1 /* 2 * Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef CPU_X86_ASSEMBLER_X86_HPP 26 #define CPU_X86_ASSEMBLER_X86_HPP 27 28 #include "asm/register.hpp" 29 #include "vm_version_x86.hpp" 30 31 class BiasedLockingCounters; 32 33 // Contains all the definitions needed for x86 assembly code generation. 34 35 // Calling convention 36 class Argument { 37 public: 38 enum { 39 #ifdef _LP64 40 #ifdef _WIN64 41 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) 42 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) 43 #else 44 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) 45 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) 46 #endif // _WIN64 47 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... 48 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... 49 #else 50 n_register_parameters = 0 // 0 registers used to pass arguments 51 #endif // _LP64 52 }; 53 }; 54 55 56 #ifdef _LP64 57 // Symbolically name the register arguments used by the c calling convention. 58 // Windows is different from linux/solaris. So much for standards... 59 60 #ifdef _WIN64 61 62 REGISTER_DECLARATION(Register, c_rarg0, rcx); 63 REGISTER_DECLARATION(Register, c_rarg1, rdx); 64 REGISTER_DECLARATION(Register, c_rarg2, r8); 65 REGISTER_DECLARATION(Register, c_rarg3, r9); 66 67 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 68 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 69 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 70 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 71 72 #else 73 74 REGISTER_DECLARATION(Register, c_rarg0, rdi); 75 REGISTER_DECLARATION(Register, c_rarg1, rsi); 76 REGISTER_DECLARATION(Register, c_rarg2, rdx); 77 REGISTER_DECLARATION(Register, c_rarg3, rcx); 78 REGISTER_DECLARATION(Register, c_rarg4, r8); 79 REGISTER_DECLARATION(Register, c_rarg5, r9); 80 81 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 82 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 83 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 84 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 85 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); 86 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); 87 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); 88 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); 89 90 #endif // _WIN64 91 92 // Symbolically name the register arguments used by the Java calling convention. 93 // We have control over the convention for java so we can do what we please. 94 // What pleases us is to offset the java calling convention so that when 95 // we call a suitable jni method the arguments are lined up and we don't 96 // have to do little shuffling. A suitable jni method is non-static and a 97 // small number of arguments (two fewer args on windows) 98 // 99 // |-------------------------------------------------------| 100 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | 101 // |-------------------------------------------------------| 102 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) 103 // | rdi rsi rdx rcx r8 r9 | solaris/linux 104 // |-------------------------------------------------------| 105 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | 106 // |-------------------------------------------------------| 107 108 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); 109 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); 110 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); 111 // Windows runs out of register args here 112 #ifdef _WIN64 113 REGISTER_DECLARATION(Register, j_rarg3, rdi); 114 REGISTER_DECLARATION(Register, j_rarg4, rsi); 115 #else 116 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); 117 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); 118 #endif /* _WIN64 */ 119 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); 120 121 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); 122 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); 123 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); 124 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); 125 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); 126 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); 127 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); 128 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); 129 130 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile 131 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile 132 133 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved 134 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved 135 136 #else 137 // rscratch1 will apear in 32bit code that is dead but of course must compile 138 // Using noreg ensures if the dead code is incorrectly live and executed it 139 // will cause an assertion failure 140 #define rscratch1 noreg 141 #define rscratch2 noreg 142 143 #endif // _LP64 144 145 // JSR 292 146 // On x86, the SP does not have to be saved when invoking method handle intrinsics 147 // or compiled lambda forms. We indicate that by setting rbp_mh_SP_save to noreg. 148 REGISTER_DECLARATION(Register, rbp_mh_SP_save, noreg); 149 150 // Address is an abstraction used to represent a memory location 151 // using any of the amd64 addressing modes with one object. 152 // 153 // Note: A register location is represented via a Register, not 154 // via an address for efficiency & simplicity reasons. 155 156 class ArrayAddress; 157 158 class Address { 159 public: 160 enum ScaleFactor { 161 no_scale = -1, 162 times_1 = 0, 163 times_2 = 1, 164 times_4 = 2, 165 times_8 = 3, 166 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) 167 }; 168 static ScaleFactor times(int size) { 169 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size"); 170 if (size == 8) return times_8; 171 if (size == 4) return times_4; 172 if (size == 2) return times_2; 173 return times_1; 174 } 175 static int scale_size(ScaleFactor scale) { 176 assert(scale != no_scale, ""); 177 assert(((1 << (int)times_1) == 1 && 178 (1 << (int)times_2) == 2 && 179 (1 << (int)times_4) == 4 && 180 (1 << (int)times_8) == 8), ""); 181 return (1 << (int)scale); 182 } 183 184 private: 185 Register _base; 186 Register _index; 187 XMMRegister _xmmindex; 188 ScaleFactor _scale; 189 int _disp; 190 bool _isxmmindex; 191 RelocationHolder _rspec; 192 193 // Easily misused constructors make them private 194 // %%% can we make these go away? 195 NOT_LP64(Address(address loc, RelocationHolder spec);) 196 Address(int disp, address loc, relocInfo::relocType rtype); 197 Address(int disp, address loc, RelocationHolder spec); 198 199 public: 200 201 int disp() { return _disp; } 202 // creation 203 Address() 204 : _base(noreg), 205 _index(noreg), 206 _xmmindex(xnoreg), 207 _scale(no_scale), 208 _disp(0), 209 _isxmmindex(false){ 210 } 211 212 // No default displacement otherwise Register can be implicitly 213 // converted to 0(Register) which is quite a different animal. 214 215 Address(Register base, int disp) 216 : _base(base), 217 _index(noreg), 218 _xmmindex(xnoreg), 219 _scale(no_scale), 220 _disp(disp), 221 _isxmmindex(false){ 222 } 223 224 Address(Register base, Register index, ScaleFactor scale, int disp = 0) 225 : _base (base), 226 _index(index), 227 _xmmindex(xnoreg), 228 _scale(scale), 229 _disp (disp), 230 _isxmmindex(false) { 231 assert(!index->is_valid() == (scale == Address::no_scale), 232 "inconsistent address"); 233 } 234 235 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) 236 : _base (base), 237 _index(index.register_or_noreg()), 238 _xmmindex(xnoreg), 239 _scale(scale), 240 _disp (disp + (index.constant_or_zero() * scale_size(scale))), 241 _isxmmindex(false){ 242 if (!index.is_register()) scale = Address::no_scale; 243 assert(!_index->is_valid() == (scale == Address::no_scale), 244 "inconsistent address"); 245 } 246 247 Address(Register base, XMMRegister index, ScaleFactor scale, int disp = 0) 248 : _base (base), 249 _index(noreg), 250 _xmmindex(index), 251 _scale(scale), 252 _disp(disp), 253 _isxmmindex(true) { 254 assert(!index->is_valid() == (scale == Address::no_scale), 255 "inconsistent address"); 256 } 257 258 Address plus_disp(int disp) const { 259 Address a = (*this); 260 a._disp += disp; 261 return a; 262 } 263 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const { 264 Address a = (*this); 265 a._disp += disp.constant_or_zero() * scale_size(scale); 266 if (disp.is_register()) { 267 assert(!a.index()->is_valid(), "competing indexes"); 268 a._index = disp.as_register(); 269 a._scale = scale; 270 } 271 return a; 272 } 273 bool is_same_address(Address a) const { 274 // disregard _rspec 275 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale; 276 } 277 278 // The following two overloads are used in connection with the 279 // ByteSize type (see sizes.hpp). They simplify the use of 280 // ByteSize'd arguments in assembly code. Note that their equivalent 281 // for the optimized build are the member functions with int disp 282 // argument since ByteSize is mapped to an int type in that case. 283 // 284 // Note: DO NOT introduce similar overloaded functions for WordSize 285 // arguments as in the optimized mode, both ByteSize and WordSize 286 // are mapped to the same type and thus the compiler cannot make a 287 // distinction anymore (=> compiler errors). 288 289 #ifdef ASSERT 290 Address(Register base, ByteSize disp) 291 : _base(base), 292 _index(noreg), 293 _xmmindex(xnoreg), 294 _scale(no_scale), 295 _disp(in_bytes(disp)), 296 _isxmmindex(false){ 297 } 298 299 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 300 : _base(base), 301 _index(index), 302 _xmmindex(xnoreg), 303 _scale(scale), 304 _disp(in_bytes(disp)), 305 _isxmmindex(false){ 306 assert(!index->is_valid() == (scale == Address::no_scale), 307 "inconsistent address"); 308 } 309 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 310 : _base (base), 311 _index(index.register_or_noreg()), 312 _xmmindex(xnoreg), 313 _scale(scale), 314 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))), 315 _isxmmindex(false) { 316 if (!index.is_register()) scale = Address::no_scale; 317 assert(!_index->is_valid() == (scale == Address::no_scale), 318 "inconsistent address"); 319 } 320 321 #endif // ASSERT 322 323 // accessors 324 bool uses(Register reg) const { return _base == reg || _index == reg; } 325 Register base() const { return _base; } 326 Register index() const { return _index; } 327 XMMRegister xmmindex() const { return _xmmindex; } 328 ScaleFactor scale() const { return _scale; } 329 int disp() const { return _disp; } 330 bool isxmmindex() const { return _isxmmindex; } 331 332 // Convert the raw encoding form into the form expected by the constructor for 333 // Address. An index of 4 (rsp) corresponds to having no index, so convert 334 // that to noreg for the Address constructor. 335 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc); 336 337 static Address make_array(ArrayAddress); 338 339 private: 340 bool base_needs_rex() const { 341 return _base != noreg && _base->encoding() >= 8; 342 } 343 344 bool index_needs_rex() const { 345 return _index != noreg &&_index->encoding() >= 8; 346 } 347 348 bool xmmindex_needs_rex() const { 349 return _xmmindex != xnoreg && _xmmindex->encoding() >= 8; 350 } 351 352 relocInfo::relocType reloc() const { return _rspec.type(); } 353 354 friend class Assembler; 355 friend class MacroAssembler; 356 friend class LIR_Assembler; // base/index/scale/disp 357 }; 358 359 // 360 // AddressLiteral has been split out from Address because operands of this type 361 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 362 // the few instructions that need to deal with address literals are unique and the 363 // MacroAssembler does not have to implement every instruction in the Assembler 364 // in order to search for address literals that may need special handling depending 365 // on the instruction and the platform. As small step on the way to merging i486/amd64 366 // directories. 367 // 368 class AddressLiteral { 369 friend class ArrayAddress; 370 RelocationHolder _rspec; 371 // Typically we use AddressLiterals we want to use their rval 372 // However in some situations we want the lval (effect address) of the item. 373 // We provide a special factory for making those lvals. 374 bool _is_lval; 375 376 // If the target is far we'll need to load the ea of this to 377 // a register to reach it. Otherwise if near we can do rip 378 // relative addressing. 379 380 address _target; 381 382 protected: 383 // creation 384 AddressLiteral() 385 : _is_lval(false), 386 _target(NULL) 387 {} 388 389 public: 390 391 392 AddressLiteral(address target, relocInfo::relocType rtype); 393 394 AddressLiteral(address target, RelocationHolder const& rspec) 395 : _rspec(rspec), 396 _is_lval(false), 397 _target(target) 398 {} 399 400 AddressLiteral addr() { 401 AddressLiteral ret = *this; 402 ret._is_lval = true; 403 return ret; 404 } 405 406 407 private: 408 409 address target() { return _target; } 410 bool is_lval() { return _is_lval; } 411 412 relocInfo::relocType reloc() const { return _rspec.type(); } 413 const RelocationHolder& rspec() const { return _rspec; } 414 415 friend class Assembler; 416 friend class MacroAssembler; 417 friend class Address; 418 friend class LIR_Assembler; 419 }; 420 421 // Convience classes 422 class RuntimeAddress: public AddressLiteral { 423 424 public: 425 426 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 427 428 }; 429 430 class ExternalAddress: public AddressLiteral { 431 private: 432 static relocInfo::relocType reloc_for_target(address target) { 433 // Sometimes ExternalAddress is used for values which aren't 434 // exactly addresses, like the card table base. 435 // external_word_type can't be used for values in the first page 436 // so just skip the reloc in that case. 437 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 438 } 439 440 public: 441 442 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} 443 444 }; 445 446 class InternalAddress: public AddressLiteral { 447 448 public: 449 450 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 451 452 }; 453 454 // x86 can do array addressing as a single operation since disp can be an absolute 455 // address amd64 can't. We create a class that expresses the concept but does extra 456 // magic on amd64 to get the final result 457 458 class ArrayAddress { 459 private: 460 461 AddressLiteral _base; 462 Address _index; 463 464 public: 465 466 ArrayAddress() {}; 467 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 468 AddressLiteral base() { return _base; } 469 Address index() { return _index; } 470 471 }; 472 473 class InstructionAttr; 474 475 // 64-bit refect the fxsave size which is 512 bytes and the new xsave area on EVEX which is another 2176 bytes 476 // See fxsave and xsave(EVEX enabled) documentation for layout 477 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY(2688 / wordSize); 478 479 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 480 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 481 // is what you get. The Assembler is generating code into a CodeBuffer. 482 483 class Assembler : public AbstractAssembler { 484 friend class AbstractAssembler; // for the non-virtual hack 485 friend class LIR_Assembler; // as_Address() 486 friend class StubGenerator; 487 488 public: 489 enum Condition { // The x86 condition codes used for conditional jumps/moves. 490 zero = 0x4, 491 notZero = 0x5, 492 equal = 0x4, 493 notEqual = 0x5, 494 less = 0xc, 495 lessEqual = 0xe, 496 greater = 0xf, 497 greaterEqual = 0xd, 498 below = 0x2, 499 belowEqual = 0x6, 500 above = 0x7, 501 aboveEqual = 0x3, 502 overflow = 0x0, 503 noOverflow = 0x1, 504 carrySet = 0x2, 505 carryClear = 0x3, 506 negative = 0x8, 507 positive = 0x9, 508 parity = 0xa, 509 noParity = 0xb 510 }; 511 512 enum Prefix { 513 // segment overrides 514 CS_segment = 0x2e, 515 SS_segment = 0x36, 516 DS_segment = 0x3e, 517 ES_segment = 0x26, 518 FS_segment = 0x64, 519 GS_segment = 0x65, 520 521 REX = 0x40, 522 523 REX_B = 0x41, 524 REX_X = 0x42, 525 REX_XB = 0x43, 526 REX_R = 0x44, 527 REX_RB = 0x45, 528 REX_RX = 0x46, 529 REX_RXB = 0x47, 530 531 REX_W = 0x48, 532 533 REX_WB = 0x49, 534 REX_WX = 0x4A, 535 REX_WXB = 0x4B, 536 REX_WR = 0x4C, 537 REX_WRB = 0x4D, 538 REX_WRX = 0x4E, 539 REX_WRXB = 0x4F, 540 541 VEX_3bytes = 0xC4, 542 VEX_2bytes = 0xC5, 543 EVEX_4bytes = 0x62, 544 Prefix_EMPTY = 0x0 545 }; 546 547 enum VexPrefix { 548 VEX_B = 0x20, 549 VEX_X = 0x40, 550 VEX_R = 0x80, 551 VEX_W = 0x80 552 }; 553 554 enum ExexPrefix { 555 EVEX_F = 0x04, 556 EVEX_V = 0x08, 557 EVEX_Rb = 0x10, 558 EVEX_X = 0x40, 559 EVEX_Z = 0x80 560 }; 561 562 enum VexSimdPrefix { 563 VEX_SIMD_NONE = 0x0, 564 VEX_SIMD_66 = 0x1, 565 VEX_SIMD_F3 = 0x2, 566 VEX_SIMD_F2 = 0x3 567 }; 568 569 enum VexOpcode { 570 VEX_OPCODE_NONE = 0x0, 571 VEX_OPCODE_0F = 0x1, 572 VEX_OPCODE_0F_38 = 0x2, 573 VEX_OPCODE_0F_3A = 0x3, 574 VEX_OPCODE_MASK = 0x1F 575 }; 576 577 enum AvxVectorLen { 578 AVX_128bit = 0x0, 579 AVX_256bit = 0x1, 580 AVX_512bit = 0x2, 581 AVX_NoVec = 0x4 582 }; 583 584 enum EvexTupleType { 585 EVEX_FV = 0, 586 EVEX_HV = 4, 587 EVEX_FVM = 6, 588 EVEX_T1S = 7, 589 EVEX_T1F = 11, 590 EVEX_T2 = 13, 591 EVEX_T4 = 15, 592 EVEX_T8 = 17, 593 EVEX_HVM = 18, 594 EVEX_QVM = 19, 595 EVEX_OVM = 20, 596 EVEX_M128 = 21, 597 EVEX_DUP = 22, 598 EVEX_ETUP = 23 599 }; 600 601 enum EvexInputSizeInBits { 602 EVEX_8bit = 0, 603 EVEX_16bit = 1, 604 EVEX_32bit = 2, 605 EVEX_64bit = 3, 606 EVEX_NObit = 4 607 }; 608 609 enum WhichOperand { 610 // input to locate_operand, and format code for relocations 611 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 612 disp32_operand = 1, // embedded 32-bit displacement or address 613 call32_operand = 2, // embedded 32-bit self-relative displacement 614 #ifndef _LP64 615 _WhichOperand_limit = 3 616 #else 617 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 618 _WhichOperand_limit = 4 619 #endif 620 }; 621 622 enum ComparisonPredicate { 623 eq = 0, 624 lt = 1, 625 le = 2, 626 _false = 3, 627 neq = 4, 628 nlt = 5, 629 nle = 6, 630 _true = 7 631 }; 632 633 634 // NOTE: The general philopsophy of the declarations here is that 64bit versions 635 // of instructions are freely declared without the need for wrapping them an ifdef. 636 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 637 // In the .cpp file the implementations are wrapped so that they are dropped out 638 // of the resulting jvm. This is done mostly to keep the footprint of MINIMAL 639 // to the size it was prior to merging up the 32bit and 64bit assemblers. 640 // 641 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 642 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 643 644 private: 645 646 bool _legacy_mode_bw; 647 bool _legacy_mode_dq; 648 bool _legacy_mode_vl; 649 bool _legacy_mode_vlbw; 650 bool _is_managed; 651 bool _vector_masking; // For stub code use only 652 653 class InstructionAttr *_attributes; 654 655 // 64bit prefixes 656 int prefix_and_encode(int reg_enc, bool byteinst = false); 657 int prefixq_and_encode(int reg_enc); 658 659 int prefix_and_encode(int dst_enc, int src_enc) { 660 return prefix_and_encode(dst_enc, false, src_enc, false); 661 } 662 int prefix_and_encode(int dst_enc, bool dst_is_byte, int src_enc, bool src_is_byte); 663 int prefixq_and_encode(int dst_enc, int src_enc); 664 665 void prefix(Register reg); 666 void prefix(Register dst, Register src, Prefix p); 667 void prefix(Register dst, Address adr, Prefix p); 668 void prefix(Address adr); 669 void prefixq(Address adr); 670 671 void prefix(Address adr, Register reg, bool byteinst = false); 672 void prefix(Address adr, XMMRegister reg); 673 void prefixq(Address adr, Register reg); 674 void prefixq(Address adr, XMMRegister reg); 675 676 void prefetch_prefix(Address src); 677 678 void rex_prefix(Address adr, XMMRegister xreg, 679 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 680 int rex_prefix_and_encode(int dst_enc, int src_enc, 681 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 682 683 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, int nds_enc, VexSimdPrefix pre, VexOpcode opc); 684 685 void evex_prefix(bool vex_r, bool vex_b, bool vex_x, bool evex_r, bool evex_v, 686 int nds_enc, VexSimdPrefix pre, VexOpcode opc); 687 688 void vex_prefix(Address adr, int nds_enc, int xreg_enc, 689 VexSimdPrefix pre, VexOpcode opc, 690 InstructionAttr *attributes); 691 692 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, 693 VexSimdPrefix pre, VexOpcode opc, 694 InstructionAttr *attributes); 695 696 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, VexSimdPrefix pre, 697 VexOpcode opc, InstructionAttr *attributes); 698 699 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, VexSimdPrefix pre, 700 VexOpcode opc, InstructionAttr *attributes); 701 702 // Helper functions for groups of instructions 703 void emit_arith_b(int op1, int op2, Register dst, int imm8); 704 705 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 706 // Force generation of a 4 byte immediate value even if it fits into 8bit 707 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32); 708 void emit_arith(int op1, int op2, Register dst, Register src); 709 710 bool emit_compressed_disp_byte(int &disp); 711 712 void emit_operand(Register reg, 713 Register base, Register index, Address::ScaleFactor scale, 714 int disp, 715 RelocationHolder const& rspec, 716 int rip_relative_correction = 0); 717 718 void emit_operand(XMMRegister reg, Register base, XMMRegister index, 719 Address::ScaleFactor scale, 720 int disp, RelocationHolder const& rspec); 721 722 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 723 724 // operands that only take the original 32bit registers 725 void emit_operand32(Register reg, Address adr); 726 727 void emit_operand(XMMRegister reg, 728 Register base, Register index, Address::ScaleFactor scale, 729 int disp, 730 RelocationHolder const& rspec); 731 732 void emit_operand(XMMRegister reg, Address adr); 733 734 void emit_operand(MMXRegister reg, Address adr); 735 736 // workaround gcc (3.2.1-7) bug 737 void emit_operand(Address adr, MMXRegister reg); 738 739 740 // Immediate-to-memory forms 741 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 742 743 void emit_farith(int b1, int b2, int i); 744 745 746 protected: 747 #ifdef ASSERT 748 void check_relocation(RelocationHolder const& rspec, int format); 749 #endif 750 751 void emit_data(jint data, relocInfo::relocType rtype, int format); 752 void emit_data(jint data, RelocationHolder const& rspec, int format); 753 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 754 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 755 756 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 757 758 // These are all easily abused and hence protected 759 760 // 32BIT ONLY SECTION 761 #ifndef _LP64 762 // Make these disappear in 64bit mode since they would never be correct 763 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 764 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 765 766 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 767 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 768 769 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 770 #else 771 // 64BIT ONLY SECTION 772 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 773 774 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 775 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 776 777 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 778 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 779 #endif // _LP64 780 781 // These are unique in that we are ensured by the caller that the 32bit 782 // relative in these instructions will always be able to reach the potentially 783 // 64bit address described by entry. Since they can take a 64bit address they 784 // don't have the 32 suffix like the other instructions in this class. 785 786 void call_literal(address entry, RelocationHolder const& rspec); 787 void jmp_literal(address entry, RelocationHolder const& rspec); 788 789 // Avoid using directly section 790 // Instructions in this section are actually usable by anyone without danger 791 // of failure but have performance issues that are addressed my enhanced 792 // instructions which will do the proper thing base on the particular cpu. 793 // We protect them because we don't trust you... 794 795 // Don't use next inc() and dec() methods directly. INC & DEC instructions 796 // could cause a partial flag stall since they don't set CF flag. 797 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 798 // which call inc() & dec() or add() & sub() in accordance with 799 // the product flag UseIncDec value. 800 801 void decl(Register dst); 802 void decl(Address dst); 803 void decq(Register dst); 804 void decq(Address dst); 805 806 void incl(Register dst); 807 void incl(Address dst); 808 void incq(Register dst); 809 void incq(Address dst); 810 811 // New cpus require use of movsd and movss to avoid partial register stall 812 // when loading from memory. But for old Opteron use movlpd instead of movsd. 813 // The selection is done in MacroAssembler::movdbl() and movflt(). 814 815 // Move Scalar Single-Precision Floating-Point Values 816 void movss(XMMRegister dst, Address src); 817 void movss(XMMRegister dst, XMMRegister src); 818 void movss(Address dst, XMMRegister src); 819 820 // Move Scalar Double-Precision Floating-Point Values 821 void movsd(XMMRegister dst, Address src); 822 void movsd(XMMRegister dst, XMMRegister src); 823 void movsd(Address dst, XMMRegister src); 824 void movlpd(XMMRegister dst, Address src); 825 826 // New cpus require use of movaps and movapd to avoid partial register stall 827 // when moving between registers. 828 void movaps(XMMRegister dst, XMMRegister src); 829 void movapd(XMMRegister dst, XMMRegister src); 830 831 // End avoid using directly 832 833 834 // Instruction prefixes 835 void prefix(Prefix p); 836 837 public: 838 839 // Creation 840 Assembler(CodeBuffer* code) : AbstractAssembler(code) { 841 init_attributes(); 842 } 843 844 // Decoding 845 static address locate_operand(address inst, WhichOperand which); 846 static address locate_next_instruction(address inst); 847 848 // Utilities 849 static bool is_polling_page_far() NOT_LP64({ return false;}); 850 static bool query_compressed_disp_byte(int disp, bool is_evex_inst, int vector_len, 851 int cur_tuple_type, int in_size_in_bits, int cur_encoding); 852 853 // Generic instructions 854 // Does 32bit or 64bit as needed for the platform. In some sense these 855 // belong in macro assembler but there is no need for both varieties to exist 856 857 void init_attributes(void) { 858 _legacy_mode_bw = (VM_Version::supports_avx512bw() == false); 859 _legacy_mode_dq = (VM_Version::supports_avx512dq() == false); 860 _legacy_mode_vl = (VM_Version::supports_avx512vl() == false); 861 _legacy_mode_vlbw = (VM_Version::supports_avx512vlbw() == false); 862 _is_managed = false; 863 _vector_masking = false; 864 _attributes = NULL; 865 } 866 867 void set_attributes(InstructionAttr *attributes) { _attributes = attributes; } 868 void clear_attributes(void) { _attributes = NULL; } 869 870 void set_managed(void) { _is_managed = true; } 871 void clear_managed(void) { _is_managed = false; } 872 bool is_managed(void) { return _is_managed; } 873 874 void lea(Register dst, Address src); 875 876 void mov(Register dst, Register src); 877 878 void pusha(); 879 void popa(); 880 881 void pushf(); 882 void popf(); 883 884 void push(int32_t imm32); 885 886 void push(Register src); 887 888 void pop(Register dst); 889 890 // These are dummies to prevent surprise implicit conversions to Register 891 void push(void* v); 892 void pop(void* v); 893 894 // These do register sized moves/scans 895 void rep_mov(); 896 void rep_stos(); 897 void rep_stosb(); 898 void repne_scan(); 899 #ifdef _LP64 900 void repne_scanl(); 901 #endif 902 903 // Vanilla instructions in lexical order 904 905 void adcl(Address dst, int32_t imm32); 906 void adcl(Address dst, Register src); 907 void adcl(Register dst, int32_t imm32); 908 void adcl(Register dst, Address src); 909 void adcl(Register dst, Register src); 910 911 void adcq(Register dst, int32_t imm32); 912 void adcq(Register dst, Address src); 913 void adcq(Register dst, Register src); 914 915 void addb(Address dst, int imm8); 916 void addw(Address dst, int imm16); 917 918 void addl(Address dst, int32_t imm32); 919 void addl(Address dst, Register src); 920 void addl(Register dst, int32_t imm32); 921 void addl(Register dst, Address src); 922 void addl(Register dst, Register src); 923 924 void addq(Address dst, int32_t imm32); 925 void addq(Address dst, Register src); 926 void addq(Register dst, int32_t imm32); 927 void addq(Register dst, Address src); 928 void addq(Register dst, Register src); 929 930 #ifdef _LP64 931 //Add Unsigned Integers with Carry Flag 932 void adcxq(Register dst, Register src); 933 934 //Add Unsigned Integers with Overflow Flag 935 void adoxq(Register dst, Register src); 936 #endif 937 938 void addr_nop_4(); 939 void addr_nop_5(); 940 void addr_nop_7(); 941 void addr_nop_8(); 942 943 // Add Scalar Double-Precision Floating-Point Values 944 void addsd(XMMRegister dst, Address src); 945 void addsd(XMMRegister dst, XMMRegister src); 946 947 // Add Scalar Single-Precision Floating-Point Values 948 void addss(XMMRegister dst, Address src); 949 void addss(XMMRegister dst, XMMRegister src); 950 951 // AES instructions 952 void aesdec(XMMRegister dst, Address src); 953 void aesdec(XMMRegister dst, XMMRegister src); 954 void aesdeclast(XMMRegister dst, Address src); 955 void aesdeclast(XMMRegister dst, XMMRegister src); 956 void aesenc(XMMRegister dst, Address src); 957 void aesenc(XMMRegister dst, XMMRegister src); 958 void aesenclast(XMMRegister dst, Address src); 959 void aesenclast(XMMRegister dst, XMMRegister src); 960 void vaesdec(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 961 void vaesdeclast(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 962 963 void andl(Address dst, int32_t imm32); 964 void andl(Register dst, int32_t imm32); 965 void andl(Register dst, Address src); 966 void andl(Register dst, Register src); 967 968 void andq(Address dst, int32_t imm32); 969 void andq(Register dst, int32_t imm32); 970 void andq(Register dst, Address src); 971 void andq(Register dst, Register src); 972 973 // BMI instructions 974 void andnl(Register dst, Register src1, Register src2); 975 void andnl(Register dst, Register src1, Address src2); 976 void andnq(Register dst, Register src1, Register src2); 977 void andnq(Register dst, Register src1, Address src2); 978 979 void blsil(Register dst, Register src); 980 void blsil(Register dst, Address src); 981 void blsiq(Register dst, Register src); 982 void blsiq(Register dst, Address src); 983 984 void blsmskl(Register dst, Register src); 985 void blsmskl(Register dst, Address src); 986 void blsmskq(Register dst, Register src); 987 void blsmskq(Register dst, Address src); 988 989 void blsrl(Register dst, Register src); 990 void blsrl(Register dst, Address src); 991 void blsrq(Register dst, Register src); 992 void blsrq(Register dst, Address src); 993 994 void bsfl(Register dst, Register src); 995 void bsrl(Register dst, Register src); 996 997 #ifdef _LP64 998 void bsfq(Register dst, Register src); 999 void bsrq(Register dst, Register src); 1000 #endif 1001 1002 void bswapl(Register reg); 1003 1004 void bswapq(Register reg); 1005 1006 void call(Label& L, relocInfo::relocType rtype); 1007 void call(Register reg); // push pc; pc <- reg 1008 void call(Address adr); // push pc; pc <- adr 1009 1010 void cdql(); 1011 1012 void cdqq(); 1013 1014 void cld(); 1015 1016 void clflush(Address adr); 1017 1018 void cmovl(Condition cc, Register dst, Register src); 1019 void cmovl(Condition cc, Register dst, Address src); 1020 1021 void cmovq(Condition cc, Register dst, Register src); 1022 void cmovq(Condition cc, Register dst, Address src); 1023 1024 1025 void cmpb(Address dst, int imm8); 1026 1027 void cmpl(Address dst, int32_t imm32); 1028 1029 void cmpl(Register dst, int32_t imm32); 1030 void cmpl(Register dst, Register src); 1031 void cmpl(Register dst, Address src); 1032 1033 void cmpq(Address dst, int32_t imm32); 1034 void cmpq(Address dst, Register src); 1035 1036 void cmpq(Register dst, int32_t imm32); 1037 void cmpq(Register dst, Register src); 1038 void cmpq(Register dst, Address src); 1039 1040 // these are dummies used to catch attempting to convert NULL to Register 1041 void cmpl(Register dst, void* junk); // dummy 1042 void cmpq(Register dst, void* junk); // dummy 1043 1044 void cmpw(Address dst, int imm16); 1045 1046 void cmpxchg8 (Address adr); 1047 1048 void cmpxchgb(Register reg, Address adr); 1049 void cmpxchgl(Register reg, Address adr); 1050 1051 void cmpxchgq(Register reg, Address adr); 1052 1053 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1054 void comisd(XMMRegister dst, Address src); 1055 void comisd(XMMRegister dst, XMMRegister src); 1056 1057 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1058 void comiss(XMMRegister dst, Address src); 1059 void comiss(XMMRegister dst, XMMRegister src); 1060 1061 // Identify processor type and features 1062 void cpuid(); 1063 1064 // CRC32C 1065 void crc32(Register crc, Register v, int8_t sizeInBytes); 1066 void crc32(Register crc, Address adr, int8_t sizeInBytes); 1067 1068 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 1069 void cvtsd2ss(XMMRegister dst, XMMRegister src); 1070 void cvtsd2ss(XMMRegister dst, Address src); 1071 1072 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 1073 void cvtsi2sdl(XMMRegister dst, Register src); 1074 void cvtsi2sdl(XMMRegister dst, Address src); 1075 void cvtsi2sdq(XMMRegister dst, Register src); 1076 void cvtsi2sdq(XMMRegister dst, Address src); 1077 1078 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 1079 void cvtsi2ssl(XMMRegister dst, Register src); 1080 void cvtsi2ssl(XMMRegister dst, Address src); 1081 void cvtsi2ssq(XMMRegister dst, Register src); 1082 void cvtsi2ssq(XMMRegister dst, Address src); 1083 1084 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 1085 void cvtdq2pd(XMMRegister dst, XMMRegister src); 1086 1087 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 1088 void cvtdq2ps(XMMRegister dst, XMMRegister src); 1089 1090 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 1091 void cvtss2sd(XMMRegister dst, XMMRegister src); 1092 void cvtss2sd(XMMRegister dst, Address src); 1093 1094 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 1095 void cvttsd2sil(Register dst, Address src); 1096 void cvttsd2sil(Register dst, XMMRegister src); 1097 void cvttsd2siq(Register dst, XMMRegister src); 1098 1099 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 1100 void cvttss2sil(Register dst, XMMRegister src); 1101 void cvttss2siq(Register dst, XMMRegister src); 1102 1103 void cvttpd2dq(XMMRegister dst, XMMRegister src); 1104 1105 // Divide Scalar Double-Precision Floating-Point Values 1106 void divsd(XMMRegister dst, Address src); 1107 void divsd(XMMRegister dst, XMMRegister src); 1108 1109 // Divide Scalar Single-Precision Floating-Point Values 1110 void divss(XMMRegister dst, Address src); 1111 void divss(XMMRegister dst, XMMRegister src); 1112 1113 void emms(); 1114 1115 void fabs(); 1116 1117 void fadd(int i); 1118 1119 void fadd_d(Address src); 1120 void fadd_s(Address src); 1121 1122 // "Alternate" versions of x87 instructions place result down in FPU 1123 // stack instead of on TOS 1124 1125 void fadda(int i); // "alternate" fadd 1126 void faddp(int i = 1); 1127 1128 void fchs(); 1129 1130 void fcom(int i); 1131 1132 void fcomp(int i = 1); 1133 void fcomp_d(Address src); 1134 void fcomp_s(Address src); 1135 1136 void fcompp(); 1137 1138 void fcos(); 1139 1140 void fdecstp(); 1141 1142 void fdiv(int i); 1143 void fdiv_d(Address src); 1144 void fdivr_s(Address src); 1145 void fdiva(int i); // "alternate" fdiv 1146 void fdivp(int i = 1); 1147 1148 void fdivr(int i); 1149 void fdivr_d(Address src); 1150 void fdiv_s(Address src); 1151 1152 void fdivra(int i); // "alternate" reversed fdiv 1153 1154 void fdivrp(int i = 1); 1155 1156 void ffree(int i = 0); 1157 1158 void fild_d(Address adr); 1159 void fild_s(Address adr); 1160 1161 void fincstp(); 1162 1163 void finit(); 1164 1165 void fist_s (Address adr); 1166 void fistp_d(Address adr); 1167 void fistp_s(Address adr); 1168 1169 void fld1(); 1170 1171 void fld_d(Address adr); 1172 void fld_s(Address adr); 1173 void fld_s(int index); 1174 void fld_x(Address adr); // extended-precision (80-bit) format 1175 1176 void fldcw(Address src); 1177 1178 void fldenv(Address src); 1179 1180 void fldlg2(); 1181 1182 void fldln2(); 1183 1184 void fldz(); 1185 1186 void flog(); 1187 void flog10(); 1188 1189 void fmul(int i); 1190 1191 void fmul_d(Address src); 1192 void fmul_s(Address src); 1193 1194 void fmula(int i); // "alternate" fmul 1195 1196 void fmulp(int i = 1); 1197 1198 void fnsave(Address dst); 1199 1200 void fnstcw(Address src); 1201 1202 void fnstsw_ax(); 1203 1204 void fprem(); 1205 void fprem1(); 1206 1207 void frstor(Address src); 1208 1209 void fsin(); 1210 1211 void fsqrt(); 1212 1213 void fst_d(Address adr); 1214 void fst_s(Address adr); 1215 1216 void fstp_d(Address adr); 1217 void fstp_d(int index); 1218 void fstp_s(Address adr); 1219 void fstp_x(Address adr); // extended-precision (80-bit) format 1220 1221 void fsub(int i); 1222 void fsub_d(Address src); 1223 void fsub_s(Address src); 1224 1225 void fsuba(int i); // "alternate" fsub 1226 1227 void fsubp(int i = 1); 1228 1229 void fsubr(int i); 1230 void fsubr_d(Address src); 1231 void fsubr_s(Address src); 1232 1233 void fsubra(int i); // "alternate" reversed fsub 1234 1235 void fsubrp(int i = 1); 1236 1237 void ftan(); 1238 1239 void ftst(); 1240 1241 void fucomi(int i = 1); 1242 void fucomip(int i = 1); 1243 1244 void fwait(); 1245 1246 void fxch(int i = 1); 1247 1248 void fxrstor(Address src); 1249 void xrstor(Address src); 1250 1251 void fxsave(Address dst); 1252 void xsave(Address dst); 1253 1254 void fyl2x(); 1255 void frndint(); 1256 void f2xm1(); 1257 void fldl2e(); 1258 1259 void hlt(); 1260 1261 void idivl(Register src); 1262 void divl(Register src); // Unsigned division 1263 1264 #ifdef _LP64 1265 void idivq(Register src); 1266 #endif 1267 1268 void imull(Register src); 1269 void imull(Register dst, Register src); 1270 void imull(Register dst, Register src, int value); 1271 void imull(Register dst, Address src); 1272 1273 #ifdef _LP64 1274 void imulq(Register dst, Register src); 1275 void imulq(Register dst, Register src, int value); 1276 void imulq(Register dst, Address src); 1277 #endif 1278 1279 // jcc is the generic conditional branch generator to run- 1280 // time routines, jcc is used for branches to labels. jcc 1281 // takes a branch opcode (cc) and a label (L) and generates 1282 // either a backward branch or a forward branch and links it 1283 // to the label fixup chain. Usage: 1284 // 1285 // Label L; // unbound label 1286 // jcc(cc, L); // forward branch to unbound label 1287 // bind(L); // bind label to the current pc 1288 // jcc(cc, L); // backward branch to bound label 1289 // bind(L); // illegal: a label may be bound only once 1290 // 1291 // Note: The same Label can be used for forward and backward branches 1292 // but it may be bound only once. 1293 1294 void jcc(Condition cc, Label& L, bool maybe_short = true); 1295 1296 // Conditional jump to a 8-bit offset to L. 1297 // WARNING: be very careful using this for forward jumps. If the label is 1298 // not bound within an 8-bit offset of this instruction, a run-time error 1299 // will occur. 1300 1301 // Use macro to record file and line number. 1302 #define jccb(cc, L) jccb_0(cc, L, __FILE__, __LINE__) 1303 1304 void jccb_0(Condition cc, Label& L, const char* file, int line); 1305 1306 void jmp(Address entry); // pc <- entry 1307 1308 // Label operations & relative jumps (PPUM Appendix D) 1309 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L 1310 1311 void jmp(Register entry); // pc <- entry 1312 1313 // Unconditional 8-bit offset jump to L. 1314 // WARNING: be very careful using this for forward jumps. If the label is 1315 // not bound within an 8-bit offset of this instruction, a run-time error 1316 // will occur. 1317 1318 // Use macro to record file and line number. 1319 #define jmpb(L) jmpb_0(L, __FILE__, __LINE__) 1320 1321 void jmpb_0(Label& L, const char* file, int line); 1322 1323 void ldmxcsr( Address src ); 1324 1325 void leal(Register dst, Address src); 1326 1327 void leaq(Register dst, Address src); 1328 1329 void lfence(); 1330 1331 void lock(); 1332 1333 void lzcntl(Register dst, Register src); 1334 1335 #ifdef _LP64 1336 void lzcntq(Register dst, Register src); 1337 #endif 1338 1339 enum Membar_mask_bits { 1340 StoreStore = 1 << 3, 1341 LoadStore = 1 << 2, 1342 StoreLoad = 1 << 1, 1343 LoadLoad = 1 << 0 1344 }; 1345 1346 // Serializes memory and blows flags 1347 void membar(Membar_mask_bits order_constraint) { 1348 // We only have to handle StoreLoad 1349 if (order_constraint & StoreLoad) { 1350 // All usable chips support "locked" instructions which suffice 1351 // as barriers, and are much faster than the alternative of 1352 // using cpuid instruction. We use here a locked add [esp-C],0. 1353 // This is conveniently otherwise a no-op except for blowing 1354 // flags, and introducing a false dependency on target memory 1355 // location. We can't do anything with flags, but we can avoid 1356 // memory dependencies in the current method by locked-adding 1357 // somewhere else on the stack. Doing [esp+C] will collide with 1358 // something on stack in current method, hence we go for [esp-C]. 1359 // It is convenient since it is almost always in data cache, for 1360 // any small C. We need to step back from SP to avoid data 1361 // dependencies with other things on below SP (callee-saves, for 1362 // example). Without a clear way to figure out the minimal safe 1363 // distance from SP, it makes sense to step back the complete 1364 // cache line, as this will also avoid possible second-order effects 1365 // with locked ops against the cache line. Our choice of offset 1366 // is bounded by x86 operand encoding, which should stay within 1367 // [-128; +127] to have the 8-byte displacement encoding. 1368 // 1369 // Any change to this code may need to revisit other places in 1370 // the code where this idiom is used, in particular the 1371 // orderAccess code. 1372 1373 int offset = -VM_Version::L1_line_size(); 1374 if (offset < -128) { 1375 offset = -128; 1376 } 1377 1378 lock(); 1379 addl(Address(rsp, offset), 0);// Assert the lock# signal here 1380 } 1381 } 1382 1383 void mfence(); 1384 1385 // Moves 1386 1387 void mov64(Register dst, int64_t imm64); 1388 1389 void movb(Address dst, Register src); 1390 void movb(Address dst, int imm8); 1391 void movb(Register dst, Address src); 1392 1393 void movddup(XMMRegister dst, XMMRegister src); 1394 1395 void kmovbl(KRegister dst, Register src); 1396 void kmovbl(Register dst, KRegister src); 1397 void kmovwl(KRegister dst, Register src); 1398 void kmovwl(KRegister dst, Address src); 1399 void kmovwl(Register dst, KRegister src); 1400 void kmovdl(KRegister dst, Register src); 1401 void kmovdl(Register dst, KRegister src); 1402 void kmovql(KRegister dst, KRegister src); 1403 void kmovql(Address dst, KRegister src); 1404 void kmovql(KRegister dst, Address src); 1405 void kmovql(KRegister dst, Register src); 1406 void kmovql(Register dst, KRegister src); 1407 1408 void knotwl(KRegister dst, KRegister src); 1409 1410 void kortestbl(KRegister dst, KRegister src); 1411 void kortestwl(KRegister dst, KRegister src); 1412 void kortestdl(KRegister dst, KRegister src); 1413 void kortestql(KRegister dst, KRegister src); 1414 1415 void ktestq(KRegister src1, KRegister src2); 1416 void ktestd(KRegister src1, KRegister src2); 1417 1418 void ktestql(KRegister dst, KRegister src); 1419 1420 void movdl(XMMRegister dst, Register src); 1421 void movdl(Register dst, XMMRegister src); 1422 void movdl(XMMRegister dst, Address src); 1423 void movdl(Address dst, XMMRegister src); 1424 1425 // Move Double Quadword 1426 void movdq(XMMRegister dst, Register src); 1427 void movdq(Register dst, XMMRegister src); 1428 1429 // Move Aligned Double Quadword 1430 void movdqa(XMMRegister dst, XMMRegister src); 1431 void movdqa(XMMRegister dst, Address src); 1432 1433 // Move Unaligned Double Quadword 1434 void movdqu(Address dst, XMMRegister src); 1435 void movdqu(XMMRegister dst, Address src); 1436 void movdqu(XMMRegister dst, XMMRegister src); 1437 1438 // Move Unaligned 256bit Vector 1439 void vmovdqu(Address dst, XMMRegister src); 1440 void vmovdqu(XMMRegister dst, Address src); 1441 void vmovdqu(XMMRegister dst, XMMRegister src); 1442 1443 // Move Unaligned 512bit Vector 1444 void evmovdqub(Address dst, XMMRegister src, int vector_len); 1445 void evmovdqub(XMMRegister dst, Address src, int vector_len); 1446 void evmovdqub(XMMRegister dst, XMMRegister src, int vector_len); 1447 void evmovdqub(XMMRegister dst, KRegister mask, Address src, int vector_len); 1448 void evmovdquw(Address dst, XMMRegister src, int vector_len); 1449 void evmovdquw(Address dst, KRegister mask, XMMRegister src, int vector_len); 1450 void evmovdquw(XMMRegister dst, Address src, int vector_len); 1451 void evmovdquw(XMMRegister dst, KRegister mask, Address src, int vector_len); 1452 void evmovdqul(Address dst, XMMRegister src, int vector_len); 1453 void evmovdqul(XMMRegister dst, Address src, int vector_len); 1454 void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len); 1455 void evmovdquq(Address dst, XMMRegister src, int vector_len); 1456 void evmovdquq(XMMRegister dst, Address src, int vector_len); 1457 void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len); 1458 1459 // Move lower 64bit to high 64bit in 128bit register 1460 void movlhps(XMMRegister dst, XMMRegister src); 1461 1462 void movl(Register dst, int32_t imm32); 1463 void movl(Address dst, int32_t imm32); 1464 void movl(Register dst, Register src); 1465 void movl(Register dst, Address src); 1466 void movl(Address dst, Register src); 1467 1468 // These dummies prevent using movl from converting a zero (like NULL) into Register 1469 // by giving the compiler two choices it can't resolve 1470 1471 void movl(Address dst, void* junk); 1472 void movl(Register dst, void* junk); 1473 1474 #ifdef _LP64 1475 void movq(Register dst, Register src); 1476 void movq(Register dst, Address src); 1477 void movq(Address dst, Register src); 1478 #endif 1479 1480 void movq(Address dst, MMXRegister src ); 1481 void movq(MMXRegister dst, Address src ); 1482 1483 #ifdef _LP64 1484 // These dummies prevent using movq from converting a zero (like NULL) into Register 1485 // by giving the compiler two choices it can't resolve 1486 1487 void movq(Address dst, void* dummy); 1488 void movq(Register dst, void* dummy); 1489 #endif 1490 1491 // Move Quadword 1492 void movq(Address dst, XMMRegister src); 1493 void movq(XMMRegister dst, Address src); 1494 1495 void movsbl(Register dst, Address src); 1496 void movsbl(Register dst, Register src); 1497 1498 #ifdef _LP64 1499 void movsbq(Register dst, Address src); 1500 void movsbq(Register dst, Register src); 1501 1502 // Move signed 32bit immediate to 64bit extending sign 1503 void movslq(Address dst, int32_t imm64); 1504 void movslq(Register dst, int32_t imm64); 1505 1506 void movslq(Register dst, Address src); 1507 void movslq(Register dst, Register src); 1508 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1509 #endif 1510 1511 void movswl(Register dst, Address src); 1512 void movswl(Register dst, Register src); 1513 1514 #ifdef _LP64 1515 void movswq(Register dst, Address src); 1516 void movswq(Register dst, Register src); 1517 #endif 1518 1519 void movw(Address dst, int imm16); 1520 void movw(Register dst, Address src); 1521 void movw(Address dst, Register src); 1522 1523 void movzbl(Register dst, Address src); 1524 void movzbl(Register dst, Register src); 1525 1526 #ifdef _LP64 1527 void movzbq(Register dst, Address src); 1528 void movzbq(Register dst, Register src); 1529 #endif 1530 1531 void movzwl(Register dst, Address src); 1532 void movzwl(Register dst, Register src); 1533 1534 #ifdef _LP64 1535 void movzwq(Register dst, Address src); 1536 void movzwq(Register dst, Register src); 1537 #endif 1538 1539 // Unsigned multiply with RAX destination register 1540 void mull(Address src); 1541 void mull(Register src); 1542 1543 #ifdef _LP64 1544 void mulq(Address src); 1545 void mulq(Register src); 1546 void mulxq(Register dst1, Register dst2, Register src); 1547 #endif 1548 1549 // Multiply Scalar Double-Precision Floating-Point Values 1550 void mulsd(XMMRegister dst, Address src); 1551 void mulsd(XMMRegister dst, XMMRegister src); 1552 1553 // Multiply Scalar Single-Precision Floating-Point Values 1554 void mulss(XMMRegister dst, Address src); 1555 void mulss(XMMRegister dst, XMMRegister src); 1556 1557 void negl(Register dst); 1558 1559 #ifdef _LP64 1560 void negq(Register dst); 1561 #endif 1562 1563 void nop(int i = 1); 1564 1565 void notl(Register dst); 1566 1567 #ifdef _LP64 1568 void notq(Register dst); 1569 #endif 1570 1571 void orl(Address dst, int32_t imm32); 1572 void orl(Register dst, int32_t imm32); 1573 void orl(Register dst, Address src); 1574 void orl(Register dst, Register src); 1575 void orl(Address dst, Register src); 1576 1577 void orb(Address dst, int imm8); 1578 1579 void orq(Address dst, int32_t imm32); 1580 void orq(Register dst, int32_t imm32); 1581 void orq(Register dst, Address src); 1582 void orq(Register dst, Register src); 1583 1584 // Pack with unsigned saturation 1585 void packuswb(XMMRegister dst, XMMRegister src); 1586 void packuswb(XMMRegister dst, Address src); 1587 void vpackuswb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1588 1589 // Pemutation of 64bit words 1590 void vpermq(XMMRegister dst, XMMRegister src, int imm8, int vector_len); 1591 void vpermq(XMMRegister dst, XMMRegister src, int imm8); 1592 void vperm2i128(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8); 1593 void vperm2f128(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8); 1594 void evpermi2q(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1595 1596 void pause(); 1597 1598 // Undefined Instruction 1599 void ud2(); 1600 1601 // SSE4.2 string instructions 1602 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1603 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1604 1605 void pcmpeqb(XMMRegister dst, XMMRegister src); 1606 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1607 void evpcmpeqb(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len); 1608 void evpcmpeqb(KRegister kdst, XMMRegister nds, Address src, int vector_len); 1609 void evpcmpeqb(KRegister kdst, KRegister mask, XMMRegister nds, Address src, int vector_len); 1610 1611 void evpcmpgtb(KRegister kdst, XMMRegister nds, Address src, int vector_len); 1612 void evpcmpgtb(KRegister kdst, KRegister mask, XMMRegister nds, Address src, int vector_len); 1613 1614 void evpcmpuw(KRegister kdst, XMMRegister nds, XMMRegister src, ComparisonPredicate vcc, int vector_len); 1615 void evpcmpuw(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, ComparisonPredicate of, int vector_len); 1616 void evpcmpuw(KRegister kdst, XMMRegister nds, Address src, ComparisonPredicate vcc, int vector_len); 1617 1618 void pcmpeqw(XMMRegister dst, XMMRegister src); 1619 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1620 void evpcmpeqw(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len); 1621 void evpcmpeqw(KRegister kdst, XMMRegister nds, Address src, int vector_len); 1622 1623 void pcmpeqd(XMMRegister dst, XMMRegister src); 1624 void vpcmpeqd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1625 void evpcmpeqd(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len); 1626 void evpcmpeqd(KRegister kdst, XMMRegister nds, Address src, int vector_len); 1627 1628 void pcmpeqq(XMMRegister dst, XMMRegister src); 1629 void vpcmpeqq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1630 void evpcmpeqq(KRegister kdst, XMMRegister nds, XMMRegister src, int vector_len); 1631 void evpcmpeqq(KRegister kdst, XMMRegister nds, Address src, int vector_len); 1632 1633 void pmovmskb(Register dst, XMMRegister src); 1634 void vpmovmskb(Register dst, XMMRegister src); 1635 1636 // SSE 4.1 extract 1637 void pextrd(Register dst, XMMRegister src, int imm8); 1638 void pextrq(Register dst, XMMRegister src, int imm8); 1639 void pextrd(Address dst, XMMRegister src, int imm8); 1640 void pextrq(Address dst, XMMRegister src, int imm8); 1641 void pextrb(Address dst, XMMRegister src, int imm8); 1642 // SSE 2 extract 1643 void pextrw(Register dst, XMMRegister src, int imm8); 1644 void pextrw(Address dst, XMMRegister src, int imm8); 1645 1646 // SSE 4.1 insert 1647 void pinsrd(XMMRegister dst, Register src, int imm8); 1648 void pinsrq(XMMRegister dst, Register src, int imm8); 1649 void pinsrd(XMMRegister dst, Address src, int imm8); 1650 void pinsrq(XMMRegister dst, Address src, int imm8); 1651 void pinsrb(XMMRegister dst, Address src, int imm8); 1652 // SSE 2 insert 1653 void pinsrw(XMMRegister dst, Register src, int imm8); 1654 void pinsrw(XMMRegister dst, Address src, int imm8); 1655 1656 // SSE4.1 packed move 1657 void pmovzxbw(XMMRegister dst, XMMRegister src); 1658 void pmovzxbw(XMMRegister dst, Address src); 1659 1660 void vpmovzxbw( XMMRegister dst, Address src, int vector_len); 1661 void vpmovzxbw(XMMRegister dst, XMMRegister src, int vector_len); 1662 void evpmovzxbw(XMMRegister dst, KRegister mask, Address src, int vector_len); 1663 1664 void evpmovwb(Address dst, XMMRegister src, int vector_len); 1665 void evpmovwb(Address dst, KRegister mask, XMMRegister src, int vector_len); 1666 1667 void vpmovzxwd(XMMRegister dst, XMMRegister src, int vector_len); 1668 1669 void evpmovdb(Address dst, XMMRegister src, int vector_len); 1670 1671 // Multiply add 1672 void pmaddwd(XMMRegister dst, XMMRegister src); 1673 void vpmaddwd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1674 // Multiply add accumulate 1675 void evpdpwssd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1676 1677 #ifndef _LP64 // no 32bit push/pop on amd64 1678 void popl(Address dst); 1679 #endif 1680 1681 #ifdef _LP64 1682 void popq(Address dst); 1683 #endif 1684 1685 void popcntl(Register dst, Address src); 1686 void popcntl(Register dst, Register src); 1687 1688 void vpopcntd(XMMRegister dst, XMMRegister src, int vector_len); 1689 1690 #ifdef _LP64 1691 void popcntq(Register dst, Address src); 1692 void popcntq(Register dst, Register src); 1693 #endif 1694 1695 // Prefetches (SSE, SSE2, 3DNOW only) 1696 1697 void prefetchnta(Address src); 1698 void prefetchr(Address src); 1699 void prefetcht0(Address src); 1700 void prefetcht1(Address src); 1701 void prefetcht2(Address src); 1702 void prefetchw(Address src); 1703 1704 // Shuffle Bytes 1705 void pshufb(XMMRegister dst, XMMRegister src); 1706 void pshufb(XMMRegister dst, Address src); 1707 void vpshufb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1708 1709 // Shuffle Packed Doublewords 1710 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1711 void pshufd(XMMRegister dst, Address src, int mode); 1712 void vpshufd(XMMRegister dst, XMMRegister src, int mode, int vector_len); 1713 1714 // Shuffle Packed Low Words 1715 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1716 void pshuflw(XMMRegister dst, Address src, int mode); 1717 1718 // Shuffle packed values at 128 bit granularity 1719 void evshufi64x2(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8, int vector_len); 1720 1721 // Shift Right by bytes Logical DoubleQuadword Immediate 1722 void psrldq(XMMRegister dst, int shift); 1723 // Shift Left by bytes Logical DoubleQuadword Immediate 1724 void pslldq(XMMRegister dst, int shift); 1725 1726 // Logical Compare 128bit 1727 void ptest(XMMRegister dst, XMMRegister src); 1728 void ptest(XMMRegister dst, Address src); 1729 // Logical Compare 256bit 1730 void vptest(XMMRegister dst, XMMRegister src); 1731 void vptest(XMMRegister dst, Address src); 1732 1733 // Interleave Low Bytes 1734 void punpcklbw(XMMRegister dst, XMMRegister src); 1735 void punpcklbw(XMMRegister dst, Address src); 1736 1737 // Interleave Low Doublewords 1738 void punpckldq(XMMRegister dst, XMMRegister src); 1739 void punpckldq(XMMRegister dst, Address src); 1740 1741 // Interleave Low Quadwords 1742 void punpcklqdq(XMMRegister dst, XMMRegister src); 1743 1744 #ifndef _LP64 // no 32bit push/pop on amd64 1745 void pushl(Address src); 1746 #endif 1747 1748 void pushq(Address src); 1749 1750 void rcll(Register dst, int imm8); 1751 1752 void rclq(Register dst, int imm8); 1753 1754 void rcrq(Register dst, int imm8); 1755 1756 void rcpps(XMMRegister dst, XMMRegister src); 1757 1758 void rcpss(XMMRegister dst, XMMRegister src); 1759 1760 void rdtsc(); 1761 1762 void ret(int imm16); 1763 1764 #ifdef _LP64 1765 void rorq(Register dst, int imm8); 1766 void rorxq(Register dst, Register src, int imm8); 1767 void rorxd(Register dst, Register src, int imm8); 1768 #endif 1769 1770 void sahf(); 1771 1772 void sarl(Register dst, int imm8); 1773 void sarl(Register dst); 1774 1775 void sarq(Register dst, int imm8); 1776 void sarq(Register dst); 1777 1778 void sbbl(Address dst, int32_t imm32); 1779 void sbbl(Register dst, int32_t imm32); 1780 void sbbl(Register dst, Address src); 1781 void sbbl(Register dst, Register src); 1782 1783 void sbbq(Address dst, int32_t imm32); 1784 void sbbq(Register dst, int32_t imm32); 1785 void sbbq(Register dst, Address src); 1786 void sbbq(Register dst, Register src); 1787 1788 void setb(Condition cc, Register dst); 1789 1790 void palignr(XMMRegister dst, XMMRegister src, int imm8); 1791 void vpalignr(XMMRegister dst, XMMRegister src1, XMMRegister src2, int imm8, int vector_len); 1792 void evalignq(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 1793 1794 void pblendw(XMMRegister dst, XMMRegister src, int imm8); 1795 1796 void sha1rnds4(XMMRegister dst, XMMRegister src, int imm8); 1797 void sha1nexte(XMMRegister dst, XMMRegister src); 1798 void sha1msg1(XMMRegister dst, XMMRegister src); 1799 void sha1msg2(XMMRegister dst, XMMRegister src); 1800 // xmm0 is implicit additional source to the following instruction. 1801 void sha256rnds2(XMMRegister dst, XMMRegister src); 1802 void sha256msg1(XMMRegister dst, XMMRegister src); 1803 void sha256msg2(XMMRegister dst, XMMRegister src); 1804 1805 void shldl(Register dst, Register src); 1806 void shldl(Register dst, Register src, int8_t imm8); 1807 1808 void shll(Register dst, int imm8); 1809 void shll(Register dst); 1810 1811 void shlq(Register dst, int imm8); 1812 void shlq(Register dst); 1813 1814 void shrdl(Register dst, Register src); 1815 1816 void shrl(Register dst, int imm8); 1817 void shrl(Register dst); 1818 1819 void shrq(Register dst, int imm8); 1820 void shrq(Register dst); 1821 1822 void smovl(); // QQQ generic? 1823 1824 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1825 void sqrtsd(XMMRegister dst, Address src); 1826 void sqrtsd(XMMRegister dst, XMMRegister src); 1827 1828 // Compute Square Root of Scalar Single-Precision Floating-Point Value 1829 void sqrtss(XMMRegister dst, Address src); 1830 void sqrtss(XMMRegister dst, XMMRegister src); 1831 1832 void std(); 1833 1834 void stmxcsr( Address dst ); 1835 1836 void subl(Address dst, int32_t imm32); 1837 void subl(Address dst, Register src); 1838 void subl(Register dst, int32_t imm32); 1839 void subl(Register dst, Address src); 1840 void subl(Register dst, Register src); 1841 1842 void subq(Address dst, int32_t imm32); 1843 void subq(Address dst, Register src); 1844 void subq(Register dst, int32_t imm32); 1845 void subq(Register dst, Address src); 1846 void subq(Register dst, Register src); 1847 1848 // Force generation of a 4 byte immediate value even if it fits into 8bit 1849 void subl_imm32(Register dst, int32_t imm32); 1850 void subq_imm32(Register dst, int32_t imm32); 1851 1852 // Subtract Scalar Double-Precision Floating-Point Values 1853 void subsd(XMMRegister dst, Address src); 1854 void subsd(XMMRegister dst, XMMRegister src); 1855 1856 // Subtract Scalar Single-Precision Floating-Point Values 1857 void subss(XMMRegister dst, Address src); 1858 void subss(XMMRegister dst, XMMRegister src); 1859 1860 void testb(Register dst, int imm8); 1861 void testb(Address dst, int imm8); 1862 1863 void testl(Register dst, int32_t imm32); 1864 void testl(Register dst, Register src); 1865 void testl(Register dst, Address src); 1866 1867 void testq(Register dst, int32_t imm32); 1868 void testq(Register dst, Register src); 1869 void testq(Register dst, Address src); 1870 1871 // BMI - count trailing zeros 1872 void tzcntl(Register dst, Register src); 1873 void tzcntq(Register dst, Register src); 1874 1875 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1876 void ucomisd(XMMRegister dst, Address src); 1877 void ucomisd(XMMRegister dst, XMMRegister src); 1878 1879 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1880 void ucomiss(XMMRegister dst, Address src); 1881 void ucomiss(XMMRegister dst, XMMRegister src); 1882 1883 void xabort(int8_t imm8); 1884 1885 void xaddb(Address dst, Register src); 1886 void xaddw(Address dst, Register src); 1887 void xaddl(Address dst, Register src); 1888 void xaddq(Address dst, Register src); 1889 1890 void xbegin(Label& abort, relocInfo::relocType rtype = relocInfo::none); 1891 1892 void xchgb(Register reg, Address adr); 1893 void xchgw(Register reg, Address adr); 1894 void xchgl(Register reg, Address adr); 1895 void xchgl(Register dst, Register src); 1896 1897 void xchgq(Register reg, Address adr); 1898 void xchgq(Register dst, Register src); 1899 1900 void xend(); 1901 1902 // Get Value of Extended Control Register 1903 void xgetbv(); 1904 1905 void xorl(Register dst, int32_t imm32); 1906 void xorl(Register dst, Address src); 1907 void xorl(Register dst, Register src); 1908 1909 void xorb(Register dst, Address src); 1910 1911 void xorq(Register dst, Address src); 1912 void xorq(Register dst, Register src); 1913 1914 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1915 1916 // AVX 3-operands scalar instructions (encoded with VEX prefix) 1917 1918 void vaddsd(XMMRegister dst, XMMRegister nds, Address src); 1919 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1920 void vaddss(XMMRegister dst, XMMRegister nds, Address src); 1921 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1922 void vdivsd(XMMRegister dst, XMMRegister nds, Address src); 1923 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1924 void vdivss(XMMRegister dst, XMMRegister nds, Address src); 1925 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1926 void vfmadd231sd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1927 void vfmadd231ss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1928 void vmulsd(XMMRegister dst, XMMRegister nds, Address src); 1929 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1930 void vmulss(XMMRegister dst, XMMRegister nds, Address src); 1931 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1932 void vsubsd(XMMRegister dst, XMMRegister nds, Address src); 1933 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1934 void vsubss(XMMRegister dst, XMMRegister nds, Address src); 1935 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1936 1937 void vmaxss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1938 void vmaxsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1939 void vminss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1940 void vminsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1941 1942 void shlxl(Register dst, Register src1, Register src2); 1943 void shlxq(Register dst, Register src1, Register src2); 1944 1945 //====================VECTOR ARITHMETIC===================================== 1946 1947 // Add Packed Floating-Point Values 1948 void addpd(XMMRegister dst, XMMRegister src); 1949 void addpd(XMMRegister dst, Address src); 1950 void addps(XMMRegister dst, XMMRegister src); 1951 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1952 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1953 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1954 void vaddps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1955 1956 // Subtract Packed Floating-Point Values 1957 void subpd(XMMRegister dst, XMMRegister src); 1958 void subps(XMMRegister dst, XMMRegister src); 1959 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1960 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1961 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1962 void vsubps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1963 1964 // Multiply Packed Floating-Point Values 1965 void mulpd(XMMRegister dst, XMMRegister src); 1966 void mulpd(XMMRegister dst, Address src); 1967 void mulps(XMMRegister dst, XMMRegister src); 1968 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1969 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1970 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1971 void vmulps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1972 1973 void vfmadd231pd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1974 void vfmadd231ps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1975 void vfmadd231pd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1976 void vfmadd231ps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1977 1978 // Divide Packed Floating-Point Values 1979 void divpd(XMMRegister dst, XMMRegister src); 1980 void divps(XMMRegister dst, XMMRegister src); 1981 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1982 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1983 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1984 void vdivps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1985 1986 // Sqrt Packed Floating-Point Values 1987 void vsqrtpd(XMMRegister dst, XMMRegister src, int vector_len); 1988 void vsqrtpd(XMMRegister dst, Address src, int vector_len); 1989 void vsqrtps(XMMRegister dst, XMMRegister src, int vector_len); 1990 void vsqrtps(XMMRegister dst, Address src, int vector_len); 1991 1992 // Bitwise Logical AND of Packed Floating-Point Values 1993 void andpd(XMMRegister dst, XMMRegister src); 1994 void andps(XMMRegister dst, XMMRegister src); 1995 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1996 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1997 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1998 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1999 2000 void unpckhpd(XMMRegister dst, XMMRegister src); 2001 void unpcklpd(XMMRegister dst, XMMRegister src); 2002 2003 // Bitwise Logical XOR of Packed Floating-Point Values 2004 void xorpd(XMMRegister dst, XMMRegister src); 2005 void xorps(XMMRegister dst, XMMRegister src); 2006 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2007 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2008 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2009 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2010 2011 // Add horizontal packed integers 2012 void vphaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2013 void vphaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2014 void phaddw(XMMRegister dst, XMMRegister src); 2015 void phaddd(XMMRegister dst, XMMRegister src); 2016 2017 // Add packed integers 2018 void paddb(XMMRegister dst, XMMRegister src); 2019 void paddw(XMMRegister dst, XMMRegister src); 2020 void paddd(XMMRegister dst, XMMRegister src); 2021 void paddd(XMMRegister dst, Address src); 2022 void paddq(XMMRegister dst, XMMRegister src); 2023 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2024 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2025 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2026 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2027 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2028 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2029 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2030 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2031 2032 // Sub packed integers 2033 void psubb(XMMRegister dst, XMMRegister src); 2034 void psubw(XMMRegister dst, XMMRegister src); 2035 void psubd(XMMRegister dst, XMMRegister src); 2036 void psubq(XMMRegister dst, XMMRegister src); 2037 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2038 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2039 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2040 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2041 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2042 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2043 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2044 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2045 2046 // Multiply packed integers (only shorts and ints) 2047 void pmullw(XMMRegister dst, XMMRegister src); 2048 void pmulld(XMMRegister dst, XMMRegister src); 2049 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2050 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2051 void vpmullq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2052 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2053 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2054 void vpmullq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2055 2056 // Shift left packed integers 2057 void psllw(XMMRegister dst, int shift); 2058 void pslld(XMMRegister dst, int shift); 2059 void psllq(XMMRegister dst, int shift); 2060 void psllw(XMMRegister dst, XMMRegister shift); 2061 void pslld(XMMRegister dst, XMMRegister shift); 2062 void psllq(XMMRegister dst, XMMRegister shift); 2063 void vpsllw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2064 void vpslld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2065 void vpsllq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2066 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2067 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2068 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2069 void vpslldq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2070 2071 // Logical shift right packed integers 2072 void psrlw(XMMRegister dst, int shift); 2073 void psrld(XMMRegister dst, int shift); 2074 void psrlq(XMMRegister dst, int shift); 2075 void psrlw(XMMRegister dst, XMMRegister shift); 2076 void psrld(XMMRegister dst, XMMRegister shift); 2077 void psrlq(XMMRegister dst, XMMRegister shift); 2078 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2079 void vpsrld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2080 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2081 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2082 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2083 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2084 void vpsrldq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2085 void evpsrlvw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2086 void evpsllvw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2087 2088 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs) 2089 void psraw(XMMRegister dst, int shift); 2090 void psrad(XMMRegister dst, int shift); 2091 void psraw(XMMRegister dst, XMMRegister shift); 2092 void psrad(XMMRegister dst, XMMRegister shift); 2093 void vpsraw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2094 void vpsrad(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2095 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2096 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2097 2098 // And packed integers 2099 void pand(XMMRegister dst, XMMRegister src); 2100 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2101 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2102 void vpandq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2103 2104 // Andn packed integers 2105 void pandn(XMMRegister dst, XMMRegister src); 2106 void vpandn(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2107 2108 // Or packed integers 2109 void por(XMMRegister dst, XMMRegister src); 2110 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2111 void vpor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2112 void vporq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2113 2114 // Xor packed integers 2115 void pxor(XMMRegister dst, XMMRegister src); 2116 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2117 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2118 void evpxorq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2119 void evpxorq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2120 2121 2122 // vinserti forms 2123 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2124 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8); 2125 void vinserti32x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2126 void vinserti32x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8); 2127 void vinserti64x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2128 2129 // vinsertf forms 2130 void vinsertf128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2131 void vinsertf128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8); 2132 void vinsertf32x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2133 void vinsertf32x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8); 2134 void vinsertf64x4(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8); 2135 void vinsertf64x4(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8); 2136 2137 // vextracti forms 2138 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8); 2139 void vextracti128(Address dst, XMMRegister src, uint8_t imm8); 2140 void vextracti32x4(XMMRegister dst, XMMRegister src, uint8_t imm8); 2141 void vextracti32x4(Address dst, XMMRegister src, uint8_t imm8); 2142 void vextracti64x2(XMMRegister dst, XMMRegister src, uint8_t imm8); 2143 void vextracti64x4(XMMRegister dst, XMMRegister src, uint8_t imm8); 2144 void vextracti64x4(Address dst, XMMRegister src, uint8_t imm8); 2145 2146 // vextractf forms 2147 void vextractf128(XMMRegister dst, XMMRegister src, uint8_t imm8); 2148 void vextractf128(Address dst, XMMRegister src, uint8_t imm8); 2149 void vextractf32x4(XMMRegister dst, XMMRegister src, uint8_t imm8); 2150 void vextractf32x4(Address dst, XMMRegister src, uint8_t imm8); 2151 void vextractf64x2(XMMRegister dst, XMMRegister src, uint8_t imm8); 2152 void vextractf64x4(XMMRegister dst, XMMRegister src, uint8_t imm8); 2153 void vextractf64x4(Address dst, XMMRegister src, uint8_t imm8); 2154 2155 // xmm/mem sourced byte/word/dword/qword replicate 2156 void vpbroadcastb(XMMRegister dst, XMMRegister src, int vector_len); 2157 void vpbroadcastb(XMMRegister dst, Address src, int vector_len); 2158 void vpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len); 2159 void vpbroadcastw(XMMRegister dst, Address src, int vector_len); 2160 void vpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len); 2161 void vpbroadcastd(XMMRegister dst, Address src, int vector_len); 2162 void vpbroadcastq(XMMRegister dst, XMMRegister src, int vector_len); 2163 void vpbroadcastq(XMMRegister dst, Address src, int vector_len); 2164 2165 void evbroadcasti64x2(XMMRegister dst, XMMRegister src, int vector_len); 2166 void evbroadcasti64x2(XMMRegister dst, Address src, int vector_len); 2167 2168 // scalar single/double precision replicate 2169 void vpbroadcastss(XMMRegister dst, XMMRegister src, int vector_len); 2170 void vpbroadcastss(XMMRegister dst, Address src, int vector_len); 2171 void vpbroadcastsd(XMMRegister dst, XMMRegister src, int vector_len); 2172 void vpbroadcastsd(XMMRegister dst, Address src, int vector_len); 2173 2174 // gpr sourced byte/word/dword/qword replicate 2175 void evpbroadcastb(XMMRegister dst, Register src, int vector_len); 2176 void evpbroadcastw(XMMRegister dst, Register src, int vector_len); 2177 void evpbroadcastd(XMMRegister dst, Register src, int vector_len); 2178 void evpbroadcastq(XMMRegister dst, Register src, int vector_len); 2179 2180 void evpgatherdd(XMMRegister dst, KRegister k1, Address src, int vector_len); 2181 2182 // Carry-Less Multiplication Quadword 2183 void pclmulqdq(XMMRegister dst, XMMRegister src, int mask); 2184 void vpclmulqdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int mask); 2185 void evpclmulqdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int mask, int vector_len); 2186 // AVX instruction which is used to clear upper 128 bits of YMM registers and 2187 // to avoid transaction penalty between AVX and SSE states. There is no 2188 // penalty if legacy SSE instructions are encoded using VEX prefix because 2189 // they always clear upper 128 bits. It should be used before calling 2190 // runtime code and native libraries. 2191 void vzeroupper(); 2192 2193 // AVX support for vectorized conditional move (float/double). The following two instructions used only coupled. 2194 void cmppd(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop, int vector_len); 2195 void blendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMMRegister src2, int vector_len); 2196 void cmpps(XMMRegister dst, XMMRegister nds, XMMRegister src, int cop, int vector_len); 2197 void blendvps(XMMRegister dst, XMMRegister nds, XMMRegister src1, XMMRegister src2, int vector_len); 2198 void vpblendd(XMMRegister dst, XMMRegister nds, XMMRegister src, int imm8, int vector_len); 2199 2200 protected: 2201 // Next instructions require address alignment 16 bytes SSE mode. 2202 // They should be called only from corresponding MacroAssembler instructions. 2203 void andpd(XMMRegister dst, Address src); 2204 void andps(XMMRegister dst, Address src); 2205 void xorpd(XMMRegister dst, Address src); 2206 void xorps(XMMRegister dst, Address src); 2207 2208 }; 2209 2210 // The Intel x86/Amd64 Assembler attributes: All fields enclosed here are to guide encoding level decisions. 2211 // Specific set functions are for specialized use, else defaults or whatever was supplied to object construction 2212 // are applied. 2213 class InstructionAttr { 2214 public: 2215 InstructionAttr( 2216 int vector_len, // The length of vector to be applied in encoding - for both AVX and EVEX 2217 bool rex_vex_w, // Width of data: if 32-bits or less, false, else if 64-bit or specially defined, true 2218 bool legacy_mode, // Details if either this instruction is conditionally encoded to AVX or earlier if true else possibly EVEX 2219 bool no_reg_mask, // when true, k0 is used when EVEX encoding is chosen, else embedded_opmask_register_specifier is used 2220 bool uses_vl) // This instruction may have legacy constraints based on vector length for EVEX 2221 : 2222 _avx_vector_len(vector_len), 2223 _rex_vex_w(rex_vex_w), 2224 _rex_vex_w_reverted(false), 2225 _legacy_mode(legacy_mode), 2226 _no_reg_mask(no_reg_mask), 2227 _uses_vl(uses_vl), 2228 _tuple_type(Assembler::EVEX_ETUP), 2229 _input_size_in_bits(Assembler::EVEX_NObit), 2230 _is_evex_instruction(false), 2231 _evex_encoding(0), 2232 _is_clear_context(true), 2233 _is_extended_context(false), 2234 _embedded_opmask_register_specifier(0), // hard code k0 2235 _current_assembler(NULL) { 2236 if (UseAVX < 3) _legacy_mode = true; 2237 } 2238 2239 ~InstructionAttr() { 2240 if (_current_assembler != NULL) { 2241 _current_assembler->clear_attributes(); 2242 } 2243 _current_assembler = NULL; 2244 } 2245 2246 private: 2247 int _avx_vector_len; 2248 bool _rex_vex_w; 2249 bool _rex_vex_w_reverted; 2250 bool _legacy_mode; 2251 bool _no_reg_mask; 2252 bool _uses_vl; 2253 int _tuple_type; 2254 int _input_size_in_bits; 2255 bool _is_evex_instruction; 2256 int _evex_encoding; 2257 bool _is_clear_context; 2258 bool _is_extended_context; 2259 int _embedded_opmask_register_specifier; 2260 2261 Assembler *_current_assembler; 2262 2263 public: 2264 // query functions for field accessors 2265 int get_vector_len(void) const { return _avx_vector_len; } 2266 bool is_rex_vex_w(void) const { return _rex_vex_w; } 2267 bool is_rex_vex_w_reverted(void) { return _rex_vex_w_reverted; } 2268 bool is_legacy_mode(void) const { return _legacy_mode; } 2269 bool is_no_reg_mask(void) const { return _no_reg_mask; } 2270 bool uses_vl(void) const { return _uses_vl; } 2271 int get_tuple_type(void) const { return _tuple_type; } 2272 int get_input_size(void) const { return _input_size_in_bits; } 2273 int is_evex_instruction(void) const { return _is_evex_instruction; } 2274 int get_evex_encoding(void) const { return _evex_encoding; } 2275 bool is_clear_context(void) const { return _is_clear_context; } 2276 bool is_extended_context(void) const { return _is_extended_context; } 2277 int get_embedded_opmask_register_specifier(void) const { return _embedded_opmask_register_specifier; } 2278 2279 // Set the vector len manually 2280 void set_vector_len(int vector_len) { _avx_vector_len = vector_len; } 2281 2282 // Set revert rex_vex_w for avx encoding 2283 void set_rex_vex_w_reverted(void) { _rex_vex_w_reverted = true; } 2284 2285 // Set rex_vex_w based on state 2286 void set_rex_vex_w(bool state) { _rex_vex_w = state; } 2287 2288 // Set the instruction to be encoded in AVX mode 2289 void set_is_legacy_mode(void) { _legacy_mode = true; } 2290 2291 // Set the current instuction to be encoded as an EVEX instuction 2292 void set_is_evex_instruction(void) { _is_evex_instruction = true; } 2293 2294 // Internal encoding data used in compressed immediate offset programming 2295 void set_evex_encoding(int value) { _evex_encoding = value; } 2296 2297 // Set the Evex.Z field to be used to clear all non directed XMM/YMM/ZMM components 2298 void reset_is_clear_context(void) { _is_clear_context = false; } 2299 2300 // Map back to current asembler so that we can manage object level assocation 2301 void set_current_assembler(Assembler *current_assembler) { _current_assembler = current_assembler; } 2302 2303 // Address modifiers used for compressed displacement calculation 2304 void set_address_attributes(int tuple_type, int input_size_in_bits) { 2305 if (VM_Version::supports_evex()) { 2306 _tuple_type = tuple_type; 2307 _input_size_in_bits = input_size_in_bits; 2308 } 2309 } 2310 2311 // Set embedded opmask register specifier. 2312 void set_embedded_opmask_register_specifier(KRegister mask) { 2313 _embedded_opmask_register_specifier = (*mask).encoding() & 0x7; 2314 } 2315 2316 }; 2317 2318 #endif // CPU_X86_ASSEMBLER_X86_HPP