,
 349                                          Register result,
 350                                          Register tmp1,
 351                                          Register tmp2,
 352                                          Register tmp3,
 353                                          Register tmp4,
 354                                          Label *L_success);
 355 
 356   void population_count(Register dst, Register src, Register tmp1, Register tmp2);
 357 
 358   // As above, but with a constant super_klass.
 359   // The result is in Register result, not the condition codes.
 360   bool lookup_secondary_supers_table_const(Register r_sub_klass,
 361                                            Register r_super_klass,
 362                                            Register result,
 363                                            Register tmp1,
 364                                            Register tmp2,
 365                                            Register tmp3,
 366                                            Register tmp4,
 367                                            u1 super_klass_slot,
 368                                            bool stub_is_near = false);
 369 
 370   void verify_secondary_supers_table(Register r_sub_klass,
 371                                      Register r_super_klass,
 372                                      Register result,
 373                                      Register tmp1,
 374                                      Register tmp2,
 375                                      Register tmp3);
 376 
 377   void lookup_secondary_supers_table_slow_path(Register r_super_klass,
 378                                                Register r_array_base,
 379                                                Register r_array_index,
 380                                                Register r_bitmap,
 381                                                Register result,
 382                                                Register tmp,
 383                                                bool is_stub = true);
 384 
 385   void check_klass_subtype(Register sub_klass,
 386                            Register super_klass,
 387                            Register tmp_reg,
 388                            Label& L_success);
 389 
 390   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
 391 
 392   // only if +VerifyOops
 393   void _verify_oop(Register reg, const char* s, const char* file, int line);
 394   void _verify_oop_addr(Address addr, const char* s, const char* file, int line);
 395 
 396   void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
 397     if (VerifyOops) {
 398       _verify_oop(reg, s, file, line);
 399     }
 400   }
 401   void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
 402     if (VerifyOops) {
 403       _verify_oop_addr(reg, s, file, line);
 404     }
 405   }
 406 
 407   void _verify_method_ptr(Register reg, const char* msg, const char* file, int line) {}
 408   void _verify_klass_ptr(Register reg, const char* msg, const char* file, int line) {}
 409 
 410 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
 411 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
 412 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
 413 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
 414 #define verify_klass_ptr(reg) _verify_method_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
 415 
 416   // A more convenient access to fence for our purposes
 417   // We used four bit to indicate the read and write bits in the predecessors and successors,
 418   // and extended i for r, o for w if UseConservativeFence enabled.
 419   enum Membar_mask_bits {
 420     StoreStore = 0b0101,               // (pred = w   + succ = w)
 421     LoadStore  = 0b1001,               // (pred = r   + succ = w)
 422     StoreLoad  = 0b0110,               // (pred = w   + succ = r)
 423     LoadLoad   = 0b1010,               // (pred = r   + succ = r)
 424     AnyAny     = LoadStore | StoreLoad // (pred = rw  + succ = rw)
 425   };
 426 
 427   void membar(uint32_t order_constraint);
 428 
 429  private:
 430 
 431   static void membar_mask_to_pred_succ(uint32_t order_constraint,
 432                                        uint32_t& predecessor, uint32_t& successor) {
 433     predecessor = (order_constraint >> 2) & 0x3;
 434     successor = order_constraint & 0x3;
 435 
 436     // extend rw -> iorw:
 437     // 01(w) -> 0101(ow)
 438     // 10(r) -> 1010(ir)
 439     // 11(rw)-> 1111(iorw)
 440     if (UseConservativeFence) {
 441       predecessor |= predecessor << 2;
 442       successor   |= successor << 2;
 443     }
 444   }
 445 
 446   static int pred_succ_to_membar_mask(uint32_t predecessor, uint32_t successor) {
 447     return ((predecessor & 0x3) << 2) | (successor & 0x3);
 448   }
 449 
 450  public:
 451 
 452   void cmodx_fence();
 453 
 454   void pause() {
 455     // Zihintpause
 456     // PAUSE is encoded as a FENCE instruction with pred=W, succ=0, fm=0, rd=x0, and rs1=x0.
 457     Assembler::fence(w, 0);
 458   }
 459 
 460   // prints msg, dumps registers and stops execution
 461   void stop(const char* msg);
 462 
 463   static void debug64(char* msg, int64_t pc, int64_t regs[]);
 464 
 465   void unimplemented(const char* what = "");
 466 
 467   void should_not_reach_here() { stop("should not reach here"); }
 468 
 469   static address target_addr_for_insn(address insn_addr);
 470 
 471   // Required platform-specific helpers for Label::patch_instructions.
 472   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 473   static int pd_patch_instruction_size(address branch, address target);
 474   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 475     pd_patch_instruction_size(branch, target);
 476   }
 477   static address pd_call_destination(address branch) {
 478     return target_addr_for_insn(branch);
 479   }
 480 
 481   static int patch_oop(address insn_addr, address o);
 482 
 483   static address get_target_of_li32(address insn_addr);
 484   static int patch_imm_in_li32(address branch, int32_t target);
 485 
 486   // Return whether code is emitted to a scratch blob.
 487   virtual bool in_scratch_emit_size() {
 488     return false;
 489   }
 490 
 491   address emit_reloc_call_address_stub(int insts_call_instruction_offset, address target);
 492   static int max_reloc_call_address_stub_size();
 493 
 494   void emit_static_call_stub();
 495   static int static_call_stub_size();
 496 
 497   // The following 4 methods return the offset of the appropriate move instruction
 498 
 499   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 500   int load_unsigned_byte(Register dst, Address src);
 501   int load_unsigned_short(Register dst, Address src);
 502 
 503   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 504   int load_signed_byte(Register dst, Address src);
 505   int load_signed_short(Register dst, Address src);
 506 
 507   // Load and store values by size and signed-ness
 508   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed);
 509   void store_sized_value(Address dst, Register src, size_t size_in_bytes);
 510 
 511   // Misaligned loads, will use the best way, according to the AvoidUnalignedAccess flag
 512   void load_short_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity = 1);
 513   void load_int_misaligned(Register dst, Address src, Register tmp, bool is_signed, int granularity = 1);
 514   void load_long_misaligned(Register dst, Address src, Register tmp, int granularity = 1);
 515 
 516  public:
 517   // Standard pseudo instructions
 518   inline void nop() {
 519     addi(x0, x0, 0);
 520   }
 521 
 522   inline void mv(Register Rd, Register Rs) {
 523     if (Rd != Rs) {
 524       addi(Rd, Rs, 0);
 525     }
 526   }
 527 
