1 /* 2 * Copyright (c) 1997, 2013, 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_VM_ASSEMBLER_X86_HPP 26 #define CPU_X86_VM_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 VALUE_OBJ_CLASS_SPEC { 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 fixed register usages: 146 REGISTER_DECLARATION(Register, rbp_mh_SP_save, rbp); 147 148 // Address is an abstraction used to represent a memory location 149 // using any of the amd64 addressing modes with one object. 150 // 151 // Note: A register location is represented via a Register, not 152 // via an address for efficiency & simplicity reasons. 153 154 class ArrayAddress; 155 156 class Address VALUE_OBJ_CLASS_SPEC { 157 public: 158 enum ScaleFactor { 159 no_scale = -1, 160 times_1 = 0, 161 times_2 = 1, 162 times_4 = 2, 163 times_8 = 3, 164 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) 165 }; 166 static ScaleFactor times(int size) { 167 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size"); 168 if (size == 8) return times_8; 169 if (size == 4) return times_4; 170 if (size == 2) return times_2; 171 return times_1; 172 } 173 static int scale_size(ScaleFactor scale) { 174 assert(scale != no_scale, ""); 175 assert(((1 << (int)times_1) == 1 && 176 (1 << (int)times_2) == 2 && 177 (1 << (int)times_4) == 4 && 178 (1 << (int)times_8) == 8), ""); 179 return (1 << (int)scale); 180 } 181 182 private: 183 Register _base; 184 Register _index; 185 ScaleFactor _scale; 186 int _disp; 187 RelocationHolder _rspec; 188 189 // Easily misused constructors make them private 190 // %%% can we make these go away? 191 NOT_LP64(Address(address loc, RelocationHolder spec);) 192 Address(int disp, address loc, relocInfo::relocType rtype); 193 Address(int disp, address loc, RelocationHolder spec); 194 195 public: 196 197 int disp() { return _disp; } 198 // creation 199 Address() 200 : _base(noreg), 201 _index(noreg), 202 _scale(no_scale), 203 _disp(0) { 204 } 205 206 // No default displacement otherwise Register can be implicitly 207 // converted to 0(Register) which is quite a different animal. 208 209 Address(Register base, int disp) 210 : _base(base), 211 _index(noreg), 212 _scale(no_scale), 213 _disp(disp) { 214 } 215 216 Address(Register base, Register index, ScaleFactor scale, int disp = 0) 217 : _base (base), 218 _index(index), 219 _scale(scale), 220 _disp (disp) { 221 assert(!index->is_valid() == (scale == Address::no_scale), 222 "inconsistent address"); 223 } 224 225 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) 226 : _base (base), 227 _index(index.register_or_noreg()), 228 _scale(scale), 229 _disp (disp + (index.constant_or_zero() * scale_size(scale))) { 230 if (!index.is_register()) scale = Address::no_scale; 231 assert(!_index->is_valid() == (scale == Address::no_scale), 232 "inconsistent address"); 233 } 234 235 Address plus_disp(int disp) const { 236 Address a = (*this); 237 a._disp += disp; 238 return a; 239 } 240 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const { 241 Address a = (*this); 242 a._disp += disp.constant_or_zero() * scale_size(scale); 243 if (disp.is_register()) { 244 assert(!a.index()->is_valid(), "competing indexes"); 245 a._index = disp.as_register(); 246 a._scale = scale; 247 } 248 return a; 249 } 250 bool is_same_address(Address a) const { 251 // disregard _rspec 252 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale; 253 } 254 255 // The following two overloads are used in connection with the 256 // ByteSize type (see sizes.hpp). They simplify the use of 257 // ByteSize'd arguments in assembly code. Note that their equivalent 258 // for the optimized build are the member functions with int disp 259 // argument since ByteSize is mapped to an int type in that case. 260 // 261 // Note: DO NOT introduce similar overloaded functions for WordSize 262 // arguments as in the optimized mode, both ByteSize and WordSize 263 // are mapped to the same type and thus the compiler cannot make a 264 // distinction anymore (=> compiler errors). 265 266 #ifdef ASSERT 267 Address(Register base, ByteSize disp) 268 : _base(base), 269 _index(noreg), 270 _scale(no_scale), 271 _disp(in_bytes(disp)) { 272 } 273 274 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 275 : _base(base), 276 _index(index), 277 _scale(scale), 278 _disp(in_bytes(disp)) { 279 assert(!index->is_valid() == (scale == Address::no_scale), 280 "inconsistent address"); 281 } 282 283 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 284 : _base (base), 285 _index(index.register_or_noreg()), 286 _scale(scale), 287 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { 288 if (!index.is_register()) scale = Address::no_scale; 289 assert(!_index->is_valid() == (scale == Address::no_scale), 290 "inconsistent address"); 291 } 292 293 #endif // ASSERT 294 295 // accessors 296 bool uses(Register reg) const { return _base == reg || _index == reg; } 297 Register base() const { return _base; } 298 Register index() const { return _index; } 299 ScaleFactor scale() const { return _scale; } 300 int disp() const { return _disp; } 301 302 // Convert the raw encoding form into the form expected by the constructor for 303 // Address. An index of 4 (rsp) corresponds to having no index, so convert 304 // that to noreg for the Address constructor. 305 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc); 306 307 static Address make_array(ArrayAddress); 308 309 private: 310 bool base_needs_rex() const { 311 return _base != noreg && _base->encoding() >= 8; 312 } 313 314 bool index_needs_rex() const { 315 return _index != noreg &&_index->encoding() >= 8; 316 } 317 318 relocInfo::relocType reloc() const { return _rspec.type(); } 319 320 friend class Assembler; 321 friend class MacroAssembler; 322 friend class LIR_Assembler; // base/index/scale/disp 323 }; 324 325 // 326 // AddressLiteral has been split out from Address because operands of this type 327 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 328 // the few instructions that need to deal with address literals are unique and the 329 // MacroAssembler does not have to implement every instruction in the Assembler 330 // in order to search for address literals that may need special handling depending 331 // on the instruction and the platform. As small step on the way to merging i486/amd64 332 // directories. 333 // 334 class AddressLiteral VALUE_OBJ_CLASS_SPEC { 335 friend class ArrayAddress; 336 RelocationHolder _rspec; 337 // Typically we use AddressLiterals we want to use their rval 338 // However in some situations we want the lval (effect address) of the item. 339 // We provide a special factory for making those lvals. 340 bool _is_lval; 341 342 // If the target is far we'll need to load the ea of this to 343 // a register to reach it. Otherwise if near we can do rip 344 // relative addressing. 345 346 address _target; 347 348 protected: 349 // creation 350 AddressLiteral() 351 : _is_lval(false), 352 _target(NULL) 353 {} 354 355 public: 356 357 358 AddressLiteral(address target, relocInfo::relocType rtype); 359 360 AddressLiteral(address target, RelocationHolder const& rspec) 361 : _rspec(rspec), 362 _is_lval(false), 363 _target(target) 364 {} 365 366 AddressLiteral addr() { 367 AddressLiteral ret = *this; 368 ret._is_lval = true; 369 return ret; 370 } 371 372 373 private: 374 375 address target() { return _target; } 376 bool is_lval() { return _is_lval; } 377 378 relocInfo::relocType reloc() const { return _rspec.type(); } 379 const RelocationHolder& rspec() const { return _rspec; } 380 381 friend class Assembler; 382 friend class MacroAssembler; 383 friend class Address; 384 friend class LIR_Assembler; 385 }; 386 387 // Convience classes 388 class RuntimeAddress: public AddressLiteral { 389 390 public: 391 392 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 393 394 }; 395 396 class ExternalAddress: public AddressLiteral { 397 private: 398 static relocInfo::relocType reloc_for_target(address target) { 399 // Sometimes ExternalAddress is used for values which aren't 400 // exactly addresses, like the card table base. 401 // external_word_type can't be used for values in the first page 402 // so just skip the reloc in that case. 403 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 404 } 405 406 public: 407 408 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} 409 410 }; 411 412 class InternalAddress: public AddressLiteral { 413 414 public: 415 416 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 417 418 }; 419 420 // x86 can do array addressing as a single operation since disp can be an absolute 421 // address amd64 can't. We create a class that expresses the concept but does extra 422 // magic on amd64 to get the final result 423 424 class ArrayAddress VALUE_OBJ_CLASS_SPEC { 425 private: 426 427 AddressLiteral _base; 428 Address _index; 429 430 public: 431 432 ArrayAddress() {}; 433 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 434 AddressLiteral base() { return _base; } 435 Address index() { return _index; } 436 437 }; 438 439 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); 440 441 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 442 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 443 // is what you get. The Assembler is generating code into a CodeBuffer. 444 445 class Assembler : public AbstractAssembler { 446 friend class AbstractAssembler; // for the non-virtual hack 447 friend class LIR_Assembler; // as_Address() 448 friend class StubGenerator; 449 450 public: 451 enum Condition { // The x86 condition codes used for conditional jumps/moves. 452 zero = 0x4, 453 notZero = 0x5, 454 equal = 0x4, 455 notEqual = 0x5, 456 less = 0xc, 457 lessEqual = 0xe, 458 greater = 0xf, 459 greaterEqual = 0xd, 460 below = 0x2, 461 belowEqual = 0x6, 462 above = 0x7, 463 aboveEqual = 0x3, 464 overflow = 0x0, 465 noOverflow = 0x1, 466 carrySet = 0x2, 467 carryClear = 0x3, 468 negative = 0x8, 469 positive = 0x9, 470 parity = 0xa, 471 noParity = 0xb 472 }; 473 474 enum Prefix { 475 // segment overrides 476 CS_segment = 0x2e, 477 SS_segment = 0x36, 478 DS_segment = 0x3e, 479 ES_segment = 0x26, 480 FS_segment = 0x64, 481 GS_segment = 0x65, 482 483 REX = 0x40, 484 485 REX_B = 0x41, 486 REX_X = 0x42, 487 REX_XB = 0x43, 488 REX_R = 0x44, 489 REX_RB = 0x45, 490 REX_RX = 0x46, 491 REX_RXB = 0x47, 492 493 REX_W = 0x48, 494 495 REX_WB = 0x49, 496 REX_WX = 0x4A, 497 REX_WXB = 0x4B, 498 REX_WR = 0x4C, 499 REX_WRB = 0x4D, 500 REX_WRX = 0x4E, 501 REX_WRXB = 0x4F, 502 503 VEX_3bytes = 0xC4, 504 VEX_2bytes = 0xC5, 505 EVEX_4bytes = 0x62 506 }; 507 508 enum VexPrefix { 509 VEX_B = 0x20, 510 VEX_X = 0x40, 511 VEX_R = 0x80, 512 VEX_W = 0x80 513 }; 514 515 enum ExexPrefix { 516 EVEX_F = 0x04, 517 EVEX_V = 0x08, 518 EVEX_Rb = 0x10, 519 EVEX_X = 0x40, 520 EVEX_Z = 0x80 521 }; 522 523 enum VexSimdPrefix { 524 VEX_SIMD_NONE = 0x0, 525 VEX_SIMD_66 = 0x1, 526 VEX_SIMD_F3 = 0x2, 527 VEX_SIMD_F2 = 0x3 528 }; 529 530 enum VexOpcode { 531 VEX_OPCODE_NONE = 0x0, 532 VEX_OPCODE_0F = 0x1, 533 VEX_OPCODE_0F_38 = 0x2, 534 VEX_OPCODE_0F_3A = 0x3 535 }; 536 537 enum AvxVectorLen { 538 AVX_128bit = 0x0, 539 AVX_256bit = 0x1, 540 AVX_512bit = 0x2, 541 AVX_NoVec = 0x4 542 }; 543 544 enum EvexTupleType { 545 EVEX_FV = 0, 546 EVEX_HV = 4, 547 EVEX_FVM = 6, 548 EVEX_T1S = 7, 549 EVEX_T1F = 11, 550 EVEX_T2 = 13, 551 EVEX_T4 = 15, 552 EVEX_T8 = 17, 553 EVEX_HVM = 18, 554 EVEX_QVM = 19, 555 EVEX_OVM = 20, 556 EVEX_M128 = 21, 557 EVEX_DUP = 22, 558 EVEX_ETUP = 23 559 }; 560 561 enum EvexInputSizeInBits { 562 EVEX_8bit = 0, 563 EVEX_16bit = 1, 564 EVEX_32bit = 2, 565 EVEX_64bit = 3 566 }; 567 568 enum WhichOperand { 569 // input to locate_operand, and format code for relocations 570 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 571 disp32_operand = 1, // embedded 32-bit displacement or address 572 call32_operand = 2, // embedded 32-bit self-relative displacement 573 #ifndef _LP64 574 _WhichOperand_limit = 3 575 #else 576 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 577 _WhichOperand_limit = 4 578 #endif 579 }; 580 581 582 583 // NOTE: The general philopsophy of the declarations here is that 64bit versions 584 // of instructions are freely declared without the need for wrapping them an ifdef. 585 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 586 // In the .cpp file the implementations are wrapped so that they are dropped out 587 // of the resulting jvm. This is done mostly to keep the footprint of MINIMAL 588 // to the size it was prior to merging up the 32bit and 64bit assemblers. 589 // 590 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 591 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 592 593 private: 594 595 int evex_encoding; 596 int input_size_in_bits; 597 int avx_vector_len; 598 int tuple_type; 599 bool is_evex_instruction; 600 601 // 64bit prefixes 602 int prefix_and_encode(int reg_enc, bool byteinst = false); 603 int prefixq_and_encode(int reg_enc); 604 605 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); 606 int prefixq_and_encode(int dst_enc, int src_enc); 607 608 void prefix(Register reg); 609 void prefix(Address adr); 610 void prefixq(Address adr); 611 612 void prefix(Address adr, Register reg, bool byteinst = false); 613 void prefix(Address adr, XMMRegister reg); 614 void prefixq(Address adr, Register reg); 615 void prefixq(Address adr, XMMRegister reg); 616 617 void prefetch_prefix(Address src); 618 619 void rex_prefix(Address adr, XMMRegister xreg, 620 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 621 int rex_prefix_and_encode(int dst_enc, int src_enc, 622 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 623 624 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, 625 int nds_enc, VexSimdPrefix pre, VexOpcode opc, 626 int vector_len); 627 628 void evex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, bool evex_r, bool evex_v, 629 int nds_enc, VexSimdPrefix pre, VexOpcode opc, 630 bool is_extended_context, bool is_merge_context, 631 int vector_len, bool no_mask_reg ); 632 633 void vex_prefix(Address adr, int nds_enc, int xreg_enc, 634 VexSimdPrefix pre, VexOpcode opc, 635 bool vex_w, int vector_len, 636 bool legacy_mode = false, bool no_mask_reg = false); 637 638 void vex_prefix(XMMRegister dst, XMMRegister nds, Address src, 639 VexSimdPrefix pre, int vector_len = AVX_128bit, 640 bool no_mask_reg = false) { 641 int dst_enc = dst->encoding(); 642 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 643 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, false, vector_len, false, no_mask_reg); 644 } 645 646 void vex_prefix_q(XMMRegister dst, XMMRegister nds, Address src, 647 VexSimdPrefix pre, int vector_len = AVX_128bit, 648 bool no_mask_reg = false) { 649 int dst_enc = dst->encoding(); 650 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 651 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, true, vector_len, false, no_mask_reg); 652 } 653 654 void vex_prefix_0F38(Register dst, Register nds, Address src, bool no_mask_reg = false) { 655 bool vex_w = false; 656 int vector_len = AVX_128bit; 657 vex_prefix(src, nds->encoding(), dst->encoding(), 658 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 659 vector_len, no_mask_reg); 660 } 661 662 void vex_prefix_0F38_q(Register dst, Register nds, Address src, bool no_mask_reg = false) { 663 bool vex_w = true; 664 int vector_len = AVX_128bit; 665 vex_prefix(src, nds->encoding(), dst->encoding(), 666 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 667 vector_len, no_mask_reg); 668 } 669 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, 670 VexSimdPrefix pre, VexOpcode opc, 671 bool vex_w, int vector_len, 672 bool legacy_mode, bool no_mask_reg); 673 674 int vex_prefix_0F38_and_encode(Register dst, Register nds, Register src, bool no_mask_reg = false) { 675 bool vex_w = false; 676 int vector_len = AVX_128bit; 677 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 678 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 679 false, no_mask_reg); 680 } 681 int vex_prefix_0F38_and_encode_q(Register dst, Register nds, Register src, bool no_mask_reg = false) { 682 bool vex_w = true; 683 int vector_len = AVX_128bit; 684 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 685 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 686 false, no_mask_reg); 687 } 688 int vex_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 689 VexSimdPrefix pre, int vector_len = AVX_128bit, 690 VexOpcode opc = VEX_OPCODE_0F, bool legacy_mode = false, 691 bool no_mask_reg = false) { 692 int src_enc = src->encoding(); 693 int dst_enc = dst->encoding(); 694 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 695 return vex_prefix_and_encode(dst_enc, nds_enc, src_enc, pre, opc, false, vector_len, legacy_mode, no_mask_reg); 696 } 697 698 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, 699 VexSimdPrefix pre, bool no_mask_reg, VexOpcode opc = VEX_OPCODE_0F, 700 bool rex_w = false, int vector_len = AVX_128bit, bool legacy_mode = false); 701 702 void simd_prefix(XMMRegister dst, Address src, VexSimdPrefix pre, 703 bool no_mask_reg, VexOpcode opc = VEX_OPCODE_0F) { 704 simd_prefix(dst, xnoreg, src, pre, no_mask_reg, opc); 705 } 706 707 void simd_prefix(Address dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg) { 708 simd_prefix(src, dst, pre, no_mask_reg); 709 } 710 void simd_prefix_q(XMMRegister dst, XMMRegister nds, Address src, 711 VexSimdPrefix pre, bool no_mask_reg = false) { 712 bool rex_w = true; 713 simd_prefix(dst, nds, src, pre, no_mask_reg, VEX_OPCODE_0F, rex_w); 714 } 715 716 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 717 VexSimdPrefix pre, bool no_mask_reg, 718 VexOpcode opc = VEX_OPCODE_0F, 719 bool rex_w = false, int vector_len = AVX_128bit, 720 bool legacy_mode = false); 721 722 int kreg_prefix_and_encode(KRegister dst, KRegister nds, KRegister src, 723 VexSimdPrefix pre, bool no_mask_reg, 724 VexOpcode opc = VEX_OPCODE_0F, 725 bool rex_w = false, int vector_len = AVX_128bit); 726 727 int kreg_prefix_and_encode(KRegister dst, KRegister nds, Register src, 728 VexSimdPrefix pre, bool no_mask_reg, 729 VexOpcode opc = VEX_OPCODE_0F, 730 bool rex_w = false, int vector_len = AVX_128bit); 731 732 // Move/convert 32-bit integer value. 733 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, Register src, 734 VexSimdPrefix pre, bool no_mask_reg) { 735 // It is OK to cast from Register to XMMRegister to pass argument here 736 // since only encoding is used in simd_prefix_and_encode() and number of 737 // Gen and Xmm registers are the same. 738 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, no_mask_reg, VEX_OPCODE_0F); 739 } 740 int simd_prefix_and_encode(XMMRegister dst, Register src, VexSimdPrefix pre, bool no_mask_reg) { 741 return simd_prefix_and_encode(dst, xnoreg, src, pre, no_mask_reg); 742 } 743 int simd_prefix_and_encode(Register dst, XMMRegister src, 744 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 745 bool no_mask_reg = false) { 746 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, no_mask_reg, opc); 747 } 748 749 // Move/convert 64-bit integer value. 750 int simd_prefix_and_encode_q(XMMRegister dst, XMMRegister nds, Register src, 751 VexSimdPrefix pre, bool no_mask_reg = false) { 752 bool rex_w = true; 753 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, no_mask_reg, VEX_OPCODE_0F, rex_w); 754 } 755 int simd_prefix_and_encode_q(XMMRegister dst, Register src, VexSimdPrefix pre, bool no_mask_reg) { 756 return simd_prefix_and_encode_q(dst, xnoreg, src, pre, no_mask_reg); 757 } 758 int simd_prefix_and_encode_q(Register dst, XMMRegister src, 759 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 760 bool no_mask_reg = false) { 761 bool rex_w = true; 762 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, no_mask_reg, opc, rex_w); 763 } 764 765 // Helper functions for groups of instructions 766 void emit_arith_b(int op1, int op2, Register dst, int imm8); 767 768 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 769 // Force generation of a 4 byte immediate value even if it fits into 8bit 770 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32); 771 void emit_arith(int op1, int op2, Register dst, Register src); 772 773 void emit_simd_arith(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 774 void emit_simd_arith_q(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 775 void emit_simd_arith(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 776 void emit_simd_arith_q(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 777 void emit_simd_arith_nonds(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 778 void emit_simd_arith_nonds_q(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 779 void emit_simd_arith_nonds(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 780 void emit_simd_arith_nonds_q(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 781 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 782 Address src, VexSimdPrefix pre, int vector_len, 783 bool no_mask_reg = false); 784 void emit_vex_arith_q(int opcode, XMMRegister dst, XMMRegister nds, 785 Address src, VexSimdPrefix pre, int vector_len, 786 bool no_mask_reg = false); 787 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 788 XMMRegister src, VexSimdPrefix pre, int vector_len, 789 bool no_mask_reg = false); 790 void emit_vex_arith_q(int opcode, XMMRegister dst, XMMRegister nds, 791 XMMRegister src, VexSimdPrefix pre, int vector_len, 792 bool no_mask_reg = false); 793 794 bool emit_compressed_disp_byte(int &disp); 795 796 void emit_operand(Register reg, 797 Register base, Register index, Address::ScaleFactor scale, 798 int disp, 799 RelocationHolder const& rspec, 800 int rip_relative_correction = 0); 801 802 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 803 804 // operands that only take the original 32bit registers 805 void emit_operand32(Register reg, Address adr); 806 807 void emit_operand(XMMRegister reg, 808 Register base, Register index, Address::ScaleFactor scale, 809 int disp, 810 RelocationHolder const& rspec); 811 812 void emit_operand(XMMRegister reg, Address adr); 813 814 void emit_operand(MMXRegister reg, Address adr); 815 816 // workaround gcc (3.2.1-7) bug 817 void emit_operand(Address adr, MMXRegister reg); 818 819 820 // Immediate-to-memory forms 821 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 822 823 void emit_farith(int b1, int b2, int i); 824 825 826 protected: 827 #ifdef ASSERT 828 void check_relocation(RelocationHolder const& rspec, int format); 829 #endif 830 831 void emit_data(jint data, relocInfo::relocType rtype, int format); 832 void emit_data(jint data, RelocationHolder const& rspec, int format); 833 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 834 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 835 836 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 837 838 // These are all easily abused and hence protected 839 840 // 32BIT ONLY SECTION 841 #ifndef _LP64 842 // Make these disappear in 64bit mode since they would never be correct 843 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 844 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 845 846 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 847 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 848 849 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 850 #else 851 // 64BIT ONLY SECTION 852 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 853 854 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 855 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 856 857 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 858 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 859 #endif // _LP64 860 861 // These are unique in that we are ensured by the caller that the 32bit 862 // relative in these instructions will always be able to reach the potentially 863 // 64bit address described by entry. Since they can take a 64bit address they 864 // don't have the 32 suffix like the other instructions in this class. 865 866 void call_literal(address entry, RelocationHolder const& rspec); 867 void jmp_literal(address entry, RelocationHolder const& rspec); 868 869 // Avoid using directly section 870 // Instructions in this section are actually usable by anyone without danger 871 // of failure but have performance issues that are addressed my enhanced 872 // instructions which will do the proper thing base on the particular cpu. 873 // We protect them because we don't trust you... 874 875 // Don't use next inc() and dec() methods directly. INC & DEC instructions 876 // could cause a partial flag stall since they don't set CF flag. 877 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 878 // which call inc() & dec() or add() & sub() in accordance with 879 // the product flag UseIncDec value. 880 881 void decl(Register dst); 882 void decl(Address dst); 883 void decq(Register dst); 884 void decq(Address dst); 885 886 void incl(Register dst); 887 void incl(Address dst); 888 void incq(Register dst); 889 void incq(Address dst); 890 891 // New cpus require use of movsd and movss to avoid partial register stall 892 // when loading from memory. But for old Opteron use movlpd instead of movsd. 893 // The selection is done in MacroAssembler::movdbl() and movflt(). 894 895 // Move Scalar Single-Precision Floating-Point Values 896 void movss(XMMRegister dst, Address src); 897 void movss(XMMRegister dst, XMMRegister src); 898 void movss(Address dst, XMMRegister src); 899 900 // Move Scalar Double-Precision Floating-Point Values 901 void movsd(XMMRegister dst, Address src); 902 void movsd(XMMRegister dst, XMMRegister src); 903 void movsd(Address dst, XMMRegister src); 904 void movlpd(XMMRegister dst, Address src); 905 906 // New cpus require use of movaps and movapd to avoid partial register stall 907 // when moving between registers. 908 void movaps(XMMRegister dst, XMMRegister src); 909 void movapd(XMMRegister dst, XMMRegister src); 910 911 // End avoid using directly 912 913 914 // Instruction prefixes 915 void prefix(Prefix p); 916 917 public: 918 919 // Creation 920 Assembler(CodeBuffer* code) : AbstractAssembler(code) { 921 init_attributes(); 922 } 923 924 // Decoding 925 static address locate_operand(address inst, WhichOperand which); 926 static address locate_next_instruction(address inst); 927 928 // Utilities 929 static bool is_polling_page_far() NOT_LP64({ return false;}); 930 931 // Generic instructions 932 // Does 32bit or 64bit as needed for the platform. In some sense these 933 // belong in macro assembler but there is no need for both varieties to exist 934 935 void init_attributes(void) { 936 evex_encoding = 0; 937 input_size_in_bits = 0; 938 avx_vector_len = AVX_NoVec; 939 tuple_type = EVEX_ETUP; 940 is_evex_instruction = false; 941 } 942 943 void lea(Register dst, Address src); 944 945 void mov(Register dst, Register src); 946 947 void pusha(); 948 void popa(); 949 950 void pushf(); 951 void popf(); 952 953 void push(int32_t imm32); 954 955 void push(Register src); 956 957 void pop(Register dst); 958 959 // These are dummies to prevent surprise implicit conversions to Register 960 void push(void* v); 961 void pop(void* v); 962 963 // These do register sized moves/scans 964 void rep_mov(); 965 void rep_stos(); 966 void rep_stosb(); 967 void repne_scan(); 968 #ifdef _LP64 969 void repne_scanl(); 970 #endif 971 972 // Vanilla instructions in lexical order 973 974 void adcl(Address dst, int32_t imm32); 975 void adcl(Address dst, Register src); 976 void adcl(Register dst, int32_t imm32); 977 void adcl(Register dst, Address src); 978 void adcl(Register dst, Register src); 979 980 void adcq(Register dst, int32_t imm32); 981 void adcq(Register dst, Address src); 982 void adcq(Register dst, Register src); 983 984 void addl(Address dst, int32_t imm32); 985 void addl(Address dst, Register src); 986 void addl(Register dst, int32_t imm32); 987 void addl(Register dst, Address src); 988 void addl(Register dst, Register src); 989 990 void addq(Address dst, int32_t imm32); 991 void addq(Address dst, Register src); 992 void addq(Register dst, int32_t imm32); 993 void addq(Register dst, Address src); 994 void addq(Register dst, Register src); 995 996 #ifdef _LP64 997 //Add Unsigned Integers with Carry Flag 998 void adcxq(Register dst, Register src); 999 1000 //Add Unsigned Integers with Overflow Flag 1001 void adoxq(Register dst, Register src); 1002 #endif 1003 1004 void addr_nop_4(); 1005 void addr_nop_5(); 1006 void addr_nop_7(); 1007 void addr_nop_8(); 1008 1009 // Add Scalar Double-Precision Floating-Point Values 1010 void addsd(XMMRegister dst, Address src); 1011 void addsd(XMMRegister dst, XMMRegister src); 1012 1013 // Add Scalar Single-Precision Floating-Point Values 1014 void addss(XMMRegister dst, Address src); 1015 void addss(XMMRegister dst, XMMRegister src); 1016 1017 // AES instructions 1018 void aesdec(XMMRegister dst, Address src); 1019 void aesdec(XMMRegister dst, XMMRegister src); 1020 void aesdeclast(XMMRegister dst, Address src); 1021 void aesdeclast(XMMRegister dst, XMMRegister src); 1022 void aesenc(XMMRegister dst, Address src); 1023 void aesenc(XMMRegister dst, XMMRegister src); 1024 void aesenclast(XMMRegister dst, Address src); 1025 void aesenclast(XMMRegister dst, XMMRegister src); 1026 1027 1028 void andl(Address dst, int32_t imm32); 1029 void andl(Register dst, int32_t imm32); 1030 void andl(Register dst, Address src); 1031 void andl(Register dst, Register src); 1032 1033 void andq(Address dst, int32_t imm32); 1034 void andq(Register dst, int32_t imm32); 1035 void andq(Register dst, Address src); 1036 void andq(Register dst, Register src); 1037 1038 // BMI instructions 1039 void andnl(Register dst, Register src1, Register src2); 1040 void andnl(Register dst, Register src1, Address src2); 1041 void andnq(Register dst, Register src1, Register src2); 1042 void andnq(Register dst, Register src1, Address src2); 1043 1044 void blsil(Register dst, Register src); 1045 void blsil(Register dst, Address src); 1046 void blsiq(Register dst, Register src); 1047 void blsiq(Register dst, Address src); 1048 1049 void blsmskl(Register dst, Register src); 1050 void blsmskl(Register dst, Address src); 1051 void blsmskq(Register dst, Register src); 1052 void blsmskq(Register dst, Address src); 1053 1054 void blsrl(Register dst, Register src); 1055 void blsrl(Register dst, Address src); 1056 void blsrq(Register dst, Register src); 1057 void blsrq(Register dst, Address src); 1058 1059 void bsfl(Register dst, Register src); 1060 void bsrl(Register dst, Register src); 1061 1062 #ifdef _LP64 1063 void bsfq(Register dst, Register src); 1064 void bsrq(Register dst, Register src); 1065 #endif 1066 1067 void bswapl(Register reg); 1068 1069 void bswapq(Register reg); 1070 1071 void call(Label& L, relocInfo::relocType rtype); 1072 void call(Register reg); // push pc; pc <- reg 1073 void call(Address adr); // push pc; pc <- adr 1074 1075 void cdql(); 1076 1077 void cdqq(); 1078 1079 void cld(); 1080 1081 void clflush(Address adr); 1082 1083 void cmovl(Condition cc, Register dst, Register src); 1084 void cmovl(Condition cc, Register dst, Address src); 1085 1086 void cmovq(Condition cc, Register dst, Register src); 1087 void cmovq(Condition cc, Register dst, Address src); 1088 1089 1090 void cmpb(Address dst, int imm8); 1091 1092 void cmpl(Address dst, int32_t imm32); 1093 1094 void cmpl(Register dst, int32_t imm32); 1095 void cmpl(Register dst, Register src); 1096 void cmpl(Register dst, Address src); 1097 1098 void cmpq(Address dst, int32_t imm32); 1099 void cmpq(Address dst, Register src); 1100 1101 void cmpq(Register dst, int32_t imm32); 1102 void cmpq(Register dst, Register src); 1103 void cmpq(Register dst, Address src); 1104 1105 // these are dummies used to catch attempting to convert NULL to Register 1106 void cmpl(Register dst, void* junk); // dummy 1107 void cmpq(Register dst, void* junk); // dummy 1108 1109 void cmpw(Address dst, int imm16); 1110 1111 void cmpxchg8 (Address adr); 1112 1113 void cmpxchgb(Register reg, Address adr); 1114 void cmpxchgl(Register reg, Address adr); 1115 1116 void cmpxchgq(Register reg, Address adr); 1117 1118 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1119 void comisd(XMMRegister dst, Address src); 1120 void comisd(XMMRegister dst, XMMRegister src); 1121 1122 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1123 void comiss(XMMRegister dst, Address src); 1124 void comiss(XMMRegister dst, XMMRegister src); 1125 1126 // Identify processor type and features 1127 void cpuid(); 1128 1129 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 1130 void cvtsd2ss(XMMRegister dst, XMMRegister src); 1131 void cvtsd2ss(XMMRegister dst, Address src); 1132 1133 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 1134 void cvtsi2sdl(XMMRegister dst, Register src); 1135 void cvtsi2sdl(XMMRegister dst, Address src); 1136 void cvtsi2sdq(XMMRegister dst, Register src); 1137 void cvtsi2sdq(XMMRegister dst, Address src); 1138 1139 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 1140 void cvtsi2ssl(XMMRegister dst, Register src); 1141 void cvtsi2ssl(XMMRegister dst, Address src); 1142 void cvtsi2ssq(XMMRegister dst, Register src); 1143 void cvtsi2ssq(XMMRegister dst, Address src); 1144 1145 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 1146 void cvtdq2pd(XMMRegister dst, XMMRegister src); 1147 1148 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 1149 void cvtdq2ps(XMMRegister dst, XMMRegister src); 1150 1151 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 1152 void cvtss2sd(XMMRegister dst, XMMRegister src); 1153 void cvtss2sd(XMMRegister dst, Address src); 1154 1155 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 1156 void cvttsd2sil(Register dst, Address src); 1157 void cvttsd2sil(Register dst, XMMRegister src); 1158 void cvttsd2siq(Register dst, XMMRegister src); 1159 1160 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 1161 void cvttss2sil(Register dst, XMMRegister src); 1162 void cvttss2siq(Register dst, XMMRegister src); 1163 1164 // Divide Scalar Double-Precision Floating-Point Values 1165 void divsd(XMMRegister dst, Address src); 1166 void divsd(XMMRegister dst, XMMRegister src); 1167 1168 // Divide Scalar Single-Precision Floating-Point Values 1169 void divss(XMMRegister dst, Address src); 1170 void divss(XMMRegister dst, XMMRegister src); 1171 1172 void emms(); 1173 1174 void fabs(); 1175 1176 void fadd(int i); 1177 1178 void fadd_d(Address src); 1179 void fadd_s(Address src); 1180 1181 // "Alternate" versions of x87 instructions place result down in FPU 1182 // stack instead of on TOS 1183 1184 void fadda(int i); // "alternate" fadd 1185 void faddp(int i = 1); 1186 1187 void fchs(); 1188 1189 void fcom(int i); 1190 1191 void fcomp(int i = 1); 1192 void fcomp_d(Address src); 1193 void fcomp_s(Address src); 1194 1195 void fcompp(); 1196 1197 void fcos(); 1198 1199 void fdecstp(); 1200 1201 void fdiv(int i); 1202 void fdiv_d(Address src); 1203 void fdivr_s(Address src); 1204 void fdiva(int i); // "alternate" fdiv 1205 void fdivp(int i = 1); 1206 1207 void fdivr(int i); 1208 void fdivr_d(Address src); 1209 void fdiv_s(Address src); 1210 1211 void fdivra(int i); // "alternate" reversed fdiv 1212 1213 void fdivrp(int i = 1); 1214 1215 void ffree(int i = 0); 1216 1217 void fild_d(Address adr); 1218 void fild_s(Address adr); 1219 1220 void fincstp(); 1221 1222 void finit(); 1223 1224 void fist_s (Address adr); 1225 void fistp_d(Address adr); 1226 void fistp_s(Address adr); 1227 1228 void fld1(); 1229 1230 void fld_d(Address adr); 1231 void fld_s(Address adr); 1232 void fld_s(int index); 1233 void fld_x(Address adr); // extended-precision (80-bit) format 1234 1235 void fldcw(Address src); 1236 1237 void fldenv(Address src); 1238 1239 void fldlg2(); 1240 1241 void fldln2(); 1242 1243 void fldz(); 1244 1245 void flog(); 1246 void flog10(); 1247 1248 void fmul(int i); 1249 1250 void fmul_d(Address src); 1251 void fmul_s(Address src); 1252 1253 void fmula(int i); // "alternate" fmul 1254 1255 void fmulp(int i = 1); 1256 1257 void fnsave(Address dst); 1258 1259 void fnstcw(Address src); 1260 1261 void fnstsw_ax(); 1262 1263 void fprem(); 1264 void fprem1(); 1265 1266 void frstor(Address src); 1267 1268 void fsin(); 1269 1270 void fsqrt(); 1271 1272 void fst_d(Address adr); 1273 void fst_s(Address adr); 1274 1275 void fstp_d(Address adr); 1276 void fstp_d(int index); 1277 void fstp_s(Address adr); 1278 void fstp_x(Address adr); // extended-precision (80-bit) format 1279 1280 void fsub(int i); 1281 void fsub_d(Address src); 1282 void fsub_s(Address src); 1283 1284 void fsuba(int i); // "alternate" fsub 1285 1286 void fsubp(int i = 1); 1287 1288 void fsubr(int i); 1289 void fsubr_d(Address src); 1290 void fsubr_s(Address src); 1291 1292 void fsubra(int i); // "alternate" reversed fsub 1293 1294 void fsubrp(int i = 1); 1295 1296 void ftan(); 1297 1298 void ftst(); 1299 1300 void fucomi(int i = 1); 1301 void fucomip(int i = 1); 1302 1303 void fwait(); 1304 1305 void fxch(int i = 1); 1306 1307 void fxrstor(Address src); 1308 1309 void fxsave(Address dst); 1310 1311 void fyl2x(); 1312 void frndint(); 1313 void f2xm1(); 1314 void fldl2e(); 1315 1316 void hlt(); 1317 1318 void idivl(Register src); 1319 void divl(Register src); // Unsigned division 1320 1321 #ifdef _LP64 1322 void idivq(Register src); 1323 #endif 1324 1325 void imull(Register dst, Register src); 1326 void imull(Register dst, Register src, int value); 1327 void imull(Register dst, Address src); 1328 1329 #ifdef _LP64 1330 void imulq(Register dst, Register src); 1331 void imulq(Register dst, Register src, int value); 1332 void imulq(Register dst, Address src); 1333 #endif 1334 1335 // jcc is the generic conditional branch generator to run- 1336 // time routines, jcc is used for branches to labels. jcc 1337 // takes a branch opcode (cc) and a label (L) and generates 1338 // either a backward branch or a forward branch and links it 1339 // to the label fixup chain. Usage: 1340 // 1341 // Label L; // unbound label 1342 // jcc(cc, L); // forward branch to unbound label 1343 // bind(L); // bind label to the current pc 1344 // jcc(cc, L); // backward branch to bound label 1345 // bind(L); // illegal: a label may be bound only once 1346 // 1347 // Note: The same Label can be used for forward and backward branches 1348 // but it may be bound only once. 1349 1350 void jcc(Condition cc, Label& L, bool maybe_short = true); 1351 1352 // Conditional jump to a 8-bit offset to L. 1353 // WARNING: be very careful using this for forward jumps. If the label is 1354 // not bound within an 8-bit offset of this instruction, a run-time error 1355 // will occur. 1356 void jccb(Condition cc, Label& L); 1357 1358 void jmp(Address entry); // pc <- entry 1359 1360 // Label operations & relative jumps (PPUM Appendix D) 1361 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L 1362 1363 void jmp(Register entry); // pc <- entry 1364 1365 // Unconditional 8-bit offset jump to L. 1366 // WARNING: be very careful using this for forward jumps. If the label is 1367 // not bound within an 8-bit offset of this instruction, a run-time error 1368 // will occur. 1369 void jmpb(Label& L); 1370 1371 void ldmxcsr( Address src ); 1372 1373 void leal(Register dst, Address src); 1374 1375 void leaq(Register dst, Address src); 1376 1377 void lfence(); 1378 1379 void lock(); 1380 1381 void lzcntl(Register dst, Register src); 1382 1383 #ifdef _LP64 1384 void lzcntq(Register dst, Register src); 1385 #endif 1386 1387 enum Membar_mask_bits { 1388 StoreStore = 1 << 3, 1389 LoadStore = 1 << 2, 1390 StoreLoad = 1 << 1, 1391 LoadLoad = 1 << 0 1392 }; 1393 1394 // Serializes memory and blows flags 1395 void membar(Membar_mask_bits order_constraint) { 1396 if (os::is_MP()) { 1397 // We only have to handle StoreLoad 1398 if (order_constraint & StoreLoad) { 1399 // All usable chips support "locked" instructions which suffice 1400 // as barriers, and are much faster than the alternative of 1401 // using cpuid instruction. We use here a locked add [esp-C],0. 1402 // This is conveniently otherwise a no-op except for blowing 1403 // flags, and introducing a false dependency on target memory 1404 // location. We can't do anything with flags, but we can avoid 1405 // memory dependencies in the current method by locked-adding 1406 // somewhere else on the stack. Doing [esp+C] will collide with 1407 // something on stack in current method, hence we go for [esp-C]. 1408 // It is convenient since it is almost always in data cache, for 1409 // any small C. We need to step back from SP to avoid data 1410 // dependencies with other things on below SP (callee-saves, for 1411 // example). Without a clear way to figure out the minimal safe 1412 // distance from SP, it makes sense to step back the complete 1413 // cache line, as this will also avoid possible second-order effects 1414 // with locked ops against the cache line. Our choice of offset 1415 // is bounded by x86 operand encoding, which should stay within 1416 // [-128; +127] to have the 8-byte displacement encoding. 1417 // 1418 // Any change to this code may need to revisit other places in 1419 // the code where this idiom is used, in particular the 1420 // orderAccess code. 1421 1422 int offset = -VM_Version::L1_line_size(); 1423 if (offset < -128) { 1424 offset = -128; 1425 } 1426 1427 lock(); 1428 addl(Address(rsp, offset), 0);// Assert the lock# signal here 1429 } 1430 } 1431 } 1432 1433 void mfence(); 1434 1435 // Moves 1436 1437 void mov64(Register dst, int64_t imm64); 1438 1439 void movb(Address dst, Register src); 1440 void movb(Address dst, int imm8); 1441 void movb(Register dst, Address src); 1442 1443 void kmovq(KRegister dst, KRegister src); 1444 void kmovql(KRegister dst, Register src); 1445 void kmovq(Address dst, KRegister src); 1446 void kmovq(KRegister dst, Address src); 1447 1448 void movdl(XMMRegister dst, Register src); 1449 void movdl(Register dst, XMMRegister src); 1450 void movdl(XMMRegister dst, Address src); 1451 void movdl(Address dst, XMMRegister src); 1452 1453 // Move Double Quadword 1454 void movdq(XMMRegister dst, Register src); 1455 void movdq(Register dst, XMMRegister src); 1456 1457 // Move Aligned Double Quadword 1458 void movdqa(XMMRegister dst, XMMRegister src); 1459 void movdqa(XMMRegister dst, Address src); 1460 1461 // Move Unaligned Double Quadword 1462 void movdqu(Address dst, XMMRegister src); 1463 void movdqu(XMMRegister dst, Address src); 1464 void movdqu(XMMRegister dst, XMMRegister src); 1465 1466 // Move Unaligned 256bit Vector 1467 void vmovdqu(Address dst, XMMRegister src); 1468 void vmovdqu(XMMRegister dst, Address src); 1469 void vmovdqu(XMMRegister dst, XMMRegister src); 1470 1471 // Move Unaligned 512bit Vector 1472 void evmovdqu(Address dst, XMMRegister src, int vector_len); 1473 void evmovdqu(XMMRegister dst, Address src, int vector_len); 1474 void evmovdqu(XMMRegister dst, XMMRegister src, int vector_len); 1475 1476 // Move lower 64bit to high 64bit in 128bit register 1477 void movlhps(XMMRegister dst, XMMRegister src); 1478 1479 void movl(Register dst, int32_t imm32); 1480 void movl(Address dst, int32_t imm32); 1481 void movl(Register dst, Register src); 1482 void movl(Register dst, Address src); 1483 void movl(Address dst, Register src); 1484 1485 // These dummies prevent using movl from converting a zero (like NULL) into Register 1486 // by giving the compiler two choices it can't resolve 1487 1488 void movl(Address dst, void* junk); 1489 void movl(Register dst, void* junk); 1490 1491 #ifdef _LP64 1492 void movq(Register dst, Register src); 1493 void movq(Register dst, Address src); 1494 void movq(Address dst, Register src); 1495 #endif 1496 1497 void movq(Address dst, MMXRegister src ); 1498 void movq(MMXRegister dst, Address src ); 1499 1500 #ifdef _LP64 1501 // These dummies prevent using movq from converting a zero (like NULL) into Register 1502 // by giving the compiler two choices it can't resolve 1503 1504 void movq(Address dst, void* dummy); 1505 void movq(Register dst, void* dummy); 1506 #endif 1507 1508 // Move Quadword 1509 void movq(Address dst, XMMRegister src); 1510 void movq(XMMRegister dst, Address src); 1511 1512 void movsbl(Register dst, Address src); 1513 void movsbl(Register dst, Register src); 1514 1515 #ifdef _LP64 1516 void movsbq(Register dst, Address src); 1517 void movsbq(Register dst, Register src); 1518 1519 // Move signed 32bit immediate to 64bit extending sign 1520 void movslq(Address dst, int32_t imm64); 1521 void movslq(Register dst, int32_t imm64); 1522 1523 void movslq(Register dst, Address src); 1524 void movslq(Register dst, Register src); 1525 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1526 #endif 1527 1528 void movswl(Register dst, Address src); 1529 void movswl(Register dst, Register src); 1530 1531 #ifdef _LP64 1532 void movswq(Register dst, Address src); 1533 void movswq(Register dst, Register src); 1534 #endif 1535 1536 void movw(Address dst, int imm16); 1537 void movw(Register dst, Address src); 1538 void movw(Address dst, Register src); 1539 1540 void movzbl(Register dst, Address src); 1541 void movzbl(Register dst, Register src); 1542 1543 #ifdef _LP64 1544 void movzbq(Register dst, Address src); 1545 void movzbq(Register dst, Register src); 1546 #endif 1547 1548 void movzwl(Register dst, Address src); 1549 void movzwl(Register dst, Register src); 1550 1551 #ifdef _LP64 1552 void movzwq(Register dst, Address src); 1553 void movzwq(Register dst, Register src); 1554 #endif 1555 1556 // Unsigned multiply with RAX destination register 1557 void mull(Address src); 1558 void mull(Register src); 1559 1560 #ifdef _LP64 1561 void mulq(Address src); 1562 void mulq(Register src); 1563 void mulxq(Register dst1, Register dst2, Register src); 1564 #endif 1565 1566 // Multiply Scalar Double-Precision Floating-Point Values 1567 void mulsd(XMMRegister dst, Address src); 1568 void mulsd(XMMRegister dst, XMMRegister src); 1569 1570 // Multiply Scalar Single-Precision Floating-Point Values 1571 void mulss(XMMRegister dst, Address src); 1572 void mulss(XMMRegister dst, XMMRegister src); 1573 1574 void negl(Register dst); 1575 1576 #ifdef _LP64 1577 void negq(Register dst); 1578 #endif 1579 1580 void nop(int i = 1); 1581 1582 void notl(Register dst); 1583 1584 #ifdef _LP64 1585 void notq(Register dst); 1586 #endif 1587 1588 void orl(Address dst, int32_t imm32); 1589 void orl(Register dst, int32_t imm32); 1590 void orl(Register dst, Address src); 1591 void orl(Register dst, Register src); 1592 1593 void orq(Address dst, int32_t imm32); 1594 void orq(Register dst, int32_t imm32); 1595 void orq(Register dst, Address src); 1596 void orq(Register dst, Register src); 1597 1598 // Pack with unsigned saturation 1599 void packuswb(XMMRegister dst, XMMRegister src); 1600 void packuswb(XMMRegister dst, Address src); 1601 void vpackuswb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1602 1603 // Pemutation of 64bit words 1604 void vpermq(XMMRegister dst, XMMRegister src, int imm8, int vector_len); 1605 1606 void pause(); 1607 1608 // SSE4.2 string instructions 1609 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1610 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1611 1612 // SSE 4.1 extract 1613 void pextrd(Register dst, XMMRegister src, int imm8); 1614 void pextrq(Register dst, XMMRegister src, int imm8); 1615 1616 // SSE 4.1 insert 1617 void pinsrd(XMMRegister dst, Register src, int imm8); 1618 void pinsrq(XMMRegister dst, Register src, int imm8); 1619 1620 // SSE4.1 packed move 1621 void pmovzxbw(XMMRegister dst, XMMRegister src); 1622 void pmovzxbw(XMMRegister dst, Address src); 1623 1624 #ifndef _LP64 // no 32bit push/pop on amd64 1625 void popl(Address dst); 1626 #endif 1627 1628 #ifdef _LP64 1629 void popq(Address dst); 1630 #endif 1631 1632 void popcntl(Register dst, Address src); 1633 void popcntl(Register dst, Register src); 1634 1635 #ifdef _LP64 1636 void popcntq(Register dst, Address src); 1637 void popcntq(Register dst, Register src); 1638 #endif 1639 1640 // Prefetches (SSE, SSE2, 3DNOW only) 1641 1642 void prefetchnta(Address src); 1643 void prefetchr(Address src); 1644 void prefetcht0(Address src); 1645 void prefetcht1(Address src); 1646 void prefetcht2(Address src); 1647 void prefetchw(Address src); 1648 1649 // Shuffle Bytes 1650 void pshufb(XMMRegister dst, XMMRegister src); 1651 void pshufb(XMMRegister dst, Address src); 1652 1653 // Shuffle Packed Doublewords 1654 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1655 void pshufd(XMMRegister dst, Address src, int mode); 1656 1657 // Shuffle Packed Low Words 1658 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1659 void pshuflw(XMMRegister dst, Address src, int mode); 1660 1661 // Shift Right by bytes Logical DoubleQuadword Immediate 1662 void psrldq(XMMRegister dst, int shift); 1663 1664 // Logical Compare 128bit 1665 void ptest(XMMRegister dst, XMMRegister src); 1666 void ptest(XMMRegister dst, Address src); 1667 // Logical Compare 256bit 1668 void vptest(XMMRegister dst, XMMRegister src); 1669 void vptest(XMMRegister dst, Address src); 1670 1671 // Interleave Low Bytes 1672 void punpcklbw(XMMRegister dst, XMMRegister src); 1673 void punpcklbw(XMMRegister dst, Address src); 1674 1675 // Interleave Low Doublewords 1676 void punpckldq(XMMRegister dst, XMMRegister src); 1677 void punpckldq(XMMRegister dst, Address src); 1678 1679 // Interleave Low Quadwords 1680 void punpcklqdq(XMMRegister dst, XMMRegister src); 1681 1682 #ifndef _LP64 // no 32bit push/pop on amd64 1683 void pushl(Address src); 1684 #endif 1685 1686 void pushq(Address src); 1687 1688 void rcll(Register dst, int imm8); 1689 1690 void rclq(Register dst, int imm8); 1691 1692 void rdtsc(); 1693 1694 void ret(int imm16); 1695 1696 #ifdef _LP64 1697 void rorq(Register dst, int imm8); 1698 void rorxq(Register dst, Register src, int imm8); 1699 #endif 1700 1701 void sahf(); 1702 1703 void sarl(Register dst, int imm8); 1704 void sarl(Register dst); 1705 1706 void sarq(Register dst, int imm8); 1707 void sarq(Register dst); 1708 1709 void sbbl(Address dst, int32_t imm32); 1710 void sbbl(Register dst, int32_t imm32); 1711 void sbbl(Register dst, Address src); 1712 void sbbl(Register dst, Register src); 1713 1714 void sbbq(Address dst, int32_t imm32); 1715 void sbbq(Register dst, int32_t imm32); 1716 void sbbq(Register dst, Address src); 1717 void sbbq(Register dst, Register src); 1718 1719 void setb(Condition cc, Register dst); 1720 1721 void shldl(Register dst, Register src); 1722 1723 void shll(Register dst, int imm8); 1724 void shll(Register dst); 1725 1726 void shlq(Register dst, int imm8); 1727 void shlq(Register dst); 1728 1729 void shrdl(Register dst, Register src); 1730 1731 void shrl(Register dst, int imm8); 1732 void shrl(Register dst); 1733 1734 void shrq(Register dst, int imm8); 1735 void shrq(Register dst); 1736 1737 void smovl(); // QQQ generic? 1738 1739 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1740 void sqrtsd(XMMRegister dst, Address src); 1741 void sqrtsd(XMMRegister dst, XMMRegister src); 1742 1743 // Compute Square Root of Scalar Single-Precision Floating-Point Value 1744 void sqrtss(XMMRegister dst, Address src); 1745 void sqrtss(XMMRegister dst, XMMRegister src); 1746 1747 void std(); 1748 1749 void stmxcsr( Address dst ); 1750 1751 void subl(Address dst, int32_t imm32); 1752 void subl(Address dst, Register src); 1753 void subl(Register dst, int32_t imm32); 1754 void subl(Register dst, Address src); 1755 void subl(Register dst, Register src); 1756 1757 void subq(Address dst, int32_t imm32); 1758 void subq(Address dst, Register src); 1759 void subq(Register dst, int32_t imm32); 1760 void subq(Register dst, Address src); 1761 void subq(Register dst, Register src); 1762 1763 // Force generation of a 4 byte immediate value even if it fits into 8bit 1764 void subl_imm32(Register dst, int32_t imm32); 1765 void subq_imm32(Register dst, int32_t imm32); 1766 1767 // Subtract Scalar Double-Precision Floating-Point Values 1768 void subsd(XMMRegister dst, Address src); 1769 void subsd(XMMRegister dst, XMMRegister src); 1770 1771 // Subtract Scalar Single-Precision Floating-Point Values 1772 void subss(XMMRegister dst, Address src); 1773 void subss(XMMRegister dst, XMMRegister src); 1774 1775 void testb(Register dst, int imm8); 1776 1777 void testl(Register dst, int32_t imm32); 1778 void testl(Register dst, Register src); 1779 void testl(Register dst, Address src); 1780 1781 void testq(Register dst, int32_t imm32); 1782 void testq(Register dst, Register src); 1783 1784 // BMI - count trailing zeros 1785 void tzcntl(Register dst, Register src); 1786 void tzcntq(Register dst, Register src); 1787 1788 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1789 void ucomisd(XMMRegister dst, Address src); 1790 void ucomisd(XMMRegister dst, XMMRegister src); 1791 1792 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1793 void ucomiss(XMMRegister dst, Address src); 1794 void ucomiss(XMMRegister dst, XMMRegister src); 1795 1796 void xabort(int8_t imm8); 1797 1798 void xaddl(Address dst, Register src); 1799 1800 void xaddq(Address dst, Register src); 1801 1802 void xbegin(Label& abort, relocInfo::relocType rtype = relocInfo::none); 1803 1804 void xchgl(Register reg, Address adr); 1805 void xchgl(Register dst, Register src); 1806 1807 void xchgq(Register reg, Address adr); 1808 void xchgq(Register dst, Register src); 1809 1810 void xend(); 1811 1812 // Get Value of Extended Control Register 1813 void xgetbv(); 1814 1815 void xorl(Register dst, int32_t imm32); 1816 void xorl(Register dst, Address src); 1817 void xorl(Register dst, Register src); 1818 1819 void xorq(Register dst, Address src); 1820 void xorq(Register dst, Register src); 1821 1822 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1823 1824 // AVX 3-operands scalar instructions (encoded with VEX prefix) 1825 1826 void vaddsd(XMMRegister dst, XMMRegister nds, Address src); 1827 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1828 void vaddss(XMMRegister dst, XMMRegister nds, Address src); 1829 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1830 void vdivsd(XMMRegister dst, XMMRegister nds, Address src); 1831 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1832 void vdivss(XMMRegister dst, XMMRegister nds, Address src); 1833 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1834 void vmulsd(XMMRegister dst, XMMRegister nds, Address src); 1835 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1836 void vmulss(XMMRegister dst, XMMRegister nds, Address src); 1837 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1838 void vsubsd(XMMRegister dst, XMMRegister nds, Address src); 1839 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1840 void vsubss(XMMRegister dst, XMMRegister nds, Address src); 1841 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1842 1843 1844 //====================VECTOR ARITHMETIC===================================== 1845 1846 // Add Packed Floating-Point Values 1847 void addpd(XMMRegister dst, XMMRegister src); 1848 void addps(XMMRegister dst, XMMRegister src); 1849 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1850 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1851 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1852 void vaddps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1853 1854 // Subtract Packed Floating-Point Values 1855 void subpd(XMMRegister dst, XMMRegister src); 1856 void subps(XMMRegister dst, XMMRegister src); 1857 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1858 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1859 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1860 void vsubps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1861 1862 // Multiply Packed Floating-Point Values 1863 void mulpd(XMMRegister dst, XMMRegister src); 1864 void mulps(XMMRegister dst, XMMRegister src); 1865 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1866 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1867 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1868 void vmulps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1869 1870 // Divide Packed Floating-Point Values 1871 void divpd(XMMRegister dst, XMMRegister src); 1872 void divps(XMMRegister dst, XMMRegister src); 1873 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1874 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1875 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1876 void vdivps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1877 1878 // Bitwise Logical AND of Packed Floating-Point Values 1879 void andpd(XMMRegister dst, XMMRegister src); 1880 void andps(XMMRegister dst, XMMRegister src); 1881 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1882 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1883 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1884 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1885 1886 // Bitwise Logical XOR of Packed Floating-Point Values 1887 void xorpd(XMMRegister dst, XMMRegister src); 1888 void xorps(XMMRegister dst, XMMRegister src); 1889 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1890 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1891 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1892 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1893 1894 // Add horizontal packed integers 1895 void vphaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1896 void vphaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1897 void phaddw(XMMRegister dst, XMMRegister src); 1898 void phaddd(XMMRegister dst, XMMRegister src); 1899 1900 // Add packed integers 1901 void paddb(XMMRegister dst, XMMRegister src); 1902 void paddw(XMMRegister dst, XMMRegister src); 1903 void paddd(XMMRegister dst, XMMRegister src); 1904 void paddq(XMMRegister dst, XMMRegister src); 1905 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1906 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1907 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1908 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1909 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1910 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1911 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1912 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1913 1914 // Sub packed integers 1915 void psubb(XMMRegister dst, XMMRegister src); 1916 void psubw(XMMRegister dst, XMMRegister src); 1917 void psubd(XMMRegister dst, XMMRegister src); 1918 void psubq(XMMRegister dst, XMMRegister src); 1919 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1920 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1921 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1922 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1923 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1924 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1925 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1926 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1927 1928 // Multiply packed integers (only shorts and ints) 1929 void pmullw(XMMRegister dst, XMMRegister src); 1930 void pmulld(XMMRegister dst, XMMRegister src); 1931 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1932 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1933 void vpmullq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1934 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1935 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1936 void vpmullq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1937 1938 // Shift left packed integers 1939 void psllw(XMMRegister dst, int shift); 1940 void pslld(XMMRegister dst, int shift); 1941 void psllq(XMMRegister dst, int shift); 1942 void psllw(XMMRegister dst, XMMRegister shift); 1943 void pslld(XMMRegister dst, XMMRegister shift); 1944 void psllq(XMMRegister dst, XMMRegister shift); 1945 void vpsllw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1946 void vpslld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1947 void vpsllq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1948 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1949 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1950 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1951 1952 // Logical shift right packed integers 1953 void psrlw(XMMRegister dst, int shift); 1954 void psrld(XMMRegister dst, int shift); 1955 void psrlq(XMMRegister dst, int shift); 1956 void psrlw(XMMRegister dst, XMMRegister shift); 1957 void psrld(XMMRegister dst, XMMRegister shift); 1958 void psrlq(XMMRegister dst, XMMRegister shift); 1959 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1960 void vpsrld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1961 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1962 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1963 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1964 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1965 1966 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs) 1967 void psraw(XMMRegister dst, int shift); 1968 void psrad(XMMRegister dst, int shift); 1969 void psraw(XMMRegister dst, XMMRegister shift); 1970 void psrad(XMMRegister dst, XMMRegister shift); 1971 void vpsraw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1972 void vpsrad(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1973 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1974 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1975 1976 // And packed integers 1977 void pand(XMMRegister dst, XMMRegister src); 1978 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1979 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1980 1981 // Or packed integers 1982 void por(XMMRegister dst, XMMRegister src); 1983 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1984 void vpor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1985 1986 // Xor packed integers 1987 void pxor(XMMRegister dst, XMMRegister src); 1988 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1989 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1990 1991 // Copy low 128bit into high 128bit of YMM registers. 1992 void vinsertf128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 1993 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 1994 void vextractf128h(XMMRegister dst, XMMRegister src); 1995 void vextracti128h(XMMRegister dst, XMMRegister src); 1996 1997 // Load/store high 128bit of YMM registers which does not destroy other half. 1998 void vinsertf128h(XMMRegister dst, Address src); 1999 void vinserti128h(XMMRegister dst, Address src); 2000 void vextractf128h(Address dst, XMMRegister src); 2001 void vextracti128h(Address dst, XMMRegister src); 2002 2003 // Copy low 256bit into high 256bit of ZMM registers. 2004 void vinserti64x4h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2005 void vinsertf64x4h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2006 void vextracti64x4h(XMMRegister dst, XMMRegister src); 2007 void vextractf64x4h(XMMRegister dst, XMMRegister src); 2008 void vextracti64x2h(XMMRegister dst, XMMRegister src, int value); 2009 void vextractf64x2h(XMMRegister dst, XMMRegister src, int value); 2010 void vextractf32x4h(XMMRegister dst, XMMRegister src, int value); 2011 2012 // duplicate 4-bytes integer data from src into 8 locations in dest 2013 void vpbroadcastd(XMMRegister dst, XMMRegister src); 2014 2015 // duplicate 4-bytes integer data from src into vector_len locations in dest 2016 void evpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len); 2017 2018 // Carry-Less Multiplication Quadword 2019 void pclmulqdq(XMMRegister dst, XMMRegister src, int mask); 2020 void vpclmulqdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int mask); 2021 2022 // AVX instruction which is used to clear upper 128 bits of YMM registers and 2023 // to avoid transaction penalty between AVX and SSE states. There is no 2024 // penalty if legacy SSE instructions are encoded using VEX prefix because 2025 // they always clear upper 128 bits. It should be used before calling 2026 // runtime code and native libraries. 2027 void vzeroupper(); 2028 2029 protected: 2030 // Next instructions require address alignment 16 bytes SSE mode. 2031 // They should be called only from corresponding MacroAssembler instructions. 2032 void andpd(XMMRegister dst, Address src); 2033 void andps(XMMRegister dst, Address src); 2034 void xorpd(XMMRegister dst, Address src); 2035 void xorps(XMMRegister dst, Address src); 2036 2037 }; 2038 2039 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP