1 /* 2 * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved. 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 6 * This code is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 only, as 8 * published by the Free Software Foundation. 9 * 10 * This code is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 13 * version 2 for more details (a copy is included in the LICENSE file that 14 * accompanied this code). 15 * 16 * You should have received a copy of the GNU General Public License version 17 * 2 along with this work; if not, write to the Free Software Foundation, 18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 21 * or visit www.oracle.com if you need additional information or have any 22 * questions. 23 * 24 */ 25 26 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP 27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP 28 29 #include "asm/assembler.inline.hpp" 30 #include "code/aotCodeCache.hpp" 31 #include "code/vmreg.hpp" 32 #include "metaprogramming/enableIf.hpp" 33 #include "oops/compressedOops.hpp" 34 #include "oops/compressedKlass.hpp" 35 #include "runtime/vm_version.hpp" 36 #include "utilities/powerOfTwo.hpp" 37 38 class OopMap; 39 40 // MacroAssembler extends Assembler by frequently used macros. 41 // 42 // Instructions for which a 'better' code sequence exists depending 43 // on arguments should also go in here. 44 45 class MacroAssembler: public Assembler { 46 friend class LIR_Assembler; 47 48 public: 49 using Assembler::mov; 50 using Assembler::movi; 51 52 protected: 53 54 // Support for VM calls 55 // 56 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 57 // may customize this version by overriding it for its purposes (e.g., to save/restore 58 // additional registers when doing a VM call). 59 virtual void call_VM_leaf_base( 60 address entry_point, // the entry point 61 int number_of_arguments, // the number of arguments to pop after the call 62 Label *retaddr = nullptr 63 ); 64 65 virtual void call_VM_leaf_base( 66 address entry_point, // the entry point 67 int number_of_arguments, // the number of arguments to pop after the call 68 Label &retaddr) { 69 call_VM_leaf_base(entry_point, number_of_arguments, &retaddr); 70 } 71 72 // This is the base routine called by the different versions of call_VM. The interpreter 73 // may customize this version by overriding it for its purposes (e.g., to save/restore 74 // additional registers when doing a VM call). 75 // 76 // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base 77 // returns the register which contains the thread upon return. If a thread register has been 78 // specified, the return value will correspond to that register. If no last_java_sp is specified 79 // (noreg) than rsp will be used instead. 80 virtual void call_VM_base( // returns the register containing the thread upon return 81 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 82 Register java_thread, // the thread if computed before ; use noreg otherwise 83 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 84 Label* return_pc, // to set up last_Java_frame; use nullptr otherwise 85 address entry_point, // the entry point 86 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 87 bool check_exceptions // whether to check for pending exceptions after return 88 ); 89 90 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 91 92 enum KlassDecodeMode { 93 KlassDecodeNone, 94 KlassDecodeZero, 95 KlassDecodeXor, 96 KlassDecodeMovk 97 }; 98 99 // Calculate decoding mode based on given parameters, used for checking then ultimately setting. 100 static KlassDecodeMode klass_decode_mode(address base, int shift, const size_t range); 101 102 private: 103 static KlassDecodeMode _klass_decode_mode; 104 105 // Returns above setting with asserts 106 static KlassDecodeMode klass_decode_mode(); 107 108 public: 109 // Checks the decode mode and returns false if not compatible with preferred decoding mode. 110 static bool check_klass_decode_mode(address base, int shift, const size_t range); 111 112 // Sets the decode mode and returns false if cannot be set. 113 static bool set_klass_decode_mode(address base, int shift, const size_t range); 114 115 public: 116 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 117 118 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 119 // The implementation is only non-empty for the InterpreterMacroAssembler, 120 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 121 virtual void check_and_handle_popframe(Register java_thread); 122 virtual void check_and_handle_earlyret(Register java_thread); 123 124 void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod, Register tmp = rscratch1); 125 void rt_call(address dest, Register tmp = rscratch1); 126 127 // Load Effective Address 128 void lea(Register r, const Address &a) { 129 InstructionMark im(this); 130 a.lea(this, r); 131 } 132 133 // Whether materializing the given address for a LDR/STR requires an 134 // additional lea instruction. 135 static bool legitimize_address_requires_lea(const Address &a, int size) { 136 return a.getMode() == Address::base_plus_offset && 137 !Address::offset_ok_for_immed(a.offset(), exact_log2(size)); 138 } 139 140 /* Sometimes we get misaligned loads and stores, usually from Unsafe 141 accesses, and these can exceed the offset range. */ 142 Address legitimize_address(const Address &a, int size, Register scratch) { 143 if (legitimize_address_requires_lea(a, size)) { 144 block_comment("legitimize_address {"); 145 lea(scratch, a); 146 block_comment("} legitimize_address"); 147 return Address(scratch); 148 } 149 return a; 150 } 151 152 void addmw(Address a, Register incr, Register scratch) { 153 ldrw(scratch, a); 154 addw(scratch, scratch, incr); 155 strw(scratch, a); 156 } 157 158 // Add constant to memory word 159 void addmw(Address a, int imm, Register scratch) { 160 ldrw(scratch, a); 161 if (imm > 0) 162 addw(scratch, scratch, (unsigned)imm); 163 else 164 subw(scratch, scratch, (unsigned)-imm); 165 strw(scratch, a); 166 } 167 168 void bind(Label& L) { 169 Assembler::bind(L); 170 code()->clear_last_insn(); 171 code()->set_last_label(pc()); 172 } 173 174 void membar(Membar_mask_bits order_constraint); 175 176 using Assembler::ldr; 177 using Assembler::str; 178 using Assembler::ldrw; 179 using Assembler::strw; 180 181 void ldr(Register Rx, const Address &adr); 182 void ldrw(Register Rw, const Address &adr); 183 void str(Register Rx, const Address &adr); 184 void strw(Register Rx, const Address &adr); 185 186 // Frame creation and destruction shared between JITs. 187 void build_frame(int framesize); 188 void remove_frame(int framesize); 189 190 virtual void _call_Unimplemented(address call_site) { 191 mov(rscratch2, call_site); 192 } 193 194 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__ 195 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention." 