1 /* 2 * Copyright (c) 1997, 2011, 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_ASSEMBLER_AARCH64_HPP 27 #define CPU_AARCH64_VM_ASSEMBLER_AARCH64_HPP 28 29 #include "asm/register.hpp" 30 31 // definitions of various symbolic names for machine registers 32 33 // First intercalls between C and Java which use 8 general registers 34 // and 8 floating registers 35 36 // we also have to copy between x86 and ARM registers but that's a 37 // secondary complication -- not all code employing C call convention 38 // executes as x86 code though -- we generate some of it 39 40 class Argument VALUE_OBJ_CLASS_SPEC { 41 public: 42 enum { 43 n_int_register_parameters_c = 8, // r0, r1, ... r7 (c_rarg0, c_rarg1, ...) 44 n_float_register_parameters_c = 8, // v0, v1, ... v7 (c_farg0, c_farg1, ... ) 45 46 n_int_register_parameters_j = 8, // r1, ... r7, r0 (rj_rarg0, j_rarg1, ... 47 n_float_register_parameters_j = 8 // v0, v1, ... v7 (j_farg0, j_farg1, ... 48 }; 49 }; 50 51 REGISTER_DECLARATION(Register, c_rarg0, r0); 52 REGISTER_DECLARATION(Register, c_rarg1, r1); 53 REGISTER_DECLARATION(Register, c_rarg2, r2); 54 REGISTER_DECLARATION(Register, c_rarg3, r3); 55 REGISTER_DECLARATION(Register, c_rarg4, r4); 56 REGISTER_DECLARATION(Register, c_rarg5, r5); 57 REGISTER_DECLARATION(Register, c_rarg6, r6); 58 REGISTER_DECLARATION(Register, c_rarg7, r7); 59 60 REGISTER_DECLARATION(FloatRegister, c_farg0, v0); 61 REGISTER_DECLARATION(FloatRegister, c_farg1, v1); 62 REGISTER_DECLARATION(FloatRegister, c_farg2, v2); 63 REGISTER_DECLARATION(FloatRegister, c_farg3, v3); 64 REGISTER_DECLARATION(FloatRegister, c_farg4, v4); 65 REGISTER_DECLARATION(FloatRegister, c_farg5, v5); 66 REGISTER_DECLARATION(FloatRegister, c_farg6, v6); 67 REGISTER_DECLARATION(FloatRegister, c_farg7, v7); 68 69 // Symbolically name the register arguments used by the Java calling convention. 70 // We have control over the convention for java so we can do what we please. 71 // What pleases us is to offset the java calling convention so that when 72 // we call a suitable jni method the arguments are lined up and we don't 73 // have to do much shuffling. A suitable jni method is non-static and a 74 // small number of arguments 75 // 76 // |--------------------------------------------------------------------| 77 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 c_rarg6 c_rarg7 | 78 // |--------------------------------------------------------------------| 79 // | r0 r1 r2 r3 r4 r5 r6 r7 | 80 // |--------------------------------------------------------------------| 81 // | j_rarg7 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 j_rarg5 j_rarg6 | 82 // |--------------------------------------------------------------------| 83 84 85 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); 86 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); 87 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); 88 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); 89 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); 90 REGISTER_DECLARATION(Register, j_rarg5, c_rarg6); 91 REGISTER_DECLARATION(Register, j_rarg6, c_rarg7); 92 REGISTER_DECLARATION(Register, j_rarg7, c_rarg0); 93 94 // Java floating args are passed as per C 95 96 REGISTER_DECLARATION(FloatRegister, j_farg0, v0); 97 REGISTER_DECLARATION(FloatRegister, j_farg1, v1); 98 REGISTER_DECLARATION(FloatRegister, j_farg2, v2); 99 REGISTER_DECLARATION(FloatRegister, j_farg3, v3); 100 REGISTER_DECLARATION(FloatRegister, j_farg4, v4); 101 REGISTER_DECLARATION(FloatRegister, j_farg5, v5); 102 REGISTER_DECLARATION(FloatRegister, j_farg6, v6); 103 REGISTER_DECLARATION(FloatRegister, j_farg7, v7); 104 105 // registers used to hold VM data either temporarily within a method 106 // or across method calls 107 108 // volatile (caller-save) registers 109 110 // r8 is used for indirect result location return 111 // we use it and r9 as scratch registers 112 REGISTER_DECLARATION(Register, rscratch1, r8); 113 REGISTER_DECLARATION(Register, rscratch2, r9); 114 115 // current method -- must be in a call-clobbered register 116 REGISTER_DECLARATION(Register, rmethod, r12); 117 118 // non-volatile (callee-save) registers are r16-29 119 // of which the following are dedicated global state 120 121 // link register 122 REGISTER_DECLARATION(Register, lr, r30); 123 // frame pointer 124 REGISTER_DECLARATION(Register, rfp, r29); 125 // current thread 126 REGISTER_DECLARATION(Register, rthread, r28); 127 // base of heap 128 REGISTER_DECLARATION(Register, rheapbase, r27); 129 // constant pool cache 130 REGISTER_DECLARATION(Register, rcpool, r26); 131 // monitors allocated on stack 132 REGISTER_DECLARATION(Register, rmonitors, r25); 133 // locals on stack 134 REGISTER_DECLARATION(Register, rlocals, r24); 135 // bytecode pointer 136 REGISTER_DECLARATION(Register, rbcp, r22); 137 // Dispatch table base 138 REGISTER_DECLARATION(Register, rdispatch, r21); 139 // Java stack pointer 140 REGISTER_DECLARATION(Register, esp, r20); 141 142 // TODO : x86 uses rbp to save SP in method handle code 143 // we may need to do the same with fp 144 // JSR 292 fixed register usages: 145 //REGISTER_DECLARATION(Register, r_mh_SP_save, r29); 146 147 #define assert_cond(ARG1) assert(ARG1, #ARG1) 148 149 namespace asm_util { 150 uint32_t encode_logical_immediate(bool is32, uint64_t imm); 151 }; 152 153 using namespace asm_util; 154 155 156 class Assembler; 157 158 class Instruction_aarch64 { 159 unsigned insn; 160 #ifdef ASSERT 161 unsigned bits; 162 #endif 163 Assembler *assem; 164 165 public: 166 167 Instruction_aarch64(class Assembler *as) { 168 #ifdef ASSERT 169 bits = 0; 170 #endif 171 insn = 0; 172 assem = as; 173 } 174 175 inline ~Instruction_aarch64(); 176 177 unsigned &get_insn() { return insn; } 178 #ifdef ASSERT 179 unsigned &get_bits() { return bits; } 180 #endif 181 182 static inline int32_t extend(unsigned val, int hi = 31, int lo = 0) { 183 union { 184 unsigned u; 185 int n; 186 }; 187 188 u = val << (31 - hi); 189 n = n >> (31 - hi + lo); 190 return n; 191 } 192 193 static inline uint32_t extract(uint32_t val, int msb, int lsb) { 194 int nbits = msb - lsb + 1; 195 assert_cond(msb >= lsb); 196 uint32_t mask = (1U << nbits) - 1; 197 uint32_t result = val >> lsb; 198 result &= mask; 199 return result; 200 } 201 202 static inline int32_t sextract(uint32_t val, int msb, int lsb) { 203 uint32_t uval = extract(val, msb, lsb); 204 return extend(uval, msb - lsb); 205 } 206 207 static void patch(address a, int msb, int lsb, unsigned long val) { 208 int nbits = msb - lsb + 1; 209 guarantee(val < (1U << nbits), "Field too big for insn"); 210 assert_cond(msb >= lsb); 211 unsigned mask = (1U << nbits) - 1; 212 val <<= lsb; 213 mask <<= lsb; 214 unsigned target = *(unsigned *)a; 215 target &= ~mask; 216 target |= val; 217 *(unsigned *)a = target; 218 } 219 220 static void spatch(address a, int msb, int lsb, long val) { 221 int nbits = msb - lsb + 1; 222 long chk = val >> (nbits - 1); 223 guarantee (chk == -1 || chk == 0, "Field too big for insn"); 224 unsigned uval = val; 225 unsigned mask = (1U << nbits) - 1; 226 uval &= mask; 227 uval <<= lsb; 228 mask <<= lsb; 229 unsigned target = *(unsigned *)a; 230 target &= ~mask; 231 target |= uval; 232 *(unsigned *)a = target; 233 } 234 235 void f(unsigned val, int msb, int lsb) { 236 int nbits = msb - lsb + 1; 237 guarantee(val < (1U << nbits), "Field too big for insn"); 238 assert_cond(msb >= lsb); 239 unsigned mask = (1U << nbits) - 1; 240 val <<= lsb; 241 mask <<= lsb; 242 insn |= val; 243 assert_cond((bits & mask) == 0); 244 #ifdef ASSERT 245 bits |= mask; 246 #endif 247 } 248 249 void f(unsigned val, int bit) { 250 f(val, bit, bit); 251 } 252 253 void sf(long val, int msb, int lsb) { 254 int nbits = msb - lsb + 1; 255 long chk = val >> (nbits - 1); 256 guarantee (chk == -1 || chk == 0, "Field too big for insn"); 257 unsigned uval = val; 258 unsigned mask = (1U << nbits) - 1; 259 uval &= mask; 260 f(uval, lsb + nbits - 1, lsb); 261 } 262 263 void rf(Register r, int lsb) { 264 f(r->encoding_nocheck(), lsb + 4, lsb); 265 } 266 267 // reg|ZR 268 void zrf(Register r, int lsb) { 269 f(r->encoding_nocheck() - (r == zr), lsb + 4, lsb); 270 } 271 272 // reg|SP 273 void srf(Register r, int lsb) { 274 f(r == sp ? 31 : r->encoding_nocheck(), lsb + 4, lsb); 275 } 276 277 void rf(FloatRegister r, int lsb) { 278 f(r->encoding_nocheck(), lsb + 4, lsb); 279 } 280 281 unsigned get(int msb = 31, int lsb = 0) { 282 int nbits = msb - lsb + 1; 283 unsigned mask = ((1U << nbits) - 1) << lsb; 284 assert_cond(bits & mask == mask); 285 return (insn & mask) >> lsb; 286 } 287 288 void fixed(unsigned value, unsigned mask) { 289 assert_cond ((mask & bits) == 0); 290 #ifdef ASSERT 291 bits |= mask; 292 #endif 293 insn |= value; 294 } 295 }; 296 297 #define starti Instruction_aarch64 do_not_use(this); set_current(&do_not_use) 298 299 class PrePost { 300 int _offset; 301 Register _r; 302 public: 303 PrePost(Register reg, int o) : _r(reg), _offset(o) { } 304 int offset() { return _offset; } 305 Register reg() { return _r; } 306 }; 307 308 class Pre : public PrePost { 309 public: 310 Pre(Register reg, int o) : PrePost(reg, o) { } 311 }; 312 class Post : public PrePost { 313 public: 314 Post(Register reg, int o) : PrePost(reg, o) { } 315 }; 316 317 namespace ext 318 { 319 enum operation { uxtb, uxth, uxtw, uxtx, sxtb, sxth, sxtw, sxtx }; 320 }; 321 322 // abs methods which cannot overflow and so are well-defined across 323 // the entire domain of integer types. 324 static inline unsigned int uabs(unsigned int n) { 325 union { 326 unsigned int result; 327 int value; 328 }; 329 result = n; 330 if (value < 0) result = -result; 331 return result; 332 } 333 static inline unsigned long uabs(unsigned long n) { 334 union { 335 unsigned long result; 336 long value; 337 }; 338 result = n; 339 if (value < 0) result = -result; 340 return result; 341 } 342 static inline unsigned long uabs(long n) { return uabs((unsigned long)n); } 343 static inline unsigned long uabs(int n) { return uabs((unsigned int)n); } 344 345 // Addressing modes 346 class Address VALUE_OBJ_CLASS_SPEC { 347 public: 348 349 enum mode { no_mode, base_plus_offset, pre, post, pcrel, 350 base_plus_offset_reg, literal }; 351 352 // Shift and extend for base reg + reg offset addressing 353 class extend { 354 int _option, _shift; 355 ext::operation _op; 356 public: 357 extend() { } 358 extend(int s, int o, ext::operation op) : _shift(s), _option(o), _op(op) { } 359 int option() const{ return _option; } 360 int shift() const { return _shift; } 361 ext::operation op() const { return _op; } 362 }; 363 class uxtw : public extend { 364 public: 365 uxtw(int shift = -1): extend(shift, 0b010, ext::uxtw) { } 366 }; 367 class lsl : public extend { 368 public: 369 lsl(int shift = -1): extend(shift, 0b011, ext::uxtx) { } 370 }; 371 class sxtw : public extend { 372 public: 373 sxtw(int shift = -1): extend(shift, 0b110, ext::sxtw) { } 374 }; 375 class sxtx : public extend { 376 public: 377 sxtx(int shift = -1): extend(shift, 0b111, ext::sxtx) { } 378 }; 379 380 private: 381 Register _base; 382 Register _index; 383 long _offset; 384 enum mode _mode; 385 extend _ext; 386 387 RelocationHolder _rspec; 388 389 // Typically we use AddressLiterals we want to use their rval 390 // However in some situations we want the lval (effect address) of 391 // the item. We provide a special factory for making those lvals. 392 bool _is_lval; 393 394 // If the target is far we'll need to load the ea of this to a 395 // register to reach it. Otherwise if near we can do PC-relative 396 // addressing. 397 address _target; 398 399 public: 400 Address() 401 : _mode(no_mode) { } 402 Address(Register r) 403 : _mode(base_plus_offset), _base(r), _offset(0), _index(noreg), _target(0) { } 404 Address(Register r, int o) 405 : _mode(base_plus_offset), _base(r), _offset(o), _index(noreg), _target(0) { } 406 Address(Register r, long o) 407 : _mode(base_plus_offset), _base(r), _offset(o), _index(noreg), _target(0) { } 408 Address(Register r, unsigned long o) 409 : _mode(base_plus_offset), _base(r), _offset(o), _index(noreg), _target(0) { } 410 #ifdef ASSERT 411 Address(Register r, ByteSize disp) 412 : _mode(base_plus_offset), _base(r), _offset(in_bytes(disp)), 413 _index(noreg), _target(0) { } 414 #endif 415 Address(Register r, Register r1, extend ext = lsl()) 416 : _mode(base_plus_offset_reg), _base(r), _index(r1), 417 _ext(ext), _offset(0), _target(0) { } 418 Address(Pre p) 419 : _mode(pre), _base(p.reg()), _offset(p.offset()) { } 420 Address(Post p) 421 : _mode(post), _base(p.reg()), _offset(p.offset()), _target(0) { } 422 Address(address target, RelocationHolder const& rspec) 423 : _mode(literal), 424 _rspec(rspec), 425 _is_lval(false), 426 _target(target) { } 427 Address(address target, relocInfo::relocType rtype = relocInfo::external_word_type); 428 Address(Register base, RegisterOrConstant index, extend ext = lsl()) 429 : _base (base), 430 _ext(ext), _offset(0), _target(0) { 431 if (index.is_register()) { 432 _mode = base_plus_offset_reg; 433 _index = index.as_register(); 434 } else { 435 guarantee(ext.option() == ext::uxtx, "should be"); 436 assert(index.is_constant(), "should be"); 437 _mode = base_plus_offset; 438 _offset = index.as_constant() << ext.shift(); 439 } 440 } 441 442 Register base() const { 443 guarantee((_mode == base_plus_offset | _mode == base_plus_offset_reg 444 | _mode == post), 445 "wrong mode"); 446 return _base; 447 } 448 long offset() const { 449 return _offset; 450 } 451 Register index() const { 452 return _index; 453 } 454 mode getMode() const { 455 return _mode; 456 } 457 bool uses(Register reg) const { return _base == reg || _index == reg; } 458 address target() const { return _target; } 459 const RelocationHolder& rspec() const { return _rspec; } 460 461 void encode(Instruction_aarch64 *i) const { 462 i->f(0b111, 29, 27); 463 i->srf(_base, 5); 464 465 switch(_mode) { 466 case base_plus_offset: 467 { 468 unsigned size = i->get(31, 30); 469 unsigned mask = (1 << size) - 1; 470 if (_offset < 0 || _offset & mask) 471 { 472 i->f(0b00, 25, 24); 473 i->f(0, 21), i->f(0b00, 11, 10); 474 i->sf(_offset, 20, 12); 475 } else { 476 i->f(0b01, 25, 24); 477 i->f(_offset >> size, 21, 10); 478 } 479 } 480 break; 481 482 case base_plus_offset_reg: 483 { 484 i->f(0b00, 25, 24); 485 i->f(1, 21); 486 i->rf(_index, 16); 487 i->f(_ext.option(), 15, 13); 488 unsigned size = i->get(31, 30); 489 if (size == 0) // It's a byte 490 i->f(_ext.shift() >= 0, 12); 491 else { 492 if (_ext.shift() > 0) 493 assert(_ext.shift() == (int)size, "bad shift"); 494 i->f(_ext.shift() > 0, 12); 495 } 496 i->f(0b10, 11, 10); 497 } 498 break; 499 500 case pre: 501 i->f(0b00, 25, 24); 502 i->f(0, 21), i->f(0b11, 11, 10); 503 i->sf(_offset, 20, 12); 504 break; 505 506 case post: 507 i->f(0b00, 25, 24); 508 i->f(0, 21), i->f(0b01, 11, 10); 509 i->sf(_offset, 20, 12); 510 break; 511 512 default: 513 ShouldNotReachHere(); 514 } 515 } 516 517 void encode_pair(Instruction_aarch64 *i) const { 518 switch(_mode) { 519 case base_plus_offset: 520 i->f(0b010, 25, 23); 521 break; 522 case pre: 523 i->f(0b011, 25, 23); 524 break; 525 case post: 526 i->f(0b001, 25, 23); 527 break; 528 default: 529 ShouldNotReachHere(); 530 } 531 532 unsigned size; // Operand shift in 32-bit words 533 534 if (i->get(26, 26)) { // float 535 switch(i->get(31, 30)) { 536 case 0b10: 537 size = 2; break; 538 case 0b01: 539 size = 1; break; 540 case 0b00: 541 size = 0; break; 542 default: 543 ShouldNotReachHere(); 544 } 545 } else { 546 size = i->get(31, 31); 547 } 548 549 size = 4 << size; 550 guarantee(_offset % size == 0, "bad offset"); 551 i->sf(_offset / size, 21, 15); 552 i->srf(_base, 5); 553 } 554 555 void encode_nontemporal_pair(Instruction_aarch64 *i) const { 556 // Only base + offset is allowed 557 i->f(0b000, 25, 23); 558 unsigned size = i->get(31, 31); 559 size = 4 << size; 560 guarantee(_offset % size == 0, "bad offset"); 561 i->sf(_offset / size, 21, 15); 562 i->srf(_base, 5); 563 guarantee(_mode == Address::base_plus_offset, 564 "Bad addressing mode for non-temporal op"); 565 } 566 567 void lea(MacroAssembler *, Register) const; 568 569 static bool offset_ok_for_immed(long offset, int shift = 0) { 570 unsigned mask = (1 << shift) - 1; 571 if (offset < 0 || offset & mask) { 572 return (uabs(offset) < (1 << (20 - 12))); // Unscaled offset 573 } else { 574 return ((offset >> shift) < (1 << (21 - 10 + 1))); // Scaled, unsigned offset 575 } 576 } 577 }; 578 579 // Convience classes 580 class RuntimeAddress: public Address { 581 582 public: 583 584 RuntimeAddress(address target) : Address(target, relocInfo::runtime_call_type) {} 585 586 }; 587 588 class OopAddress: public Address { 589 590 public: 591 592 OopAddress(address target) : Address(target, relocInfo::oop_type){} 593 594 }; 595 596 class ExternalAddress: public Address { 597 private: 598 static relocInfo::relocType reloc_for_target(address target) { 599 // Sometimes ExternalAddress is used for values which aren't 600 // exactly addresses, like the card table base. 601 // external_word_type can't be used for values in the first page 602 // so just skip the reloc in that case. 603 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 604 } 605 606 public: 607 608 ExternalAddress(address target) : Address(target, reloc_for_target(target)) {} 609 610 }; 611 612 class InternalAddress: public Address { 613 614 public: 615 616 InternalAddress(address target) : Address(target, relocInfo::internal_word_type) {} 617 }; 618 619 const int FPUStateSizeInWords = 32 * 2; 620 typedef enum { 621 PLDL1KEEP = 0b00000, PLDL1STRM, PLDL2KEEP, PLDL2STRM, PLDL3KEEP, PLDL3STRM, 622 PSTL1KEEP = 0b10000, PSTL1STRM, PSTL2KEEP, PSTL2STRM, PSTL3KEEP, PSTL3STRM, 623 PLIL1KEEP = 0b01000, PLIL1STRM, PLIL2KEEP, PLIL2STRM, PLIL3KEEP, PLIL3STRM 624 } prfop; 625 626 class Assembler : public AbstractAssembler { 627 628 #ifndef PRODUCT 629 static const unsigned long asm_bp; 630 631 void emit_long(jint x) { 632 if ((unsigned long)pc() == asm_bp) 633 asm volatile ("nop"); 634 AbstractAssembler::emit_int32(x); 635 } 636 #else 637 void emit_long(jint x) { 638 AbstractAssembler::emit_int32(x); 639 } 640 #endif 641 642 public: 643 644 enum { instruction_size = 4 }; 645 646 Address adjust(Register base, int offset, bool preIncrement) { 647 if (preIncrement) 648 return Address(Pre(base, offset)); 649 else 650 return Address(Post(base, offset)); 651 } 652 653 Address pre(Register base, int offset) { 654 return adjust(base, offset, true); 655 } 656 657 Address post (Register base, int offset) { 658 return adjust(base, offset, false); 659 } 660 661 Instruction_aarch64* current; 662 663 void set_current(Instruction_aarch64* i) { current = i; } 664 665 void f(unsigned val, int msb, int lsb) { 666 current->f(val, msb, lsb); 667 } 668 void f(unsigned val, int msb) { 669 current->f(val, msb, msb); 670 } 671 void sf(long val, int msb, int lsb) { 672 current->sf(val, msb, lsb); 673 } 674 void rf(Register reg, int lsb) { 675 current->rf(reg, lsb); 676 } 677 void srf(Register reg, int lsb) { 678 current->srf(reg, lsb); 679 } 680 void zrf(Register reg, int lsb) { 681 current->zrf(reg, lsb); 682 } 683 void rf(FloatRegister reg, int lsb) { 684 current->rf(reg, lsb); 685 } 686 void fixed(unsigned value, unsigned mask) { 687 current->fixed(value, mask); 688 } 689 690 void emit() { 691 emit_long(current->get_insn()); 692 assert_cond(current->get_bits() == 0xffffffff); 693 current = NULL; 694 } 695 696 typedef void (Assembler::* uncond_branch_insn)(address dest); 697 typedef void (Assembler::* compare_and_branch_insn)(Register Rt, address dest); 698 typedef void (Assembler::* test_and_branch_insn)(Register Rt, int bitpos, address dest); 699 typedef void (Assembler::* prefetch_insn)(address target, prfop); 700 701 void wrap_label(Label &L, uncond_branch_insn insn); 702 void wrap_label(Register r, Label &L, compare_and_branch_insn insn); 703 void wrap_label(Register r, int bitpos, Label &L, test_and_branch_insn insn); 704 void wrap_label(Label &L, prfop, prefetch_insn insn); 705 706 // PC-rel. addressing 707 708 void adr(Register Rd, address dest); 709 void _adrp(Register Rd, address dest); 710 711 void adr(Register Rd, const Address &dest); 712 void _adrp(Register Rd, const Address &dest); 713 714 void adr(Register Rd, Label &L) { 715 wrap_label(Rd, L, &Assembler::Assembler::adr); 716 } 717 void _adrp(Register Rd, Label &L) { 718 wrap_label(Rd, L, &Assembler::_adrp); 719 } 720 721 void adrp(Register Rd, const Address &dest, unsigned long &offset); 722 723 #undef INSN 724 725 void add_sub_immediate(Register Rd, Register Rn, unsigned uimm, int op, 726 int negated_op); 727 728 // Add/subtract (immediate) 729 #define INSN(NAME, decode, negated) \ 730 void NAME(Register Rd, Register Rn, unsigned imm, unsigned shift) { \ 731 starti; \ 732 f(decode, 31, 29), f(0b10001, 28, 24), f(shift, 23, 22), f(imm, 21, 10); \ 733 zrf(Rd, 0), srf(Rn, 5); \ 734 } \ 735 \ 736 void NAME(Register Rd, Register Rn, unsigned imm) { \ 737 starti; \ 738 add_sub_immediate(Rd, Rn, imm, decode, negated); \ 739 } 740 741 INSN(addsw, 0b001, 0b011); 742 INSN(subsw, 0b011, 0b001); 743 INSN(adds, 0b101, 0b111); 744 INSN(subs, 0b111, 0b101); 745 746 #undef INSN 747 748 #define INSN(NAME, decode, negated) \ 749 void NAME(Register Rd, Register Rn, unsigned imm) { \ 750 starti; \ 751 add_sub_immediate(Rd, Rn, imm, decode, negated); \ 752 } 753 754 INSN(addw, 0b000, 0b010); 755 INSN(subw, 0b010, 0b000); 756 INSN(add, 0b100, 0b110); 757 INSN(sub, 0b110, 0b100); 758 759 #undef INSN 760 761 // Logical (immediate) 762 #define INSN(NAME, decode, is32) \ 763 void NAME(Register Rd, Register Rn, uint64_t imm) { \ 764 starti; \ 765 uint32_t val = encode_logical_immediate(is32, imm); \ 766 f(decode, 31, 29), f(0b100100, 28, 23), f(val, 22, 10); \ 767 srf(Rd, 0), zrf(Rn, 5); \ 768 } 769 770 INSN(andw, 0b000, true); 771 INSN(orrw, 0b001, true); 772 INSN(eorw, 0b010, true); 773 INSN(andr, 0b100, false); 774 INSN(orr, 0b101, false); 775 INSN(eor, 0b110, false); 776 777 #undef INSN 778 779 #define INSN(NAME, decode, is32) \ 780 void NAME(Register Rd, Register Rn, uint64_t imm) { \ 781 starti; \ 782 uint32_t val = encode_logical_immediate(is32, imm); \ 783 f(decode, 31, 29), f(0b100100, 28, 23), f(val, 22, 10); \ 784 zrf(Rd, 0), zrf(Rn, 5); \ 785 } 786 787 INSN(ands, 0b111, false); 788 INSN(andsw, 0b011, true); 789 790 #undef INSN 791 792 // Move wide (immediate) 793 #define INSN(NAME, opcode) \ 794 void NAME(Register Rd, unsigned imm, unsigned shift = 0) { \ 795 assert_cond((shift/16)*16 == shift); \ 796 starti; \ 797 f(opcode, 31, 29), f(0b100101, 28, 23), f(shift/16, 22, 21), \ 798 f(imm, 20, 5); \ 799 rf(Rd, 0); \ 800 } 801 802 INSN(movnw, 0b000); 803 INSN(movzw, 0b010); 804 INSN(movkw, 0b011); 805 INSN(movn, 0b100); 806 INSN(movz, 0b110); 807 INSN(movk, 0b111); 808 809 #undef INSN 810 811 // Bitfield 812 #define INSN(NAME, opcode) \ 813 void NAME(Register Rd, Register Rn, unsigned immr, unsigned imms) { \ 814 starti; \ 815 f(opcode, 31, 22), f(immr, 21, 16), f(imms, 15, 10); \ 816 rf(Rn, 5), rf(Rd, 0); \ 817 } 818 819 INSN(sbfmw, 0b0001001100); 820 INSN(bfmw, 0b0011001100); 821 INSN(ubfmw, 0b0101001100); 822 INSN(sbfm, 0b1001001101); 823 INSN(bfm, 0b1011001101); 824 INSN(ubfm, 0b1101001101); 825 826 #undef INSN 827 828 // Extract 829 #define INSN(NAME, opcode) \ 830 void NAME(Register Rd, Register Rn, Register Rm, unsigned imms) { \ 831 starti; \ 832 f(opcode, 31, 21), f(imms, 15, 10); \ 833 rf(Rm, 16), rf(Rn, 5), rf(Rd, 0); \ 834 } 835 836 INSN(extrw, 0b00010011100); 837 INSN(extr, 0b10010011110); 838 839 #undef INSN 840 841 // The maximum range of a branch is fixed for the AArch64 842 // architecture. In debug mode we shrink it in order to test 843 // trampolines, but not so small that branches in the interpreter 844 // are out of range. 845 static const unsigned long branch_range = NOT_DEBUG(128 * M) DEBUG_ONLY(2 * M); 846 847 static bool reachable_from_branch_at(address branch, address target) { 848 return uabs(target - branch) < branch_range; 849 } 850 851 // Unconditional branch (immediate) 852 #define INSN(NAME, opcode) \ 853 void NAME(address dest) { \ 854 starti; \ 855 long offset = (dest - pc()) >> 2; \ 856 DEBUG_ONLY(assert(reachable_from_branch_at(pc(), dest), "debug only")); \ 857 f(opcode, 31), f(0b00101, 30, 26), sf(offset, 25, 0); \ 858 } \ 859 void NAME(Label &L) { \ 860 wrap_label(L, &Assembler::NAME); \ 861 } \ 862 void NAME(const Address &dest); 863 864 INSN(b, 0); 865 INSN(bl, 1); 866 867 #undef INSN 868 869 // Compare & branch (immediate) 870 #define INSN(NAME, opcode) \ 871 void NAME(Register Rt, address dest) { \ 872 long offset = (dest - pc()) >> 2; \ 873 starti; \ 874 f(opcode, 31, 24), sf(offset, 23, 5), rf(Rt, 0); \ 875 } \ 876 void NAME(Register Rt, Label &L) { \ 877 wrap_label(Rt, L, &Assembler::NAME); \ 878 } 879 880 INSN(cbzw, 0b00110100); 881 INSN(cbnzw, 0b00110101); 882 INSN(cbz, 0b10110100); 883 INSN(cbnz, 0b10110101); 884 885 #undef INSN 886 887 // Test & branch (immediate) 888 #define INSN(NAME, opcode) \ 889 void NAME(Register Rt, int bitpos, address dest) { \ 890 long offset = (dest - pc()) >> 2; \ 891 int b5 = bitpos >> 5; \ 892 bitpos &= 0x1f; \ 893 starti; \ 894 f(b5, 31), f(opcode, 30, 24), f(bitpos, 23, 19), sf(offset, 18, 5); \ 895 rf(Rt, 0); \ 896 } \ 897 void NAME(Register Rt, int bitpos, Label &L) { \ 898 wrap_label(Rt, bitpos, L, &Assembler::NAME); \ 899 } 900 901 INSN(tbz, 0b0110110); 902 INSN(tbnz, 0b0110111); 903 904 #undef INSN 905 906 // Conditional branch (immediate) 907 enum Condition 908 {EQ, NE, HS, CS=HS, LO, CC=LO, MI, PL, VS, VC, HI, LS, GE, LT, GT, LE, AL, NV}; 909 910 void br(Condition cond, address dest) { 911 long offset = (dest - pc()) >> 2; 912 starti; 913 f(0b0101010, 31, 25), f(0, 24), sf(offset, 23, 5), f(0, 4), f(cond, 3, 0); 914 } 915 916 #define INSN(NAME, cond) \ 917 void NAME(address dest) { \ 918 br(cond, dest); \ 919 } 920 921 INSN(beq, EQ); 922 INSN(bne, NE); 923 INSN(bhs, HS); 924 INSN(bcs, CS); 925 INSN(blo, LO); 926 INSN(bcc, CC); 927 INSN(bmi, MI); 928 INSN(bpl, PL); 929 INSN(bvs, VS); 930 INSN(bvc, VC); 931 INSN(bhi, HI); 932 INSN(bls, LS); 933 INSN(bge, GE); 934 INSN(blt, LT); 935 INSN(bgt, GT); 936 INSN(ble, LE); 937 INSN(bal, AL); 938 INSN(bnv, NV); 939 940 void br(Condition cc, Label &L); 941 942 #undef INSN 943 944 // Exception generation 945 void generate_exception(int opc, int op2, int LL, unsigned imm) { 946 starti; 947 f(0b11010100, 31, 24); 948 f(opc, 23, 21), f(imm, 20, 5), f(op2, 4, 2), f(LL, 1, 0); 949 } 950 951 #define INSN(NAME, opc, op2, LL) \ 952 void NAME(unsigned imm) { \ 953 generate_exception(opc, op2, LL, imm); \ 954 } 955 956 INSN(svc, 0b000, 0, 0b01); 957 INSN(hvc, 0b000, 0, 0b10); 958 INSN(smc, 0b000, 0, 0b11); 959 INSN(brk, 0b001, 0, 0b00); 960 INSN(hlt, 0b010, 0, 0b00); 961 INSN(dpcs1, 0b101, 0, 0b01); 962 INSN(dpcs2, 0b101, 0, 0b10); 963 INSN(dpcs3, 0b101, 0, 0b11); 964 965 #undef INSN 966 967 // System 968 void system(int op0, int op1, int CRn, int CRm, int op2, 969 Register rt = (Register)0b11111) 970 { 971 starti; 972 f(0b11010101000, 31, 21); 973 f(op0, 20, 19); 974 f(op1, 18, 16); 975 f(CRn, 15, 12); 976 f(CRm, 11, 8); 977 f(op2, 7, 5); 978 rf(rt, 0); 979 } 980 981 void hint(int imm) { 982 system(0b00, 0b011, 0b0010, imm, 0b000); 983 } 984 985 void nop() { 986 hint(0); 987 } 988 // we only provide mrs and msr for the special purpose system 989 // registers where op1 (instr[20:19]) == 11 and, (currently) only 990 // use it for FPSR n.b msr has L (instr[21]) == 0 mrs has L == 1 991 992 void msr(int op1, int CRn, int CRm, int op2, Register rt) { 993 starti; 994 f(0b1101010100011, 31, 19); 995 f(op1, 18, 16); 996 f(CRn, 15, 12); 997 f(CRm, 11, 8); 998 f(op2, 7, 5); 999 // writing zr is ok 1000 zrf(rt, 0); 1001 } 1002 1003 void mrs(int op1, int CRn, int CRm, int op2, Register rt) { 1004 starti; 1005 f(0b1101010100111, 31, 19); 1006 f(op1, 18, 16); 1007 f(CRn, 15, 12); 1008 f(CRm, 11, 8); 1009 f(op2, 7, 5); 1010 // reading to zr is a mistake 1011 rf(rt, 0); 1012 } 1013 1014 enum barrier {OSHLD = 0b0001, OSHST, OSH, NSHLD=0b0101, NSHST, NSH, 1015 ISHLD = 0b1001, ISHST, ISH, LD=0b1101, ST, SY}; 1016 1017 void dsb(barrier imm) { 1018 system(0b00, 0b011, 0b00011, imm, 0b100); 1019 } 1020 1021 void dmb(barrier imm) { 1022 system(0b00, 0b011, 0b00011, imm, 0b101); 1023 } 1024 1025 void isb() { 1026 system(0b00, 0b011, 0b00011, SY, 0b110); 1027 } 1028 1029 void dc(Register Rt) { 1030 system(0b01, 0b011, 0b0111, 0b1011, 0b001, Rt); 1031 } 1032 1033 void ic(Register Rt) { 1034 system(0b01, 0b011, 0b0111, 0b0101, 0b001, Rt); 1035 } 1036 1037 // A more convenient access to dmb for our purposes 1038 enum Membar_mask_bits { 1039 // We can use ISH for a barrier because the ARM ARM says "This 1040 // architecture assumes that all Processing Elements that use the 1041 // same operating system or hypervisor are in the same Inner 1042 // Shareable shareability domain." 1043 StoreStore = ISHST, 1044 LoadStore = ISHLD, 1045 LoadLoad = ISHLD, 1046 StoreLoad = ISH, 1047 AnyAny = ISH 1048 }; 1049 1050 void membar(Membar_mask_bits order_constraint) { 1051 dmb(Assembler::barrier(order_constraint)); 1052 } 1053 1054 // Unconditional branch (register) 1055 void branch_reg(Register R, int opc) { 1056 starti; 1057 f(0b1101011, 31, 25); 1058 f(opc, 24, 21); 1059 f(0b11111000000, 20, 10); 1060 rf(R, 5); 1061 f(0b00000, 4, 0); 1062 } 1063 1064 #define INSN(NAME, opc) \ 1065 void NAME(Register R) { \ 1066 branch_reg(R, opc); \ 1067 } 1068 1069 INSN(br, 0b0000); 1070 INSN(blr, 0b0001); 1071 INSN(ret, 0b0010); 1072 1073 void ret(void *p); // This forces a compile-time error for ret(0) 1074 1075 #undef INSN 1076 1077 #define INSN(NAME, opc) \ 1078 void NAME() { \ 1079 branch_reg((Register)0b11111, opc); \ 1080 } 1081 1082 INSN(eret, 0b0100); 1083 INSN(drps, 0b0101); 1084 1085 #undef INSN 1086 1087 // Load/store exclusive 1088 enum operand_size { byte, halfword, word, xword }; 1089 1090 void load_store_exclusive(Register Rs, Register Rt1, Register Rt2, 1091 Register Rn, enum operand_size sz, int op, int o0) { 1092 starti; 1093 f(sz, 31, 30), f(0b001000, 29, 24), f(op, 23, 21); 1094 rf(Rs, 16), f(o0, 15), rf(Rt2, 10), rf(Rn, 5), rf(Rt1, 0); 1095 } 1096 1097 #define INSN4(NAME, sz, op, o0) /* Four registers */ \ 1098 void NAME(Register Rs, Register Rt1, Register Rt2, Register Rn) { \ 1099 load_store_exclusive(Rs, Rt1, Rt2, Rn, sz, op, o0); \ 1100 } 1101 1102 #define INSN3(NAME, sz, op, o0) /* Three registers */ \ 1103 void NAME(Register Rs, Register Rt, Register Rn) { \ 1104 load_store_exclusive(Rs, Rt, (Register)0b11111, Rn, sz, op, o0); \ 1105 } 1106 1107 #define INSN2(NAME, sz, op, o0) /* Two registers */ \ 1108 void NAME(Register Rt, Register Rn) { \ 1109 load_store_exclusive((Register)0b11111, Rt, (Register)0b11111, \ 1110 Rn, sz, op, o0); \ 1111 } 1112 1113 #define INSN_FOO(NAME, sz, op, o0) /* Three registers, encoded differently */ \ 1114 void NAME(Register Rt1, Register Rt2, Register Rn) { \ 1115 load_store_exclusive((Register)0b11111, Rt1, Rt2, Rn, sz, op, o0); \ 1116 } 1117 1118 // bytes 1119 INSN3(stxrb, byte, 0b000, 0); 1120 INSN3(stlxrb, byte, 0b000, 1); 1121 INSN2(ldxrb, byte, 0b010, 0); 1122 INSN2(ldaxrb, byte, 0b010, 1); 1123 INSN2(stlrb, byte, 0b100, 1); 1124 INSN2(ldarb, byte, 0b110, 1); 1125 1126 // halfwords 1127 INSN3(stxrh, halfword, 0b000, 0); 1128 INSN3(stlxrh, halfword, 0b000, 1); 1129 INSN2(ldxrh, halfword, 0b010, 0); 1130 INSN2(ldaxrh, halfword, 0b010, 1); 1131 INSN2(stlrh, halfword, 0b100, 1); 1132 INSN2(ldarh, halfword, 0b110, 1); 1133 1134 // words 1135 INSN3(stxrw, word, 0b000, 0); 1136 INSN3(stlxrw, word, 0b000, 1); 1137 INSN4(stxpw, word, 0b001, 0); 1138 INSN4(stlxpw, word, 0b001, 1); 1139 INSN2(ldxrw, word, 0b010, 0); 1140 INSN2(ldaxrw, word, 0b010, 1); 1141 INSN_FOO(ldxpw, word, 0b011, 0); 1142 INSN_FOO(ldaxpw, word, 0b011, 1); 1143 INSN2(stlrw, word, 0b100, 1); 1144 INSN2(ldarw, word, 0b110, 1); 1145 1146 // xwords 1147 INSN3(stxr, xword, 0b000, 0); 1148 INSN3(stlxr, xword, 0b000, 1); 1149 INSN4(stxp, xword, 0b001, 0); 1150 INSN4(stlxp, xword, 0b001, 1); 1151 INSN2(ldxr, xword, 0b010, 0); 1152 INSN2(ldaxr, xword, 0b010, 1); 1153 INSN_FOO(ldxp, xword, 0b011, 0); 1154 INSN_FOO(ldaxp, xword, 0b011, 1); 1155 INSN2(stlr, xword, 0b100, 1); 1156 INSN2(ldar, xword, 0b110, 1); 1157 1158 #undef INSN2 1159 #undef INSN3 1160 #undef INSN4 1161 #undef INSN_FOO 1162 1163 // Load register (literal) 1164 #define INSN(NAME, opc, V) \ 1165 void NAME(Register Rt, address dest) { \ 1166 long offset = (dest - pc()) >> 2; \ 1167 starti; \ 1168 f(opc, 31, 30), f(0b011, 29, 27), f(V, 26), f(0b00, 25, 24), \ 1169 sf(offset, 23, 5); \ 1170 rf(Rt, 0); \ 1171 } \ 1172 void NAME(Register Rt, address dest, relocInfo::relocType rtype) { \ 1173 InstructionMark im(this); \ 1174 guarantee(rtype == relocInfo::internal_word_type, \ 1175 "only internal_word_type relocs make sense here"); \ 1176 code_section()->relocate(inst_mark(), InternalAddress(dest).rspec()); \ 1177 NAME(Rt, dest); \ 1178 } \ 1179 void NAME(Register Rt, Label &L) { \ 1180 wrap_label(Rt, L, &Assembler::NAME); \ 1181 } 1182 1183 INSN(ldrw, 0b00, 0); 1184 INSN(ldr, 0b01, 0); 1185 INSN(ldrsw, 0b10, 0); 1186 1187 #undef INSN 1188 1189 #define INSN(NAME, opc, V) \ 1190 void NAME(FloatRegister Rt, address dest) { \ 1191 long offset = (dest - pc()) >> 2; \ 1192 starti; \ 1193 f(opc, 31, 30), f(0b011, 29, 27), f(V, 26), f(0b00, 25, 24), \ 1194 sf(offset, 23, 5); \ 1195 rf((Register)Rt, 0); \ 1196 } 1197 1198 INSN(ldrs, 0b00, 1); 1199 INSN(ldrd, 0b01, 1); 1200 INSN(ldrq, 0x10, 1); 1201 1202 #undef INSN 1203 1204 #define INSN(NAME, opc, V) \ 1205 void NAME(address dest, prfop op = PLDL1KEEP) { \ 1206 long offset = (dest - pc()) >> 2; \ 1207 starti; \ 1208 f(opc, 31, 30), f(0b011, 29, 27), f(V, 26), f(0b00, 25, 24), \ 1209 sf(offset, 23, 5); \ 1210 f(op, 4, 0); \ 1211 } \ 1212 void NAME(Label &L, prfop op = PLDL1KEEP) { \ 1213 wrap_label(L, op, &Assembler::NAME); \ 1214 } 1215 1216 INSN(prfm, 0b11, 0); 1217 1218 #undef INSN 1219 1220 // Load/store 1221 void ld_st1(int opc, int p1, int V, int L, 1222 Register Rt1, Register Rt2, Address adr, bool no_allocate) { 1223 starti; 1224 f(opc, 31, 30), f(p1, 29, 27), f(V, 26), f(L, 22); 1225 zrf(Rt2, 10), zrf(Rt1, 0); 1226 if (no_allocate) { 1227 adr.encode_nontemporal_pair(current); 1228 } else { 1229 adr.encode_pair(current); 1230 } 1231 } 1232 1233 // Load/store register pair (offset) 1234 #define INSN(NAME, size, p1, V, L, no_allocate) \ 1235 void NAME(Register Rt1, Register Rt2, Address adr) { \ 1236 ld_st1(size, p1, V, L, Rt1, Rt2, adr, no_allocate); \ 1237 } 1238 1239 INSN(stpw, 0b00, 0b101, 0, 0, false); 1240 INSN(ldpw, 0b00, 0b101, 0, 1, false); 1241 INSN(ldpsw, 0b01, 0b101, 0, 1, false); 1242 INSN(stp, 0b10, 0b101, 0, 0, false); 1243 INSN(ldp, 0b10, 0b101, 0, 1, false); 1244 1245 // Load/store no-allocate pair (offset) 1246 INSN(stnpw, 0b00, 0b101, 0, 0, true); 1247 INSN(ldnpw, 0b00, 0b101, 0, 1, true); 1248 INSN(stnp, 0b10, 0b101, 0, 0, true); 1249 INSN(ldnp, 0b10, 0b101, 0, 1, true); 1250 1251 #undef INSN 1252 1253 #define INSN(NAME, size, p1, V, L, no_allocate) \ 1254 void NAME(FloatRegister Rt1, FloatRegister Rt2, Address adr) { \ 1255 ld_st1(size, p1, V, L, (Register)Rt1, (Register)Rt2, adr, no_allocate); \ 1256 } 1257 1258 INSN(stps, 0b00, 0b101, 1, 0, false); 1259 INSN(ldps, 0b00, 0b101, 1, 1, false); 1260 INSN(stpd, 0b01, 0b101, 1, 0, false); 1261 INSN(ldpd, 0b01, 0b101, 1, 1, false); 1262 INSN(stpq, 0b10, 0b101, 1, 0, false); 1263 INSN(ldpq, 0b10, 0b101, 1, 1, false); 1264 1265 #undef INSN 1266 1267 // Load/store register (all modes) 1268 void ld_st2(Register Rt, const Address &adr, int size, int op, int V = 0) { 1269 starti; 1270 1271 f(V, 26); // general reg? 1272 zrf(Rt, 0); 1273 1274 // Encoding for literal loads is done here (rather than pushed 1275 // down into Address::encode) because the encoding of this 1276 // instruction is too different from all of the other forms to 1277 // make it worth sharing. 1278 if (adr.getMode() == Address::literal) { 1279 assert(size == 0b10 || size == 0b11, "bad operand size in ldr"); 1280 assert(op == 0b01, "literal form can only be used with loads"); 1281 f(size & 0b01, 31, 30), f(0b011, 29, 27), f(0b00, 25, 24); 1282 long offset = (adr.target() - pc()) >> 2; 1283 sf(offset, 23, 5); 1284 code_section()->relocate(pc(), adr.rspec()); 1285 return; 1286 } 1287 1288 f(size, 31, 30); 1289 f(op, 23, 22); // str 1290 adr.encode(current); 1291 } 1292 1293 #define INSN(NAME, size, op) \ 1294 void NAME(Register Rt, const Address &adr) { \ 1295 ld_st2(Rt, adr, size, op); \ 1296 } \ 1297 1298 INSN(str, 0b11, 0b00); 1299 INSN(strw, 0b10, 0b00); 1300 INSN(strb, 0b00, 0b00); 1301 INSN(strh, 0b01, 0b00); 1302 1303 INSN(ldr, 0b11, 0b01); 1304 INSN(ldrw, 0b10, 0b01); 1305 INSN(ldrb, 0b00, 0b01); 1306 INSN(ldrh, 0b01, 0b01); 1307 1308 INSN(ldrsb, 0b00, 0b10); 1309 INSN(ldrsbw, 0b00, 0b11); 1310 INSN(ldrsh, 0b01, 0b10); 1311 INSN(ldrshw, 0b01, 0b11); 1312 INSN(ldrsw, 0b10, 0b10); 1313 1314 #undef INSN 1315 1316 #define INSN(NAME, size, op) \ 1317 void NAME(const Address &adr, prfop pfop = PLDL1KEEP) { \ 1318 ld_st2((Register)pfop, adr, size, op); \ 1319 } 1320 1321 INSN(prfm, 0b11, 0b10); // FIXME: PRFM should not be used with 1322 // writeback modes, but the assembler 1323 // doesn't enfore that. 1324 1325 #undef INSN 1326 1327 #define INSN(NAME, size, op) \ 1328 void NAME(FloatRegister Rt, const Address &adr) { \ 1329 ld_st2((Register)Rt, adr, size, op, 1); \ 1330 } 1331 1332 INSN(strd, 0b11, 0b00); 1333 INSN(strs, 0b10, 0b00); 1334 INSN(ldrd, 0b11, 0b01); 1335 INSN(ldrs, 0b10, 0b01); 1336 INSN(strq, 0b00, 0b10); 1337 INSN(ldrq, 0x00, 0b11); 1338 1339 #undef INSN 1340 1341 enum shift_kind { LSL, LSR, ASR, ROR }; 1342 1343 void op_shifted_reg(unsigned decode, 1344 enum shift_kind kind, unsigned shift, 1345 unsigned size, unsigned op) { 1346 f(size, 31); 1347 f(op, 30, 29); 1348 f(decode, 28, 24); 1349 f(shift, 15, 10); 1350 f(kind, 23, 22); 1351 } 1352 1353 // Logical (shifted register) 1354 #define INSN(NAME, size, op, N) \ 1355 void NAME(Register Rd, Register Rn, Register Rm, \ 1356 enum shift_kind kind = LSL, unsigned shift = 0) { \ 1357 starti; \ 1358 f(N, 21); \ 1359 zrf(Rm, 16), zrf(Rn, 5), zrf(Rd, 0); \ 1360 op_shifted_reg(0b01010, kind, shift, size, op); \ 1361 } 1362 1363 INSN(andr, 1, 0b00, 0); 1364 INSN(orr, 1, 0b01, 0); 1365 INSN(eor, 1, 0b10, 0); 1366 INSN(ands, 1, 0b11, 0); 1367 INSN(andw, 0, 0b00, 0); 1368 INSN(orrw, 0, 0b01, 0); 1369 INSN(eorw, 0, 0b10, 0); 1370 INSN(andsw, 0, 0b11, 0); 1371 1372 INSN(bic, 1, 0b00, 1); 1373 INSN(orn, 1, 0b01, 1); 1374 INSN(eon, 1, 0b10, 1); 1375 INSN(bics, 1, 0b11, 1); 1376 INSN(bicw, 0, 0b00, 1); 1377 INSN(ornw, 0, 0b01, 1); 1378 INSN(eonw, 0, 0b10, 1); 1379 INSN(bicsw, 0, 0b11, 1); 1380 1381 #undef INSN 1382 1383 // Add/subtract (shifted register) 1384 #define INSN(NAME, size, op) \ 1385 void NAME(Register Rd, Register Rn, Register Rm, \ 1386 enum shift_kind kind, unsigned shift = 0) { \ 1387 starti; \ 1388 f(0, 21); \ 1389 assert_cond(kind != ROR); \ 1390 zrf(Rd, 0), zrf(Rn, 5), zrf(Rm, 16); \ 1391 op_shifted_reg(0b01011, kind, shift, size, op); \ 1392 } 1393 1394 INSN(add, 1, 0b000); 1395 INSN(sub, 1, 0b10); 1396 INSN(addw, 0, 0b000); 1397 INSN(subw, 0, 0b10); 1398 1399 INSN(adds, 1, 0b001); 1400 INSN(subs, 1, 0b11); 1401 INSN(addsw, 0, 0b001); 1402 INSN(subsw, 0, 0b11); 1403 1404 #undef INSN 1405 1406 // Add/subtract (extended register) 1407 #define INSN(NAME, op) \ 1408 void NAME(Register Rd, Register Rn, Register Rm, \ 1409 ext::operation option, int amount = 0) { \ 1410 starti; \ 1411 zrf(Rm, 16), srf(Rn, 5), srf(Rd, 0); \ 1412 add_sub_extended_reg(op, 0b01011, Rd, Rn, Rm, 0b00, option, amount); \ 1413 } 1414 1415 void add_sub_extended_reg(unsigned op, unsigned decode, 1416 Register Rd, Register Rn, Register Rm, 1417 unsigned opt, ext::operation option, unsigned imm) { 1418 guarantee(imm <= 4, "shift amount must be < 4"); 1419 f(op, 31, 29), f(decode, 28, 24), f(opt, 23, 22), f(1, 21); 1420 f(option, 15, 13), f(imm, 12, 10); 1421 } 1422 1423 INSN(addw, 0b000); 1424 INSN(subw, 0b010); 1425 INSN(add, 0b100); 1426 INSN(sub, 0b110); 1427 1428 #undef INSN 1429 1430 #define INSN(NAME, op) \ 1431 void NAME(Register Rd, Register Rn, Register Rm, \ 1432 ext::operation option, int amount = 0) { \ 1433 starti; \ 1434 zrf(Rm, 16), srf(Rn, 5), zrf(Rd, 0); \ 1435 add_sub_extended_reg(op, 0b01011, Rd, Rn, Rm, 0b00, option, amount); \ 1436 } 1437 1438 INSN(addsw, 0b001); 1439 INSN(subsw, 0b011); 1440 INSN(adds, 0b101); 1441 INSN(subs, 0b111); 1442 1443 #undef INSN 1444 1445 // Aliases for short forms of add and sub 1446 #define INSN(NAME) \ 1447 void NAME(Register Rd, Register Rn, Register Rm) { \ 1448 if (Rd == sp || Rn == sp) \ 1449 NAME(Rd, Rn, Rm, ext::uxtx); \ 1450 else \ 1451 NAME(Rd, Rn, Rm, LSL); \ 1452 } 1453 1454 INSN(addw); 1455 INSN(subw); 1456 INSN(add); 1457 INSN(sub); 1458 1459 INSN(addsw); 1460 INSN(subsw); 1461 INSN(adds); 1462 INSN(subs); 1463 1464 #undef INSN 1465 1466 // Add/subtract (with carry) 1467 void add_sub_carry(unsigned op, Register Rd, Register Rn, Register Rm) { 1468 starti; 1469 f(op, 31, 29); 1470 f(0b11010000, 28, 21); 1471 f(0b000000, 15, 10); 1472 zrf(Rm, 16), zrf(Rn, 5), zrf(Rd, 0); 1473 } 1474 1475 #define INSN(NAME, op) \ 1476 void NAME(Register Rd, Register Rn, Register Rm) { \ 1477 add_sub_carry(op, Rd, Rn, Rm); \ 1478 } 1479 1480 INSN(adcw, 0b000); 1481 INSN(adcsw, 0b001); 1482 INSN(sbcw, 0b010); 1483 INSN(sbcsw, 0b011); 1484 INSN(adc, 0b100); 1485 INSN(adcs, 0b101); 1486 INSN(sbc,0b110); 1487 INSN(sbcs, 0b111); 1488 1489 #undef INSN 1490 1491 // Conditional compare (both kinds) 1492 void conditional_compare(unsigned op, int o2, int o3, 1493 Register Rn, unsigned imm5, unsigned nzcv, 1494 unsigned cond) { 1495 f(op, 31, 29); 1496 f(0b11010010, 28, 21); 1497 f(cond, 15, 12); 1498 f(o2, 10); 1499 f(o3, 4); 1500 f(nzcv, 3, 0); 1501 f(imm5, 20, 16), rf(Rn, 5); 1502 } 1503 1504 #define INSN(NAME, op) \ 1505 void NAME(Register Rn, Register Rm, int imm, Condition cond) { \ 1506 starti; \ 1507 f(0, 11); \ 1508 conditional_compare(op, 0, 0, Rn, (uintptr_t)Rm, imm, cond); \ 1509 } \ 1510 \ 1511 void NAME(Register Rn, int imm5, int imm, Condition cond) { \ 1512 starti; \ 1513 f(1, 11); \ 1514 conditional_compare(op, 0, 0, Rn, imm5, imm, cond); \ 1515 } 1516 1517 INSN(ccmnw, 0b001); 1518 INSN(ccmpw, 0b011); 1519 INSN(ccmn, 0b101); 1520 INSN(ccmp, 0b111); 1521 1522 #undef INSN 1523 1524 // Conditional select 1525 void conditional_select(unsigned op, unsigned op2, 1526 Register Rd, Register Rn, Register Rm, 1527 unsigned cond) { 1528 starti; 1529 f(op, 31, 29); 1530 f(0b11010100, 28, 21); 1531 f(cond, 15, 12); 1532 f(op2, 11, 10); 1533 zrf(Rm, 16), zrf(Rn, 5), rf(Rd, 0); 1534 } 1535 1536 #define INSN(NAME, op, op2) \ 1537 void NAME(Register Rd, Register Rn, Register Rm, Condition cond) { \ 1538 conditional_select(op, op2, Rd, Rn, Rm, cond); \ 1539 } 1540 1541 INSN(cselw, 0b000, 0b00); 1542 INSN(csincw, 0b000, 0b01); 1543 INSN(csinvw, 0b010, 0b00); 1544 INSN(csnegw, 0b010, 0b01); 1545 INSN(csel, 0b100, 0b00); 1546 INSN(csinc, 0b100, 0b01); 1547 INSN(csinv, 0b110, 0b00); 1548 INSN(csneg, 0b110, 0b01); 1549 1550 #undef INSN 1551 1552 // Data processing 1553 void data_processing(unsigned op29, unsigned opcode, 1554 Register Rd, Register Rn) { 1555 f(op29, 31, 29), f(0b11010110, 28, 21); 1556 f(opcode, 15, 10); 1557 rf(Rn, 5), rf(Rd, 0); 1558 } 1559 1560 // (1 source) 1561 #define INSN(NAME, op29, opcode2, opcode) \ 1562 void NAME(Register Rd, Register Rn) { \ 1563 starti; \ 1564 f(opcode2, 20, 16); \ 1565 data_processing(op29, opcode, Rd, Rn); \ 1566 } 1567 1568 INSN(rbitw, 0b010, 0b00000, 0b00000); 1569 INSN(rev16w, 0b010, 0b00000, 0b00001); 1570 INSN(revw, 0b010, 0b00000, 0b00010); 1571 INSN(clzw, 0b010, 0b00000, 0b00100); 1572 INSN(clsw, 0b010, 0b00000, 0b00101); 1573 1574 INSN(rbit, 0b110, 0b00000, 0b00000); 1575 INSN(rev16, 0b110, 0b00000, 0b00001); 1576 INSN(rev32, 0b110, 0b00000, 0b00010); 1577 INSN(rev, 0b110, 0b00000, 0b00011); 1578 INSN(clz, 0b110, 0b00000, 0b00100); 1579 INSN(cls, 0b110, 0b00000, 0b00101); 1580 1581 #undef INSN 1582 1583 // (2 sources) 1584 #define INSN(NAME, op29, opcode) \ 1585 void NAME(Register Rd, Register Rn, Register Rm) { \ 1586 starti; \ 1587 rf(Rm, 16); \ 1588 data_processing(op29, opcode, Rd, Rn); \ 1589 } 1590 1591 INSN(udivw, 0b000, 0b000010); 1592 INSN(sdivw, 0b000, 0b000011); 1593 INSN(lslvw, 0b000, 0b001000); 1594 INSN(lsrvw, 0b000, 0b001001); 1595 INSN(asrvw, 0b000, 0b001010); 1596 INSN(rorvw, 0b000, 0b001011); 1597 1598 INSN(udiv, 0b100, 0b000010); 1599 INSN(sdiv, 0b100, 0b000011); 1600 INSN(lslv, 0b100, 0b001000); 1601 INSN(lsrv, 0b100, 0b001001); 1602 INSN(asrv, 0b100, 0b001010); 1603 INSN(rorv, 0b100, 0b001011); 1604 1605 #undef INSN 1606 1607 // (3 sources) 1608 void data_processing(unsigned op54, unsigned op31, unsigned o0, 1609 Register Rd, Register Rn, Register Rm, 1610 Register Ra) { 1611 starti; 1612 f(op54, 31, 29), f(0b11011, 28, 24); 1613 f(op31, 23, 21), f(o0, 15); 1614 zrf(Rm, 16), zrf(Ra, 10), zrf(Rn, 5), zrf(Rd, 0); 1615 } 1616 1617 #define INSN(NAME, op54, op31, o0) \ 1618 void NAME(Register Rd, Register Rn, Register Rm, Register Ra) { \ 1619 data_processing(op54, op31, o0, Rd, Rn, Rm, Ra); \ 1620 } 1621 1622 INSN(maddw, 0b000, 0b000, 0); 1623 INSN(msubw, 0b000, 0b000, 1); 1624 INSN(madd, 0b100, 0b000, 0); 1625 INSN(msub, 0b100, 0b000, 1); 1626 INSN(smaddl, 0b100, 0b001, 0); 1627 INSN(smsubl, 0b100, 0b001, 1); 1628 INSN(umaddl, 0b100, 0b101, 0); 1629 INSN(umsubl, 0b100, 0b101, 1); 1630 1631 #undef INSN 1632 1633 #define INSN(NAME, op54, op31, o0) \ 1634 void NAME(Register Rd, Register Rn, Register Rm) { \ 1635 data_processing(op54, op31, o0, Rd, Rn, Rm, (Register)31); \ 1636 } 1637 1638 INSN(smulh, 0b100, 0b010, 0); 1639 INSN(umulh, 0b100, 0b110, 0); 1640 1641 #undef INSN 1642 1643 // Floating-point data-processing (1 source) 1644 void data_processing(unsigned op31, unsigned type, unsigned opcode, 1645 FloatRegister Vd, FloatRegister Vn) { 1646 starti; 1647 f(op31, 31, 29); 1648 f(0b11110, 28, 24); 1649 f(type, 23, 22), f(1, 21), f(opcode, 20, 15), f(0b10000, 14, 10); 1650 rf(Vn, 5), rf(Vd, 0); 1651 } 1652 1653 #define INSN(NAME, op31, type, opcode) \ 1654 void NAME(FloatRegister Vd, FloatRegister Vn) { \ 1655 data_processing(op31, type, opcode, Vd, Vn); \ 1656 } 1657 1658 private: 1659 INSN(i_fmovs, 0b000, 0b00, 0b000000); 1660 public: 1661 INSN(fabss, 0b000, 0b00, 0b000001); 1662 INSN(fnegs, 0b000, 0b00, 0b000010); 1663 INSN(fsqrts, 0b000, 0b00, 0b000011); 1664 INSN(fcvts, 0b000, 0b00, 0b000101); // Single-precision to double-precision 1665 1666 private: 1667 INSN(i_fmovd, 0b000, 0b01, 0b000000); 1668 public: 1669 INSN(fabsd, 0b000, 0b01, 0b000001); 1670 INSN(fnegd, 0b000, 0b01, 0b000010); 1671 INSN(fsqrtd, 0b000, 0b01, 0b000011); 1672 INSN(fcvtd, 0b000, 0b01, 0b000100); // Double-precision to single-precision 1673 1674 void fmovd(FloatRegister Vd, FloatRegister Vn) { 1675 assert(Vd != Vn, "should be"); 1676 i_fmovd(Vd, Vn); 1677 } 1678 1679 void fmovs(FloatRegister Vd, FloatRegister Vn) { 1680 assert(Vd != Vn, "should be"); 1681 i_fmovs(Vd, Vn); 1682 } 1683 1684 #undef INSN 1685 1686 // Floating-point data-processing (2 source) 1687 void data_processing(unsigned op31, unsigned type, unsigned opcode, 1688 FloatRegister Vd, FloatRegister Vn, FloatRegister Vm) { 1689 starti; 1690 f(op31, 31, 29); 1691 f(0b11110, 28, 24); 1692 f(type, 23, 22), f(1, 21), f(opcode, 15, 12), f(0b10, 11, 10); 1693 rf(Vm, 16), rf(Vn, 5), rf(Vd, 0); 1694 } 1695 1696 #define INSN(NAME, op31, type, opcode) \ 1697 void NAME(FloatRegister Vd, FloatRegister Vn, FloatRegister Vm) { \ 1698 data_processing(op31, type, opcode, Vd, Vn, Vm); \ 1699 } 1700 1701 INSN(fmuls, 0b000, 0b00, 0b0000); 1702 INSN(fdivs, 0b000, 0b00, 0b0001); 1703 INSN(fadds, 0b000, 0b00, 0b0010); 1704 INSN(fsubs, 0b000, 0b00, 0b0011); 1705 INSN(fnmuls, 0b000, 0b00, 0b1000); 1706 1707 INSN(fmuld, 0b000, 0b01, 0b0000); 1708 INSN(fdivd, 0b000, 0b01, 0b0001); 1709 INSN(faddd, 0b000, 0b01, 0b0010); 1710 INSN(fsubd, 0b000, 0b01, 0b0011); 1711 INSN(fnmuld, 0b000, 0b01, 0b1000); 1712 1713 #undef INSN 1714 1715 // Floating-point data-processing (3 source) 1716 void data_processing(unsigned op31, unsigned type, unsigned o1, unsigned o0, 1717 FloatRegister Vd, FloatRegister Vn, FloatRegister Vm, 1718 FloatRegister Va) { 1719 starti; 1720 f(op31, 31, 29); 1721 f(0b11111, 28, 24); 1722 f(type, 23, 22), f(o1, 21), f(o0, 15); 1723 rf(Vm, 16), rf(Va, 10), rf(Vn, 5), rf(Vd, 0); 1724 } 1725 1726 #define INSN(NAME, op31, type, o1, o0) \ 1727 void NAME(FloatRegister Vd, FloatRegister Vn, FloatRegister Vm, \ 1728 FloatRegister Va) { \ 1729 data_processing(op31, type, o1, o0, Vd, Vn, Vm, Va); \ 1730 } 1731 1732 INSN(fmadds, 0b000, 0b00, 0, 0); 1733 INSN(fmsubs, 0b000, 0b00, 0, 1); 1734 INSN(fnmadds, 0b000, 0b00, 1, 0); 1735 INSN(fnmsubs, 0b000, 0b00, 1, 1); 1736 1737 INSN(fmaddd, 0b000, 0b01, 0, 0); 1738 INSN(fmsubd, 0b000, 0b01, 0, 1); 1739 INSN(fnmaddd, 0b000, 0b01, 1, 0); 1740 INSN(fnmsub, 0b000, 0b01, 1, 1); 1741 1742 #undef INSN 1743 1744 // Floating-point conditional select 1745 void fp_conditional_select(unsigned op31, unsigned type, 1746 unsigned op1, unsigned op2, 1747 Condition cond, FloatRegister Vd, 1748 FloatRegister Vn, FloatRegister Vm) { 1749 starti; 1750 f(op31, 31, 29); 1751 f(0b11110, 28, 24); 1752 f(type, 23, 22); 1753 f(op1, 21, 21); 1754 f(op2, 11, 10); 1755 f(cond, 15, 12); 1756 rf(Vm, 16), rf(Vn, 5), rf(Vd, 0); 1757 } 1758 1759 #define INSN(NAME, op31, type, op1, op2) \ 1760 void NAME(FloatRegister Vd, FloatRegister Vn, \ 1761 FloatRegister Vm, Condition cond) { \ 1762 fp_conditional_select(op31, type, op1, op2, cond, Vd, Vn, Vm); \ 1763 } 1764 1765 INSN(fcsels, 0b000, 0b00, 0b1, 0b11); 1766 INSN(fcseld, 0b000, 0b01, 0b1, 0b11); 1767 1768 #undef INSN 1769 1770 // Floating-point<->integer conversions 1771 void float_int_convert(unsigned op31, unsigned type, 1772 unsigned rmode, unsigned opcode, 1773 Register Rd, Register Rn) { 1774 starti; 1775 f(op31, 31, 29); 1776 f(0b11110, 28, 24); 1777 f(type, 23, 22), f(1, 21), f(rmode, 20, 19); 1778 f(opcode, 18, 16), f(0b000000, 15, 10); 1779 zrf(Rn, 5), zrf(Rd, 0); 1780 } 1781 1782 #define INSN(NAME, op31, type, rmode, opcode) \ 1783 void NAME(Register Rd, FloatRegister Vn) { \ 1784 float_int_convert(op31, type, rmode, opcode, Rd, (Register)Vn); \ 1785 } 1786 1787 INSN(fcvtzsw, 0b000, 0b00, 0b11, 0b000); 1788 INSN(fcvtzs, 0b100, 0b00, 0b11, 0b000); 1789 INSN(fcvtzdw, 0b000, 0b01, 0b11, 0b000); 1790 INSN(fcvtzd, 0b100, 0b01, 0b11, 0b000); 1791 1792 INSN(fmovs, 0b000, 0b00, 0b00, 0b110); 1793 INSN(fmovd, 0b100, 0b01, 0b00, 0b110); 1794 1795 // INSN(fmovhid, 0b100, 0b10, 0b01, 0b110); 1796 1797 #undef INSN 1798 1799 #define INSN(NAME, op31, type, rmode, opcode) \ 1800 void NAME(FloatRegister Vd, Register Rn) { \ 1801 float_int_convert(op31, type, rmode, opcode, (Register)Vd, Rn); \ 1802 } 1803 1804 INSN(fmovs, 0b000, 0b00, 0b00, 0b111); 1805 INSN(fmovd, 0b100, 0b01, 0b00, 0b111); 1806 1807 INSN(scvtfws, 0b000, 0b00, 0b00, 0b010); 1808 INSN(scvtfs, 0b100, 0b00, 0b00, 0b010); 1809 INSN(scvtfwd, 0b000, 0b01, 0b00, 0b010); 1810 INSN(scvtfd, 0b100, 0b01, 0b00, 0b010); 1811 1812 // INSN(fmovhid, 0b100, 0b10, 0b01, 0b111); 1813 1814 #undef INSN 1815 1816 // Floating-point compare 1817 void float_compare(unsigned op31, unsigned type, 1818 unsigned op, unsigned op2, 1819 FloatRegister Vn, FloatRegister Vm = (FloatRegister)0) { 1820 starti; 1821 f(op31, 31, 29); 1822 f(0b11110, 28, 24); 1823 f(type, 23, 22), f(1, 21); 1824 f(op, 15, 14), f(0b1000, 13, 10), f(op2, 4, 0); 1825 rf(Vn, 5), rf(Vm, 16); 1826 } 1827 1828 1829 #define INSN(NAME, op31, type, op, op2) \ 1830 void NAME(FloatRegister Vn, FloatRegister Vm) { \ 1831 float_compare(op31, type, op, op2, Vn, Vm); \ 1832 } 1833 1834 #define INSN1(NAME, op31, type, op, op2) \ 1835 void NAME(FloatRegister Vn, double d) { \ 1836 assert_cond(d == 0.0); \ 1837 float_compare(op31, type, op, op2, Vn); \ 1838 } 1839 1840 INSN(fcmps, 0b000, 0b00, 0b00, 0b00000); 1841 INSN1(fcmps, 0b000, 0b00, 0b00, 0b01000); 1842 // INSN(fcmpes, 0b000, 0b00, 0b00, 0b10000); 1843 // INSN1(fcmpes, 0b000, 0b00, 0b00, 0b11000); 1844 1845 INSN(fcmpd, 0b000, 0b01, 0b00, 0b00000); 1846 INSN1(fcmpd, 0b000, 0b01, 0b00, 0b01000); 1847 // INSN(fcmped, 0b000, 0b01, 0b00, 0b10000); 1848 // INSN1(fcmped, 0b000, 0b01, 0b00, 0b11000); 1849 1850 #undef INSN 1851 #undef INSN1 1852 1853 // Floating-point Move (immediate) 1854 private: 1855 unsigned pack(double value); 1856 1857 void fmov_imm(FloatRegister Vn, double value, unsigned size) { 1858 starti; 1859 f(0b00011110, 31, 24), f(size, 23, 22), f(1, 21); 1860 f(pack(value), 20, 13), f(0b10000000, 12, 5); 1861 rf(Vn, 0); 1862 } 1863 1864 public: 1865 1866 void fmovs(FloatRegister Vn, double value) { 1867 if (value) 1868 fmov_imm(Vn, value, 0b00); 1869 else 1870 fmovs(Vn, zr); 1871 } 1872 void fmovd(FloatRegister Vn, double value) { 1873 if (value) 1874 fmov_imm(Vn, value, 0b01); 1875 else 1876 fmovd(Vn, zr); 1877 } 1878 1879 /* SIMD extensions 1880 * 1881 * We just use FloatRegister in the following. They are exactly the same 1882 * as SIMD registers. 1883 */ 1884 public: 1885 1886 enum SIMD_Arrangement { 1887 T8B, T16B, T4H, T8H, T2S, T4S, T1D, T2D 1888 }; 1889 1890 enum SIMD_RegVariant { 1891 S32, D64, Q128 1892 }; 1893 1894 1895 private: 1896 1897 void ld_st(FloatRegister Vt, SIMD_Arrangement T, Register Xn, int op1, int op2) { 1898 starti; 1899 f(0,31), f((int)T & 1, 30); 1900 f(op1, 29, 21), f(0, 20, 16), f(op2, 15, 12); 1901 f((int)T >> 1, 11, 10), rf(Xn, 5), rf(Vt, 0); 1902 } 1903 void ld_st(FloatRegister Vt, SIMD_Arrangement T, Register Xn, 1904 int imm, int op1, int op2) { 1905 starti; 1906 f(0,31), f((int)T & 1, 30); 1907 f(op1 | 0b100, 29, 21), f(0b11111, 20, 16), f(op2, 15, 12); 1908 f((int)T >> 1, 11, 10), rf(Xn, 5), rf(Vt, 0); 1909 } 1910 void ld_st(FloatRegister Vt, SIMD_Arrangement T, Register Xn, 1911 Register Xm, int op1, int op2) { 1912 starti; 1913 f(0,31), f((int)T & 1, 30); 1914 f(op1 | 0b100, 29, 21), rf(Xm, 16), f(op2, 15, 12); 1915 f((int)T >> 1, 11, 10), rf(Xn, 5), rf(Vt, 0); 1916 } 1917 1918 void ld_st(FloatRegister Vt, SIMD_Arrangement T, Address a, int op1, int op2) { 1919 switch (a.getMode()) { 1920 case Address::base_plus_offset: 1921 guarantee(a.offset() == 0, "no offset allowed here"); 1922 ld_st(Vt, T, a.base(), op1, op2); 1923 break; 1924 case Address::post: 1925 ld_st(Vt, T, a.base(), a.offset(), op1, op2); 1926 break; 1927 case Address::base_plus_offset_reg: 1928 ld_st(Vt, T, a.base(), a.index(), op1, op2); 1929 break; 1930 default: 1931 ShouldNotReachHere(); 1932 } 1933 } 1934 1935 public: 1936 1937 #define INSN1(NAME, op1, op2) \ 1938 void NAME(FloatRegister Vt, SIMD_Arrangement T, const Address &a) { \ 1939 ld_st(Vt, T, a, op1, op2); \ 1940 } 1941 1942 #define INSN2(NAME, op1, op2) \ 1943 void NAME(FloatRegister Vt, FloatRegister Vt2, SIMD_Arrangement T, const Address &a) { \ 1944 assert(Vt->successor() == Vt2, "Registers must be ordered"); \ 1945 ld_st(Vt, T, a, op1, op2); \ 1946 } 1947 1948 #define INSN3(NAME, op1, op2) \ 1949 void NAME(FloatRegister Vt, FloatRegister Vt2, FloatRegister Vt3, \ 1950 SIMD_Arrangement T, const Address &a) { \ 1951 assert(Vt->successor() == Vt2 && Vt2->successor() == Vt3, \ 1952 "Registers must be ordered"); \ 1953 ld_st(Vt, T, a, op1, op2); \ 1954 } 1955 1956 #define INSN4(NAME, op1, op2) \ 1957 void NAME(FloatRegister Vt, FloatRegister Vt2, FloatRegister Vt3, \ 1958 FloatRegister Vt4, SIMD_Arrangement T, const Address &a) { \ 1959 assert(Vt->successor() == Vt2 && Vt2->successor() == Vt3 && \ 1960 Vt3->successor() == Vt4, "Registers must be ordered"); \ 1961 ld_st(Vt, T, a, op1, op2); \ 1962 } 1963 1964 INSN1(ld1, 0b001100010, 0b0111); 1965 INSN2(ld1, 0b001100010, 0b1010); 1966 INSN3(ld1, 0b001100010, 0b0110); 1967 INSN4(ld1, 0b001100010, 0b0010); 1968 1969 INSN2(ld2, 0b001100010, 0b1000); 1970 INSN3(ld3, 0b001100010, 0b0100); 1971 INSN4(ld4, 0b001100010, 0b0000); 1972 1973 INSN1(st1, 0b001100000, 0b0111); 1974 INSN2(st1, 0b001100000, 0b1010); 1975 INSN3(st1, 0b001100000, 0b0110); 1976 INSN4(st1, 0b001100000, 0b0010); 1977 1978 INSN2(st2, 0b001100000, 0b1000); 1979 INSN3(st3, 0b001100000, 0b0100); 1980 INSN4(st4, 0b001100000, 0b0000); 1981 1982 INSN1(ld1r, 0b001101010, 0b1100); 1983 INSN2(ld2r, 0b001101011, 0b1100); 1984 INSN3(ld3r, 0b001101010, 0b1110); 1985 INSN4(ld4r, 0b001101011, 0b1110); 1986 1987 #undef INSN1 1988 #undef INSN2 1989 #undef INSN3 1990 #undef INSN4 1991 1992 #define INSN(NAME, opc) \ 1993 void NAME(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn, FloatRegister Vm) { \ 1994 starti; \ 1995 assert(T == T8B || T == T16B, "must be T8B or T16B"); \ 1996 f(0, 31), f((int)T & 1, 30), f(opc, 29, 21); \ 1997 rf(Vm, 16), f(0b000111, 15, 10), rf(Vn, 5), rf(Vd, 0); \ 1998 } 1999 2000 INSN(eor, 0b101110001); 2001 INSN(orr, 0b001110101); 2002 INSN(andr, 0b001110001); 2003 INSN(bic, 0b001110011); 2004 INSN(bif, 0b101110111); 2005 INSN(bit, 0b101110101); 2006 INSN(bsl, 0b101110011); 2007 INSN(orn, 0b001110111); 2008 2009 #undef INSN 2010 2011 #define INSN(NAME, opc) \ 2012 void NAME(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn, FloatRegister Vm) { \ 2013 starti; \ 2014 f(0, 31), f((int)T & 1, 30), f(opc, 29), f(0b01110, 28, 24); \ 2015 f((int)T >> 1, 23, 22), f(1, 21), rf(Vm, 16), f(0b100001, 15, 10); \ 2016 rf(Vn, 5), rf(Vd, 0); \ 2017 } 2018 2019 INSN(addv, 0); 2020 INSN(subv, 1); 2021 2022 #undef INSN 2023 2024 #define INSN(NAME, opc) \ 2025 void NAME(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn, FloatRegister Vm) { \ 2026 starti; \ 2027 assert(T == T4S, "arrangement must be T4S"); \ 2028 f(0b01011110000, 31, 21), rf(Vm, 16), f(opc, 15, 10), rf(Vn, 5), rf(Vd, 0); \ 2029 } 2030 2031 INSN(sha1c, 0b000000); 2032 INSN(sha1m, 0b001000); 2033 INSN(sha1p, 0b000100); 2034 INSN(sha1su0, 0b001100); 2035 INSN(sha256h2, 0b010100); 2036 INSN(sha256h, 0b010000); 2037 INSN(sha256su1, 0b011000); 2038 2039 #undef INSN 2040 2041 #define INSN(NAME, opc) \ 2042 void NAME(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) { \ 2043 starti; \ 2044 assert(T == T4S, "arrangement must be T4S"); \ 2045 f(0b0101111000101000, 31, 16), f(opc, 15, 10), rf(Vn, 5), rf(Vd, 0); \ 2046 } 2047 2048 INSN(sha1h, 0b000010); 2049 INSN(sha1su1, 0b000110); 2050 INSN(sha256su0, 0b001010); 2051 2052 #undef INSN 2053 2054 #define INSN(NAME, opc) \ 2055 void NAME(FloatRegister Vd, FloatRegister Vn) { \ 2056 starti; \ 2057 f(opc, 31, 10), rf(Vn, 5), rf(Vd, 0); \ 2058 } 2059 2060 INSN(aese, 0b0100111000101000010010); 2061 INSN(aesd, 0b0100111000101000010110); 2062 INSN(aesmc, 0b0100111000101000011010); 2063 INSN(aesimc, 0b0100111000101000011110); 2064 2065 #undef INSN 2066 2067 void shl(FloatRegister Vd, FloatRegister Vn, SIMD_Arrangement T, int shift){ 2068 starti; 2069 /* The encodings for the immh:immb fields (bits 22:16) are 2070 * 0001 xxx 8B/16B, shift = xxx 2071 * 001x xxx 4H/8H, shift = xxxx 2072 * 01xx xxx 2S/4S, shift = xxxxx 2073 * 1xxx xxx 1D/2D, shift = xxxxxx (1D is RESERVED) 2074 */ 2075 assert((1 << ((T>>1)+3)) > shift, "Invalid Shift value"); 2076 f(0, 31), f(T & 1, 30), f(0b0011110, 29, 23), f((1 << ((T>>1)+3))|shift, 22, 16); 2077 f(0b010101, 15, 10), rf(Vn, 5), rf(Vd, 0); 2078 } 2079 2080 void ushll(FloatRegister Vd, SIMD_Arrangement Ta, FloatRegister Vn, SIMD_Arrangement Tb, int shift) { 2081 starti; 2082 /* The encodings for the immh:immb fields (bits 22:16) are 2083 * 0001 xxx 8H, 8B/16b shift = xxx 2084 * 001x xxx 4S, 4H/8H shift = xxxx 2085 * 01xx xxx 2D, 2S/4S shift = xxxxx 2086 * 1xxx xxx RESERVED 2087 */ 2088 assert((Tb >> 1) + 1 == (Ta >> 1), "Incompatible arrangement"); 2089 assert((1 << ((Tb>>1)+3)) > shift, "Invalid shift value"); 2090 f(0, 31), f(Tb & 1, 30), f(0b1011110, 29, 23), f((1 << ((Tb>>1)+3))|shift, 22, 16); 2091 f(0b101001, 15, 10), rf(Vn, 5), rf(Vd, 0); 2092 } 2093 void ushll2(FloatRegister Vd, SIMD_Arrangement Ta, FloatRegister Vn, SIMD_Arrangement Tb, int shift) { 2094 ushll(Vd, Ta, Vn, Tb, shift); 2095 } 2096 2097 void uzp1(FloatRegister Vd, FloatRegister Vn, FloatRegister Vm, SIMD_Arrangement T, int op = 0){ 2098 starti; 2099 f(0, 31), f((T & 0x1), 30), f(0b001110, 29, 24), f((T >> 1), 23, 22), f(0, 21); 2100 rf(Vm, 16), f(0, 15), f(op, 14), f(0b0110, 13, 10), rf(Vn, 5), rf(Vd, 0); 2101 } 2102 void uzp2(FloatRegister Vd, FloatRegister Vn, FloatRegister Vm, SIMD_Arrangement T){ 2103 uzp1(Vd, Vn, Vm, T, 1); 2104 } 2105 2106 // Move from general purpose register 2107 // mov Vd.T[index], Rn 2108 void mov(FloatRegister Vd, SIMD_Arrangement T, int index, Register Xn) { 2109 starti; 2110 f(0b01001110000, 31, 21), f(((1 << (T >> 1)) | (index << ((T >> 1) + 1))), 20, 16); 2111 f(0b000111, 15, 10), rf(Xn, 5), rf(Vd, 0); 2112 } 2113 2114 // Move to general purpose register 2115 // mov Rd, Vn.T[index] 2116 void mov(Register Xd, FloatRegister Vn, SIMD_Arrangement T, int index) { 2117 starti; 2118 f(0, 31), f((T >= T1D) ? 1:0, 30), f(0b001110000, 29, 21); 2119 f(((1 << (T >> 1)) | (index << ((T >> 1) + 1))), 20, 16); 2120 f(0b001111, 15, 10), rf(Vn, 5), rf(Xd, 0); 2121 } 2122 2123 // We do not handle the 1Q arrangement. 2124 void pmull(FloatRegister Vd, SIMD_Arrangement Ta, FloatRegister Vn, FloatRegister Vm, SIMD_Arrangement Tb) { 2125 starti; 2126 assert(Ta == T8H && (Tb == T8B || Tb == T16B), "Invalid Size specifier"); 2127 f(0, 31), f(Tb & 1, 30), f(0b001110001, 29, 21), rf(Vm, 16), f(0b111000, 15, 10); 2128 rf(Vn, 5), rf(Vd, 0); 2129 } 2130 void pmull2(FloatRegister Vd, SIMD_Arrangement Ta, FloatRegister Vn, FloatRegister Vm, SIMD_Arrangement Tb) { 2131 pmull(Vd, Ta, Vn, Vm, Tb); 2132 } 2133 2134 void uqxtn(FloatRegister Vd, SIMD_Arrangement Tb, FloatRegister Vn, SIMD_Arrangement Ta) { 2135 starti; 2136 int size_b = (int)Tb >> 1; 2137 int size_a = (int)Ta >> 1; 2138 assert(size_b < 3 && size_b == size_a - 1, "Invalid size specifier"); 2139 f(0, 31), f(Tb & 1, 30), f(0b101110, 29, 24), f(size_b, 23, 22); 2140 f(0b100001010010, 21, 10), rf(Vn, 5), rf(Vd, 0); 2141 } 2142 2143 void rev32(FloatRegister Vd, SIMD_Arrangement T, FloatRegister Vn) 2144 { 2145 starti; 2146 assert(T <= T8H, "must be one of T8B, T16B, T4H, T8H"); 2147 f(0, 31), f((int)T & 1, 30), f(0b101110, 29, 24); 2148 f(T <= T16B ? 0b00 : 0b01, 23, 22), f(0b100000000010, 21, 10); 2149 rf(Vn, 5), rf(Vd, 0); 2150 } 2151 2152 // CRC32 instructions 2153 #define INSN(NAME, sf, sz) \ 2154 void NAME(Register Rd, Register Rn, Register Rm) { \ 2155 starti; \ 2156 f(sf, 31), f(0b0011010110, 30, 21), f(0b0100, 15, 12), f(sz, 11, 10); \ 2157 rf(Rm, 16), rf(Rn, 5), rf(Rd, 0); \ 2158 } 2159 2160 INSN(crc32b, 0, 0b00); 2161 INSN(crc32h, 0, 0b01); 2162 INSN(crc32w, 0, 0b10); 2163 INSN(crc32x, 1, 0b11); 2164 2165 #undef INSN 2166 2167 2168 /* Simulator extensions to the ISA 2169 2170 haltsim 2171 2172 takes no arguments, causes the sim to enter a debug break and then 2173 return from the simulator run() call with STATUS_HALT? The linking 2174 code will call fatal() when it sees STATUS_HALT. 2175 2176 blrt Xn, Wm 2177 blrt Xn, #gpargs, #fpargs, #type 2178 Xn holds the 64 bit x86 branch_address 2179 call format is encoded either as immediate data in the call 2180 or in register Wm. In the latter case 2181 Wm[13..6] = #gpargs, 2182 Wm[5..2] = #fpargs, 2183 Wm[1,0] = #type 2184 2185 calls the x86 code address 'branch_address' supplied in Xn passing 2186 arguments taken from the general and floating point registers according 2187 to the supplied counts 'gpargs' and 'fpargs'. may return a result in r0 2188 or v0 according to the the return type #type' where 2189 2190 address branch_address; 2191 uimm4 gpargs; 2192 uimm4 fpargs; 2193 enum ReturnType type; 2194 2195 enum ReturnType 2196 { 2197 void_ret = 0, 2198 int_ret = 1, 2199 long_ret = 1, 2200 obj_ret = 1, // i.e. same as long 2201 float_ret = 2, 2202 double_ret = 3 2203 } 2204 2205 notify 2206 2207 notifies the simulator of a transfer of control. instr[14:0] 2208 identifies the type of change of control. 2209 2210 0 ==> initial entry to a method. 2211 2212 1 ==> return into a method from a submethod call. 2213 2214 2 ==> exit out of Java method code. 2215 2216 3 ==> start execution for a new bytecode. 2217 2218 in cases 1 and 2 the simulator is expected to use a JVM callback to 2219 identify the name of the specific method being executed. in case 4 2220 the simulator is expected to use a JVM callback to identify the 2221 bytecode index. 2222 2223 Instruction encodings 2224 --------------------- 2225 2226 These are encoded in the space with instr[28:25] = 00 which is 2227 unallocated. Encodings are 2228 2229 10987654321098765432109876543210 2230 PSEUDO_HALT = 0x11100000000000000000000000000000 2231 PSEUDO_BLRT = 0x11000000000000000_______________ 2232 PSEUDO_BLRTR = 0x1100000000000000100000__________ 2233 PSEUDO_NOTIFY = 0x10100000000000000_______________ 2234 2235 instr[31,29] = op1 : 111 ==> HALT, 110 ==> BLRT/BLRTR, 101 ==> NOTIFY 2236 2237 for BLRT 2238 instr[14,11] = #gpargs, instr[10,7] = #fpargs 2239 instr[6,5] = #type, instr[4,0] = Rn 2240 for BLRTR 2241 instr[9,5] = Rm, instr[4,0] = Rn 2242 for NOTIFY 2243 instr[14:0] = type : 0 ==> entry, 1 ==> reentry, 2 ==> exit, 3 ==> bcstart 2244 */ 2245 2246 enum NotifyType { method_entry, method_reentry, method_exit, bytecode_start }; 2247 2248 virtual void notify(int type) { 2249 if (UseBuiltinSim) { 2250 starti; 2251 // 109 2252 f(0b101, 31, 29); 2253 // 87654321098765 2254 f(0b00000000000000, 28, 15); 2255 f(type, 14, 0); 2256 } 2257 } 2258 2259 void blrt(Register Rn, int gpargs, int fpargs, int type) { 2260 if (UseBuiltinSim) { 2261 starti; 2262 f(0b110, 31 ,29); 2263 f(0b00, 28, 25); 2264 // 4321098765 2265 f(0b0000000000, 24, 15); 2266 f(gpargs, 14, 11); 2267 f(fpargs, 10, 7); 2268 f(type, 6, 5); 2269 rf(Rn, 0); 2270 } else { 2271 blr(Rn); 2272 } 2273 } 2274 2275 void blrt(Register Rn, Register Rm) { 2276 if (UseBuiltinSim) { 2277 starti; 2278 f(0b110, 31 ,29); 2279 f(0b00, 28, 25); 2280 // 4321098765 2281 f(0b0000000001, 24, 15); 2282 // 43210 2283 f(0b00000, 14, 10); 2284 rf(Rm, 5); 2285 rf(Rn, 0); 2286 } else { 2287 blr(Rn); 2288 } 2289 } 2290 2291 void haltsim() { 2292 starti; 2293 f(0b111, 31 ,29); 2294 f(0b00, 28, 27); 2295 // 654321098765432109876543210 2296 f(0b000000000000000000000000000, 26, 0); 2297 } 2298 2299 Assembler(CodeBuffer* code) : AbstractAssembler(code) { 2300 } 2301 2302 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 2303 Register tmp, 2304 int offset) { 2305 ShouldNotCallThis(); 2306 return RegisterOrConstant(); 2307 } 2308 2309 // Stack overflow checking 2310 virtual void bang_stack_with_offset(int offset); 2311 2312 static bool operand_valid_for_logical_immediate(bool is32, uint64_t imm); 2313 static bool operand_valid_for_add_sub_immediate(long imm); 2314 static bool operand_valid_for_float_immediate(double imm); 2315 2316 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 2317 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 2318 }; 2319 2320 inline Assembler::Membar_mask_bits operator|(Assembler::Membar_mask_bits a, 2321 Assembler::Membar_mask_bits b) { 2322 return Assembler::Membar_mask_bits(unsigned(a)|unsigned(b)); 2323 } 2324 2325 Instruction_aarch64::~Instruction_aarch64() { 2326 assem->emit(); 2327 } 2328 2329 #undef starti 2330 2331 // Invert a condition 2332 inline const Assembler::Condition operator~(const Assembler::Condition cond) { 2333 return Assembler::Condition(int(cond) ^ 1); 2334 } 2335 2336 class BiasedLockingCounters; 2337 2338 extern "C" void das(uint64_t start, int len); 2339 2340 #endif // CPU_AARCH64_VM_ASSEMBLER_AARCH64_HPP