1 /* 2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved. 3 * Copyright (c) 2014, 2015, 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_VM_MACROASSEMBLER_AARCH64_HPP 27 #define CPU_AARCH64_VM_MACROASSEMBLER_AARCH64_HPP 28 29 #include "asm/assembler.hpp" 30 31 #ifdef ASSERT 32 #include "gc/shared/collectedHeap.hpp" 33 #endif 34 35 // MacroAssembler extends Assembler by frequently used macros. 36 // 37 // Instructions for which a 'better' code sequence exists depending 38 // on arguments should also go in here. 39 40 class MacroAssembler: public Assembler { 41 friend class LIR_Assembler; 42 43 public: 44 using Assembler::mov; 45 using Assembler::movi; 46 47 protected: 48 49 // Support for VM calls 50 // 51 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 52 // may customize this version by overriding it for its purposes (e.g., to save/restore 53 // additional registers when doing a VM call). 54 virtual void call_VM_leaf_base( 55 address entry_point, // the entry point 56 int number_of_arguments, // the number of arguments to pop after the call 57 Label *retaddr = NULL 58 ); 59 60 virtual void call_VM_leaf_base( 61 address entry_point, // the entry point 62 int number_of_arguments, // the number of arguments to pop after the call 63 Label &retaddr) { 64 call_VM_leaf_base(entry_point, number_of_arguments, &retaddr); 65 } 66 67 // This is the base routine called by the different versions of call_VM. The interpreter 68 // may customize this version by overriding it for its purposes (e.g., to save/restore 69 // additional registers when doing a VM call). 70 // 71 // If no java_thread register is specified (noreg) than rthread will be used instead. call_VM_base 72 // returns the register which contains the thread upon return. If a thread register has been 73 // specified, the return value will correspond to that register. If no last_java_sp is specified 74 // (noreg) than rsp will be used instead. 75 virtual void call_VM_base( // returns the register containing the thread upon return 76 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 77 Register java_thread, // the thread if computed before ; use noreg otherwise 78 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 79 address entry_point, // the entry point 80 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 81 bool check_exceptions // whether to check for pending exceptions after return 82 ); 83 84 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 85 // The implementation is only non-empty for the InterpreterMacroAssembler, 86 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 87 virtual void check_and_handle_popframe(Register java_thread); 88 virtual void check_and_handle_earlyret(Register java_thread); 89 90 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 91 92 // Maximum size of class area in Metaspace when compressed 93 uint64_t use_XOR_for_compressed_class_base; 94 95 public: 96 MacroAssembler(CodeBuffer* code) : Assembler(code) { 97 use_XOR_for_compressed_class_base 98 = (operand_valid_for_logical_immediate(false /*is32*/, 99 (uint64_t)Universe::narrow_klass_base()) 100 && ((uint64_t)Universe::narrow_klass_base() 101 > (1u << log2_intptr(CompressedClassSpaceSize)))); 102 } 103 104 // Biased locking support 105 // lock_reg and obj_reg must be loaded up with the appropriate values. 106 // swap_reg is killed. 107 // tmp_reg must be supplied and must not be rscratch1 or rscratch2 108 // Optional slow case is for implementations (interpreter and C1) which branch to 109 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 110 // Returns offset of first potentially-faulting instruction for null 111 // check info (currently consumed only by C1). If 112 // swap_reg_contains_mark is true then returns -1 as it is assumed 113 // the calling code has already passed any potential faults. 114 int biased_locking_enter(Register lock_reg, Register obj_reg, 115 Register swap_reg, Register tmp_reg, 116 bool swap_reg_contains_mark, 117 Label& done, Label* slow_case = NULL, 118 BiasedLockingCounters* counters = NULL); 119 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 120 121 122 // Helper functions for statistics gathering. 123 // Unconditional atomic increment. 124 void atomic_incw(Register counter_addr, Register tmp, Register tmp2); 125 void atomic_incw(Address counter_addr, Register tmp1, Register tmp2, Register tmp3) { 126 lea(tmp1, counter_addr); 127 atomic_incw(tmp1, tmp2, tmp3); 128 } 129 // Load Effective Address 130 void lea(Register r, const Address &a) { 131 InstructionMark im(this); 132 code_section()->relocate(inst_mark(), a.rspec()); 133 a.lea(this, r); 134 } 135 136 void addmw(Address a, Register incr, Register scratch) { 137 ldrw(scratch, a); 138 addw(scratch, scratch, incr); 139 strw(scratch, a); 140 } 141 142 // Add constant to memory word 143 void addmw(Address a, int imm, Register scratch) { 144 ldrw(scratch, a); 145 if (imm > 0) 146 addw(scratch, scratch, (unsigned)imm); 147 else 148 subw(scratch, scratch, (unsigned)-imm); 149 strw(scratch, a); 150 } 151 152 void bind(Label& L) { 153 Assembler::bind(L); 154 code()->clear_last_membar(); 155 } 156 157 void membar(Membar_mask_bits order_constraint); 158 159 // Frame creation and destruction shared between JITs. 160 void build_frame(int framesize); 161 void remove_frame(int framesize); 162 163 virtual void _call_Unimplemented(address call_site) { 164 mov(rscratch2, call_site); 165 haltsim(); 166 } 167 168 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__) 169 170 virtual void notify(int type); 171 172 // aliases defined in AARCH64 spec 173 174 template<class T> 175 inline void cmpw(Register Rd, T imm) { subsw(zr, Rd, imm); } 176 inline void cmp(Register Rd, unsigned imm) { subs(zr, Rd, imm); } 177 178 inline void cmnw(Register Rd, unsigned imm) { addsw(zr, Rd, imm); } 179 inline void cmn(Register Rd, unsigned imm) { adds(zr, Rd, imm); } 180 181 void cset(Register Rd, Assembler::Condition cond) { 182 csinc(Rd, zr, zr, ~cond); 183 } 184 void csetw(Register Rd, Assembler::Condition cond) { 185 csincw(Rd, zr, zr, ~cond); 186 } 187 188 void cneg(Register Rd, Register Rn, Assembler::Condition cond) { 189 csneg(Rd, Rn, Rn, ~cond); 190 } 191 void cnegw(Register Rd, Register Rn, Assembler::Condition cond) { 192 csnegw(Rd, Rn, Rn, ~cond); 193 } 194 195 inline void movw(Register Rd, Register Rn) { 196 if (Rd == sp || Rn == sp) { 197 addw(Rd, Rn, 0U); 198 } else { 199 orrw(Rd, zr, Rn); 200 } 201 } 202 inline void mov(Register Rd, Register Rn) { 203 assert(Rd != r31_sp && Rn != r31_sp, "should be"); 204 if (Rd == Rn) { 205 } else if (Rd == sp || Rn == sp) { 206 add(Rd, Rn, 0U); 207 } else { 208 orr(Rd, zr, Rn); 209 } 210 } 211 212 inline void moviw(Register Rd, unsigned imm) { orrw(Rd, zr, imm); } 213 inline void movi(Register Rd, unsigned imm) { orr(Rd, zr, imm); } 214 215 inline void tstw(Register Rd, Register Rn) { andsw(zr, Rd, Rn); } 216 inline void tst(Register Rd, Register Rn) { ands(zr, Rd, Rn); } 217 218 inline void tstw(Register Rd, uint64_t imm) { andsw(zr, Rd, imm); } 219 inline void tst(Register Rd, uint64_t imm) { ands(zr, Rd, imm); } 220 221 inline void bfiw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 222 bfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 223 } 224 inline void bfi(Register Rd, Register Rn, unsigned lsb, unsigned width) { 225 bfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 226 } 227 228 inline void bfxilw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 229 bfmw(Rd, Rn, lsb, (lsb + width - 1)); 230 } 231 inline void bfxil(Register Rd, Register Rn, unsigned lsb, unsigned width) { 232 bfm(Rd, Rn, lsb , (lsb + width - 1)); 233 } 234 235 inline void sbfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 236 sbfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 237 } 238 inline void sbfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) { 239 sbfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 240 } 241 242 inline void sbfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 243 sbfmw(Rd, Rn, lsb, (lsb + width - 1)); 244 } 245 inline void sbfx(Register Rd, Register Rn, unsigned lsb, unsigned width) { 246 sbfm(Rd, Rn, lsb , (lsb + width - 1)); 247 } 248 249 inline void ubfizw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 250 ubfmw(Rd, Rn, ((32 - lsb) & 31), (width - 1)); 251 } 252 inline void ubfiz(Register Rd, Register Rn, unsigned lsb, unsigned width) { 253 ubfm(Rd, Rn, ((64 - lsb) & 63), (width - 1)); 254 } 255 256 inline void ubfxw(Register Rd, Register Rn, unsigned lsb, unsigned width) { 257 ubfmw(Rd, Rn, lsb, (lsb + width - 1)); 258 } 259 inline void ubfx(Register Rd, Register Rn, unsigned lsb, unsigned width) { 260 ubfm(Rd, Rn, lsb , (lsb + width - 1)); 261 } 262 263 inline void asrw(Register Rd, Register Rn, unsigned imm) { 264 sbfmw(Rd, Rn, imm, 31); 265 } 266 267 inline void asr(Register Rd, Register Rn, unsigned imm) { 268 sbfm(Rd, Rn, imm, 63); 269 } 270 271 inline void lslw(Register Rd, Register Rn, unsigned imm) { 272 ubfmw(Rd, Rn, ((32 - imm) & 31), (31 - imm)); 273 } 274 275 inline void lsl(Register Rd, Register Rn, unsigned imm) { 276 ubfm(Rd, Rn, ((64 - imm) & 63), (63 - imm)); 277 } 278 279 inline void lsrw(Register Rd, Register Rn, unsigned imm) { 280 ubfmw(Rd, Rn, imm, 31); 281 } 282 283 inline void lsr(Register Rd, Register Rn, unsigned imm) { 284 ubfm(Rd, Rn, imm, 63); 285 } 286 287 inline void rorw(Register Rd, Register Rn, unsigned imm) { 288 extrw(Rd, Rn, Rn, imm); 289 } 290 291 inline void ror(Register Rd, Register Rn, unsigned imm) { 292 extr(Rd, Rn, Rn, imm); 293 } 294 295 inline void sxtbw(Register Rd, Register Rn) { 296 sbfmw(Rd, Rn, 0, 7); 297 } 298 inline void sxthw(Register Rd, Register Rn) { 299 sbfmw(Rd, Rn, 0, 15); 300 } 301 inline void sxtb(Register Rd, Register Rn) { 302 sbfm(Rd, Rn, 0, 7); 303 } 304 inline void sxth(Register Rd, Register Rn) { 305 sbfm(Rd, Rn, 0, 15); 306 } 307 inline void sxtw(Register Rd, Register Rn) { 308 sbfm(Rd, Rn, 0, 31); 309 } 310 311 inline void uxtbw(Register Rd, Register Rn) { 312 ubfmw(Rd, Rn, 0, 7); 313 } 314 inline void uxthw(Register Rd, Register Rn) { 315 ubfmw(Rd, Rn, 0, 15); 316 } 317 inline void uxtb(Register Rd, Register Rn) { 318 ubfm(Rd, Rn, 0, 7); 319 } 320 inline void uxth(Register Rd, Register Rn) { 321 ubfm(Rd, Rn, 0, 15); 322 } 323 inline void uxtw(Register Rd, Register Rn) { 324 ubfm(Rd, Rn, 0, 31); 325 } 326 327 inline void cmnw(Register Rn, Register Rm) { 328 addsw(zr, Rn, Rm); 329 } 330 inline void cmn(Register Rn, Register Rm) { 331 adds(zr, Rn, Rm); 332 } 333 334 inline void cmpw(Register Rn, Register Rm) { 335 subsw(zr, Rn, Rm); 336 } 337 inline void cmp(Register Rn, Register Rm) { 338 subs(zr, Rn, Rm); 339 } 340 341 inline void negw(Register Rd, Register Rn) { 342 subw(Rd, zr, Rn); 343 } 344 345 inline void neg(Register Rd, Register Rn) { 346 sub(Rd, zr, Rn); 347 } 348 349 inline void negsw(Register Rd, Register Rn) { 350 subsw(Rd, zr, Rn); 351 } 352 353 inline void negs(Register Rd, Register Rn) { 354 subs(Rd, zr, Rn); 355 } 356 357 inline void cmnw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 358 addsw(zr, Rn, Rm, kind, shift); 359 } 360 inline void cmn(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 361 adds(zr, Rn, Rm, kind, shift); 362 } 363 364 inline void cmpw(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 365 subsw(zr, Rn, Rm, kind, shift); 366 } 367 inline void cmp(Register Rn, Register Rm, enum shift_kind kind, unsigned shift = 0) { 368 subs(zr, Rn, Rm, kind, shift); 369 } 370 371 inline void negw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 372 subw(Rd, zr, Rn, kind, shift); 373 } 374 375 inline void neg(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 376 sub(Rd, zr, Rn, kind, shift); 377 } 378 379 inline void negsw(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 380 subsw(Rd, zr, Rn, kind, shift); 381 } 382 383 inline void negs(Register Rd, Register Rn, enum shift_kind kind, unsigned shift = 0) { 384 subs(Rd, zr, Rn, kind, shift); 385 } 386 387 inline void mnegw(Register Rd, Register Rn, Register Rm) { 388 msubw(Rd, Rn, Rm, zr); 389 } 390 inline void mneg(Register Rd, Register Rn, Register Rm) { 391 msub(Rd, Rn, Rm, zr); 392 } 393 394 inline void mulw(Register Rd, Register Rn, Register Rm) { 395 maddw(Rd, Rn, Rm, zr); 396 } 397 inline void mul(Register Rd, Register Rn, Register Rm) { 398 madd(Rd, Rn, Rm, zr); 399 } 400 401 inline void smnegl(Register Rd, Register Rn, Register Rm) { 402 smsubl(Rd, Rn, Rm, zr); 403 } 404 inline void smull(Register Rd, Register Rn, Register Rm) { 405 smaddl(Rd, Rn, Rm, zr); 406 } 407 408 inline void umnegl(Register Rd, Register Rn, Register Rm) { 409 umsubl(Rd, Rn, Rm, zr); 410 } 411 inline void umull(Register Rd, Register Rn, Register Rm) { 412 umaddl(Rd, Rn, Rm, zr); 413 } 414 415 #define WRAP(INSN) \ 416 void INSN(Register Rd, Register Rn, Register Rm, Register Ra) { \ 417 if ((VM_Version::features() & VM_Version::CPU_A53MAC) && Ra != zr) \ 418 nop(); \ 419 Assembler::INSN(Rd, Rn, Rm, Ra); \ 420 } 421 422 WRAP(madd) WRAP(msub) WRAP(maddw) WRAP(msubw) 423 WRAP(smaddl) WRAP(smsubl) WRAP(umaddl) WRAP(umsubl) 424 #undef WRAP 425 426 427 // macro assembly operations needed for aarch64 428 429 // first two private routines for loading 32 bit or 64 bit constants 430 private: 431 432 void mov_immediate64(Register dst, u_int64_t imm64); 433 void mov_immediate32(Register dst, u_int32_t imm32); 434 435 int push(unsigned int bitset, Register stack); 436 int pop(unsigned int bitset, Register stack); 437 438 void mov(Register dst, Address a); 439 440 public: 441 void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); } 442 void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); } 443 444 // Push and pop everything that might be clobbered by a native 445 // runtime call except rscratch1 and rscratch2. (They are always 446 // scratch, so we don't have to protect them.) Only save the lower 447 // 64 bits of each vector register. 448 void push_call_clobbered_registers(); 449 void pop_call_clobbered_registers(); 450 451 // now mov instructions for loading absolute addresses and 32 or 452 // 64 bit integers 453 454 inline void mov(Register dst, address addr) { 455 assert(Universe::heap() == NULL 456 || !Universe::heap()->is_in(addr), "use movptr for oop pointers"); 457 mov_immediate64(dst, (uintptr_t)addr); 458 } 459 460 inline void mov(Register dst, u_int64_t imm64) 461 { 462 mov_immediate64(dst, imm64); 463 } 464 465 inline void movw(Register dst, u_int32_t imm32) 466 { 467 mov_immediate32(dst, imm32); 468 } 469 470 inline void mov(Register dst, long l) 471 { 472 mov(dst, (u_int64_t)l); 473 } 474 475 inline void mov(Register dst, int i) 476 { 477 mov(dst, (long)i); 478 } 479 480 void mov(Register dst, RegisterOrConstant src) { 481 if (src.is_register()) 482 mov(dst, src.as_register()); 483 else 484 mov(dst, src.as_constant()); 485 } 486 487 void movptr(Register r, uintptr_t imm64); 488 489 void mov(FloatRegister Vd, SIMD_Arrangement T, u_int32_t imm32); 490 491 void mov(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) { 492 orr(Vd, T, Vn, Vn); 493 } 494 495 public: 496 497 // Generalized Test Bit And Branch, including a "far" variety which 498 // spans more than 32KiB. 499 void tbr(Condition cond, Register Rt, int bitpos, Label &dest, bool far = false) { 500 assert(cond == EQ || cond == NE, "must be"); 501 502 if (far) 503 cond = ~cond; 504 505 void (Assembler::* branch)(Register Rt, int bitpos, Label &L); 506 if (cond == Assembler::EQ) 507 branch = &Assembler::tbz; 508 else 509 branch = &Assembler::tbnz; 510 511 if (far) { 512 Label L; 513 (this->*branch)(Rt, bitpos, L); 514 b(dest); 515 bind(L); 516 } else { 517 (this->*branch)(Rt, bitpos, dest); 518 } 519 } 520 521 // macro instructions for accessing and updating floating point 522 // status register 523 // 524 // FPSR : op1 == 011 525 // CRn == 0100 526 // CRm == 0100 527 // op2 == 001 528 529 inline void get_fpsr(Register reg) 530 { 531 mrs(0b11, 0b0100, 0b0100, 0b001, reg); 532 } 533 534 inline void set_fpsr(Register reg) 535 { 536 msr(0b011, 0b0100, 0b0100, 0b001, reg); 537 } 538 539 inline void clear_fpsr() 540 { 541 msr(0b011, 0b0100, 0b0100, 0b001, zr); 542 } 543 544 // Macro instructions for accessing and updating the condition flags 545 inline void get_cflags(Register reg) 546 { 547 mrs(0b011, 0b0100, 0b0010, 0b000, reg); 548 } 549 550 inline void set_cflags(Register reg) 551 { 552 msr(0b011, 0b0100, 0b0010, 0b000, reg); 553 } 554 555 // DCZID_EL0: op1 == 011 556 // CRn == 0000 557 // CRm == 0000 558 // op2 == 111 559 inline void get_dczid_el0(Register reg) 560 { 561 mrs(0b011, 0b0000, 0b0000, 0b111, reg); 562 } 563 564 // CTR_EL0: op1 == 011 565 // CRn == 0000 566 // CRm == 0000 567 // op2 == 001 568 inline void get_ctr_el0(Register reg) 569 { 570 mrs(0b011, 0b0000, 0b0000, 0b001, reg); 571 } 572 573 // idiv variant which deals with MINLONG as dividend and -1 as divisor 574 int corrected_idivl(Register result, Register ra, Register rb, 575 bool want_remainder, Register tmp = rscratch1); 576 int corrected_idivq(Register result, Register ra, Register rb, 577 bool want_remainder, Register tmp = rscratch1); 578 579 // Support for NULL-checks 580 // 581 // Generates code that causes a NULL OS exception if the content of reg is NULL. 582 // If the accessed location is M[reg + offset] and the offset is known, provide the 583 // offset. No explicit code generation is needed if the offset is within a certain 584 // range (0 <= offset <= page_size). 585 586 virtual void null_check(Register reg, int offset = -1); 587 static bool needs_explicit_null_check(intptr_t offset); 588 589 static address target_addr_for_insn(address insn_addr, unsigned insn); 590 static address target_addr_for_insn(address insn_addr) { 591 unsigned insn = *(unsigned*)insn_addr; 592 return target_addr_for_insn(insn_addr, insn); 593 } 594 595 // Required platform-specific helpers for Label::patch_instructions. 596 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 597 static int pd_patch_instruction_size(address branch, address target); 598 static void pd_patch_instruction(address branch, address target) { 599 pd_patch_instruction_size(branch, target); 600 } 601 static address pd_call_destination(address branch) { 602 return target_addr_for_insn(branch); 603 } 604 #ifndef PRODUCT 605 static void pd_print_patched_instruction(address branch); 606 #endif 607 608 static int patch_oop(address insn_addr, address o); 609 610 address emit_trampoline_stub(int insts_call_instruction_offset, address target); 611 612 // The following 4 methods return the offset of the appropriate move instruction 613 614 // Support for fast byte/short loading with zero extension (depending on particular CPU) 615 int load_unsigned_byte(Register dst, Address src); 616 int load_unsigned_short(Register dst, Address src); 617 618 // Support for fast byte/short loading with sign extension (depending on particular CPU) 619 int load_signed_byte(Register dst, Address src); 620 int load_signed_short(Register dst, Address src); 621 622 int load_signed_byte32(Register dst, Address src); 623 int load_signed_short32(Register dst, Address src); 624 625 // Support for sign-extension (hi:lo = extend_sign(lo)) 626 void extend_sign(Register hi, Register lo); 627 628 // Load and store values by size and signed-ness 629 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); 630 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); 631 632 // Support for inc/dec with optimal instruction selection depending on value 633 634 // x86_64 aliases an unqualified register/address increment and 635 // decrement to call incrementq and decrementq but also supports 636 // explicitly sized calls to incrementq/decrementq or 637 // incrementl/decrementl 638 639 // for aarch64 the proper convention would be to use 640 // increment/decrement for 64 bit operatons and 641 // incrementw/decrementw for 32 bit operations. so when porting 642 // x86_64 code we can leave calls to increment/decrement as is, 643 // replace incrementq/decrementq with increment/decrement and 644 // replace incrementl/decrementl with incrementw/decrementw. 645 646 // n.b. increment/decrement calls with an Address destination will 647 // need to use a scratch register to load the value to be 648 // incremented. increment/decrement calls which add or subtract a 649 // constant value greater than 2^12 will need to use a 2nd scratch 650 // register to hold the constant. so, a register increment/decrement 651 // may trash rscratch2 and an address increment/decrement trash 652 // rscratch and rscratch2 653 654 void decrementw(Address dst, int value = 1); 655 void decrementw(Register reg, int value = 1); 656 657 void decrement(Register reg, int value = 1); 658 void decrement(Address dst, int value = 1); 659 660 void incrementw(Address dst, int value = 1); 661 void incrementw(Register reg, int value = 1); 662 663 void increment(Register reg, int value = 1); 664 void increment(Address dst, int value = 1); 665 666 667 // Alignment 668 void align(int modulus); 669 670 // Stack frame creation/removal 671 void enter() 672 { 673 stp(rfp, lr, Address(pre(sp, -2 * wordSize))); 674 mov(rfp, sp); 675 } 676 void leave() 677 { 678 mov(sp, rfp); 679 ldp(rfp, lr, Address(post(sp, 2 * wordSize))); 680 } 681 682 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 683 // The pointer will be loaded into the thread register. 684 void get_thread(Register thread); 685 686 687 // Support for VM calls 688 // 689 // It is imperative that all calls into the VM are handled via the call_VM macros. 690 // They make sure that the stack linkage is setup correctly. call_VM's correspond 691 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 692 693 694 void call_VM(Register oop_result, 695 address entry_point, 696 bool check_exceptions = true); 697 void call_VM(Register oop_result, 698 address entry_point, 699 Register arg_1, 700 bool check_exceptions = true); 701 void call_VM(Register oop_result, 702 address entry_point, 703 Register arg_1, Register arg_2, 704 bool check_exceptions = true); 705 void call_VM(Register oop_result, 706 address entry_point, 707 Register arg_1, Register arg_2, Register arg_3, 708 bool check_exceptions = true); 709 710 // Overloadings with last_Java_sp 711 void call_VM(Register oop_result, 712 Register last_java_sp, 713 address entry_point, 714 int number_of_arguments = 0, 715 bool check_exceptions = true); 716 void call_VM(Register oop_result, 717 Register last_java_sp, 718 address entry_point, 719 Register arg_1, bool 720 check_exceptions = true); 721 void call_VM(Register oop_result, 722 Register last_java_sp, 723 address entry_point, 724 Register arg_1, Register arg_2, 725 bool check_exceptions = true); 726 void call_VM(Register oop_result, 727 Register last_java_sp, 728 address entry_point, 729 Register arg_1, Register arg_2, Register arg_3, 730 bool check_exceptions = true); 731 732 void get_vm_result (Register oop_result, Register thread); 733 void get_vm_result_2(Register metadata_result, Register thread); 734 735 // These always tightly bind to MacroAssembler::call_VM_base 736 // bypassing the virtual implementation 737 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); 738 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); 739 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); 740 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); 741 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); 742 743 void call_VM_leaf(address entry_point, 744 int number_of_arguments = 0); 745 void call_VM_leaf(address entry_point, 746 Register arg_1); 747 void call_VM_leaf(address entry_point, 748 Register arg_1, Register arg_2); 749 void call_VM_leaf(address entry_point, 750 Register arg_1, Register arg_2, Register arg_3); 751 752 // These always tightly bind to MacroAssembler::call_VM_leaf_base 753 // bypassing the virtual implementation 754 void super_call_VM_leaf(address entry_point); 755 void super_call_VM_leaf(address entry_point, Register arg_1); 756 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); 757 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); 758 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); 759 760 // last Java Frame (fills frame anchor) 761 void set_last_Java_frame(Register last_java_sp, 762 Register last_java_fp, 763 address last_java_pc, 764 Register scratch); 765 766 void set_last_Java_frame(Register last_java_sp, 767 Register last_java_fp, 768 Label &last_java_pc, 769 Register scratch); 770 771 void set_last_Java_frame(Register last_java_sp, 772 Register last_java_fp, 773 Register last_java_pc, 774 Register scratch); 775 776 void reset_last_Java_frame(Register thread); 777 778 // thread in the default location (rthread) 779 void reset_last_Java_frame(bool clear_fp); 780 781 // Stores 782 void store_check(Register obj); // store check for obj - register is destroyed afterwards 783 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 784 785 #if INCLUDE_ALL_GCS 786 787 void g1_write_barrier_pre(Register obj, 788 Register pre_val, 789 Register thread, 790 Register tmp, 791 bool tosca_live, 792 bool expand_call); 793 794 void g1_write_barrier_post(Register store_addr, 795 Register new_val, 796 Register thread, 797 Register tmp, 798 Register tmp2); 799 800 #endif // INCLUDE_ALL_GCS 801 802 // oop manipulations 803 void load_klass(Register dst, Register src); 804 void store_klass(Register dst, Register src); 805 void cmp_klass(Register oop, Register trial_klass, Register tmp); 806 807 void load_mirror(Register dst, Register method); 808 809 void load_heap_oop(Register dst, Address src); 810 811 void load_heap_oop_not_null(Register dst, Address src); 812 void store_heap_oop(Address dst, Register src); 813 814 // currently unimplemented 815 // Used for storing NULL. All other oop constants should be 816 // stored using routines that take a jobject. 817 void store_heap_oop_null(Address dst); 818 819 void load_prototype_header(Register dst, Register src); 820 821 void store_klass_gap(Register dst, Register src); 822 823 // This dummy is to prevent a call to store_heap_oop from 824 // converting a zero (like NULL) into a Register by giving 825 // the compiler two choices it can't resolve 826 827 void store_heap_oop(Address dst, void* dummy); 828 829 void encode_heap_oop(Register d, Register s); 830 void encode_heap_oop(Register r) { encode_heap_oop(r, r); } 831 void decode_heap_oop(Register d, Register s); 832 void decode_heap_oop(Register r) { decode_heap_oop(r, r); } 833 void encode_heap_oop_not_null(Register r); 834 void decode_heap_oop_not_null(Register r); 835 void encode_heap_oop_not_null(Register dst, Register src); 836 void decode_heap_oop_not_null(Register dst, Register src); 837 838 void set_narrow_oop(Register dst, jobject obj); 839 840 void encode_klass_not_null(Register r); 841 void decode_klass_not_null(Register r); 842 void encode_klass_not_null(Register dst, Register src); 843 void decode_klass_not_null(Register dst, Register src); 844 845 void set_narrow_klass(Register dst, Klass* k); 846 847 // if heap base register is used - reinit it with the correct value 848 void reinit_heapbase(); 849 850 DEBUG_ONLY(void verify_heapbase(const char* msg);) 851 852 void push_CPU_state(bool save_vectors = false); 853 void pop_CPU_state(bool restore_vectors = false) ; 854 855 // Round up to a power of two 856 void round_to(Register reg, int modulus); 857 858 // allocation 859 void eden_allocate( 860 Register obj, // result: pointer to object after successful allocation 861 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 862 int con_size_in_bytes, // object size in bytes if known at compile time 863 Register t1, // temp register 864 Label& slow_case // continuation point if fast allocation fails 865 ); 866 void tlab_allocate( 867 Register obj, // result: pointer to object after successful allocation 868 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 869 int con_size_in_bytes, // object size in bytes if known at compile time 870 Register t1, // temp register 871 Register t2, // temp register 872 Label& slow_case // continuation point if fast allocation fails 873 ); 874 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address 875 void verify_tlab(); 876 877 void incr_allocated_bytes(Register thread, 878 Register var_size_in_bytes, int con_size_in_bytes, 879 Register t1 = noreg); 880 881 // interface method calling 882 void lookup_interface_method(Register recv_klass, 883 Register intf_klass, 884 RegisterOrConstant itable_index, 885 Register method_result, 886 Register scan_temp, 887 Label& no_such_interface); 888 889 // virtual method calling 890 // n.b. x86 allows RegisterOrConstant for vtable_index 891 void lookup_virtual_method(Register recv_klass, 892 RegisterOrConstant vtable_index, 893 Register method_result); 894 895 // Test sub_klass against super_klass, with fast and slow paths. 896 897 // The fast path produces a tri-state answer: yes / no / maybe-slow. 898 // One of the three labels can be NULL, meaning take the fall-through. 899 // If super_check_offset is -1, the value is loaded up from super_klass. 900 // No registers are killed, except temp_reg. 901 void check_klass_subtype_fast_path(Register sub_klass, 902 Register super_klass, 903 Register temp_reg, 904 Label* L_success, 905 Label* L_failure, 906 Label* L_slow_path, 907 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 908 909 // The rest of the type check; must be wired to a corresponding fast path. 910 // It does not repeat the fast path logic, so don't use it standalone. 911 // The temp_reg and temp2_reg can be noreg, if no temps are available. 912 // Updates the sub's secondary super cache as necessary. 913 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 914 void check_klass_subtype_slow_path(Register sub_klass, 915 Register super_klass, 916 Register temp_reg, 917 Register temp2_reg, 918 Label* L_success, 919 Label* L_failure, 920 bool set_cond_codes = false); 921 922 // Simplified, combined version, good for typical uses. 923 // Falls through on failure. 924 void check_klass_subtype(Register sub_klass, 925 Register super_klass, 926 Register temp_reg, 927 Label& L_success); 928 929 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 930 931 932 // Debugging 933 934 // only if +VerifyOops 935 void verify_oop(Register reg, const char* s = "broken oop"); 936 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 937 938 // TODO: verify method and klass metadata (compare against vptr?) 939 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {} 940 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){} 941 942 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__) 943 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__) 944 945 // only if +VerifyFPU 946 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 947 948 // prints msg, dumps registers and stops execution 949 void stop(const char* msg, Label *l = NULL); 950 951 // prints msg and continues 952 void warn(const char* msg); 953 954 static void debug64(char* msg, int64_t pc, int64_t regs[]); 955 956 void untested() { stop("untested"); } 957 958 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } 959 960 void should_not_reach_here() { stop("should not reach here"); } 961 962 // Stack overflow checking 963 void bang_stack_with_offset(int offset) { 964 // stack grows down, caller passes positive offset 965 assert(offset > 0, "must bang with negative offset"); 966 mov(rscratch2, -offset); 967 str(zr, Address(sp, rscratch2)); 968 } 969 970 // Writes to stack successive pages until offset reached to check for 971 // stack overflow + shadow pages. Also, clobbers tmp 972 void bang_stack_size(Register size, Register tmp); 973 974 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 975 Register tmp, 976 int offset); 977 978 // Support for serializing memory accesses between threads 979 void serialize_memory(Register thread, Register tmp); 980 981 // Arithmetics 982 983 void addptr(const Address &dst, int32_t src); 984 void cmpptr(Register src1, Address src2); 985 986 // Various forms of CAS 987 988 void cmpxchgptr(Register oldv, Register newv, Register addr, Register tmp, 989 Label &suceed, Label *fail); 990 991 void cmpxchgw(Register oldv, Register newv, Register addr, Register tmp, 992 Label &suceed, Label *fail); 993 994 void atomic_add(Register prev, RegisterOrConstant incr, Register addr); 995 void atomic_addw(Register prev, RegisterOrConstant incr, Register addr); 996 void atomic_addal(Register prev, RegisterOrConstant incr, Register addr); 997 void atomic_addalw(Register prev, RegisterOrConstant incr, Register addr); 998 999 void atomic_xchg(Register prev, Register newv, Register addr); 1000 void atomic_xchgw(Register prev, Register newv, Register addr); 1001 void atomic_xchgal(Register prev, Register newv, Register addr); 1002 void atomic_xchgalw(Register prev, Register newv, Register addr); 1003 1004 void orptr(Address adr, RegisterOrConstant src) { 1005 ldr(rscratch2, adr); 1006 if (src.is_register()) 1007 orr(rscratch2, rscratch2, src.as_register()); 1008 else 1009 orr(rscratch2, rscratch2, src.as_constant()); 1010 str(rscratch2, adr); 1011 } 1012 1013 // A generic CAS; success or failure is in the EQ flag. 1014 // Clobbers rscratch1 1015 void cmpxchg(Register addr, Register expected, Register new_val, 1016 enum operand_size size, 1017 bool acquire, bool release, bool weak, 1018 Register result); 1019 1020 void cmpxchg_oop_shenandoah(Register res, Register addr, Register expected, Register new_val, 1021 bool narrow, 1022 bool acquire, bool release, 1023 Register tmp1 = rscratch1, Register tmp2 = rscratch2); 1024 // Calls 1025 1026 address trampoline_call(Address entry, CodeBuffer *cbuf = NULL); 1027 1028 static bool far_branches() { 1029 return ReservedCodeCacheSize > branch_range; 1030 } 1031 1032 // Jumps that can reach anywhere in the code cache. 1033 // Trashes tmp. 1034 void far_call(Address entry, CodeBuffer *cbuf = NULL, Register tmp = rscratch1); 1035 void far_jump(Address entry, CodeBuffer *cbuf = NULL, Register tmp = rscratch1); 1036 1037 static int far_branch_size() { 1038 if (far_branches()) { 1039 return 3 * 4; // adrp, add, br 1040 } else { 1041 return 4; 1042 } 1043 } 1044 1045 // Emit the CompiledIC call idiom 1046 address ic_call(address entry, jint method_index = 0); 1047 1048 public: 1049 1050 // Data 1051 1052 void mov_metadata(Register dst, Metadata* obj); 1053 Address allocate_metadata_address(Metadata* obj); 1054 Address constant_oop_address(jobject obj); 1055 1056 void movoop(Register dst, jobject obj, bool immediate = false); 1057 1058 // CRC32 code for java.util.zip.CRC32::updateBytes() instrinsic. 1059 void kernel_crc32(Register crc, Register buf, Register len, 1060 Register table0, Register table1, Register table2, Register table3, 1061 Register tmp, Register tmp2, Register tmp3); 1062 // CRC32 code for java.util.zip.CRC32C::updateBytes() instrinsic. 1063 void kernel_crc32c(Register crc, Register buf, Register len, 1064 Register table0, Register table1, Register table2, Register table3, 1065 Register tmp, Register tmp2, Register tmp3); 1066 1067 // Stack push and pop individual 64 bit registers 1068 void push(Register src); 1069 void pop(Register dst); 1070 1071 // push all registers onto the stack 1072 void pusha(); 1073 void popa(); 1074 1075 void repne_scan(Register addr, Register value, Register count, 1076 Register scratch); 1077 void repne_scanw(Register addr, Register value, Register count, 1078 Register scratch); 1079 1080 typedef void (MacroAssembler::* add_sub_imm_insn)(Register Rd, Register Rn, unsigned imm); 1081 typedef void (MacroAssembler::* add_sub_reg_insn)(Register Rd, Register Rn, Register Rm, enum shift_kind kind, unsigned shift); 1082 1083 // If a constant does not fit in an immediate field, generate some 1084 // number of MOV instructions and then perform the operation 1085 void wrap_add_sub_imm_insn(Register Rd, Register Rn, unsigned imm, 1086 add_sub_imm_insn insn1, 1087 add_sub_reg_insn insn2); 1088 // Seperate vsn which sets the flags 1089 void wrap_adds_subs_imm_insn(Register Rd, Register Rn, unsigned imm, 1090 add_sub_imm_insn insn1, 1091 add_sub_reg_insn insn2); 1092 1093 #define WRAP(INSN) \ 1094 void INSN(Register Rd, Register Rn, unsigned imm) { \ 1095 wrap_add_sub_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN); \ 1096 } \ 1097 \ 1098 void INSN(Register Rd, Register Rn, Register Rm, \ 1099 enum shift_kind kind, unsigned shift = 0) { \ 1100 Assembler::INSN(Rd, Rn, Rm, kind, shift); \ 1101 } \ 1102 \ 1103 void INSN(Register Rd, Register Rn, Register Rm) { \ 1104 Assembler::INSN(Rd, Rn, Rm); \ 1105 } \ 1106 \ 1107 void INSN(Register Rd, Register Rn, Register Rm, \ 1108 ext::operation option, int amount = 0) { \ 1109 Assembler::INSN(Rd, Rn, Rm, option, amount); \ 1110 } 1111 1112 WRAP(add) WRAP(addw) WRAP(sub) WRAP(subw) 1113 1114 #undef WRAP 1115 #define WRAP(INSN) \ 1116 void INSN(Register Rd, Register Rn, unsigned imm) { \ 1117 wrap_adds_subs_imm_insn(Rd, Rn, imm, &Assembler::INSN, &Assembler::INSN); \ 1118 } \ 1119 \ 1120 void INSN(Register Rd, Register Rn, Register Rm, \ 1121 enum shift_kind kind, unsigned shift = 0) { \ 1122 Assembler::INSN(Rd, Rn, Rm, kind, shift); \ 1123 } \ 1124 \ 1125 void INSN(Register Rd, Register Rn, Register Rm) { \ 1126 Assembler::INSN(Rd, Rn, Rm); \ 1127 } \ 1128 \ 1129 void INSN(Register Rd, Register Rn, Register Rm, \ 1130 ext::operation option, int amount = 0) { \ 1131 Assembler::INSN(Rd, Rn, Rm, option, amount); \ 1132 } 1133 1134 WRAP(adds) WRAP(addsw) WRAP(subs) WRAP(subsw) 1135 1136 void add(Register Rd, Register Rn, RegisterOrConstant increment); 1137 void addw(Register Rd, Register Rn, RegisterOrConstant increment); 1138 void sub(Register Rd, Register Rn, RegisterOrConstant decrement); 1139 void subw(Register Rd, Register Rn, RegisterOrConstant decrement); 1140 1141 void adrp(Register reg1, const Address &dest, unsigned long &byte_offset); 1142 1143 void tableswitch(Register index, jint lowbound, jint highbound, 1144 Label &jumptable, Label &jumptable_end, int stride = 1) { 1145 adr(rscratch1, jumptable); 1146 subsw(rscratch2, index, lowbound); 1147 subsw(zr, rscratch2, highbound - lowbound); 1148 br(Assembler::HS, jumptable_end); 1149 add(rscratch1, rscratch1, rscratch2, 1150 ext::sxtw, exact_log2(stride * Assembler::instruction_size)); 1151 br(rscratch1); 1152 } 1153 1154 // Form an address from base + offset in Rd. Rd may or may not 1155 // actually be used: you must use the Address that is returned. It 1156 // is up to you to ensure that the shift provided matches the size 1157 // of your data. 1158 Address form_address(Register Rd, Register base, long byte_offset, int shift); 1159 1160 // Return true iff an address is within the 48-bit AArch64 address 1161 // space. 1162 bool is_valid_AArch64_address(address a) { 1163 return ((uint64_t)a >> 48) == 0; 1164 } 1165 1166 // Load the base of the cardtable byte map into reg. 1167 void load_byte_map_base(Register reg); 1168 1169 // Prolog generator routines to support switch between x86 code and 1170 // generated ARM code 1171 1172 // routine to generate an x86 prolog for a stub function which 1173 // bootstraps into the generated ARM code which directly follows the 1174 // stub 1175 // 1176 1177 public: 1178 // enum used for aarch64--x86 linkage to define return type of x86 function 1179 enum ret_type { ret_type_void, ret_type_integral, ret_type_float, ret_type_double}; 1180 1181 #ifdef BUILTIN_SIM 1182 void c_stub_prolog(int gp_arg_count, int fp_arg_count, int ret_type, address *prolog_ptr = NULL); 1183 #else 1184 void c_stub_prolog(int gp_arg_count, int fp_arg_count, int ret_type) { } 1185 #endif 1186 1187 // special version of call_VM_leaf_base needed for aarch64 simulator 1188 // where we need to specify both the gp and fp arg counts and the 1189 // return type so that the linkage routine from aarch64 to x86 and 1190 // back knows which aarch64 registers to copy to x86 registers and 1191 // which x86 result register to copy back to an aarch64 register 1192 1193 void call_VM_leaf_base1( 1194 address entry_point, // the entry point 1195 int number_of_gp_arguments, // the number of gp reg arguments to pass 1196 int number_of_fp_arguments, // the number of fp reg arguments to pass 1197 ret_type type, // the return type for the call 1198 Label* retaddr = NULL 1199 ); 1200 1201 void ldr_constant(Register dest, const Address &const_addr) { 1202 if (NearCpool) { 1203 ldr(dest, const_addr); 1204 } else { 1205 unsigned long offset; 1206 adrp(dest, InternalAddress(const_addr.target()), offset); 1207 ldr(dest, Address(dest, offset)); 1208 } 1209 } 1210 1211 address read_polling_page(Register r, address page, relocInfo::relocType rtype); 1212 address read_polling_page(Register r, relocInfo::relocType rtype); 1213 1214 // CRC32 code for java.util.zip.CRC32::updateBytes() instrinsic. 1215 void update_byte_crc32(Register crc, Register val, Register table); 1216 void update_word_crc32(Register crc, Register v, Register tmp, 1217 Register table0, Register table1, Register table2, Register table3, 1218 bool upper = false); 1219 1220 void string_compare(Register str1, Register str2, 1221 Register cnt1, Register cnt2, Register result, 1222 Register tmp1, 1223 FloatRegister vtmp, FloatRegister vtmpZ, int ae); 1224 1225 void arrays_equals(Register a1, Register a2, 1226 Register result, Register cnt1, 1227 int elem_size, bool is_string); 1228 1229 void fill_words(Register base, Register cnt, Register value); 1230 void zero_words(Register base, u_int64_t cnt); 1231 void zero_words(Register base, Register cnt); 1232 void block_zero(Register base, Register cnt, bool is_large = false); 1233 1234 void byte_array_inflate(Register src, Register dst, Register len, 1235 FloatRegister vtmp1, FloatRegister vtmp2, 1236 FloatRegister vtmp3, Register tmp4); 1237 1238 void char_array_compress(Register src, Register dst, Register len, 1239 FloatRegister tmp1Reg, FloatRegister tmp2Reg, 1240 FloatRegister tmp3Reg, FloatRegister tmp4Reg, 1241 Register result); 1242 1243 void encode_iso_array(Register src, Register dst, 1244 Register len, Register result, 1245 FloatRegister Vtmp1, FloatRegister Vtmp2, 1246 FloatRegister Vtmp3, FloatRegister Vtmp4); 1247 void string_indexof(Register str1, Register str2, 1248 Register cnt1, Register cnt2, 1249 Register tmp1, Register tmp2, 1250 Register tmp3, Register tmp4, 1251 int int_cnt1, Register result, int ae); 1252 1253 void in_heap_check(Register r, Register tmp, Label &nope); 1254 1255 private: 1256 void shenandoah_cset_check(Register obj, Register tmp1, Register tmp2, Label& done); 1257 1258 public: 1259 void _shenandoah_store_addr_check(Register addr, const char* msg, const char* file, int line); 1260 void _shenandoah_store_addr_check(Address addr, const char* msg, const char* file, int line); 1261 #define shenandoah_store_addr_check(reg) _shenandoah_store_addr_check(reg, "oop not safe for writing", __FILE__, __LINE__) 1262 1263 void _shenandoah_store_check(Address addr, Register value, const char* msg, const char* file, int line); 1264 void _shenandoah_store_check(Register addr, Register value, const char* msg, const char* file, int line); 1265 #define shenandoah_store_check(addr, value) _shenandoah_store_check(addr, value, "oop not safe for writing", __FILE__, __LINE__) 1266 1267 private: 1268 void add2_with_carry(Register final_dest_hi, Register dest_hi, Register dest_lo, 1269 Register src1, Register src2); 1270 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2) { 1271 add2_with_carry(dest_hi, dest_hi, dest_lo, src1, src2); 1272 } 1273 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart, 1274 Register y, Register y_idx, Register z, 1275 Register carry, Register product, 1276 Register idx, Register kdx); 1277 void multiply_128_x_128_loop(Register y, Register z, 1278 Register carry, Register carry2, 1279 Register idx, Register jdx, 1280 Register yz_idx1, Register yz_idx2, 1281 Register tmp, Register tmp3, Register tmp4, 1282 Register tmp7, Register product_hi); 1283 public: 1284 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, 1285 Register zlen, Register tmp1, Register tmp2, Register tmp3, 1286 Register tmp4, Register tmp5, Register tmp6, Register tmp7); 1287 // ISB may be needed because of a safepoint 1288 void maybe_isb() { isb(); } 1289 1290 private: 1291 // Return the effective address r + (r1 << ext) + offset. 1292 // Uses rscratch2. 1293 Address offsetted_address(Register r, Register r1, Address::extend ext, 1294 int offset, int size); 1295 1296 private: 1297 // Returns an address on the stack which is reachable with a ldr/str of size 1298 // Uses rscratch2 if the address is not directly reachable 1299 Address spill_address(int size, int offset, Register tmp=rscratch2); 1300 1301 public: 1302 void spill(Register Rx, bool is64, int offset) { 1303 if (is64) { 1304 str(Rx, spill_address(8, offset)); 1305 } else { 1306 strw(Rx, spill_address(4, offset)); 1307 } 1308 } 1309 void spill(FloatRegister Vx, SIMD_RegVariant T, int offset) { 1310 str(Vx, T, spill_address(1 << (int)T, offset)); 1311 } 1312 void unspill(Register Rx, bool is64, int offset) { 1313 if (is64) { 1314 ldr(Rx, spill_address(8, offset)); 1315 } else { 1316 ldrw(Rx, spill_address(4, offset)); 1317 } 1318 } 1319 void unspill(FloatRegister Vx, SIMD_RegVariant T, int offset) { 1320 ldr(Vx, T, spill_address(1 << (int)T, offset)); 1321 } 1322 void spill_copy128(int src_offset, int dst_offset, 1323 Register tmp1=rscratch1, Register tmp2=rscratch2) { 1324 if (src_offset < 512 && (src_offset & 7) == 0 && 1325 dst_offset < 512 && (dst_offset & 7) == 0) { 1326 ldp(tmp1, tmp2, Address(sp, src_offset)); 1327 stp(tmp1, tmp2, Address(sp, dst_offset)); 1328 } else { 1329 unspill(tmp1, true, src_offset); 1330 spill(tmp1, true, dst_offset); 1331 unspill(tmp1, true, src_offset+8); 1332 spill(tmp1, true, dst_offset+8); 1333 } 1334 } 1335 }; 1336 1337 #ifdef ASSERT 1338 inline bool AbstractAssembler::pd_check_instruction_mark() { return false; } 1339 #endif 1340 1341 /** 1342 * class SkipIfEqual: 1343 * 1344 * Instantiating this class will result in assembly code being output that will 1345 * jump around any code emitted between the creation of the instance and it's 1346 * automatic destruction at the end of a scope block, depending on the value of 1347 * the flag passed to the constructor, which will be checked at run-time. 1348 */ 1349 class SkipIfEqual { 1350 private: 1351 MacroAssembler* _masm; 1352 Label _label; 1353 1354 public: 1355 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 1356 ~SkipIfEqual(); 1357 }; 1358 1359 struct tableswitch { 1360 Register _reg; 1361 int _insn_index; jint _first_key; jint _last_key; 1362 Label _after; 1363 Label _branches; 1364 }; 1365 1366 #endif // CPU_AARCH64_VM_MACROASSEMBLER_AARCH64_HPP