196 // https://reviews.llvm.org/D3311 197 198 #ifdef _WIN64 199 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__) 200 #else 201 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__) 202 #endif 203 204 // aliases defined in AARCH64 spec 205 206 template<class T> 207 inline void cmpw(Register Rd, T imm) { subsw(zr, Rd, imm); } 208 209 inline void cmp(Register Rd, unsigned char imm8) { subs(zr, Rd, imm8); } 210 inline void cmp(Register Rd, unsigned imm) = delete; 211 212 template<class T> 213 inline void cmnw(Register Rd, T imm) { addsw(zr, Rd, imm); } 214 215 inline void cmn(Register Rd, unsigned char imm8) { adds(zr, Rd, imm8); } 216 inline void cmn(Register Rd, unsigned imm) = delete; 217 218 void cset(Register Rd, Assembler::Condition cond) { 219 csinc(Rd, zr, zr, ~cond); 220 } 221 void csetw(Register Rd, Assembler::Condition cond) { 222 csincw(Rd, zr, zr, ~cond); 223 } 224 225 void cneg(Register Rd, Register Rn, Assembler::Condition cond) { 226 csneg(Rd, Rn, Rn, ~cond); 227 } 228 void cnegw(Register Rd, Register Rn, Assembler::Condition cond) { 229 csnegw(Rd, Rn, Rn, ~cond); 230 } 231 232 inline void movw(Register Rd, Register Rn) { 233 if (Rd == sp || Rn == sp) { 234 Assembler::addw(Rd, Rn, 0U); 235 } else { 236 orrw(Rd, zr, Rn); 237 } 238 } 239 inline void mov(Register Rd, Register Rn) { 240 assert(Rd != r31_sp && Rn != r31_sp, "should be"); 241 if (Rd == Rn) { 242 } else if (Rd == sp || Rn == sp) { 243 Assembler::add(Rd, Rn, 0U); 244 } else { 245 orr(Rd, zr, Rn); 246 } 247 } 248 249 inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); } 250 inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); } 251 252 inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); } 253 inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); } 254 255 inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); } 256 inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); } 257 258 inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 259 bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 260 } 261 inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) { 262 bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 263 } 264 265 inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 266 bfmw(Rd, Rn, lsb, (lsb + width - 1)); 267 } 268 inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) { 269 bfm(Rd, Rn, lsb , (lsb + width - 1)); 270 } 271 272 inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 273 sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 274 } 275 inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) { 276 sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 277 } 278 279 inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 280 sbfmw(Rd, Rn, lsb, (lsb + width - 1)); 281 } 282 inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) { 283 sbfm(Rd, Rn, lsb , (lsb + width - 1)); 284 } 285 286 inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 287 ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 288 } 289 inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) { 290 ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 291 } 292 293 inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 294 ubfmw(Rd, Rn, lsb, (lsb + width - 1)); 295 } 296 inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) { 297 ubfm(Rd, Rn, lsb , (lsb + width - 1)); 298 } 299 300 inline void asrw(Register Rd, Register Rn, unsigned imm) { 301 sbfmw(Rd, Rn, imm, 31); 302 } 303 304 inline void asr(Register Rd, Register Rn, unsigned imm) { 305 sbfm(Rd, Rn, imm, 63); 306 } 307 308 inline void lslw(Register Rd, Register Rn, unsigned imm) { 309 ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm)); 310 } 311 312 inline void lsl(Register Rd, Register Rn, unsigned imm) { 313 ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm)); 314 } 315 316 inline void lsrw(Register Rd, Register Rn, unsigned imm) { 317 ubfmw(Rd, Rn, imm, 31); 318 } 319 320 inline void lsr(Register Rd, Register Rn, unsigned imm) { 321 ubfm(Rd, Rn, imm, 63); 322 } 323 324 inline void rorw(Register Rd, Register Rn, unsigned imm) { 325 extrw(Rd, Rn, Rn, imm); 326 } 327 328 inline void ror(Register Rd, Register Rn, unsigned imm) { 329 extr(Rd, Rn, Rn, imm); 330 } 331 332 inline void rolw(Register Rd, Register Rn, unsigned imm) { 333 extrw(Rd, Rn, Rn, (32 - imm)); 334 } 335 336 inline void rol(Register Rd, Register Rn, unsigned imm) { 337 extr(Rd, Rn, Rn, (64 - imm)); 338 } 339 340 using Assembler::rax1; 341 using Assembler::eor3; 342 343 inline void rax1(Register Rd, Register Rn, Register Rm) { 344 eor(Rd, Rn, Rm, ROR, 63); // Rd = Rn ^ rol(Rm, 1) 345 } 346 347 inline void eor3(Register Rd, Register Rn, Register Rm, Register Rk) { 348 assert(Rd != Rn, "Use tmp register"); 349 eor(Rd, Rm, Rk); 350 eor(Rd, Rd, Rn); 351 } 352 353 inline void sxtbw(Register Rd, Register Rn) { 354 sbfmw(Rd, Rn, 0, 7); 355 } 356 inline void sxthw(Register Rd, Register Rn) { 357 sbfmw(Rd, Rn, 0, 15); 358 } 359 inline void sxtb(Register Rd, Register Rn) { 360 sbfm(Rd, Rn, 0, 7); 361 } 362 inline void sxth(Register Rd, Register Rn) { 363 sbfm(Rd, Rn, 0, 15); 364 } 365 inline void sxtw(Register Rd, Register Rn) { 366 sbfm(Rd, Rn, 0, 31); 367 } 368 369 inline void uxtbw(Register Rd, Register Rn) { 370 ubfmw(Rd, Rn, 0, 7); 371 } 372 inline void uxthw(Register Rd, Register Rn) { 373 ubfmw(Rd, Rn, 0, 15); 374 } 375 inline void uxtb(Register Rd, Register Rn) { 376 ubfm(Rd, Rn, 0, 7); 377 } 378 inline void uxth(Register Rd, Register Rn) { 379 ubfm(Rd, Rn, 0, 15); 380 } 381 inline void uxtw(Register Rd, Register Rn) { 382 ubfm(Rd, Rn, 0, 31); 383 } 384 385 inline void cmnw(Register Rn, Register Rm) { 386 addsw(zr, Rn, Rm); 387 } 388 inline void cmn(Register Rn, Register Rm) { 389 adds(zr, Rn, Rm); 390 } 391 392 inline void cmpw(Register Rn, Register Rm) { 393 subsw(zr, Rn, Rm); 394 } 395 inline void cmp(Register Rn, Register Rm) { 396 subs(zr, Rn, Rm); 397 } 398 399 inline void negw(Register Rd, Register Rn) { 400 subw(Rd, zr, Rn); 401 } 402 403 inline void neg(Register Rd, Register Rn) { 404 sub(Rd, zr, Rn); 405 } 406 407 inline void negsw(Register Rd, Register Rn) { 408 subsw(Rd, zr, Rn); 409 } 410 411 inline void negs(Register Rd, Register Rn) { 412 subs(Rd, zr, Rn); 413 } 414 415 inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 416 addsw(zr, Rn, Rm, kind, shift); 417 } 418 inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 419 adds(zr, Rn, Rm, kind, shift); 420 } 421 422 inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 423 subsw(zr, Rn, Rm, kind, shift); 424 } 425 inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 426 subs(zr, Rn, Rm, kind, shift); 427 } 428 429 inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 430 subw(Rd, zr, Rn, kind, shift); 431 } 432 433 inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 434 sub(Rd, zr, Rn, kind, shift); 435 } 436 437 inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 438 subsw(Rd, zr, Rn, kind, shift); 439 } 440 441 inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 442 subs(Rd, zr, Rn, kind, shift); 443 } 444 445 inline void mnegw(Register Rd, Register Rn, Register Rm) { 446 msubw(Rd, Rn, Rm, zr); 447 } 448 inline void mneg(Register Rd, Register Rn, Register Rm) { 449 msub(Rd, Rn, Rm, zr); 450 } 451 452 inline void mulw(Register Rd, Register Rn, Register Rm) { 453 maddw(Rd, Rn, Rm, zr); 454 } 455 inline void mul(Register Rd, Register Rn, Register Rm) { 456 madd(Rd, Rn, Rm, zr); 457 } 458 459 inline void smnegl(Register Rd, Register Rn, Register Rm) { 460 smsubl(Rd, Rn, Rm, zr); 461 } 462 inline void smull(Register Rd, Register Rn, Register Rm) { 463 smaddl(Rd, Rn, Rm, zr); 464 } 465 466 inline void umnegl(Register Rd, Register Rn, Register Rm) { 467 umsubl(Rd, Rn, Rm, zr); 468 } 469 inline void umull(Register Rd, Register Rn, Register Rm) { 470 umaddl(Rd, Rn, Rm, zr); 471 } 472 473 #define WRAP(INSN) \ 474 void INSN(Register Rd, Register Rn, Register Rm, Register Ra) { \ 475 if (VM_Version::supports_a53mac() && Ra != zr) \ 476 nop(); \ 477 Assembler::INSN(Rd, Rn, Rm, Ra); \ 478 } 479 480 WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw) 481 WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl) 482 #undef WRAP 483 484 485 // macro assembly operations needed for aarch64 486 487 public: 488 489 enum FpPushPopMode { 490 PushPopFull, 491 PushPopSVE, 492 PushPopNeon, 493 PushPopFp 494 }; 495 496 // first two private routines for loading 32 bit or 64 bit constants 497 private: 498 499 void mov_immediate64(Register dst, uint64_t imm64); 500 void mov_immediate32(Register dst, uint32_t imm32); 501 502 int push(unsigned int bitset, Register stack); 503 int pop(unsigned int bitset, Register stack); 504 505 int push_fp(unsigned int bitset, Register stack, FpPushPopMode mode); 506 int pop_fp(unsigned int bitset, Register stack, FpPushPopMode mode); 507 508 int push_p(unsigned int bitset, Register stack); 509 int pop_p(unsigned int bitset, Register stack); 510 511 void mov(Register dst, Address a); 512 513 public: 514 515 void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); } 516 void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); } 517 518 void push_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) push_fp(regs.bits(), stack, mode); } 519 void pop_fp(FloatRegSet regs, Register stack, FpPushPopMode mode = PushPopFull) { if (regs.bits()) pop_fp(regs.bits(), stack, mode); } 520 521 static RegSet call_clobbered_gp_registers(); 522 523 void push_p(PRegSet regs, Register stack) { if (regs.bits()) push_p(regs.bits(), stack); } 524 void pop_p(PRegSet regs, Register stack) { if (regs.bits()) pop_p(regs.bits(), stack); } 525 526 // Push and pop everything that might be clobbered by a native 527 // runtime call except rscratch1 and rscratch2. (They are always 528 // scratch, so we don't have to protect them.) Only save the lower 529 // 64 bits of each vector register. Additional registers can be excluded 530 // in a passed RegSet. 531 void push_call_clobbered_registers_except(RegSet exclude); 532 void pop_call_clobbered_registers_except(RegSet exclude); 533 534 void push_call_clobbered_registers() { 535 push_call_clobbered_registers_except(RegSet()); 536 } 537 void pop_call_clobbered_registers() { 538 pop_call_clobbered_registers_except(RegSet()); 539 } 540 541 542 // now mov instructions for loading absolute addresses and 32 or 543 // 64 bit integers 544 545 inline void mov(Register dst, address addr) { mov_immediate64(dst, (uint64_t)addr); } 546 547 template<typename T, ENABLE_IF(std::is_integral<T>::value)> 548 inline void mov(Register dst, T o) { mov_immediate64(dst, (uint64_t)o); } 549 550 inline void movw(Register dst, uint32_t imm32) { mov_immediate32(dst, imm32); } 551 552 void mov(Register dst, RegisterOrConstant src) { 553 if (src.is_register()) 554 mov(dst, src.as_register()); 555 else 556 mov(dst, src.as_constant()); 557 } 558 559 void movptr(Register r, uintptr_t imm64); 560 561 void mov(FloatRegister Vd, SIMD_Arrangement T, uint64_t imm64); 562 563 void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) { 564 orr(Vd, T, Vn, Vn); 565 } 566 567 void flt_to_flt16(Register dst, FloatRegister src, FloatRegister tmp) { 568 fcvtsh(tmp, src); 569 smov(dst, tmp, H, 0); 570 } 571 572 void flt16_to_flt(FloatRegister dst, Register src, FloatRegister tmp) { 573 mov(tmp, H, 0, src); 574 fcvths(dst, tmp); 575 } 576 577 // Generalized Test Bit And Branch, including a "far" variety which 578 // spans more than 32KiB. 579 void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool isfar = false) { 580 assert(cond == EQ || cond == NE, "must be"); 581 582 if (isfar) 583 cond = ~cond; 584 585 void (Assembler::* branch)(Register Rt, int bitpos, Label &L); 586 if (cond == Assembler::EQ) 587 branch = &Assembler::tbz; 588 else 589 branch = &Assembler::tbnz; 590 591 if (isfar) { 592 Label L; 593 (this->*branch)(Rt, bitpos, L); 594 b(dest); 595 bind(L); 596 } else { 597 (this->*branch)(Rt, bitpos, dest); 598 } 599 } 600 601 // macro instructions for accessing and updating floating point 602 // status register 603 // 604 // FPSR : op1 == 011 605 // CRn == 0100 606 // CRm == 0100 607 // op2 == 001 608 609 inline void get_fpsr(Register reg) 610 { 611 mrs(0b11, 0b0100, 0b0100, 0b001, reg); 612 } 613 614 inline void set_fpsr(Register reg) 615 { 616 msr(0b011, 0b0100, 0b0100, 0b001, reg); 617 } 618 619 inline void clear_fpsr() 620 { 621 msr(0b011, 0b0100, 0b0100, 0b001, zr); 622 } 623 624 // FPCR : op1 == 011 625 // CRn == 0100 626 // CRm == 0100 627 // op2 == 000 628 629 inline void get_fpcr(Register reg) { 630 mrs(0b11, 0b0100, 0b0100, 0b000, reg); 631 } 632 633 inline void set_fpcr(Register reg) { 634 msr(0b011, 0b0100, 0b0100, 0b000, reg); 635 } 636 637 // DCZID_EL0: op1 == 011 638 // CRn == 0000 639 // CRm == 0000 640 // op2 == 111 641 inline void get_dczid_el0(Register reg) 642 { 643 mrs(0b011, 0b0000, 0b0000, 0b111, reg); 644 } 645 646 // CTR_EL0: op1 == 011 647 // CRn == 0000 648 // CRm == 0000 649 // op2 == 001 650 inline void get_ctr_el0(Register reg) 651 { 652 mrs(0b011, 0b0000, 0b0000, 0b001, reg); 653 } 654 655 inline void get_nzcv(Register reg) { 656 mrs(0b011, 0b0100, 0b0010, 0b000, reg); 657 } 658 659 inline void set_nzcv(Register reg) { 660 msr(0b011, 0b0100, 0b0010, 0b000, reg); 661 } 662 663 // idiv variant which deals with MINLONG as dividend and -1 as divisor 664 int corrected_idivl(Register result, Register ra, Register rb, 665 bool want_remainder, Register tmp = rscratch1); 666 int corrected_idivq(Register result, Register ra, Register rb, 667 bool want_remainder, Register tmp = rscratch1); 668 669 // Support for null-checks 670 // 671 // Generates code that causes a null OS exception if the content of reg is null. 672 // If the accessed location is M[reg + offset] and the offset is known, provide the 673 // offset. No explicit code generation is needed if the offset is within a certain 674 // range (0 <= offset <= page_size). 675 676 virtual void null_check(Register reg, int offset = -1); 677 static bool needs_explicit_null_check(intptr_t offset); 678 static bool uses_implicit_null_check(void* address); 679 680 static address target_addr_for_insn(address insn_addr, unsigned insn); 681 static address target_addr_for_insn_or_null(address insn_addr, unsigned insn); 682 static address target_addr_for_insn(address insn_addr) { 683 unsigned insn = *(unsigned*)insn_addr; 684 return target_addr_for_insn(insn_addr, insn); 685 } 686 static address target_addr_for_insn_or_null(address insn_addr) { 687 unsigned insn = *(unsigned*)insn_addr; 688 return target_addr_for_insn_or_null(insn_addr, insn); 689 } 690 691 // Required platform-specific helpers for Label::patch_instructions. 692 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 693 static int pd_patch_instruction_size(address branch, address target); 694 static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) { 695 pd_patch_instruction_size(branch, target); 696 } 697 static address pd_call_destination(address branch) { 698 return target_addr_for_insn(branch); 699 } 700 #ifndef PRODUCT 701 static void pd_print_patched_instruction(address branch); 702 #endif 703 704 static int patch_oop(address insn_addr, address o); 705 static int patch_narrow_klass(address insn_addr, narrowKlass n); 706 707 // Return whether code is emitted to a scratch blob. 708 virtual bool in_scratch_emit_size() { 709 return false; 710 } 711 address emit_trampoline_stub(int insts_call_instruction_offset, address target); 712 static int max_trampoline_stub_size(); 713 void emit_static_call_stub(); 714 static int static_call_stub_size(); 715 716 // The following 4 methods return the offset of the appropriate move instruction 717 718 // Support for fast byte/short loading with zero extension (depending on particular CPU) 719 int load_unsigned_byte(Register dst, Address src); 720 int load_unsigned_short(Register dst, Address src); 721 722 // Support for fast byte/short loading with sign extension (depending on particular CPU) 723 int load_signed_byte(Register dst, Address src); 724 int load_signed_short(Register dst, Address src); 725 726 int load_signed_byte32(Register dst, Address src); 727 int load_signed_short32(Register dst, Address src); 728 729 // Support for sign-extension (hi:lo = extend_sign(lo)) 730 void extend_sign(Register hi, Register lo); 731 732 // Load and store values by size and signed-ness 733 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed); 734 void store_sized_value(Address dst, Register src, size_t size_in_bytes); 735 736 // Support for inc/dec with optimal instruction selection depending on value 737 738 // x86_64 aliases an unqualified register/address increment and 739 // decrement to call incrementq and decrementq but also supports 740 // explicitly sized calls to incrementq/decrementq or 741 // incrementl/decrementl 742 743 // for aarch64 the proper convention would be to use 744 // increment/decrement for 64 bit operations and 745 // incrementw/decrementw for 32 bit operations. so when porting 746 // x86_64 code we can leave calls to increment/decrement as is, 747 // replace incrementq/decrementq with increment/decrement and 748 // replace incrementl/decrementl with incrementw/decrementw. 749 750 // n.b. increment/decrement calls with an Address destination will 751 // need to use a scratch register to load the value to be 752 // incremented. increment/decrement calls which add or subtract a 753 // constant value greater than 2^12 will need to use a 2nd scratch 754 // register to hold the constant. so, a register increment/decrement 755 // may trash rscratch2 and an address increment/decrement trash 756 // rscratch and rscratch2 757 758 void decrementw(Address dst, int value = 1); 759 void decrementw(Register reg, int value = 1); 760 761 void decrement(Register reg, int value = 1); 762 void decrement(Address dst, int value = 1); 763 764 void incrementw(Address dst, int value = 1); 765 void incrementw(Register reg, int value = 1); 766 767 void increment(Register reg, int value = 1); 768 void increment(Address dst, int value = 1); 769 770 771 // Alignment 772 void align(int modulus); 773 void align(int modulus, int target); 774 775 // nop 776 void post_call_nop(); 777 778 // Stack frame creation/removal 779 void enter(bool strip_ret_addr = false); 780 void leave(); 781 782 // ROP Protection 783 void protect_return_address(); 784 void protect_return_address(Register return_reg); 785 void authenticate_return_address(); 786 void authenticate_return_address(Register return_reg); 787 void strip_return_address(); 788 void check_return_address(Register return_reg=lr) PRODUCT_RETURN; 789 790 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 791 // The pointer will be loaded into the thread register. 792 void get_thread(Register thread); 793 794 // support for argument shuffling 795 void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rscratch1); 796 void float_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1); 797 void long_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1); 798 void double_move(VMRegPair src, VMRegPair dst, Register tmp = rscratch1); 799 void object_move( 800 OopMap* map, 801 int oop_handle_offset, 802 int framesize_in_slots, 803 VMRegPair src, 804 VMRegPair dst, 805 bool is_receiver, 806 int* receiver_offset); 807 808 809 // Support for VM calls 810 // 811 // It is imperative that all calls into the VM are handled via the call_VM macros. 812 // They make sure that the stack linkage is setup correctly. call_VM's correspond 813 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 814 815 816 void call_VM(Register oop_result, 817 address entry_point, 818 bool check_exceptions = true); 819 void call_VM(Register oop_result, 820 address entry_point, 821 Register arg_1, 822 bool check_exceptions = true); 823 void call_VM(Register oop_result, 824 address entry_point, 825 Register arg_1, Register arg_2, 826 bool check_exceptions = true); 827 void call_VM(Register oop_result, 828 address entry_point, 829 Register arg_1, Register arg_2, Register arg_3, 830 bool check_exceptions = true); 831 832 // Overloadings with last_Java_sp 833 void call_VM(Register oop_result, 834 Register last_java_sp, 835 address entry_point, 836 int number_of_arguments = 0, 837 bool check_exceptions = true); 838 void call_VM(Register oop_result, 839 Register last_java_sp, 840 address entry_point, 841 Register arg_1, bool 842 check_exceptions = true); 843 void call_VM(Register oop_result, 844 Register last_java_sp, 845 address entry_point, 846 Register arg_1, Register arg_2, 847 bool check_exceptions = true); 848 void call_VM(Register oop_result, 849 Register last_java_sp, 850 address entry_point, 851 Register arg_1, Register arg_2, Register arg_3, 852 bool check_exceptions = true); 853 854 void get_vm_result_oop(Register oop_result, Register thread); 855 void get_vm_result_metadata(Register metadata_result, Register thread); 856 857 // These always tightly bind to MacroAssembler::call_VM_base 858 // bypassing the virtual implementation 859 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); 860 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); 861 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); 862 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true); 863 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true); 864 865 void call_VM_leaf(address entry_point, 866 int number_of_arguments = 0); 867 void call_VM_leaf(address entry_point, 868 Register arg_1); 869 void call_VM_leaf(address entry_point, 870 Register arg_1, Register arg_2); 871 void call_VM_leaf(address entry_point, 872 Register arg_1, Register arg_2, Register arg_3); 873 874 // These always tightly bind to MacroAssembler::call_VM_leaf_base 875 // bypassing the virtual implementation 876 void super_call_VM_leaf(address entry_point); 877 void super_call_VM_leaf(address entry_point, Register arg_1); 878 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); 879 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); 880 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); 881 882 // last Java Frame (fills frame anchor) 883 void set_last_Java_frame(Register last_java_sp, 884 Register last_java_fp, 885 address last_java_pc, 886 Register scratch); 887 888 void set_last_Java_frame(Register last_java_sp, 889 Register last_java_fp, 890 Label &last_java_pc, 891 Register scratch); 892 893 void set_last_Java_frame(Register last_java_sp, 894 Register last_java_fp, 895 Register last_java_pc, 896 Register scratch); 897 898 void reset_last_Java_frame(Register thread); 899 900 // thread in the default location (rthread) 901 void reset_last_Java_frame(bool clear_fp); 902 903 // Stores 904 void store_check(Register obj); // store check for obj - register is destroyed afterwards 905 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 906 907 void resolve_jobject(Register value, Register tmp1, Register tmp2); 908 void resolve_global_jobject(Register value, Register tmp1, Register tmp2); 909 910 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 911 void c2bool(Register x); 912 913 void load_method_holder_cld(Register rresult, Register rmethod); 914 void load_method_holder(Register holder, Register method); 915 916 // oop manipulations 917 void load_narrow_klass_compact(Register dst, Register src); 918 void load_klass(Register dst, Register src); 919 void store_klass(Register dst, Register src); 920 void cmp_klass(Register obj, Register klass, Register tmp); 921 void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2); 922 923 void resolve_weak_handle(Register result, Register tmp1, Register tmp2); 924 void resolve_oop_handle(Register result, Register tmp1, Register tmp2); 925 void load_mirror(Register dst, Register method, Register tmp1, Register tmp2); 926 927 void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src, 928 Register tmp1, Register tmp2); 929 930 void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val, 931 Register tmp1, Register tmp2, Register tmp3); 932 933 void load_heap_oop(Register dst, Address src, Register tmp1, 934 Register tmp2, DecoratorSet decorators = 0); 935 936 void load_heap_oop_not_null(Register dst, Address src, Register tmp1, 937 Register tmp2, DecoratorSet decorators = 0); 938 void store_heap_oop(Address dst, Register val, Register tmp1, 939 Register tmp2, Register tmp3, DecoratorSet decorators = 0); 940 941 // currently unimplemented 942 // Used for storing null. All other oop constants should be 943 // stored using routines that take a jobject. 944 void store_heap_oop_null(Address dst); 945 946 void store_klass_gap(Register dst, Register src); 947 948 // This dummy is to prevent a call to store_heap_oop from 949 // converting a zero (like null) into a Register by giving 950 // the compiler two choices it can't resolve 951 952 void store_heap_oop(Address dst, void* dummy); 953 954 void encode_heap_oop(Register d, Register s); 955 void encode_heap_oop(Register r) { encode_heap_oop(r, r); } 956 void decode_heap_oop(Register d, Register s); 957 void decode_heap_oop(Register r) { decode_heap_oop(r, r); } 958 void encode_heap_oop_not_null(Register r); 959 void decode_heap_oop_not_null(Register r); 960 void encode_heap_oop_not_null(Register dst, Register src); 961 void decode_heap_oop_not_null(Register dst, Register src); 962 963 void set_narrow_oop(Register dst, jobject obj); 964 965 void decode_klass_not_null_for_aot(Register dst, Register src); 966 void encode_klass_not_null_for_aot(Register dst, Register src); 967 void encode_klass_not_null(Register r); 968 void decode_klass_not_null(Register r); 969 void encode_klass_not_null(Register dst, Register src); 970 void decode_klass_not_null(Register dst, Register src); 971 972 void set_narrow_klass(Register dst, Klass* k); 973 974 // if heap base register is used - reinit it with the correct value 975 void reinit_heapbase(); 976 977 DEBUG_ONLY(void verify_heapbase(const char* msg);) 978 979 void push_CPU_state(bool save_vectors = false, bool use_sve = false, 980 int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0); 981 void pop_CPU_state(bool restore_vectors = false, bool use_sve = false, 982 int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0); 983 984 void push_cont_fastpath(Register java_thread = rthread); 985 void pop_cont_fastpath(Register java_thread = rthread); 986 987 void inc_held_monitor_count(Register tmp); 988 void dec_held_monitor_count(Register tmp); 989 990 // Round up to a power of two 991 void round_to(Register reg, int modulus); 992 993 // java.lang.Math::round intrinsics 994 void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp); 995 void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp); 996 997 // allocation 998 void tlab_allocate( 999 Register obj, // result: pointer to object after successful allocation 1000 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 1001 int con_size_in_bytes, // object size in bytes if known at compile time 1002 Register t1, // temp register 1003 Register t2, // temp register 1004 Label& slow_case // continuation point if fast allocation fails 1005 ); 1006 void verify_tlab(); 1007 1008 // interface method calling 1009 void lookup_interface_method(Register recv_klass, 1010 Register intf_klass, 1011 RegisterOrConstant itable_index, 1012 Register method_result, 1013 Register scan_temp, 1014 Label& no_such_interface, 1015 bool return_method = true); 1016 1017 void lookup_interface_method_stub(Register recv_klass, 1018 Register holder_klass, 1019 Register resolved_klass, 1020 Register method_result, 1021 Register temp_reg, 1022 Register temp_reg2, 1023 int itable_index, 1024 Label& L_no_such_interface); 1025 1026 // virtual method calling 1027 // n.b. x86 allows RegisterOrConstant for vtable_index 1028 void lookup_virtual_method(Register recv_klass, 1029 RegisterOrConstant vtable_index, 1030 Register method_result); 1031 1032 // Test sub_klass against super_klass, with fast and slow paths. 1033 1034 // The fast path produces a tri-state answer: yes / no / maybe-slow. 1035 // One of the three labels can be null, meaning take the fall-through. 1036 // If super_check_offset is -1, the value is loaded up from super_klass. 1037 // No registers are killed, except temp_reg. 1038 void check_klass_subtype_fast_path(Register sub_klass, 1039 Register super_klass, 1040 Register temp_reg, 1041 Label* L_success, 1042 Label* L_failure, 1043 Label* L_slow_path, 1044 Register super_check_offset = noreg); 1045 1046 // The rest of the type check; must be wired to a corresponding fast path. 1047 // It does not repeat the fast path logic, so don't use it standalone. 1048 // The temp_reg and temp2_reg can be noreg, if no temps are available. 1049 // Updates the sub's secondary super cache as necessary. 1050 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 1051 void check_klass_subtype_slow_path(Register sub_klass, 1052 Register super_klass, 1053 Register temp_reg, 1054 Register temp2_reg, 1055 Label* L_success, 1056 Label* L_failure, 1057 bool set_cond_codes = false); 1058 1059 void check_klass_subtype_slow_path_linear(Register sub_klass, 1060 Register super_klass, 1061 Register temp_reg, 1062 Register temp2_reg, 1063 Label* L_success, 1064 Label* L_failure, 1065 bool set_cond_codes = false); 1066 1067 void check_klass_subtype_slow_path_table(Register sub_klass, 1068 Register super_klass, 1069 Register temp_reg, 1070 Register temp2_reg, 1071 Register temp3_reg, 1072 Register result_reg, 1073 FloatRegister vtemp_reg, 1074 Label* L_success, 1075 Label* L_failure, 1076 bool set_cond_codes = false); 1077 1078 // If r is valid, return r. 1079 // If r is invalid, remove a register r2 from available_regs, add r2 1080 // to regs_to_push, then return r2. 1081 Register allocate_if_noreg(const Register r, 1082 RegSetIterator<Register> &available_regs, 1083 RegSet ®s_to_push); 1084 1085 // Secondary subtype checking 1086 void lookup_secondary_supers_table_var(Register sub_klass, 1087 Register r_super_klass, 1088 Register temp1, 1089 Register temp2, 1090 Register temp3, 1091 FloatRegister vtemp, 1092 Register result, 1093 Label *L_success); 1094 1095 1096 // As above, but with a constant super_klass. 1097 // The result is in Register result, not the condition codes. 1098 bool lookup_secondary_supers_table_const(Register r_sub_klass, 1099 Register r_super_klass, 1100 Register temp1, 1101 Register temp2, 1102 Register temp3, 1103 FloatRegister vtemp, 1104 Register result, 1105 u1 super_klass_slot, 1106 bool stub_is_near = false); 1107 1108 void verify_secondary_supers_table(Register r_sub_klass, 1109 Register r_super_klass, 1110 Register temp1, 1111 Register temp2, 1112 Register result); 1113 1114 void lookup_secondary_supers_table_slow_path(Register r_super_klass, 1115 Register r_array_base, 1116 Register r_array_index, 1117 Register r_bitmap, 1118 Register temp1, 1119 Register result, 1120 bool is_stub = true); 1121 1122 // Simplified, combined version, good for typical uses. 1123 // Falls through on failure. 1124 void check_klass_subtype(Register sub_klass, 1125 Register super_klass, 1126 Register temp_reg, 1127 Label& L_success); 1128 1129 void clinit_barrier(Register klass, 1130 Register thread, 1131 Label* L_fast_path = nullptr, 1132 Label* L_slow_path = nullptr); 1133 1134 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 1135 1136 void verify_sve_vector_length(Register tmp = rscratch1); 1137 void reinitialize_ptrue() { 1138 if (UseSVE > 0) { 1139 sve_ptrue(ptrue, B); 1140 } 1141 } 1142 void verify_ptrue(); 1143 1144 // Debugging 1145 1146 // only if +VerifyOops 1147 void _verify_oop(Register reg, const char* s, const char* file, int line); 1148 void _verify_oop_addr(Address addr, const char * s, const char* file, int line); 1149 1150 void _verify_oop_checked(Register reg, const char* s, const char* file, int line) { 1151 if (VerifyOops) { 1152 _verify_oop(reg, s, file, line); 1153 } 1154 } 1155 void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) { 1156 if (VerifyOops) { 1157 _verify_oop_addr(reg, s, file, line); 1158 } 1159 } 1160 1161 // TODO: verify method and klass metadata (compare against vptr?) 1162 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {} 1163 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){} 1164 1165 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__) 1166 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__) 1167 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__) 1168 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__) 1169 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__) 1170 1171 // Restore cpu control state after JNI call 1172 void restore_cpu_control_state_after_jni(Register tmp1, Register tmp2); 1173 1174 // prints msg, dumps registers and stops execution 1175 void stop(const char* msg); 1176 1177 static void debug64(char* msg, int64_t pc, int64_t regs[]); 1178 1179 void untested() { stop("untested"); } 1180 1181 void unimplemented(const char* what = ""); 1182 1183 void should_not_reach_here() { stop("should not reach here"); } 1184 1185 void _assert_asm(Condition cc, const char* msg); 1186 #define assert_asm0(cc, msg) _assert_asm(cc, FILE_AND_LINE ": " msg) 1187 #define assert_asm(masm, command, cc, msg) DEBUG_ONLY(command; (masm)->_assert_asm(cc, FILE_AND_LINE ": " #command " " #cc ": " msg)) 1188 1189 // Stack overflow checking 1190 void bang_stack_with_offset(int offset) { 1191 // stack grows down, caller passes positive offset 1192 assert(offset > 0, "must bang with negative offset"); 1193 sub(rscratch2, sp, offset); 1194 str(zr, Address(rscratch2)); 1195 } 1196 1197 // Writes to stack successive pages until offset reached to check for 1198 // stack overflow + shadow pages. Also, clobbers tmp 1199 void bang_stack_size(Register size, Register tmp); 1200 1201 // Check for reserved stack access in method being exited (for JIT) 1202 void reserved_stack_check(); 1203 1204 // Arithmetics 1205 1206 // Clobber: rscratch1, rscratch2 1207 void addptr(const Address &dst, int32_t src); 1208 1209 // Clobber: rscratch1 1210 void cmpptr(Register src1, Address src2); 1211 1212 void cmpoop(Register obj1, Register obj2); 1213 1214 // Various forms of CAS 1215 1216 void cmpxchg_obj_header(Register oldv, Register newv, Register obj, Register tmp, 1217 Label &succeed, Label *fail); 1218 void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp, 1219 Label &succeed, Label *fail); 1220 1221 void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp, 1222 Label &succeed, Label *fail); 1223 1224 void atomic_add(Register prev, RegisterOrConstant incr, Register addr); 1225 void atomic_addw(Register prev, RegisterOrConstant incr, Register addr); 1226 void atomic_addal(Register prev, RegisterOrConstant incr, Register addr); 1227 void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr); 1228 1229 void atomic_xchg(Register prev, Register newv, Register addr); 1230 void atomic_xchgw(Register prev, Register newv, Register addr); 1231 void atomic_xchgl(Register prev, Register newv, Register addr); 1232 void atomic_xchglw(Register prev, Register newv, Register addr); 1233 void atomic_xchgal(Register prev, Register newv, Register addr); 1234 void atomic_xchgalw(Register prev, Register newv, Register addr); 1235 1236 void orptr(Address adr, RegisterOrConstant src) { 1237 ldr(rscratch1, adr); 1238 if (src.is_register()) 1239 orr(rscratch1, rscratch1, src.as_register()); 1240 else 1241 orr(rscratch1, rscratch1, src.as_constant()); 1242 str(rscratch1, adr); 1243 } 1244 1245 // A generic CAS; success or failure is in the EQ flag. 1246 // Clobbers rscratch1 1247 void cmpxchg(Register addr, Register expected, Register new_val, 1248 enum operand_size size, 1249 bool acquire, bool release, bool weak, 1250 Register result); 1251 1252 #ifdef ASSERT 1253 // Template short-hand support to clean-up after a failed call to trampoline 1254 // call generation (see trampoline_call() below), when a set of Labels must 1255 // be reset (before returning). 1256 template<typename Label, typename... More> 1257 void reset_labels(Label &lbl, More&... more) { 1258 lbl.reset(); reset_labels(more...); 1259 } 1260 template<typename Label> 1261 void reset_labels(Label &lbl) { 1262 lbl.reset(); 1263 } 1264 #endif 1265 1266 private: 1267 void compare_eq(Register rn, Register rm, enum operand_size size); 1268 1269 public: 1270 // AArch64 OpenJDK uses four different types of calls: 1271 // - direct call: bl pc_relative_offset 1272 // This is the shortest and the fastest, but the offset has the range: 1273 // +/-128MB for the release build, +/-2MB for the debug build. 1274 // 1275 // - far call: adrp reg, pc_relative_offset; add; bl reg 1276 // This is longer than a direct call. The offset has 1277 // the range +/-4GB. As the code cache size is limited to 4GB, 1278 // far calls can reach anywhere in the code cache. If a jump is 1279 // needed rather than a call, a far jump 'b reg' can be used instead. 1280 // All instructions are embedded at a call site. 1281 // 1282 // - trampoline call: 1283 // This is only available in C1/C2-generated code (nmethod). It is a combination 1284 // of a direct call, which is used if the destination of a call is in range, 1285 // and a register-indirect call. It has the advantages of reaching anywhere in 1286 // the AArch64 address space and being patchable at runtime when the generated 1287 // code is being executed by other threads. 1288 // 1289 // [Main code section] 1290 // bl trampoline 1291 // [Stub code section] 1292 // trampoline: 1293 // ldr reg, pc + 8 1294 // br reg 1295 // <64-bit destination address> 1296 // 1297 // If the destination is in range when the generated code is moved to the code 1298 // cache, 'bl trampoline' is replaced with 'bl destination' and the trampoline 1299 // is not used. 1300 // The optimization does not remove the trampoline from the stub section. 1301 // This is necessary because the trampoline may well be redirected later when 1302 // code is patched, and the new destination may not be reachable by a simple BR 1303 // instruction. 1304 // 1305 // - indirect call: move reg, address; blr reg 1306 // This too can reach anywhere in the address space, but it cannot be 1307 // patched while code is running, so it must only be modified at a safepoint. 1308 // This form of call is most suitable for targets at fixed addresses, which 1309 // will never be patched. 1310 // 1311 // The patching we do conforms to the "Concurrent modification and 1312 // execution of instructions" section of the Arm Architectural 1313 // Reference Manual, which only allows B, BL, BRK, HVC, ISB, NOP, SMC, 1314 // or SVC instructions to be modified while another thread is 1315 // executing them. 1316 // 1317 // To patch a trampoline call when the BL can't reach, we first modify 1318 // the 64-bit destination address in the trampoline, then modify the 1319 // BL to point to the trampoline, then flush the instruction cache to 1320 // broadcast the change to all executing threads. See 1321 // NativeCall::set_destination_mt_safe for the details. 1322 // 1323 // There is a benign race in that the other thread might observe the 1324 // modified BL before it observes the modified 64-bit destination 1325 // address. That does not matter because the destination method has been 1326 // invalidated, so there will be a trap at its start. 1327 // For this to work, the destination address in the trampoline is 1328 // always updated, even if we're not using the trampoline. 1329 1330 // Emit a direct call if the entry address will always be in range, 1331 // otherwise a trampoline call. 1332 // Supported entry.rspec(): 1333 // - relocInfo::runtime_call_type 1334 // - relocInfo::opt_virtual_call_type 1335 // - relocInfo::static_call_type 1336 // - relocInfo::virtual_call_type 1337 // 1338 // Return: the call PC or null if CodeCache is full. 1339 // Clobbers: rscratch1 1340 address trampoline_call(Address entry); 1341 1342 static bool far_branches() { 1343 return ReservedCodeCacheSize > branch_range; 1344 } 1345 1346 // Check if branches to the non nmethod section require a far jump 1347 static bool codestub_branch_needs_far_jump() { 1348 if (AOTCodeCache::is_on_for_dump()) { 1349 // To calculate far_codestub_branch_size correctly. 1350 return true; 1351 } 1352 return CodeCache::max_distance_to_non_nmethod() > branch_range; 1353 } 1354 1355 // Emit a direct call/jump if the entry address will always be in range, 1356 // otherwise a far call/jump. 1357 // The address must be inside the code cache. 1358 // Supported entry.rspec(): 1359 // - relocInfo::external_word_type 1360 // - relocInfo::runtime_call_type 1361 // - relocInfo::none 1362 // In the case of a far call/jump, the entry address is put in the tmp register. 1363 // The tmp register is invalidated. 