 528   inline void notr(Register Rd, Register Rs) {
 529     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 530       c_not(Rd);
 531     } else {
 532       xori(Rd, Rs, -1);
 533     }
 534   }
 535 
 536   inline void neg(Register Rd, Register Rs) {
 537     sub(Rd, x0, Rs);
 538   }
 539 
 540   inline void negw(Register Rd, Register Rs) {
 541     subw(Rd, x0, Rs);
 542   }
 543 
 544   inline void sext_w(Register Rd, Register Rs) {
 545     addiw(Rd, Rs, 0);
 546   }
 547 
 548   inline void zext_b(Register Rd, Register Rs) {
 549     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 550       c_zext_b(Rd);
 551     } else {
 552       andi(Rd, Rs, 0xFF);
 553     }
 554   }
 555 
 556   inline void seqz(Register Rd, Register Rs) {
 557     sltiu(Rd, Rs, 1);
 558   }
 559 
 560   inline void snez(Register Rd, Register Rs) {
 561     sltu(Rd, x0, Rs);
 562   }
 563 
 564   inline void sltz(Register Rd, Register Rs) {
 565     slt(Rd, Rs, x0);
 566   }
 567 
 568   inline void sgtz(Register Rd, Register Rs) {
 569     slt(Rd, x0, Rs);
 570   }
 571 
 572   // Bit-manipulation extension pseudo instructions
 573   // zero extend word
 574   inline void zext_w(Register Rd, Register Rs) {
 575     assert(UseZba, "must be");
 576     if (do_compress_zcb(Rd, Rs) && (Rd == Rs)) {
 577       c_zext_w(Rd);
 578     } else {
 579       add_uw(Rd, Rs, zr);
 580     }
 581   }
 582 
 583   // Floating-point data-processing pseudo instructions
 584   inline void fmv_s(FloatRegister Rd, FloatRegister Rs) {
 585     if (Rd != Rs) {
 586       fsgnj_s(Rd, Rs, Rs);
 587     }
 588   }
 589 
 590   inline void fabs_s(FloatRegister Rd, FloatRegister Rs) {
 591     fsgnjx_s(Rd, Rs, Rs);
 592   }
 593 
 594   inline void fneg_s(FloatRegister Rd, FloatRegister Rs) {
 595     fsgnjn_s(Rd, Rs, Rs);
 596   }
 597 
 598   inline void fmv_d(FloatRegister Rd, FloatRegister Rs) {
 599     if (Rd != Rs) {
 600       fsgnj_d(Rd, Rs, Rs);
 601     }
 602   }
 603 
 604   inline void fabs_d(FloatRegister Rd, FloatRegister Rs) {
 605     fsgnjx_d(Rd, Rs, Rs);
 606   }
 607 
 608   inline void fneg_d(FloatRegister Rd, FloatRegister Rs) {
 609     fsgnjn_d(Rd, Rs, Rs);
 610   }
 611 
 612   // Control and status pseudo instructions
 613   void csrr(Register Rd, unsigned csr);         // read csr
 614   void csrw(unsigned csr, Register Rs);         // write csr
 615   void csrs(unsigned csr, Register Rs);         // set bits in csr
 616   void csrc(unsigned csr, Register Rs);         // clear bits in csr
 617   void csrwi(unsigned csr, unsigned imm);
 618   void csrsi(unsigned csr, unsigned imm);
 619   void csrci(unsigned csr, unsigned imm);
 620   void frcsr(Register Rd) { csrr(Rd, CSR_FCSR); }; // read float-point csr
 621   void fscsr(Register Rd, Register Rs);            // swap float-point csr
 622   void fscsr(Register Rs);                         // write float-point csr
 623   void frrm(Register Rd) { csrr(Rd, CSR_FRM); };   // read float-point rounding mode
 624   void fsrm(Register Rd, Register Rs);             // swap float-point rounding mode
 625   void fsrm(Register Rs);                          // write float-point rounding mode
 626   void fsrmi(Register Rd, unsigned imm);
 627   void fsrmi(unsigned imm);
 628   void frflags(Register Rd) { csrr(Rd, CSR_FFLAGS); }; // read float-point exception flags
 629   void fsflags(Register Rd, Register Rs);              // swap float-point exception flags
 630   void fsflags(Register Rs);                           // write float-point exception flags
 631   void fsflagsi(Register Rd, unsigned imm);
 632   void fsflagsi(unsigned imm);
 633   // Requires Zicntr
 634   void rdinstret(Register Rd) { csrr(Rd, CSR_INSTRET); }; // read instruction-retired counter
 635   void rdcycle(Register Rd)   { csrr(Rd, CSR_CYCLE); };   // read cycle counter
 636   void rdtime(Register Rd)    { csrr(Rd, CSR_TIME); };    // read time
 637 
 638   // Restore cpu control state after JNI call
 639   void restore_cpu_control_state_after_jni(Register tmp);
 640 
 641   // Control transfer pseudo instructions
 642   void beqz(Register Rs, const address dest);
 643   void bnez(Register Rs, const address dest);
 644   void blez(Register Rs, const address dest);
 645   void bgez(Register Rs, const address dest);
 646   void bltz(Register Rs, const address dest);
 647   void bgtz(Register Rs, const address dest);
 648 
 649   void cmov_eq(Register cmp1, Register cmp2, Register dst, Register src);
 650   void cmov_ne(Register cmp1, Register cmp2, Register dst, Register src);
 651   void cmov_le(Register cmp1, Register cmp2, Register dst, Register src);
 652   void cmov_leu(Register cmp1, Register cmp2, Register dst, Register src);
 653   void cmov_ge(Register cmp1, Register cmp2, Register dst, Register src);
 654   void cmov_geu(Register cmp1, Register cmp2, Register dst, Register src);
 655   void cmov_lt(Register cmp1, Register cmp2, Register dst, Register src);
 656   void cmov_ltu(Register cmp1, Register cmp2, Register dst, Register src);
 657   void cmov_gt(Register cmp1, Register cmp2, Register dst, Register src);
 658   void cmov_gtu(Register cmp1, Register cmp2, Register dst, Register src);
 659 
 660   void cmov_cmp_fp_eq(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 661   void cmov_cmp_fp_ne(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 662   void cmov_cmp_fp_le(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 663   void cmov_cmp_fp_ge(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 664   void cmov_cmp_fp_lt(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 665   void cmov_cmp_fp_gt(FloatRegister cmp1, FloatRegister cmp2, Register dst, Register src, bool is_single);
 666 
 667  public:
 668   // We try to follow risc-v asm menomics.
 669   // But as we don't layout a reachable GOT,
 670   // we often need to resort to movptr, li <48imm>.
 671   // https://github.com/riscv-non-isa/riscv-asm-manual/blob/main/src/asm-manual.adoc
 672 
 673   // Hotspot only use the standard calling convention using x1/ra.
 674   // The alternative calling convection using x5/t0 is not used.
 675   // Using x5 as a temp causes the CPU to mispredict returns.
 676 
 677   // JALR, return address stack updates:
 678   // | rd is x1/x5 | rs1 is x1/x5 | rd=rs1 | RAS action
 679   // | ----------- | ------------ | ------ |-------------
 680   // |     No      |      No      |   -    | None
 681   // |     No      |      Yes     |   -    | Pop
 682   // |     Yes     |      No      |   -    | Push
 683   // |     Yes     |      Yes     |   No   | Pop, then push
 684   // |     Yes     |      Yes     |   Yes  | Push
 685   //
 686   // JAL, return address stack updates:
 687   // | rd is x1/x5 | RAS action
 688   // | ----------- | ----------
 689   // |     Yes     | Push
 690   // |     No      | None
 691   //
 692   // JUMPs   uses Rd = x0/zero and Rs = x6/t1 or imm
 693   // CALLS   uses Rd = x1/ra   and Rs = x6/t1 or imm (or x1/ra*)
 694   // RETURNS uses Rd = x0/zero and Rs = x1/ra
 695   // *use of x1/ra should not normally be used, special case only.
 696 
 697   // jump: jal x0, offset
 698   // For long reach uses temp register for:
 699   // la + jr
 700   void j(const address dest, Register temp = t1);
 701   void j(const Address &dest, Register temp = t1);
 702   void j(Label &l, Register temp = noreg);
 703 
 704   // jump register: jalr x0, offset(rs)
 705   void jr(Register Rd, int32_t offset = 0);
 706 
 707   // call: la + jalr x1
 708   void call(const address dest, Register temp = t1);
 709 
 710   // jalr: jalr x1, offset(rs)
 711   void jalr(Register Rs, int32_t offset = 0);
 712 
 713   // Emit a runtime call. Only invalidates the tmp register which
 714   // is used to keep the entry address for jalr/movptr.
 715   // Uses call() for intra code cache, else movptr + jalr.
 716   // Clobebrs t1
 717   void rt_call(address dest, Register tmp = t1);
 718 
 719   // ret: jalr x0, 0(x1)
 720   inline void ret() {
 721     Assembler::jalr(x0, x1, 0);
 722   }
 723 
 724   //label
 725   void beqz(Register Rs, Label &l, bool is_far = false);
 726   void bnez(Register Rs, Label &l, bool is_far = false);
 727   void blez(Register Rs, Label &l, bool is_far = false);
 728   void bgez(Register Rs, Label &l, bool is_far = false);
 729   void bltz(Register Rs, Label &l, bool is_far = false);
 730   void bgtz(Register Rs, Label &l, bool is_far = false);
 731 
 732   void beq (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 733   void bne (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 734   void blt (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 735   void bge (Register Rs1, Register Rs2, Label &L, bool is_far = false);
 736   void bltu(Register Rs1, Register Rs2, Label &L, bool is_far = false);
 737   void bgeu(Register Rs1, Register Rs2, Label &L, bool is_far = false);
 738 
 739   void bgt (Register Rs, Register Rt, const address dest);
 740   void ble (Register Rs, Register Rt, const address dest);
 741   void bgtu(Register Rs, Register Rt, const address dest);
 742   void bleu(Register Rs, Register Rt, const address dest);
 743 
 744   void bgt (Register Rs, Register Rt, Label &l, bool is_far = false);
 745   void ble (Register Rs, Register Rt, Label &l, bool is_far = false);
 746   void bgtu(Register Rs, Register Rt, Label &l, bool is_far = false);
 747   void bleu(Register Rs, Register Rt, Label &l, bool is_far = false);
 748 
 749 #define INSN_ENTRY_RELOC(result_type, header)                               \
 750   result_type header {                                                      \
 751     guarantee(rtype == relocInfo::internal_word_type,                       \
 752               "only internal_word_type relocs make sense here");            \
 753     relocate(InternalAddress(dest).rspec());                                \
 754     IncompressibleScope scope(this);  /* relocations */
 755 
 756 #define INSN(NAME)                                                                                       \
 757   void NAME(Register Rs1, Register Rs2, const address dest) {                                            \
 758     assert_cond(dest != nullptr);                                                                        \
 759     int64_t offset = dest - pc();                                                                        \
 760     guarantee(is_simm13(offset) && is_even(offset),                                                      \
 761               "offset is invalid: is_simm_13: %s offset: " INT64_FORMAT,                                 \
 762               BOOL_TO_STR(is_simm13(offset)), offset);                                                   \
 763     Assembler::NAME(Rs1, Rs2, offset);                                                                   \
 764   }                                                                                                      \
 765   INSN_ENTRY_RELOC(void, NAME(Register Rs1, Register Rs2, address dest, relocInfo::relocType rtype))     \
 766     NAME(Rs1, Rs2, dest);                                                                                \
 767   }
 768 
 769   INSN(beq);
 770   INSN(bne);
 771   INSN(bge);
 772   INSN(bgeu);
 773   INSN(blt);
 774   INSN(bltu);
 775 
 776 #undef INSN
 777 
 778 #undef INSN_ENTRY_RELOC
 779 
 780   void float_beq(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 781   void float_bne(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 782   void float_ble(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 783   void float_bge(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 784   void float_blt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 785   void float_bgt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 786 
 787   void double_beq(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 788   void double_bne(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 789   void double_ble(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 790   void double_bge(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 791   void double_blt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 792   void double_bgt(FloatRegister Rs1, FloatRegister Rs2, Label &l, bool is_far = false, bool is_unordered = false);
 793 
 794 private:
 795   int push_reg(unsigned int bitset, Register stack);
 796   int pop_reg(unsigned int bitset, Register stack);
 797   int push_fp(unsigned int bitset, Register stack);
 798   int pop_fp(unsigned int bitset, Register stack);
 799 #ifdef COMPILER2
 800   int push_v(unsigned int bitset, Register stack);
 801   int pop_v(unsigned int bitset, Register stack);
 802 #endif // COMPILER2
 803 
 804   // The signed 20-bit upper imm can materialize at most negative 0xF...F80000000, two G.
 805   // The following signed 12-bit imm can at max subtract 0x800, two K, from that previously loaded two G.
 806   bool is_valid_32bit_offset(int64_t x) {
 807     constexpr int64_t twoG = (2 * G);
 808     constexpr int64_t twoK = (2 * K);
 809     return x < (twoG - twoK) && x >= (-twoG - twoK);
 810   }
 811 
 812   // Ensure that the auipc can reach the destination at x from anywhere within
 813   // the code cache so that if it is relocated we know it will still reach.
 814   bool is_32bit_offset_from_codecache(int64_t x) {
 815     int64_t low  = (int64_t)CodeCache::low_bound();
 816     int64_t high = (int64_t)CodeCache::high_bound();
 817     return is_valid_32bit_offset(x - low) && is_valid_32bit_offset(x - high);
 818   }
 819 
 820 public:
 821   void push_reg(Register Rs);
 822   void pop_reg(Register Rd);
 823   void push_reg(RegSet regs, Register stack) { if (regs.bits()) push_reg(regs.bits(), stack); }
 824   void pop_reg(RegSet regs, Register stack)  { if (regs.bits()) pop_reg(regs.bits(), stack); }
 825   void push_fp(FloatRegSet regs, Register stack) { if (regs.bits()) push_fp(regs.bits(), stack); }
 826   void pop_fp(FloatRegSet regs, Register stack)  { if (regs.bits()) pop_fp(regs.bits(), stack); }
 827 #ifdef COMPILER2
 828   void push_v(VectorRegSet regs, Register stack) { if (regs.bits()) push_v(regs.bits(), stack); }
 829   void pop_v(VectorRegSet regs, Register stack)  { if (regs.bits()) pop_v(regs.bits(), stack); }
 830 #endif // COMPILER2
 831 
 832   // Push and pop everything that might be clobbered by a native
 833   // runtime call except t0 and t1. (They are always
 834   // temporary registers, so we don't have to protect them.)
 835   // Additional registers can be excluded in a passed RegSet.
 836   void push_call_clobbered_registers_except(RegSet exclude);
 837   void pop_call_clobbered_registers_except(RegSet exclude);
 838 
 839   void push_call_clobbered_registers() {
 840     push_call_clobbered_registers_except(RegSet());
 841   }
 842   void pop_call_clobbered_registers() {
 843     pop_call_clobbered_registers_except(RegSet());
 844   }
 845 
 846   void push_CPU_state(bool save_vectors = false, int vector_size_in_bytes = 0);
 847   void pop_CPU_state(bool restore_vectors = false, int vector_size_in_bytes = 0);
 848 
 849   void push_cont_fastpath(Register java_thread = xthread);
 850   void pop_cont_fastpath(Register java_thread = xthread);
 851 
 852   // if heap base register is used - reinit it with the correct value
 853   void reinit_heapbase();
 854 
 855   void bind(Label& L) {
 856     Assembler::bind(L);
 857     // fences across basic blocks should not be merged
 858     code()->clear_last_insn();
 859   }
 860 
 861   typedef void (MacroAssembler::* compare_and_branch_insn)(Register Rs1, Register Rs2, const address dest);
 862   typedef void (MacroAssembler::* compare_and_branch_label_insn)(Register Rs1, Register Rs2, Label &L, bool is_far);
 863   typedef void (MacroAssembler::* jal_jalr_insn)(Register Rt, address dest);
 864 
 865   void wrap_label(Register r, Label &L, jal_jalr_insn insn);
 866   void wrap_label(Register r1, Register r2, Label &L,
 867                   compare_and_branch_insn insn,
 868                   compare_and_branch_label_insn neg_insn, bool is_far = false);
 869 
 870   void la(Register Rd, Label &label);
 871   void la(Register Rd, const address addr);
 872   void la(Register Rd, const address addr, int32_t &offset);
 873   void la(Register Rd, const Address &adr);
 874 
 875   void li16u(Register Rd, uint16_t imm);
 876   void li32(Register Rd, int32_t imm);
 877   void li  (Register Rd, int64_t imm);  // optimized load immediate
 878 
 879   // mv
 880   void mv(Register Rd, address addr)                  { li(Rd, (int64_t)addr); }
 881   void mv(Register Rd, address addr, int32_t &offset) {
 882     // Split address into a lower 12-bit sign-extended offset and the remainder,
 883     // so that the offset could be encoded in jalr or load/store instruction.
 884     offset = ((int32_t)(int64_t)addr << 20) >> 20;
 885     li(Rd, (int64_t)addr - offset);
 886   }
 887 
 888   template<typename T, ENABLE_IF(std::is_integral<T>::value)>
 889   inline void mv(Register Rd, T o)                    { li(Rd, (int64_t)o); }
 890 
 891   void mv(Register Rd, RegisterOrConstant src) {
 892     if (src.is_register()) {
 893       mv(Rd, src.as_register());
 894     } else {
 895       mv(Rd, src.as_constant());
 896     }
 897   }
 898 
 899   // Generates a load of a 48-bit constant which can be
 900   // patched to any 48-bit constant, i.e. address.
 901   // If common case supply additional temp register
 902   // to shorten the instruction sequence.
 903   void movptr(Register Rd, const Address &addr, Register tmp = noreg);
 904   void movptr(Register Rd, address addr, Register tmp = noreg);
 905   void movptr(Register Rd, address addr, int32_t &offset, Register tmp = noreg);
 906 
 907  private:
 908   void movptr1(Register Rd, uintptr_t addr, int32_t &offset);
 909   void movptr2(Register Rd, uintptr_t addr, int32_t &offset, Register tmp);
 910  public:
 911   // float imm move
 912   static bool can_hf_imm_load(short imm);
 913   static bool can_fp_imm_load(float imm);
 914   static bool can_dp_imm_load(double imm);
 915   void fli_h(FloatRegister Rd, short imm);
 916   void fli_s(FloatRegister Rd, float imm);
 917   void fli_d(FloatRegister Rd, double imm);
 918 
 919   // arith
 920   void add (Register Rd, Register Rn, int64_t increment, Register tmp = t0);
 921   void sub (Register Rd, Register Rn, int64_t decrement, Register tmp = t0);
 922   void addw(Register Rd, Register Rn, int64_t increment, Register tmp = t0);
 923   void subw(Register Rd, Register Rn, int64_t decrement, Register tmp = t0);
 924 
 925   void subi(Register Rd, Register Rn, int64_t decrement) {
 926     assert(is_simm12(-decrement), "Must be");
 927     addi(Rd, Rn, -decrement);
 928   }
 929 
 930   void subiw(Register Rd, Register Rn, int64_t decrement) {
 931     assert(is_simm12(-decrement), "Must be");
 932     addiw(Rd, Rn, -decrement);
 933   }
 934 
 935 #define INSN(NAME)                                               \
 936   inline void NAME(Register Rd, Register Rs1, Register Rs2) {    \
 937     Assembler::NAME(Rd, Rs1, Rs2);                               \
 938   }
 939 
 940   INSN(add);
 941   INSN(addw);
 942   INSN(sub);
 943   INSN(subw);
 944 
 945 #undef INSN
 946 
 947   // logic
 948   void andrw(Register Rd, Register Rs1, Register Rs2);
 949   void orrw(Register Rd, Register Rs1, Register Rs2);
 950   void xorrw(Register Rd, Register Rs1, Register Rs2);
 951 
 952   // logic with negate
 953   void andn(Register Rd, Register Rs1, Register Rs2);
 954   void orn(Register Rd, Register Rs1, Register Rs2);
 955 
 956   // reverse bytes
 957   void revbw(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2= t1);  // reverse bytes in lower word, sign-extend
 958   void revb(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);  // reverse bytes in doubleword
 959 
 960   void ror(Register dst, Register src, Register shift, Register tmp = t0);
 961   void ror(Register dst, Register src, uint32_t shift, Register tmp = t0);
 962   void rolw(Register dst, Register src, uint32_t shift, Register tmp = t0);
 963 
 964   void orptr(Address adr, RegisterOrConstant src, Register tmp1 = t0, Register tmp2 = t1);
 965 
 966 // Load and Store Instructions
 967 #define INSN_ENTRY_RELOC(result_type, header)                               \
 968   result_type header {                                                      \
 969     guarantee(rtype == relocInfo::internal_word_type,                       \
 970               "only internal_word_type relocs make sense here");            \
 971     relocate(InternalAddress(dest).rspec());                                \
 972     IncompressibleScope scope(this);  /* relocations */
 973 
 974 #define INSN(NAME)                                                                                 \
 975   void NAME(Register Rd, address dest) {                                                           \
 976     assert_cond(dest != nullptr);                                                                  \
 977     if (CodeCache::contains(dest)) {                                                               \
 978       int64_t distance = dest - pc();                                                              \
 979       assert(is_valid_32bit_offset(distance), "Must be");                                          \
 980       auipc(Rd, (int32_t)distance + 0x800);                                                        \
 981       Assembler::NAME(Rd, Rd, ((int32_t)distance << 20) >> 20);                                    \
 982     } else {                                                                                       \
 983       int32_t offset = 0;                                                                          \
 984       movptr(Rd, dest, offset);                                                                    \
 985       Assembler::NAME(Rd, Rd, offset);                                                             \
 986     }                                                                                              \
 987   }                                                                                                \
 988   INSN_ENTRY_RELOC(void, NAME(Register Rd, address dest, relocInfo::relocType rtype))              \
 989     NAME(Rd, dest);                                                                                \
 990   }                                                                                                \
 991   void NAME(Register Rd, const Address &adr, Register temp = t0) {                                 \
 992     switch (adr.getMode()) {                                                                       \
 993       case Address::literal: {                                                                     \
 994         relocate(adr.rspec(), [&] {                                                                \
 995           NAME(Rd, adr.target());                                                                  \
 996         });                                                                                        \
 997         break;                                                                                     \
 998       }                                                                                            \
 999       case Address::base_plus_offset: {                                                            \
1000         if (is_simm12(adr.offset())) {                                                             \
1001           Assembler::NAME(Rd, adr.base(), adr.offset());                                           \
1002         } else {                                                                                   \
1003           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1004           if (Rd == adr.base()) {                                                                  \
1005             la(temp, Address(adr.base(), adr.offset() - offset));                                  \
1006             Assembler::NAME(Rd, temp, offset);                                                     \
1007           } else {                                                                                 \
1008             la(Rd, Address(adr.base(), adr.offset() - offset));                                    \
1009             Assembler::NAME(Rd, Rd, offset);                                                       \
1010           }                                                                                        \
1011         }                                                                                          \
1012         break;                                                                                     \
1013       }                                                                                            \
1014       default:                                                                                     \
1015         ShouldNotReachHere();                                                                      \
1016     }                                                                                              \
1017   }                                                                                                \
1018   void NAME(Register Rd, Label &L) {                                                               \
1019     wrap_label(Rd, L, &MacroAssembler::NAME);                                                      \
1020   }
1021 
1022   INSN(lb);
1023   INSN(lbu);
1024   INSN(lh);
1025   INSN(lhu);
1026   INSN(lw);
1027   INSN(lwu);
1028   INSN(ld);
1029 
1030 #undef INSN
1031 
1032 #define INSN(NAME)                                                                                 \
1033   void NAME(FloatRegister Rd, address dest, Register temp = t0) {                                  \
1034     assert_cond(dest != nullptr);                                                                  \
1035     if (CodeCache::contains(dest)) {                                                               \
1036       int64_t distance = dest - pc();                                                              \
1037       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1038       auipc(temp, (int32_t)distance + 0x800);                                                      \
1039       Assembler::NAME(Rd, temp, ((int32_t)distance << 20) >> 20);                                  \
1040     } else {                                                                                       \
1041       int32_t offset = 0;                                                                          \
1042       movptr(temp, dest, offset);                                                                  \
1043       Assembler::NAME(Rd, temp, offset);                                                           \
1044     }                                                                                              \
1045   }                                                                                                \
1046   INSN_ENTRY_RELOC(void, NAME(FloatRegister Rd, address dest,                                      \
1047                               relocInfo::relocType rtype, Register temp = t0))                     \
1048     NAME(Rd, dest, temp);                                                                          \
1049   }                                                                                                \
1050   void NAME(FloatRegister Rd, const Address &adr, Register temp = t0) {                            \
1051     switch (adr.getMode()) {                                                                       \
1052       case Address::literal: {                                                                     \
1053         relocate(adr.rspec(), [&] {                                                                \
1054           NAME(Rd, adr.target(), temp);                                                            \
1055         });                                                                                        \
1056         break;                                                                                     \
1057       }                                                                                            \
1058       case Address::base_plus_offset: {                                                            \
1059         if (is_simm12(adr.offset())) {                                                             \
1060           Assembler::NAME(Rd, adr.base(), adr.offset());                                           \
1061         } else {                                                                                   \
1062           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1063           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1064           Assembler::NAME(Rd, temp, offset);                                                       \
1065         }                                                                                          \
1066         break;                                                                                     \
1067       }                                                                                            \
1068       default:                                                                                     \
1069         ShouldNotReachHere();                                                                      \
1070     }                                                                                              \
1071   }
1072 
1073   INSN(flh);
1074   INSN(flw);
1075   INSN(fld);
1076 
1077 #undef INSN
1078 
1079 #define INSN(NAME, REGISTER)                                                                       \
1080   INSN_ENTRY_RELOC(void, NAME(REGISTER Rs, address dest,                                           \
1081                               relocInfo::relocType rtype, Register temp = t0))                     \
1082     NAME(Rs, dest, temp);                                                                          \
1083   }
1084 
1085   INSN(sb,  Register);
1086   INSN(sh,  Register);
1087   INSN(sw,  Register);
1088   INSN(sd,  Register);
1089   INSN(fsw, FloatRegister);
1090   INSN(fsd, FloatRegister);
1091 
1092 #undef INSN
1093 
1094 #define INSN(NAME)                                                                                 \
1095   void NAME(Register Rs, address dest, Register temp = t0) {                                       \
1096     assert_cond(dest != nullptr);                                                                  \
1097     assert_different_registers(Rs, temp);                                                          \
1098     if (CodeCache::contains(dest)) {                                                               \
1099       int64_t distance = dest - pc();                                                              \
1100       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1101       auipc(temp, (int32_t)distance + 0x800);                                                      \
1102       Assembler::NAME(Rs, temp, ((int32_t)distance << 20) >> 20);                                  \
1103     } else {                                                                                       \
1104       int32_t offset = 0;                                                                          \
1105       movptr(temp, dest, offset);                                                                  \
1106       Assembler::NAME(Rs, temp, offset);                                                           \
1107     }                                                                                              \
1108   }                                                                                                \
1109   void NAME(Register Rs, const Address &adr, Register temp = t0) {                                 \
1110     switch (adr.getMode()) {                                                                       \
1111       case Address::literal: {                                                                     \
1112         assert_different_registers(Rs, temp);                                                      \
1113         relocate(adr.rspec(), [&] {                                                                \
1114           NAME(Rs, adr.target(), temp);                                                            \
1115         });                                                                                        \
1116         break;                                                                                     \
1117       }                                                                                            \
1118       case Address::base_plus_offset: {                                                            \
1119         if (is_simm12(adr.offset())) {                                                             \
1120           Assembler::NAME(Rs, adr.base(), adr.offset());                                           \
1121         } else {                                                                                   \
1122           assert_different_registers(Rs, temp);                                                    \
1123           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1124           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1125           Assembler::NAME(Rs, temp, offset);                                                       \
1126         }                                                                                          \
1127         break;                                                                                     \
1128       }                                                                                            \
1129       default:                                                                                     \
1130         ShouldNotReachHere();                                                                      \
1131     }                                                                                              \
1132   }
1133 
1134   INSN(sb);
1135   INSN(sh);
1136   INSN(sw);
1137   INSN(sd);
1138 
1139 #undef INSN
1140 
1141 #define INSN(NAME)                                                                                 \
1142   void NAME(FloatRegister Rs, address dest, Register temp = t0) {                                  \
1143     assert_cond(dest != nullptr);                                                                  \
1144     if (CodeCache::contains(dest)) {                                                               \
1145       int64_t distance = dest - pc();                                                              \
1146       assert(is_valid_32bit_offset(distance), "Must be");                                          \
1147       auipc(temp, (int32_t)distance + 0x800);                                                      \
1148       Assembler::NAME(Rs, temp, ((int32_t)distance << 20) >> 20);                                  \
1149     } else {                                                                                       \
1150       int32_t offset = 0;                                                                          \
1151       movptr(temp, dest, offset);                                                                  \
1152       Assembler::NAME(Rs, temp, offset);                                                           \
1153     }                                                                                              \
1154   }                                                                                                \
1155   void NAME(FloatRegister Rs, const Address &adr, Register temp = t0) {                            \
1156     switch (adr.getMode()) {                                                                       \
1157       case Address::literal: {                                                                     \
1158         relocate(adr.rspec(), [&] {                                                                \
1159           NAME(Rs, adr.target(), temp);                                                            \
1160         });                                                                                        \
1161         break;                                                                                     \
1162       }                                                                                            \
1163       case Address::base_plus_offset: {                                                            \
1164         if (is_simm12(adr.offset())) {                                                             \
1165           Assembler::NAME(Rs, adr.base(), adr.offset());                                           \
1166         } else {                                                                                   \
1167           int32_t offset = ((int32_t)adr.offset() << 20) >> 20;                                    \
1168           la(temp, Address(adr.base(), adr.offset() - offset));                                    \
1169           Assembler::NAME(Rs, temp, offset);                                                       \
1170         }                                                                                          \
1171         break;                                                                                     \
1172       }                                                                                            \
1173       default:                                                                                     \
1174         ShouldNotReachHere();                                                                      \
1175     }                                                                                              \
1176   }
1177 
1178   INSN(fsw);
1179   INSN(fsd);
1180 
1181 #undef INSN
1182 
1183 #undef INSN_ENTRY_RELOC
1184 
1185   void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp, Label &succeed, Label *fail);
1186   void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp, Label &succeed, Label *fail);
1187   void cmpxchg(Register addr, Register expected,
1188                Register new_val,
1189                Assembler::operand_size size,
1190                Assembler::Aqrl acquire, Assembler::Aqrl release,
1191                Register result, bool result_as_bool = false);
1192   void weak_cmpxchg(Register addr, Register expected,
1193                     Register new_val,
1194                     Assembler::operand_size size,
1195                     Assembler::Aqrl acquire, Assembler::Aqrl release,
1196                     Register result);
1197   void cmpxchg_narrow_value_helper(Register addr, Register expected, Register new_val,
1198                                    Assembler::operand_size size,
1199                                    Register shift, Register mask, Register aligned_addr);
1200   void cmpxchg_narrow_value(Register addr, Register expected,
1201                             Register new_val,
1202                             Assembler::operand_size size,
1203                             Assembler::Aqrl acquire, Assembler::Aqrl release,
1204                             Register result, bool result_as_bool,
1205                             Register tmp1, Register tmp2, Register tmp3);
1206   void weak_cmpxchg_narrow_value(Register addr, Register expected,
1207                                  Register new_val,
1208                                  Assembler::operand_size size,
1209                                  Assembler::Aqrl acquire, Assembler::Aqrl release,
1210                                  Register result,
1211                                  Register tmp1, Register tmp2, Register tmp3);
1212 
1213   void atomic_add(Register prev, RegisterOrConstant incr, Register addr);
1214   void atomic_addw(Register prev, RegisterOrConstant incr, Register addr);
1215   void atomic_addal(Register prev, RegisterOrConstant incr, Register addr);
1216   void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr);
1217 
1218   void atomic_xchg(Register prev, Register newv, Register addr);
1219   void atomic_xchgw(Register prev, Register newv, Register addr);
1220   void atomic_xchgal(Register prev, Register newv, Register addr);
1221   void atomic_xchgalw(Register prev, Register newv, Register addr);
1222   void atomic_xchgwu(Register prev, Register newv, Register addr);
1223   void atomic_xchgalwu(Register prev, Register newv, Register addr);
1224 
1225   void atomic_cas(Register prev, Register newv, Register addr, Assembler::operand_size size,
1226               Assembler::Aqrl acquire = Assembler::relaxed, Assembler::Aqrl release = Assembler::relaxed);
1227 
1228   // Emit a far call/jump. Only invalidates the tmp register which
1229   // is used to keep the entry address for jalr.
1230   // The address must be inside the code cache.
1231   // Supported entry.rspec():
1232   // - relocInfo::external_word_type
1233   // - relocInfo::runtime_call_type
1234   // - relocInfo::none
1235   // Clobbers t1 default.
1236   void far_call(const Address &entry, Register tmp = t1);
1237   void far_jump(const Address &entry, Register tmp = t1);
1238 
1239   static int far_branch_size() {
1240       return 2 * MacroAssembler::instruction_size;  // auipc + jalr, see far_call() & far_jump()
1241   }
1242 
1243   void load_byte_map_base(Register reg);
1244 
1245   void bang_stack_with_offset(int offset) {
1246     // stack grows down, caller passes positive offset
1247     assert(offset > 0, "must bang with negative offset");
1248     sub(t0, sp, offset);
1249     sd(zr, Address(t0));
1250   }
1251 
1252   virtual void _call_Unimplemented(address call_site) {
1253     mv(t1, call_site);
1254   }
1255 
1256   #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
1257 
1258   // Frame creation and destruction shared between JITs.
1259   void build_frame(int framesize);
1260   void remove_frame(int framesize);
1261 
1262   void reserved_stack_check();
1263 
1264   void get_polling_page(Register dest, relocInfo::relocType rtype);
1265   void read_polling_page(Register r, int32_t offset, relocInfo::relocType rtype);
1266 
1267   // RISCV64 OpenJDK uses three different types of calls:
1268   //
1269   //   - far call: auipc reg, pc_relative_offset; jalr ra, reg, offset
1270   //     The offset has the range [-(2G + 2K), 2G - 2K). Addresses out of the
1271   //     range in the code cache requires indirect call.
1272   //     If a jump is needed rather than a call, a far jump 'jalr x0, reg, offset'
1273   //     can be used instead.
1274   //     All instructions are embedded at a call site.
1275   //
1276   //   - indirect call: movptr + jalr
1277   //     This can reach anywhere in the address space, but it cannot be patched
1278   //     while code is running, so it must only be modified at a safepoint.
1279   //     This form of call is most suitable for targets at fixed addresses,
1280   //     which will never be patched.
1281   //
1282   //   - reloc call:
1283   //     This too can reach anywhere in the address space but is only available
1284   //     in C1/C2-generated code (nmethod).
1285   //
1286   //     [Main code section]
1287   //       auipc
1288   //       ld <address_from_stub_section>
1289   //       jalr
1290   //
1291   //     [Stub section]
1292   //     address stub:
1293   //       <64-bit destination address>
1294   //
1295   //    To change the destination we simply atomically store the new
1296   //    address in the stub section.
1297   //    There is a benign race in that the other thread might observe the old
1298   //    64-bit destination address before it observes the new address. That does
1299   //    not matter because the destination method has been invalidated, so there
1300   //    will be a trap at its start.
1301 
1302   // Emit a reloc call and create a stub to hold the entry point address.
1303   // Supported entry.rspec():
1304   // - relocInfo::runtime_call_type
1305   // - relocInfo::opt_virtual_call_type
1306   // - relocInfo::static_call_type
1307   // - relocInfo::virtual_call_type
1308   //
1309   // Return: the call PC or nullptr if CodeCache is full.
1310   address reloc_call(Address entry, Register tmp = t1);
1311 
1312   address ic_call(address entry, jint method_index = 0);
1313   static int ic_check_size();
1314   int ic_check(int end_alignment = MacroAssembler::instruction_size);
1315 
1316   // Support for memory inc/dec
1317   // n.b. increment/decrement calls with an Address destination will
1318   // need to use a scratch register to load the value to be
1319   // incremented. increment/decrement calls which add or subtract a
1320   // constant value other than sign-extended 12-bit immediate will need
1321   // to use a 2nd scratch register to hold the constant. so, an address
1322   // increment/decrement may trash both t0 and t1.
1323 
1324   void increment(const Address dst, int64_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1325   void incrementw(const Address dst, int32_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1326 
1327   void decrement(const Address dst, int64_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1328   void decrementw(const Address dst, int32_t value = 1, Register tmp1 = t0, Register tmp2 = t1);
1329 
1330   void cmpptr(Register src1, const Address &src2, Label& equal, Register tmp = t0);
1331 
1332   void clinit_barrier(Register klass, Register tmp, Label* L_fast_path = nullptr, Label* L_slow_path = nullptr);
1333   void load_method_holder_cld(Register result, Register method);
1334   void load_method_holder(Register holder, Register method);
1335 
1336   void compute_index(Register str1, Register trailing_zeros, Register match_mask,
1337                      Register result, Register char_tmp, Register tmp,
1338                      bool haystack_isL);
1339   void compute_match_mask(Register src, Register pattern, Register match_mask,
1340                           Register mask1, Register mask2);
1341 
1342   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1343   void kernel_crc32(Register crc, Register buf, Register len,
1344         Register table0, Register table1, Register table2, Register table3,
1345         Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register tmp6);
1346   void update_word_crc32(Register crc, Register v, Register tmp1, Register tmp2, Register tmp3,
1347         Register table0, Register table1, Register table2, Register table3,
1348         bool upper);
1349   void update_byte_crc32(Register crc, Register val, Register table);
1350 
1351 #ifdef COMPILER2
1352   void vector_update_crc32(Register crc, Register buf, Register len,
1353                            Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
1354                            Register table0, Register table3);
1355   void kernel_crc32_vclmul_fold(Register crc, Register buf, Register len,
1356               Register table0, Register table1, Register table2, Register table3,
1357               Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1358   void crc32_vclmul_fold_to_16_bytes_vectorsize_32(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1359                             VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4);
1360   void kernel_crc32_vclmul_fold_vectorsize_32(Register crc, Register buf, Register len,
1361                                               Register vclmul_table, Register tmp1, Register tmp2);
1362   void crc32_vclmul_fold_16_bytes_vectorsize_16(VectorRegister vx, VectorRegister vt,
1363                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1364                       Register buf, Register tmp, const int STEP);
1365   void crc32_vclmul_fold_16_bytes_vectorsize_16_2(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1366                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1367                       Register tmp);
1368   void crc32_vclmul_fold_16_bytes_vectorsize_16_3(VectorRegister vx, VectorRegister vy, VectorRegister vt,
1369                       VectorRegister vtmp1, VectorRegister vtmp2, VectorRegister vtmp3, VectorRegister vtmp4,
1370                       Register tmp);
1371   void kernel_crc32_vclmul_fold_vectorsize_16(Register crc, Register buf, Register len,
1372                                               Register vclmul_table, Register tmp1, Register tmp2);
1373 
1374   void mul_add(Register out, Register in, Register offset,
1375                Register len, Register k, Register tmp);
1376   void wide_mul(Register prod_lo, Register prod_hi, Register n, Register m);
1377   void wide_madd(Register sum_lo, Register sum_hi, Register n,
1378                  Register m, Register tmp1, Register tmp2);
1379   void cad(Register dst, Register src1, Register src2, Register carry);
1380   void cadc(Register dst, Register src1, Register src2, Register carry);
1381   void adc(Register dst, Register src1, Register src2, Register carry);
1382   void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo,
1383                        Register src1, Register src2, Register carry);
1384   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1385                              Register y, Register y_idx, Register z,
1386                              Register carry, Register product,
1387                              Register idx, Register kdx);
1388   void multiply_128_x_128_loop(Register y, Register z,
1389                                Register carry, Register carry2,
1390                                Register idx, Register jdx,
1391                                Register yz_idx1, Register yz_idx2,
1392                                Register tmp, Register tmp3, Register tmp4,
1393                                Register tmp6, Register product_hi);
1394   void multiply_to_len(Register x, Register xlen, Register y, Register ylen,
1395                        Register z, Register tmp0,
1396                        Register tmp1, Register tmp2, Register tmp3, Register tmp4,
1397                        Register tmp5, Register tmp6, Register product_hi);
1398 
1399 #endif // COMPILER2
1400 
1401   void inflate_lo32(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);
1402   void inflate_hi32(Register Rd, Register Rs, Register tmp1 = t0, Register tmp2 = t1);
1403 
1404   void ctzc_bits(Register Rd, Register Rs, bool isLL = false,
1405                  Register tmp1 = t0, Register tmp2 = t1);
1406 
1407   void zero_words(Register base, uint64_t cnt);
1408   address zero_words(Register ptr, Register cnt);
1409   void fill_words(Register base, Register cnt, Register value);
1410   void zero_memory(Register addr, Register len, Register tmp);
1411   void zero_dcache_blocks(Register base, Register cnt, Register tmp1, Register tmp2);
1412 
1413   // shift left by shamt and add
1414   void shadd(Register Rd, Register Rs1, Register Rs2, Register tmp, int shamt);
1415 
1416   // test single bit in Rs, result is set to Rd
1417   void test_bit(Register Rd, Register Rs, uint32_t bit_pos);
1418 
1419   // Here the float instructions with safe deal with some exceptions.
1420   // e.g. convert from NaN, +Inf, -Inf to int, float, double
1421   // will trigger exception, we need to deal with these situations
1422   // to get correct results.
1423   void fcvt_w_s_safe(Register dst, FloatRegister src, Register tmp = t0);
1424   void fcvt_l_s_safe(Register dst, FloatRegister src, Register tmp = t0);
1425   void fcvt_w_d_safe(Register dst, FloatRegister src, Register tmp = t0);
1426   void fcvt_l_d_safe(Register dst, FloatRegister src, Register tmp = t0);
1427 
1428   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1429   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1430 
1431   // Helper routine processing the slow path of NaN when converting float to float16
1432   void float_to_float16_NaN(Register dst, FloatRegister src, Register tmp1, Register tmp2);
1433 
1434   // vector load/store unit-stride instructions
1435   void vlex_v(VectorRegister vd, Register base, Assembler::SEW sew, VectorMask vm = unmasked) {
1436     switch (sew) {
1437       case Assembler::e64:
1438         vle64_v(vd, base, vm);
1439         break;
1440       case Assembler::e32:
1441         vle32_v(vd, base, vm);
1442         break;
1443       case Assembler::e16:
1444         vle16_v(vd, base, vm);
1445         break;
1446       case Assembler::e8: // fall through
1447       default:
1448         vle8_v(vd, base, vm);
1449         break;
1450     }
1451   }
1452 
1453   void vsex_v(VectorRegister store_data, Register base, Assembler::SEW sew, VectorMask vm = unmasked) {
1454     switch (sew) {
1455       case Assembler::e64:
1456         vse64_v(store_data, base, vm);
1457         break;
1458       case Assembler::e32:
1459         vse32_v(store_data, base, vm);
1460         break;
1461       case Assembler::e16:
1462         vse16_v(store_data, base, vm);
1463         break;
1464       case Assembler::e8: // fall through
1465       default:
1466         vse8_v(store_data, base, vm);
1467         break;
1468     }
1469   }
1470 
1471   // vector pseudo instructions
1472   // rotate vector register left with shift bits, 32-bit version
1473   inline void vrole32_vi(VectorRegister vd, uint32_t shift, VectorRegister tmp_vr) {
1474     vsrl_vi(tmp_vr, vd, 32 - shift);
1475     vsll_vi(vd, vd, shift);
1476     vor_vv(vd, vd, tmp_vr);
1477   }
1478 
1479   inline void vl1r_v(VectorRegister vd, Register rs) {
1480     vl1re8_v(vd, rs);
1481   }
1482 
1483   inline void vmnot_m(VectorRegister vd, VectorRegister vs) {
1484     vmnand_mm(vd, vs, vs);
1485   }
1486 
1487   inline void vncvt_x_x_w(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1488     vnsrl_wx(vd, vs, x0, vm);
1489   }
1490 
1491   inline void vneg_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1492     vrsub_vx(vd, vs, x0, vm);
1493   }
1494 
1495   inline void vfneg_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1496     vfsgnjn_vv(vd, vs, vs, vm);
1497   }
1498 
1499   inline void vfabs_v(VectorRegister vd, VectorRegister vs, VectorMask vm = unmasked) {
1500     vfsgnjx_vv(vd, vs, vs, vm);
1501   }
1502 
1503   inline void vmsgt_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1504     vmslt_vv(vd, vs1, vs2, vm);
1505   }
1506 
1507   inline void vmsgtu_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1508     vmsltu_vv(vd, vs1, vs2, vm);
1509   }
1510 
1511   inline void vmsge_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1512     vmsle_vv(vd, vs1, vs2, vm);
1513   }
1514 
1515   inline void vmsgeu_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1516     vmsleu_vv(vd, vs1, vs2, vm);
1517   }
1518 
1519   inline void vmfgt_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1520     vmflt_vv(vd, vs1, vs2, vm);
1521   }
1522 
1523   inline void vmfge_vv(VectorRegister vd, VectorRegister vs2, VectorRegister vs1, VectorMask vm = unmasked) {
1524     vmfle_vv(vd, vs1, vs2, vm);
1525   }
1526 
1527   inline void vmsltu_vi(VectorRegister Vd, VectorRegister Vs2, uint32_t imm, VectorMask vm = unmasked) {
1528     guarantee(imm >= 1 && imm <= 16, "imm is invalid");
1529     vmsleu_vi(Vd, Vs2, imm-1, vm);
1530   }
1531 
1532   inline void vmsgeu_vi(VectorRegister Vd, VectorRegister Vs2, uint32_t imm, VectorMask vm = unmasked) {
1533     guarantee(imm >= 1 && imm <= 16, "imm is invalid");
1534     vmsgtu_vi(Vd, Vs2, imm-1, vm);
1535   }
1536 
1537   // Copy mask register
1538   inline void vmmv_m(VectorRegister vd, VectorRegister vs) {
1539     vmand_mm(vd, vs, vs);
1540   }
1541 
1542   // Clear mask register
1543   inline void vmclr_m(VectorRegister vd) {
1544     vmxor_mm(vd, vd, vd);
1545   }
1546 
1547   // Set mask register
1548   inline void vmset_m(VectorRegister vd) {
1549     vmxnor_mm(vd, vd, vd);
1550   }
1551 
1552   inline void vnot_v(VectorRegister Vd, VectorRegister Vs, VectorMask vm = unmasked) {
1553     vxor_vi(Vd, Vs, -1, vm);
1554   }
1555 
1556   static const int zero_words_block_size;
1557 
1558   void cast_primitive_type(BasicType type, Register Rt) {
1559     switch (type) {
1560       case T_BOOLEAN:
1561         sltu(Rt, zr, Rt);
1562         break;
1563       case T_CHAR   :
1564         zext(Rt, Rt, 16);
1565         break;
1566       case T_BYTE   :
1567         sext(Rt, Rt, 8);
1568         break;
1569       case T_SHORT  :
1570         sext(Rt, Rt, 16);
1571         break;
1572       case T_INT    :
1573         sext(Rt, Rt, 32);
1574         break;
1575       case T_LONG   : /* nothing to do */        break;
1576       case T_VOID   : /* nothing to do */        break;
1577       case T_FLOAT  : /* nothing to do */        break;
1578       case T_DOUBLE : /* nothing to do */        break;
1579       default: ShouldNotReachHere();
1580     }
1581   }
1582 
1583   // float cmp with unordered_result
1584   void float_compare(Register result, FloatRegister Rs1, FloatRegister Rs2, int unordered_result);
1585   void double_compare(Register result, FloatRegister Rs1, FloatRegister Rs2, int unordered_result);
1586 
1587   // Zero/Sign-extend
1588   void zext(Register dst, Register src, int bits);
1589   void sext(Register dst, Register src, int bits);
1590 
1591 private:
1592   void cmp_x2i(Register dst, Register src1, Register src2, Register tmp, bool is_signed = true);
1593 
1594 public:
1595   // compare src1 and src2 and get -1/0/1 in dst.
1596   // if [src1 > src2], dst = 1;
1597   // if [src1 == src2], dst = 0;
1598   // if [src1 < src2], dst = -1;
1599   void cmp_l2i(Register dst, Register src1, Register src2, Register tmp = t0);
1600   void cmp_ul2i(Register dst, Register src1, Register src2, Register tmp = t0);
1601   void cmp_uw2i(Register dst, Register src1, Register src2, Register tmp = t0);
1602 
1603   // support for argument shuffling
1604   void move32_64(VMRegPair src, VMRegPair dst, Register tmp = t0);
1605   void float_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1606   void long_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1607   void double_move(VMRegPair src, VMRegPair dst, Register tmp = t0);
1608   void object_move(OopMap* map,
1609                    int oop_handle_offset,
1610                    int framesize_in_slots,
1611                    VMRegPair src,
1612                    VMRegPair dst,
1613                    bool is_receiver,
1614                    int* receiver_offset);
1615 
1616 #ifdef ASSERT
1617   // Template short-hand support to clean-up after a failed call to trampoline
1618   // call generation (see trampoline_call() below), when a set of Labels must
1619   // be reset (before returning).
1620   template<typename Label, typename... More>
1621   void reset_labels(Label& lbl, More&... more) {
1622     lbl.reset(); reset_labels(more...);
1623   }
1624   template<typename Label>
1625   void reset_labels(Label& lbl) {
1626     lbl.reset();
1627   }
1628 #endif
1629 
1630 private:
1631 
1632   void repne_scan(Register addr, Register value, Register count, Register tmp);
1633 
1634   int bitset_to_regs(unsigned int bitset, unsigned char* regs);
1635   Address add_memory_helper(const Address dst, Register tmp);
1636 
1637   void load_reserved(Register dst, Register addr, Assembler::operand_size size, Assembler::Aqrl acquire);
1638   void store_conditional(Register dst, Register new_val, Register addr, Assembler::operand_size size, Assembler::Aqrl release);
1639 
1640 public:
1641   void lightweight_lock(Register basic_lock, Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow);
1642   void lightweight_unlock(Register obj, Register tmp1, Register tmp2, Register tmp3, Label& slow);
1643 
1644 public:
1645   enum {
1646     // movptr
1647     movptr1_instruction_size = 6 * MacroAssembler::instruction_size, // lui, addi, slli, addi, slli, addi.  See movptr1().
1648     movptr2_instruction_size = 5 * MacroAssembler::instruction_size, // lui, lui, slli, add, addi.  See movptr2().
1649     load_pc_relative_instruction_size = 2 * MacroAssembler::instruction_size // auipc, ld
1650   };
1651 
1652   static bool is_load_pc_relative_at(address branch);
1653   static bool is_li16u_at(address instr);
1654 
1655   static bool is_jal_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1101111; }
1656   static bool is_jalr_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1100111 && extract_funct3(instr) == 0b000; }
1657   static bool is_branch_at(address instr)     { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b1100011; }
1658   static bool is_ld_at(address instr)         { assert_cond(instr != nullptr); return is_load_at(instr) && extract_funct3(instr) == 0b011; }
1659   static bool is_load_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0000011; }
1660   static bool is_float_load_at(address instr) { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0000111; }
1661   static bool is_auipc_at(address instr)      { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0010111; }
1662   static bool is_jump_at(address instr)       { assert_cond(instr != nullptr); return is_branch_at(instr) || is_jal_at(instr) || is_jalr_at(instr); }
1663   static bool is_add_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0110011 && extract_funct3(instr) == 0b000; }
1664   static bool is_addi_at(address instr)       { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0010011 && extract_funct3(instr) == 0b000; }
1665   static bool is_addiw_at(address instr)      { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0011011 && extract_funct3(instr) == 0b000; }
1666   static bool is_addiw_to_zr_at(address instr){ assert_cond(instr != nullptr); return is_addiw_at(instr) && extract_rd(instr) == zr; }
1667   static bool is_lui_at(address instr)        { assert_cond(instr != nullptr); return extract_opcode(instr) == 0b0110111; }
1668   static bool is_lui_to_zr_at(address instr)  { assert_cond(instr != nullptr); return is_lui_at(instr) && extract_rd(instr) == zr; }
1669 
1670   static bool is_srli_at(address instr) {
1671     assert_cond(instr != nullptr);
1672     return extract_opcode(instr) == 0b0010011 &&
1673            extract_funct3(instr) == 0b101 &&
1674            Assembler::extract(((unsigned*)instr)[0], 31, 26) == 0b000000;
1675   }
1676 
1677   static bool is_slli_shift_at(address instr, uint32_t shift) {
1678     assert_cond(instr != nullptr);
1679     return (extract_opcode(instr) == 0b0010011 && // opcode field
1680             extract_funct3(instr) == 0b001 &&     // funct3 field, select the type of operation
1681             Assembler::extract(Assembler::ld_instr(instr), 25, 20) == shift);    // shamt field
1682   }
1683 
1684   static bool is_movptr1_at(address instr);
1685   static bool is_movptr2_at(address instr);
1686 
1687   static bool is_lwu_to_zr(address instr);
1688 
1689   static Register extract_rs1(address instr);
1690   static Register extract_rs2(address instr);
1691   static Register extract_rd(address instr);
1692   static uint32_t extract_opcode(address instr);
1693   static uint32_t extract_funct3(address instr);
1694 
1695   // the instruction sequence of movptr is as below:
1696   //     lui
1697   //     addi
1698   //     slli
1699   //     addi
1700   //     slli
1701   //     addi/jalr/load
1702   static bool check_movptr1_data_dependency(address instr) {
1703     address lui = instr;
1704     address addi1 = lui + MacroAssembler::instruction_size;
1705     address slli1 = addi1 + MacroAssembler::instruction_size;
1706     address addi2 = slli1 + MacroAssembler::instruction_size;
1707     address slli2 = addi2 + MacroAssembler::instruction_size;
1708     address last_instr = slli2 + MacroAssembler::instruction_size;
1709     return extract_rs1(addi1) == extract_rd(lui) &&
1710            extract_rs1(addi1) == extract_rd(addi1) &&
1711            extract_rs1(slli1) == extract_rd(addi1) &&
1712            extract_rs1(slli1) == extract_rd(slli1) &&
1713            extract_rs1(addi2) == extract_rd(slli1) &&
1714            extract_rs1(addi2) == extract_rd(addi2) &&
1715            extract_rs1(slli2) == extract_rd(addi2) &&
1716            extract_rs1(slli2) == extract_rd(slli2) &&
1717            extract_rs1(last_instr) == extract_rd(slli2);
1718   }
1719 
1720   // the instruction sequence of movptr2 is as below:
1721   //     lui
1722   //     lui
1723   //     slli
1724   //     add
1725   //     addi/jalr/load
1726   static bool check_movptr2_data_dependency(address instr) {
1727     address lui1 = instr;
1728     address lui2 = lui1 + MacroAssembler::instruction_size;
1729     address slli = lui2 + MacroAssembler::instruction_size;
1730     address add  = slli + MacroAssembler::instruction_size;
1731     address last_instr = add + MacroAssembler::instruction_size;
1732     return extract_rd(add) == extract_rd(lui2) &&
1733            extract_rs1(add) == extract_rd(lui2) &&
1734            extract_rs2(add) == extract_rd(slli) &&
1735            extract_rs1(slli) == extract_rd(lui1) &&
1736            extract_rd(slli) == extract_rd(lui1) &&
1737            extract_rs1(last_instr) == extract_rd(add);
1738   }
1739 
1740   // the instruction sequence of li16u is as below:
1741   //     lui
1742   //     srli
1743   static bool check_li16u_data_dependency(address instr) {
1744     address lui = instr;
1745     address srli = lui + MacroAssembler::instruction_size;
1746 
1747     return extract_rs1(srli) == extract_rd(lui) &&
1748            extract_rs1(srli) == extract_rd(srli);
1749   }
1750 
1751   // the instruction sequence of li32 is as below:
1752   //     lui
1753   //     addiw
1754   static bool check_li32_data_dependency(address instr) {
1755     address lui = instr;
1756     address addiw = lui + MacroAssembler::instruction_size;
1757 
1758     return extract_rs1(addiw) == extract_rd(lui) &&
1759            extract_rs1(addiw) == extract_rd(addiw);
1760   }
1761 
1762   // the instruction sequence of pc-relative is as below:
1763   //     auipc
1764   //     jalr/addi/load/float_load
1765   static bool check_pc_relative_data_dependency(address instr) {
1766     address auipc = instr;
1767     address last_instr = auipc + MacroAssembler::instruction_size;
1768 
1769     return extract_rs1(last_instr) == extract_rd(auipc);
1770   }
1771 
1772   // the instruction sequence of load_label is as below:
1773   //     auipc
1774   //     load
1775   static bool check_load_pc_relative_data_dependency(address instr) {
1776     address auipc = instr;
1777     address load = auipc + MacroAssembler::instruction_size;
1778 
1779     return extract_rd(load) == extract_rd(auipc) &&
1780            extract_rs1(load) == extract_rd(load);
1781   }
1782 
1783   static bool is_li32_at(address instr);
1784   static bool is_pc_relative_at(address branch);
1785 
1786   static bool is_membar(address addr) {
1787     return (Bytes::get_native_u4(addr) & 0x7f) == 0b1111 && extract_funct3(addr) == 0;
1788   }
1789   static uint32_t get_membar_kind(address addr);
1790   static void set_membar_kind(address addr, uint32_t order_kind);
1791 };
1792 
1793 #ifdef ASSERT
1794 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; }
1795 #endif
1796 
1797 #endif // CPU_RISCV_MACROASSEMBLER_RISCV_HPP