1364 // 1365 // Far_jump returns the amount of the emitted code. 1366 void far_call(Address entry, Register tmp = rscratch1); 1367 int far_jump(Address entry, Register tmp = rscratch1); 1368 1369 static int far_codestub_branch_size() { 1370 if (codestub_branch_needs_far_jump()) { 1371 return 3 * 4; // adrp, add, br 1372 } else { 1373 return 4; 1374 } 1375 } 1376 1377 // Emit the CompiledIC call idiom 1378 address ic_call(address entry, jint method_index = 0); 1379 static int ic_check_size(); 1380 int ic_check(int end_alignment); 1381 1382 public: 1383 1384 // Data 1385 1386 void mov_metadata(Register dst, Metadata* obj); 1387 Address allocate_metadata_address(Metadata* obj); 1388 Address constant_oop_address(jobject obj); 1389 1390 void movoop(Register dst, jobject obj); 1391 1392 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic. 1393 void kernel_crc32(Register crc, Register buf, Register len, 1394 Register table0, Register table1, Register table2, Register table3, 1395 Register tmp, Register tmp2, Register tmp3); 1396 // CRC32 code for java.util.zip.CRC32C::updateBytes() intrinsic. 1397 void kernel_crc32c(Register crc, Register buf, Register len, 1398 Register table0, Register table1, Register table2, Register table3, 1399 Register tmp, Register tmp2, Register tmp3); 1400 1401 // Stack push and pop individual 64 bit registers 1402 void push(Register src); 1403 void pop(Register dst); 1404 1405 void repne_scan(Register addr, Register value, Register count, 1406 Register scratch); 1407 void repne_scanw(Register addr, Register value, Register count, 1408 Register scratch); 1409 1410 typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm); 1411 typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift); 1412 1413 // If a constant does not fit in an immediate field, generate some 1414 // number of MOV instructions and then perform the operation 1415 void wrap_add_sub_imm_insn(Register Rd, Register Rn, uint64_t imm, 1416 add_sub_imm_insn insn1, 1417 add_sub_reg_insn insn2, bool is32); 1418 // Separate vsn which sets the flags 1419 void wrap_adds_subs_imm_insn(Register Rd, Register Rn, uint64_t imm, 1420 add_sub_imm_insn insn1, 1421 add_sub_reg_insn insn2, bool is32); 1422 1423 #define WRAP(INSN, is32) \ 1424 void INSN(Register Rd, Register Rn, uint64_t imm) { \ 1425 wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \ 1426 } \ 1427 \ 1428 void INSN(Register Rd, Register Rn, Register Rm, \ 1429 enum shift_kind kind, unsigned shift = 0) { \ 1430 Assembler::INSN(Rd, Rn, Rm, kind, shift); \ 1431 } \ 1432 \ 1433 void INSN(Register Rd, Register Rn, Register Rm) { \ 1434 Assembler::INSN(Rd, Rn, Rm); \ 1435 } \ 1436 \ 1437 void INSN(Register Rd, Register Rn, Register Rm, \ 1438 ext::operation option, int amount = 0) { \ 1439 Assembler::INSN(Rd, Rn, Rm, option, amount); \ 1440 } 1441 1442 WRAP(add, false) WRAP(addw, true) WRAP(sub, false) WRAP(subw, true) 1443 1444 #undef WRAP 1445 #define WRAP(INSN, is32) \ 1446 void INSN(Register Rd, Register Rn, uint64_t imm) { \ 1447 wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN, is32); \ 1448 } \ 1449 \ 1450 void INSN(Register Rd, Register Rn, Register Rm, \ 1451 enum shift_kind kind, unsigned shift = 0) { \ 1452 Assembler::INSN(Rd, Rn, Rm, kind, shift); \ 1453 } \ 1454 \ 1455 void INSN(Register Rd, Register Rn, Register Rm) { \ 1456 Assembler::INSN(Rd, Rn, Rm); \ 1457 } \ 1458 \ 1459 void INSN(Register Rd, Register Rn, Register Rm, \ 1460 ext::operation option, int amount = 0) { \ 1461 Assembler::INSN(Rd, Rn, Rm, option, amount); \ 1462 } 1463 1464 WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true) 1465 1466 void add(Register Rd, Register Rn, RegisterOrConstant increment); 1467 void addw(Register Rd, Register Rn, RegisterOrConstant increment); 1468 void sub(Register Rd, Register Rn, RegisterOrConstant decrement); 1469 void subw(Register Rd, Register Rn, RegisterOrConstant decrement); 1470 1471 void adrp(Register reg1, const Address &dest, uint64_t &byte_offset); 1472 1473 void tableswitch(Register index, jint lowbound, jint highbound, 1474 Label &jumptable, Label &jumptable_end, int stride = 1) { 1475 adr(rscratch1, jumptable); 1476 subsw(rscratch2, index, lowbound); 1477 subsw(zr, rscratch2, highbound - lowbound); 1478 br(Assembler::HS, jumptable_end); 1479 add(rscratch1, rscratch1, rscratch2, 1480 ext::sxtw, exact_log2(stride * Assembler::instruction_size)); 1481 br(rscratch1); 1482 } 1483 1484 // Form an address from base + offset in Rd. Rd may or may not 1485 // actually be used: you must use the Address that is returned. It 1486 // is up to you to ensure that the shift provided matches the size 1487 // of your data. 1488 Address form_address(Register Rd, Register base, int64_t byte_offset, int shift); 1489 1490 // Return true iff an address is within the 48-bit AArch64 address 1491 // space. 1492 bool is_valid_AArch64_address(address a) { 1493 return ((uint64_t)a >> 48) == 0; 1494 } 1495 1496 // Load the base of the cardtable byte map into reg. 1497 void load_byte_map_base(Register reg); 1498 1499 // Prolog generator routines to support switch between x86 code and 1500 // generated ARM code 1501 1502 // routine to generate an x86 prolog for a stub function which 1503 // bootstraps into the generated ARM code which directly follows the 1504 // stub 1505 // 1506 1507 public: 1508 1509 address read_polling_page(Register r, relocInfo::relocType rtype); 1510 void get_polling_page(Register dest, relocInfo::relocType rtype); 1511 1512 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic. 1513 void update_byte_crc32(Register crc, Register val, Register table); 1514 void update_word_crc32(Register crc, Register v, Register tmp, 1515 Register table0, Register table1, Register table2, Register table3, 1516 bool upper = false); 1517 1518 address count_positives(Register ary1, Register len, Register result); 1519 1520 address arrays_equals(Register a1, Register a2, Register result, Register cnt1, 1521 Register tmp1, Register tmp2, Register tmp3, int elem_size); 1522 1523 // Ensure that the inline code and the stub use the same registers. 1524 #define ARRAYS_HASHCODE_REGISTERS \ 1525 do { \ 1526 assert(result == r0 && \ 1527 ary == r1 && \ 1528 cnt == r2 && \ 1529 vdata0 == v3 && \ 1530 vdata1 == v2 && \ 1531 vdata2 == v1 && \ 1532 vdata3 == v0 && \ 1533 vmul0 == v4 && \ 1534 vmul1 == v5 && \ 1535 vmul2 == v6 && \ 1536 vmul3 == v7 && \ 1537 vpow == v12 && \ 1538 vpowm == v13, "registers must match aarch64.ad"); \ 1539 } while (0) 1540 1541 void string_equals(Register a1, Register a2, Register result, Register cnt1); 1542 1543 void fill_words(Register base, Register cnt, Register value); 1544 address zero_words(Register base, uint64_t cnt); 1545 address zero_words(Register ptr, Register cnt); 1546 void zero_dcache_blocks(Register base, Register cnt); 1547 1548 static const int zero_words_block_size; 1549 1550 address byte_array_inflate(Register src, Register dst, Register len, 1551 FloatRegister vtmp1, FloatRegister vtmp2, 1552 FloatRegister vtmp3, Register tmp4); 1553 1554 void char_array_compress(Register src, Register dst, Register len, 1555 Register res, 1556 FloatRegister vtmp0, FloatRegister vtmp1, 1557 FloatRegister vtmp2, FloatRegister vtmp3, 1558 FloatRegister vtmp4, FloatRegister vtmp5); 1559 1560 void encode_iso_array(Register src, Register dst, 1561 Register len, Register res, bool ascii, 1562 FloatRegister vtmp0, FloatRegister vtmp1, 1563 FloatRegister vtmp2, FloatRegister vtmp3, 1564 FloatRegister vtmp4, FloatRegister vtmp5); 1565 1566 void generate_dsin_dcos(bool isCos, address npio2_hw, address two_over_pi, 1567 address pio2, address dsin_coef, address dcos_coef); 1568 private: 1569 // begin trigonometric functions support block 1570 void generate__ieee754_rem_pio2(address npio2_hw, address two_over_pi, address pio2); 1571 void generate__kernel_rem_pio2(address two_over_pi, address pio2); 1572 void generate_kernel_sin(FloatRegister x, bool iyIsOne, address dsin_coef); 1573 void generate_kernel_cos(FloatRegister x, address dcos_coef); 1574 // end trigonometric functions support block 1575 void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo, 1576 Register src1, Register src2); 1577 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) { 1578 add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2); 1579 } 1580 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart, 1581 Register y, Register y_idx, Register z, 1582 Register carry, Register product, 1583 Register idx, Register kdx); 1584 void multiply_128_x_128_loop(Register y, Register z, 1585 Register carry, Register carry2, 1586 Register idx, Register jdx, 1587 Register yz_idx1, Register yz_idx2, 1588 Register tmp, Register tmp3, Register tmp4, 1589 Register tmp7, Register product_hi); 1590 void kernel_crc32_using_crypto_pmull(Register crc, Register buf, 1591 Register len, Register tmp0, Register tmp1, Register tmp2, 1592 Register tmp3); 1593 void kernel_crc32_using_crc32(Register crc, Register buf, 1594 Register len, Register tmp0, Register tmp1, Register tmp2, 1595 Register tmp3); 1596 void kernel_crc32c_using_crypto_pmull(Register crc, Register buf, 1597 Register len, Register tmp0, Register tmp1, Register tmp2, 1598 Register tmp3); 1599 void kernel_crc32c_using_crc32c(Register crc, Register buf, 1600 Register len, Register tmp0, Register tmp1, Register tmp2, 1601 Register tmp3); 1602 void kernel_crc32_common_fold_using_crypto_pmull(Register crc, Register buf, 1603 Register len, Register tmp0, Register tmp1, Register tmp2, 1604 size_t table_offset); 1605 1606 void ghash_modmul (FloatRegister result, 1607 FloatRegister result_lo, FloatRegister result_hi, FloatRegister b, 1608 FloatRegister a, FloatRegister vzr, FloatRegister a1_xor_a0, FloatRegister p, 1609 FloatRegister t1, FloatRegister t2, FloatRegister t3); 1610 void ghash_load_wide(int index, Register data, FloatRegister result, FloatRegister state); 1611 public: 1612 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, 1613 Register tmp0, Register tmp1, Register tmp2, Register tmp3, 1614 Register tmp4, Register tmp5, Register tmp6, Register tmp7); 1615 void mul_add(Register out, Register in, Register offs, Register len, Register k); 1616 void ghash_multiply(FloatRegister result_lo, FloatRegister result_hi, 1617 FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0, 1618 FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3); 1619 void ghash_multiply_wide(int index, 1620 FloatRegister result_lo, FloatRegister result_hi, 1621 FloatRegister a, FloatRegister b, FloatRegister a1_xor_a0, 1622 FloatRegister tmp1, FloatRegister tmp2, FloatRegister tmp3); 1623 void ghash_reduce(FloatRegister result, FloatRegister lo, FloatRegister hi, 1624 FloatRegister p, FloatRegister z, FloatRegister t1); 1625 void ghash_reduce_wide(int index, FloatRegister result, FloatRegister lo, FloatRegister hi, 1626 FloatRegister p, FloatRegister z, FloatRegister t1); 1627 void ghash_processBlocks_wide(Label& p, Register state, Register subkeyH, 1628 Register data, Register blocks, int unrolls); 1629 1630 1631 void aesenc_loadkeys(Register key, Register keylen); 1632 void aesecb_encrypt(Register from, Register to, Register keylen, 1633 FloatRegister data = v0, int unrolls = 1); 1634 void aesecb_decrypt(Register from, Register to, Register key, Register keylen); 1635 void aes_round(FloatRegister input, FloatRegister subkey); 1636 1637 // ChaCha20 functions support block 1638 void cc20_qr_add4(FloatRegister (&addFirst)[4], 1639 FloatRegister (&addSecond)[4]); 1640 void cc20_qr_xor4(FloatRegister (&firstElem)[4], 1641 FloatRegister (&secondElem)[4], FloatRegister (&result)[4]); 1642 void cc20_qr_lrot4(FloatRegister (&sourceReg)[4], 1643 FloatRegister (&destReg)[4], int bits, FloatRegister table); 1644 void cc20_set_qr_registers(FloatRegister (&vectorSet)[4], 1645 const FloatRegister (&stateVectors)[16], int idx1, int idx2, 1646 int idx3, int idx4); 1647 1648 // Place an ISB after code may have been modified due to a safepoint. 1649 void safepoint_isb(); 1650 1651 private: 1652 // Return the effective address r + (r1 << ext) + offset. 1653 // Uses rscratch2. 1654 Address offsetted_address(Register r, Register r1, Address::extend ext, 1655 int offset, int size); 1656 1657 private: 1658 // Returns an address on the stack which is reachable with a ldr/str of size 1659 // Uses rscratch2 if the address is not directly reachable 1660 Address spill_address(int size, int offset, Register tmp=rscratch2); 1661 Address sve_spill_address(int sve_reg_size_in_bytes, int offset, Register tmp=rscratch2); 1662 1663 bool merge_alignment_check(Register base, size_t size, int64_t cur_offset, int64_t prev_offset) const; 1664 1665 // Check whether two loads/stores can be merged into ldp/stp. 1666 bool ldst_can_merge(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store) const; 1667 1668 // Merge current load/store with previous load/store into ldp/stp. 1669 void merge_ldst(Register rx, const Address &adr, size_t cur_size_in_bytes, bool is_store); 1670 1671 // Try to merge two loads/stores into ldp/stp. If success, returns true else false. 1672 bool try_merge_ldst(Register rt, const Address &adr, size_t cur_size_in_bytes, bool is_store); 1673 1674 public: 1675 void spill(Register Rx, bool is64, int offset) { 1676 if (is64) { 1677 str(Rx, spill_address(8, offset)); 1678 } else { 1679 strw(Rx, spill_address(4, offset)); 1680 } 1681 } 1682 void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) { 1683 str(Vx, T, spill_address(1 << (int)T, offset)); 1684 } 1685 1686 void spill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) { 1687 sve_str(Zx, sve_spill_address(vector_reg_size_in_bytes, offset)); 1688 } 1689 void spill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) { 1690 sve_str(pr, sve_spill_address(predicate_reg_size_in_bytes, offset)); 1691 } 1692 1693 void unspill(Register Rx, bool is64, int offset) { 1694 if (is64) { 1695 ldr(Rx, spill_address(8, offset)); 1696 } else { 1697 ldrw(Rx, spill_address(4, offset)); 1698 } 1699 } 1700 void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) { 1701 ldr(Vx, T, spill_address(1 << (int)T, offset)); 1702 } 1703 1704 void unspill_sve_vector(FloatRegister Zx, int offset, int vector_reg_size_in_bytes) { 1705 sve_ldr(Zx, sve_spill_address(vector_reg_size_in_bytes, offset)); 1706 } 1707 void unspill_sve_predicate(PRegister pr, int offset, int predicate_reg_size_in_bytes) { 1708 sve_ldr(pr, sve_spill_address(predicate_reg_size_in_bytes, offset)); 1709 } 1710 1711 void spill_copy128(int src_offset, int dst_offset, 1712 Register tmp1=rscratch1, Register tmp2=rscratch2) { 1713 if (src_offset < 512 && (src_offset & 7) == 0 && 1714 dst_offset < 512 && (dst_offset & 7) == 0) { 1715 ldp(tmp1, tmp2, Address(sp, src_offset)); 1716 stp(tmp1, tmp2, Address(sp, dst_offset)); 1717 } else { 1718 unspill(tmp1, true, src_offset); 1719 spill(tmp1, true, dst_offset); 1720 unspill(tmp1, true, src_offset+8); 1721 spill(tmp1, true, dst_offset+8); 1722 } 1723 } 1724 void spill_copy_sve_vector_stack_to_stack(int src_offset, int dst_offset, 1725 int sve_vec_reg_size_in_bytes) { 1726 assert(sve_vec_reg_size_in_bytes % 16 == 0, "unexpected sve vector reg size"); 1727 for (int i = 0; i < sve_vec_reg_size_in_bytes / 16; i++) { 1728 spill_copy128(src_offset, dst_offset); 1729 src_offset += 16; 1730 dst_offset += 16; 1731 } 1732 } 1733 void spill_copy_sve_predicate_stack_to_stack(int src_offset, int dst_offset, 1734 int sve_predicate_reg_size_in_bytes) { 1735 sve_ldr(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, src_offset)); 1736 sve_str(ptrue, sve_spill_address(sve_predicate_reg_size_in_bytes, dst_offset)); 1737 reinitialize_ptrue(); 1738 } 1739 void cache_wb(Address line); 1740 void cache_wbsync(bool is_pre); 1741 1742 // Code for java.lang.Thread::onSpinWait() intrinsic. 1743 void spin_wait(); 1744 1745 void lightweight_lock(Register basic_lock, Register obj, Register t1, Register t2, Register t3, Label& slow); 1746 void lightweight_unlock(Register obj, Register t1, Register t2, Register t3, Label& slow); 1747 1748 private: 1749 // Check the current thread doesn't need a cross modify fence. 1750 void verify_cross_modify_fence_not_required() PRODUCT_RETURN; 1751 1752 }; 1753 1754 #ifdef ASSERT 1755 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; } 1756 #endif 1757 1758 struct tableswitch { 1759 Register _reg; 1760 int _insn_index; jint _first_key; jint _last_key; 1761 Label _after; 1762 Label _branches; 1763 }; 1764 1765 #endif // CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP