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 292 146 // 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 VALUE_OBJ_CLASS_SPEC { 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 ScaleFactor _scale; 188 int _disp; 189 RelocationHolder _rspec; 190 191 // Easily misused constructors make them private 192 // %%% can we make these go away? 193 NOT_LP64(Address(address loc, RelocationHolder spec);) 194 Address(int disp, address loc, relocInfo::relocType rtype); 195 Address(int disp, address loc, RelocationHolder spec); 196 197 public: 198 199 int disp() { return _disp; } 200 // creation 201 Address() 202 : _base(noreg), 203 _index(noreg), 204 _scale(no_scale), 205 _disp(0) { 206 } 207 208 // No default displacement otherwise Register can be implicitly 209 // converted to 0(Register) which is quite a different animal. 210 211 Address(Register base, int disp) 212 : _base(base), 213 _index(noreg), 214 _scale(no_scale), 215 _disp(disp) { 216 } 217 218 Address(Register base, Register index, ScaleFactor scale, int disp = 0) 219 : _base (base), 220 _index(index), 221 _scale(scale), 222 _disp (disp) { 223 assert(!index->is_valid() == (scale == Address::no_scale), 224 "inconsistent address"); 225 } 226 227 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) 228 : _base (base), 229 _index(index.register_or_noreg()), 230 _scale(scale), 231 _disp (disp + (index.constant_or_zero() * scale_size(scale))) { 232 if (!index.is_register()) scale = Address::no_scale; 233 assert(!_index->is_valid() == (scale == Address::no_scale), 234 "inconsistent address"); 235 } 236 237 Address plus_disp(int disp) const { 238 Address a = (*this); 239 a._disp += disp; 240 return a; 241 } 242 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const { 243 Address a = (*this); 244 a._disp += disp.constant_or_zero() * scale_size(scale); 245 if (disp.is_register()) { 246 assert(!a.index()->is_valid(), "competing indexes"); 247 a._index = disp.as_register(); 248 a._scale = scale; 249 } 250 return a; 251 } 252 bool is_same_address(Address a) const { 253 // disregard _rspec 254 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale; 255 } 256 257 // The following two overloads are used in connection with the 258 // ByteSize type (see sizes.hpp). They simplify the use of 259 // ByteSize'd arguments in assembly code. Note that their equivalent 260 // for the optimized build are the member functions with int disp 261 // argument since ByteSize is mapped to an int type in that case. 262 // 263 // Note: DO NOT introduce similar overloaded functions for WordSize 264 // arguments as in the optimized mode, both ByteSize and WordSize 265 // are mapped to the same type and thus the compiler cannot make a 266 // distinction anymore (=> compiler errors). 267 268 #ifdef ASSERT 269 Address(Register base, ByteSize disp) 270 : _base(base), 271 _index(noreg), 272 _scale(no_scale), 273 _disp(in_bytes(disp)) { 274 } 275 276 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 277 : _base(base), 278 _index(index), 279 _scale(scale), 280 _disp(in_bytes(disp)) { 281 assert(!index->is_valid() == (scale == Address::no_scale), 282 "inconsistent address"); 283 } 284 285 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 286 : _base (base), 287 _index(index.register_or_noreg()), 288 _scale(scale), 289 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { 290 if (!index.is_register()) scale = Address::no_scale; 291 assert(!_index->is_valid() == (scale == Address::no_scale), 292 "inconsistent address"); 293 } 294 295 #endif // ASSERT 296 297 // accessors 298 bool uses(Register reg) const { return _base == reg || _index == reg; } 299 Register base() const { return _base; } 300 Register index() const { return _index; } 301 ScaleFactor scale() const { return _scale; } 302 int disp() const { return _disp; } 303 304 // Convert the raw encoding form into the form expected by the constructor for 305 // Address. An index of 4 (rsp) corresponds to having no index, so convert 306 // that to noreg for the Address constructor. 307 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc); 308 309 static Address make_array(ArrayAddress); 310 311 private: 312 bool base_needs_rex() const { 313 return _base != noreg && _base->encoding() >= 8; 314 } 315 316 bool index_needs_rex() const { 317 return _index != noreg &&_index->encoding() >= 8; 318 } 319 320 relocInfo::relocType reloc() const { return _rspec.type(); } 321 322 friend class Assembler; 323 friend class MacroAssembler; 324 friend class LIR_Assembler; // base/index/scale/disp 325 }; 326 327 // 328 // AddressLiteral has been split out from Address because operands of this type 329 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 330 // the few instructions that need to deal with address literals are unique and the 331 // MacroAssembler does not have to implement every instruction in the Assembler 332 // in order to search for address literals that may need special handling depending 333 // on the instruction and the platform. As small step on the way to merging i486/amd64 334 // directories. 335 // 336 class AddressLiteral VALUE_OBJ_CLASS_SPEC { 337 friend class ArrayAddress; 338 RelocationHolder _rspec; 339 // Typically we use AddressLiterals we want to use their rval 340 // However in some situations we want the lval (effect address) of the item. 341 // We provide a special factory for making those lvals. 342 bool _is_lval; 343 344 // If the target is far we'll need to load the ea of this to 345 // a register to reach it. Otherwise if near we can do rip 346 // relative addressing. 347 348 address _target; 349 350 protected: 351 // creation 352 AddressLiteral() 353 : _is_lval(false), 354 _target(NULL) 355 {} 356 357 public: 358 359 360 AddressLiteral(address target, relocInfo::relocType rtype); 361 362 AddressLiteral(address target, RelocationHolder const& rspec) 363 : _rspec(rspec), 364 _is_lval(false), 365 _target(target) 366 {} 367 368 AddressLiteral addr() { 369 AddressLiteral ret = *this; 370 ret._is_lval = true; 371 return ret; 372 } 373 374 375 private: 376 377 address target() { return _target; } 378 bool is_lval() { return _is_lval; } 379 380 relocInfo::relocType reloc() const { return _rspec.type(); } 381 const RelocationHolder& rspec() const { return _rspec; } 382 383 friend class Assembler; 384 friend class MacroAssembler; 385 friend class Address; 386 friend class LIR_Assembler; 387 }; 388 389 // Convience classes 390 class RuntimeAddress: public AddressLiteral { 391 392 public: 393 394 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 395 396 }; 397 398 class ExternalAddress: public AddressLiteral { 399 private: 400 static relocInfo::relocType reloc_for_target(address target) { 401 // Sometimes ExternalAddress is used for values which aren't 402 // exactly addresses, like the card table base. 403 // external_word_type can't be used for values in the first page 404 // so just skip the reloc in that case. 405 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 406 } 407 408 public: 409 410 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} 411 412 }; 413 414 class InternalAddress: public AddressLiteral { 415 416 public: 417 418 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 419 420 }; 421 422 // x86 can do array addressing as a single operation since disp can be an absolute 423 // address amd64 can't. We create a class that expresses the concept but does extra 424 // magic on amd64 to get the final result 425 426 class ArrayAddress VALUE_OBJ_CLASS_SPEC { 427 private: 428 429 AddressLiteral _base; 430 Address _index; 431 432 public: 433 434 ArrayAddress() {}; 435 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 436 AddressLiteral base() { return _base; } 437 Address index() { return _index; } 438 439 }; 440 441 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512*2 / wordSize); 442 443 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 444 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 445 // is what you get. The Assembler is generating code into a CodeBuffer. 446 447 class Assembler : public AbstractAssembler { 448 friend class AbstractAssembler; // for the non-virtual hack 449 friend class LIR_Assembler; // as_Address() 450 friend class StubGenerator; 451 452 public: 453 enum Condition { // The x86 condition codes used for conditional jumps/moves. 454 zero = 0x4, 455 notZero = 0x5, 456 equal = 0x4, 457 notEqual = 0x5, 458 less = 0xc, 459 lessEqual = 0xe, 460 greater = 0xf, 461 greaterEqual = 0xd, 462 below = 0x2, 463 belowEqual = 0x6, 464 above = 0x7, 465 aboveEqual = 0x3, 466 overflow = 0x0, 467 noOverflow = 0x1, 468 carrySet = 0x2, 469 carryClear = 0x3, 470 negative = 0x8, 471 positive = 0x9, 472 parity = 0xa, 473 noParity = 0xb 474 }; 475 476 enum Prefix { 477 // segment overrides 478 CS_segment = 0x2e, 479 SS_segment = 0x36, 480 DS_segment = 0x3e, 481 ES_segment = 0x26, 482 FS_segment = 0x64, 483 GS_segment = 0x65, 484 485 REX = 0x40, 486 487 REX_B = 0x41, 488 REX_X = 0x42, 489 REX_XB = 0x43, 490 REX_R = 0x44, 491 REX_RB = 0x45, 492 REX_RX = 0x46, 493 REX_RXB = 0x47, 494 495 REX_W = 0x48, 496 497 REX_WB = 0x49, 498 REX_WX = 0x4A, 499 REX_WXB = 0x4B, 500 REX_WR = 0x4C, 501 REX_WRB = 0x4D, 502 REX_WRX = 0x4E, 503 REX_WRXB = 0x4F, 504 505 VEX_3bytes = 0xC4, 506 VEX_2bytes = 0xC5, 507 EVEX_4bytes = 0x62 508 }; 509 510 enum VexPrefix { 511 VEX_B = 0x20, 512 VEX_X = 0x40, 513 VEX_R = 0x80, 514 VEX_W = 0x80 515 }; 516 517 enum ExexPrefix { 518 EVEX_F = 0x04, 519 EVEX_V = 0x08, 520 EVEX_Rb = 0x10, 521 EVEX_X = 0x40, 522 EVEX_Z = 0x80 523 }; 524 525 enum VexSimdPrefix { 526 VEX_SIMD_NONE = 0x0, 527 VEX_SIMD_66 = 0x1, 528 VEX_SIMD_F3 = 0x2, 529 VEX_SIMD_F2 = 0x3 530 }; 531 532 enum VexOpcode { 533 VEX_OPCODE_NONE = 0x0, 534 VEX_OPCODE_0F = 0x1, 535 VEX_OPCODE_0F_38 = 0x2, 536 VEX_OPCODE_0F_3A = 0x3 537 }; 538 539 enum AvxVectorLen { 540 AVX_128bit = 0x0, 541 AVX_256bit = 0x1, 542 AVX_512bit = 0x2, 543 AVX_NoVec = 0x4 544 }; 545 546 enum EvexTupleType { 547 EVEX_FV = 0, 548 EVEX_HV = 4, 549 EVEX_FVM = 6, 550 EVEX_T1S = 7, 551 EVEX_T1F = 11, 552 EVEX_T2 = 13, 553 EVEX_T4 = 15, 554 EVEX_T8 = 17, 555 EVEX_HVM = 18, 556 EVEX_QVM = 19, 557 EVEX_OVM = 20, 558 EVEX_M128 = 21, 559 EVEX_DUP = 22, 560 EVEX_ETUP = 23 561 }; 562 563 enum EvexInputSizeInBits { 564 EVEX_8bit = 0, 565 EVEX_16bit = 1, 566 EVEX_32bit = 2, 567 EVEX_64bit = 3 568 }; 569 570 enum WhichOperand { 571 // input to locate_operand, and format code for relocations 572 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 573 disp32_operand = 1, // embedded 32-bit displacement or address 574 call32_operand = 2, // embedded 32-bit self-relative displacement 575 #ifndef _LP64 576 _WhichOperand_limit = 3 577 #else 578 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 579 _WhichOperand_limit = 4 580 #endif 581 }; 582 583 584 585 // NOTE: The general philopsophy of the declarations here is that 64bit versions 586 // of instructions are freely declared without the need for wrapping them an ifdef. 587 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 588 // In the .cpp file the implementations are wrapped so that they are dropped out 589 // of the resulting jvm. This is done mostly to keep the footprint of MINIMAL 590 // to the size it was prior to merging up the 32bit and 64bit assemblers. 591 // 592 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 593 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 594 595 private: 596 597 int evex_encoding; 598 int input_size_in_bits; 599 int avx_vector_len; 600 int tuple_type; 601 bool is_evex_instruction; 602 603 // 64bit prefixes 604 int prefix_and_encode(int reg_enc, bool byteinst = false); 605 int prefixq_and_encode(int reg_enc); 606 607 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); 608 int prefixq_and_encode(int dst_enc, int src_enc); 609 610 void prefix(Register reg); 611 void prefix(Address adr); 612 void prefixq(Address adr); 613 614 void prefix(Address adr, Register reg, bool byteinst = false); 615 void prefix(Address adr, XMMRegister reg); 616 void prefixq(Address adr, Register reg); 617 void prefixq(Address adr, XMMRegister reg); 618 619 void prefetch_prefix(Address src); 620 621 void rex_prefix(Address adr, XMMRegister xreg, 622 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 623 int rex_prefix_and_encode(int dst_enc, int src_enc, 624 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 625 626 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, 627 int nds_enc, VexSimdPrefix pre, VexOpcode opc, 628 int vector_len); 629 630 void evex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, bool evex_r, bool evex_v, 631 int nds_enc, VexSimdPrefix pre, VexOpcode opc, 632 bool is_extended_context, bool is_merge_context, 633 int vector_len, bool no_mask_reg ); 634 635 void vex_prefix(Address adr, int nds_enc, int xreg_enc, 636 VexSimdPrefix pre, VexOpcode opc, 637 bool vex_w, int vector_len, 638 bool legacy_mode = false, bool no_mask_reg = false); 639 640 void vex_prefix(XMMRegister dst, XMMRegister nds, Address src, 641 VexSimdPrefix pre, int vector_len = AVX_128bit, 642 bool no_mask_reg = false, bool legacy_mode = false) { 643 int dst_enc = dst->encoding(); 644 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 645 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, false, vector_len, legacy_mode, no_mask_reg); 646 } 647 648 void vex_prefix_q(XMMRegister dst, XMMRegister nds, Address src, 649 VexSimdPrefix pre, int vector_len = AVX_128bit, 650 bool no_mask_reg = false) { 651 int dst_enc = dst->encoding(); 652 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 653 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, true, vector_len, false, no_mask_reg); 654 } 655 656 void vex_prefix_0F38(Register dst, Register nds, Address src, bool no_mask_reg = false) { 657 bool vex_w = false; 658 int vector_len = AVX_128bit; 659 vex_prefix(src, nds->encoding(), dst->encoding(), 660 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 661 vector_len, no_mask_reg); 662 } 663 664 void vex_prefix_0F38_legacy(Register dst, Register nds, Address src, bool no_mask_reg = false) { 665 bool vex_w = false; 666 int vector_len = AVX_128bit; 667 vex_prefix(src, nds->encoding(), dst->encoding(), 668 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 669 vector_len, true, no_mask_reg); 670 } 671 672 void vex_prefix_0F38_q(Register dst, Register nds, Address src, bool no_mask_reg = false) { 673 bool vex_w = true; 674 int vector_len = AVX_128bit; 675 vex_prefix(src, nds->encoding(), dst->encoding(), 676 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 677 vector_len, no_mask_reg); 678 } 679 680 void vex_prefix_0F38_q_legacy(Register dst, Register nds, Address src, bool no_mask_reg = false) { 681 bool vex_w = true; 682 int vector_len = AVX_128bit; 683 vex_prefix(src, nds->encoding(), dst->encoding(), 684 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, 685 vector_len, true, no_mask_reg); 686 } 687 688 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, 689 VexSimdPrefix pre, VexOpcode opc, 690 bool vex_w, int vector_len, 691 bool legacy_mode, bool no_mask_reg); 692 693 int vex_prefix_0F38_and_encode(Register dst, Register nds, Register src, bool no_mask_reg = false) { 694 bool vex_w = false; 695 int vector_len = AVX_128bit; 696 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 697 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 698 false, no_mask_reg); 699 } 700 701 int vex_prefix_0F38_and_encode_legacy(Register dst, Register nds, Register src, bool no_mask_reg = false) { 702 bool vex_w = false; 703 int vector_len = AVX_128bit; 704 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 705 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 706 true, no_mask_reg); 707 } 708 709 int vex_prefix_0F38_and_encode_q(Register dst, Register nds, Register src, bool no_mask_reg = false) { 710 bool vex_w = true; 711 int vector_len = AVX_128bit; 712 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 713 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 714 false, no_mask_reg); 715 } 716 717 int vex_prefix_0F38_and_encode_q_legacy(Register dst, Register nds, Register src, bool no_mask_reg = false) { 718 bool vex_w = true; 719 int vector_len = AVX_128bit; 720 return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), 721 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len, 722 true, no_mask_reg); 723 } 724 725 int vex_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 726 VexSimdPrefix pre, int vector_len = AVX_128bit, 727 VexOpcode opc = VEX_OPCODE_0F, bool legacy_mode = false, 728 bool no_mask_reg = false) { 729 int src_enc = src->encoding(); 730 int dst_enc = dst->encoding(); 731 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 732 return vex_prefix_and_encode(dst_enc, nds_enc, src_enc, pre, opc, false, vector_len, legacy_mode, no_mask_reg); 733 } 734 735 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, 736 VexSimdPrefix pre, bool no_mask_reg, VexOpcode opc = VEX_OPCODE_0F, 737 bool rex_w = false, int vector_len = AVX_128bit, bool legacy_mode = false); 738 739 void simd_prefix(XMMRegister dst, Address src, VexSimdPrefix pre, 740 bool no_mask_reg, VexOpcode opc = VEX_OPCODE_0F) { 741 simd_prefix(dst, xnoreg, src, pre, no_mask_reg, opc); 742 } 743 744 void simd_prefix(Address dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg) { 745 simd_prefix(src, dst, pre, no_mask_reg); 746 } 747 void simd_prefix_q(XMMRegister dst, XMMRegister nds, Address src, 748 VexSimdPrefix pre, bool no_mask_reg = false) { 749 bool rex_w = true; 750 simd_prefix(dst, nds, src, pre, no_mask_reg, VEX_OPCODE_0F, rex_w); 751 } 752 753 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 754 VexSimdPrefix pre, bool no_mask_reg, 755 VexOpcode opc = VEX_OPCODE_0F, 756 bool rex_w = false, int vector_len = AVX_128bit, 757 bool legacy_mode = false); 758 759 int kreg_prefix_and_encode(KRegister dst, KRegister nds, KRegister src, 760 VexSimdPrefix pre, bool no_mask_reg, 761 VexOpcode opc = VEX_OPCODE_0F, 762 bool rex_w = false, int vector_len = AVX_128bit); 763 764 int kreg_prefix_and_encode(KRegister dst, KRegister nds, Register src, 765 VexSimdPrefix pre, bool no_mask_reg, 766 VexOpcode opc = VEX_OPCODE_0F, 767 bool rex_w = false, int vector_len = AVX_128bit); 768 769 // Move/convert 32-bit integer value. 770 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, Register src, 771 VexSimdPrefix pre, bool no_mask_reg) { 772 // It is OK to cast from Register to XMMRegister to pass argument here 773 // since only encoding is used in simd_prefix_and_encode() and number of 774 // Gen and Xmm registers are the same. 775 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, no_mask_reg, VEX_OPCODE_0F); 776 } 777 int simd_prefix_and_encode(XMMRegister dst, Register src, VexSimdPrefix pre, bool no_mask_reg) { 778 return simd_prefix_and_encode(dst, xnoreg, src, pre, no_mask_reg); 779 } 780 int simd_prefix_and_encode(Register dst, XMMRegister src, 781 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 782 bool no_mask_reg = false) { 783 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, no_mask_reg, opc); 784 } 785 786 // Move/convert 64-bit integer value. 787 int simd_prefix_and_encode_q(XMMRegister dst, XMMRegister nds, Register src, 788 VexSimdPrefix pre, bool no_mask_reg = false) { 789 bool rex_w = true; 790 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, no_mask_reg, VEX_OPCODE_0F, rex_w); 791 } 792 int simd_prefix_and_encode_q(XMMRegister dst, Register src, VexSimdPrefix pre, bool no_mask_reg) { 793 return simd_prefix_and_encode_q(dst, xnoreg, src, pre, no_mask_reg); 794 } 795 int simd_prefix_and_encode_q(Register dst, XMMRegister src, 796 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 797 bool no_mask_reg = false) { 798 bool rex_w = true; 799 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, no_mask_reg, opc, rex_w); 800 } 801 802 // Helper functions for groups of instructions 803 void emit_arith_b(int op1, int op2, Register dst, int imm8); 804 805 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 806 // Force generation of a 4 byte immediate value even if it fits into 8bit 807 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32); 808 void emit_arith(int op1, int op2, Register dst, Register src); 809 810 void emit_simd_arith(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false, bool legacy_mode = false); 811 void emit_simd_arith_q(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 812 void emit_simd_arith(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false, bool legacy_mode = false); 813 void emit_simd_arith_q(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 814 void emit_simd_arith_nonds(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 815 void emit_simd_arith_nonds_q(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre, bool no_mask_reg = false); 816 void emit_simd_arith_nonds(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false, bool legacy_mode = false); 817 void emit_simd_arith_nonds_q(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre, bool no_mask_reg = false); 818 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 819 Address src, VexSimdPrefix pre, int vector_len, 820 bool no_mask_reg = false, bool legacy_mode = false); 821 void emit_vex_arith_q(int opcode, XMMRegister dst, XMMRegister nds, 822 Address src, VexSimdPrefix pre, int vector_len, 823 bool no_mask_reg = false); 824 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 825 XMMRegister src, VexSimdPrefix pre, int vector_len, 826 bool no_mask_reg = false, bool legacy_mode = false); 827 void emit_vex_arith_q(int opcode, XMMRegister dst, XMMRegister nds, 828 XMMRegister src, VexSimdPrefix pre, int vector_len, 829 bool no_mask_reg = false); 830 831 bool emit_compressed_disp_byte(int &disp); 832 833 void emit_operand(Register reg, 834 Register base, Register index, Address::ScaleFactor scale, 835 int disp, 836 RelocationHolder const& rspec, 837 int rip_relative_correction = 0); 838 839 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 840 841 // operands that only take the original 32bit registers 842 void emit_operand32(Register reg, Address adr); 843 844 void emit_operand(XMMRegister reg, 845 Register base, Register index, Address::ScaleFactor scale, 846 int disp, 847 RelocationHolder const& rspec); 848 849 void emit_operand(XMMRegister reg, Address adr); 850 851 void emit_operand(MMXRegister reg, Address adr); 852 853 // workaround gcc (3.2.1-7) bug 854 void emit_operand(Address adr, MMXRegister reg); 855 856 857 // Immediate-to-memory forms 858 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 859 860 void emit_farith(int b1, int b2, int i); 861 862 863 protected: 864 #ifdef ASSERT 865 void check_relocation(RelocationHolder const& rspec, int format); 866 #endif 867 868 void emit_data(jint data, relocInfo::relocType rtype, int format); 869 void emit_data(jint data, RelocationHolder const& rspec, int format); 870 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 871 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 872 873 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 874 875 // These are all easily abused and hence protected 876 877 // 32BIT ONLY SECTION 878 #ifndef _LP64 879 // Make these disappear in 64bit mode since they would never be correct 880 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 881 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 882 883 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 884 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 885 886 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 887 #else 888 // 64BIT ONLY SECTION 889 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 890 891 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 892 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 893 894 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 895 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 896 #endif // _LP64 897 898 // These are unique in that we are ensured by the caller that the 32bit 899 // relative in these instructions will always be able to reach the potentially 900 // 64bit address described by entry. Since they can take a 64bit address they 901 // don't have the 32 suffix like the other instructions in this class. 902 903 void call_literal(address entry, RelocationHolder const& rspec); 904 void jmp_literal(address entry, RelocationHolder const& rspec); 905 906 // Avoid using directly section 907 // Instructions in this section are actually usable by anyone without danger 908 // of failure but have performance issues that are addressed my enhanced 909 // instructions which will do the proper thing base on the particular cpu. 910 // We protect them because we don't trust you... 911 912 // Don't use next inc() and dec() methods directly. INC & DEC instructions 913 // could cause a partial flag stall since they don't set CF flag. 914 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 915 // which call inc() & dec() or add() & sub() in accordance with 916 // the product flag UseIncDec value. 917 918 void decl(Register dst); 919 void decl(Address dst); 920 void decq(Register dst); 921 void decq(Address dst); 922 923 void incl(Register dst); 924 void incl(Address dst); 925 void incq(Register dst); 926 void incq(Address dst); 927 928 // New cpus require use of movsd and movss to avoid partial register stall 929 // when loading from memory. But for old Opteron use movlpd instead of movsd. 930 // The selection is done in MacroAssembler::movdbl() and movflt(). 931 932 // Move Scalar Single-Precision Floating-Point Values 933 void movss(XMMRegister dst, Address src); 934 void movss(XMMRegister dst, XMMRegister src); 935 void movss(Address dst, XMMRegister src); 936 937 // Move Scalar Double-Precision Floating-Point Values 938 void movsd(XMMRegister dst, Address src); 939 void movsd(XMMRegister dst, XMMRegister src); 940 void movsd(Address dst, XMMRegister src); 941 void movlpd(XMMRegister dst, Address src); 942 943 // New cpus require use of movaps and movapd to avoid partial register stall 944 // when moving between registers. 945 void movaps(XMMRegister dst, XMMRegister src); 946 void movapd(XMMRegister dst, XMMRegister src); 947 948 // End avoid using directly 949 950 951 // Instruction prefixes 952 void prefix(Prefix p); 953 954 public: 955 956 // Creation 957 Assembler(CodeBuffer* code) : AbstractAssembler(code) { 958 init_attributes(); 959 } 960 961 // Decoding 962 static address locate_operand(address inst, WhichOperand which); 963 static address locate_next_instruction(address inst); 964 965 // Utilities 966 static bool is_polling_page_far() NOT_LP64({ return false;}); 967 static bool query_compressed_disp_byte(int disp, bool is_evex_inst, int vector_len, 968 int cur_tuple_type, int in_size_in_bits, int cur_encoding); 969 970 // Generic instructions 971 // Does 32bit or 64bit as needed for the platform. In some sense these 972 // belong in macro assembler but there is no need for both varieties to exist 973 974 void init_attributes(void) { 975 evex_encoding = 0; 976 input_size_in_bits = 0; 977 avx_vector_len = AVX_NoVec; 978 tuple_type = EVEX_ETUP; 979 is_evex_instruction = false; 980 } 981 982 void lea(Register dst, Address src); 983 984 void mov(Register dst, Register src); 985 986 void pusha(); 987 void popa(); 988 989 void pushf(); 990 void popf(); 991 992 void push(int32_t imm32); 993 994 void push(Register src); 995 996 void pop(Register dst); 997 998 // These are dummies to prevent surprise implicit conversions to Register 999 void push(void* v); 1000 void pop(void* v); 1001 1002 // These do register sized moves/scans 1003 void rep_mov(); 1004 void rep_stos(); 1005 void rep_stosb(); 1006 void repne_scan(); 1007 #ifdef _LP64 1008 void repne_scanl(); 1009 #endif 1010 1011 // Vanilla instructions in lexical order 1012 1013 void adcl(Address dst, int32_t imm32); 1014 void adcl(Address dst, Register src); 1015 void adcl(Register dst, int32_t imm32); 1016 void adcl(Register dst, Address src); 1017 void adcl(Register dst, Register src); 1018 1019 void adcq(Register dst, int32_t imm32); 1020 void adcq(Register dst, Address src); 1021 void adcq(Register dst, Register src); 1022 1023 void addl(Address dst, int32_t imm32); 1024 void addl(Address dst, Register src); 1025 void addl(Register dst, int32_t imm32); 1026 void addl(Register dst, Address src); 1027 void addl(Register dst, Register src); 1028 1029 void addq(Address dst, int32_t imm32); 1030 void addq(Address dst, Register src); 1031 void addq(Register dst, int32_t imm32); 1032 void addq(Register dst, Address src); 1033 void addq(Register dst, Register src); 1034 1035 #ifdef _LP64 1036 //Add Unsigned Integers with Carry Flag 1037 void adcxq(Register dst, Register src); 1038 1039 //Add Unsigned Integers with Overflow Flag 1040 void adoxq(Register dst, Register src); 1041 #endif 1042 1043 void addr_nop_4(); 1044 void addr_nop_5(); 1045 void addr_nop_7(); 1046 void addr_nop_8(); 1047 1048 // Add Scalar Double-Precision Floating-Point Values 1049 void addsd(XMMRegister dst, Address src); 1050 void addsd(XMMRegister dst, XMMRegister src); 1051 1052 // Add Scalar Single-Precision Floating-Point Values 1053 void addss(XMMRegister dst, Address src); 1054 void addss(XMMRegister dst, XMMRegister src); 1055 1056 // AES instructions 1057 void aesdec(XMMRegister dst, Address src); 1058 void aesdec(XMMRegister dst, XMMRegister src); 1059 void aesdeclast(XMMRegister dst, Address src); 1060 void aesdeclast(XMMRegister dst, XMMRegister src); 1061 void aesenc(XMMRegister dst, Address src); 1062 void aesenc(XMMRegister dst, XMMRegister src); 1063 void aesenclast(XMMRegister dst, Address src); 1064 void aesenclast(XMMRegister dst, XMMRegister src); 1065 1066 1067 void andl(Address dst, int32_t imm32); 1068 void andl(Register dst, int32_t imm32); 1069 void andl(Register dst, Address src); 1070 void andl(Register dst, Register src); 1071 1072 void andq(Address dst, int32_t imm32); 1073 void andq(Register dst, int32_t imm32); 1074 void andq(Register dst, Address src); 1075 void andq(Register dst, Register src); 1076 1077 // BMI instructions 1078 void andnl(Register dst, Register src1, Register src2); 1079 void andnl(Register dst, Register src1, Address src2); 1080 void andnq(Register dst, Register src1, Register src2); 1081 void andnq(Register dst, Register src1, Address src2); 1082 1083 void blsil(Register dst, Register src); 1084 void blsil(Register dst, Address src); 1085 void blsiq(Register dst, Register src); 1086 void blsiq(Register dst, Address src); 1087 1088 void blsmskl(Register dst, Register src); 1089 void blsmskl(Register dst, Address src); 1090 void blsmskq(Register dst, Register src); 1091 void blsmskq(Register dst, Address src); 1092 1093 void blsrl(Register dst, Register src); 1094 void blsrl(Register dst, Address src); 1095 void blsrq(Register dst, Register src); 1096 void blsrq(Register dst, Address src); 1097 1098 void bsfl(Register dst, Register src); 1099 void bsrl(Register dst, Register src); 1100 1101 #ifdef _LP64 1102 void bsfq(Register dst, Register src); 1103 void bsrq(Register dst, Register src); 1104 #endif 1105 1106 void bswapl(Register reg); 1107 1108 void bswapq(Register reg); 1109 1110 void call(Label& L, relocInfo::relocType rtype); 1111 void call(Register reg); // push pc; pc <- reg 1112 void call(Address adr); // push pc; pc <- adr 1113 1114 void cdql(); 1115 1116 void cdqq(); 1117 1118 void cld(); 1119 1120 void clflush(Address adr); 1121 1122 void cmovl(Condition cc, Register dst, Register src); 1123 void cmovl(Condition cc, Register dst, Address src); 1124 1125 void cmovq(Condition cc, Register dst, Register src); 1126 void cmovq(Condition cc, Register dst, Address src); 1127 1128 1129 void cmpb(Address dst, int imm8); 1130 1131 void cmpl(Address dst, int32_t imm32); 1132 1133 void cmpl(Register dst, int32_t imm32); 1134 void cmpl(Register dst, Register src); 1135 void cmpl(Register dst, Address src); 1136 1137 void cmpq(Address dst, int32_t imm32); 1138 void cmpq(Address dst, Register src); 1139 1140 void cmpq(Register dst, int32_t imm32); 1141 void cmpq(Register dst, Register src); 1142 void cmpq(Register dst, Address src); 1143 1144 // these are dummies used to catch attempting to convert NULL to Register 1145 void cmpl(Register dst, void* junk); // dummy 1146 void cmpq(Register dst, void* junk); // dummy 1147 1148 void cmpw(Address dst, int imm16); 1149 1150 void cmpxchg8 (Address adr); 1151 1152 void cmpxchgb(Register reg, Address adr); 1153 void cmpxchgl(Register reg, Address adr); 1154 1155 void cmpxchgq(Register reg, Address adr); 1156 1157 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1158 void comisd(XMMRegister dst, Address src); 1159 void comisd(XMMRegister dst, XMMRegister src); 1160 1161 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1162 void comiss(XMMRegister dst, Address src); 1163 void comiss(XMMRegister dst, XMMRegister src); 1164 1165 // Identify processor type and features 1166 void cpuid(); 1167 1168 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 1169 void cvtsd2ss(XMMRegister dst, XMMRegister src); 1170 void cvtsd2ss(XMMRegister dst, Address src); 1171 1172 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 1173 void cvtsi2sdl(XMMRegister dst, Register src); 1174 void cvtsi2sdl(XMMRegister dst, Address src); 1175 void cvtsi2sdq(XMMRegister dst, Register src); 1176 void cvtsi2sdq(XMMRegister dst, Address src); 1177 1178 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 1179 void cvtsi2ssl(XMMRegister dst, Register src); 1180 void cvtsi2ssl(XMMRegister dst, Address src); 1181 void cvtsi2ssq(XMMRegister dst, Register src); 1182 void cvtsi2ssq(XMMRegister dst, Address src); 1183 1184 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 1185 void cvtdq2pd(XMMRegister dst, XMMRegister src); 1186 1187 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 1188 void cvtdq2ps(XMMRegister dst, XMMRegister src); 1189 1190 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 1191 void cvtss2sd(XMMRegister dst, XMMRegister src); 1192 void cvtss2sd(XMMRegister dst, Address src); 1193 1194 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 1195 void cvttsd2sil(Register dst, Address src); 1196 void cvttsd2sil(Register dst, XMMRegister src); 1197 void cvttsd2siq(Register dst, XMMRegister src); 1198 1199 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 1200 void cvttss2sil(Register dst, XMMRegister src); 1201 void cvttss2siq(Register dst, XMMRegister src); 1202 1203 // Divide Scalar Double-Precision Floating-Point Values 1204 void divsd(XMMRegister dst, Address src); 1205 void divsd(XMMRegister dst, XMMRegister src); 1206 1207 // Divide Scalar Single-Precision Floating-Point Values 1208 void divss(XMMRegister dst, Address src); 1209 void divss(XMMRegister dst, XMMRegister src); 1210 1211 void emms(); 1212 1213 void fabs(); 1214 1215 void fadd(int i); 1216 1217 void fadd_d(Address src); 1218 void fadd_s(Address src); 1219 1220 // "Alternate" versions of x87 instructions place result down in FPU 1221 // stack instead of on TOS 1222 1223 void fadda(int i); // "alternate" fadd 1224 void faddp(int i = 1); 1225 1226 void fchs(); 1227 1228 void fcom(int i); 1229 1230 void fcomp(int i = 1); 1231 void fcomp_d(Address src); 1232 void fcomp_s(Address src); 1233 1234 void fcompp(); 1235 1236 void fcos(); 1237 1238 void fdecstp(); 1239 1240 void fdiv(int i); 1241 void fdiv_d(Address src); 1242 void fdivr_s(Address src); 1243 void fdiva(int i); // "alternate" fdiv 1244 void fdivp(int i = 1); 1245 1246 void fdivr(int i); 1247 void fdivr_d(Address src); 1248 void fdiv_s(Address src); 1249 1250 void fdivra(int i); // "alternate" reversed fdiv 1251 1252 void fdivrp(int i = 1); 1253 1254 void ffree(int i = 0); 1255 1256 void fild_d(Address adr); 1257 void fild_s(Address adr); 1258 1259 void fincstp(); 1260 1261 void finit(); 1262 1263 void fist_s (Address adr); 1264 void fistp_d(Address adr); 1265 void fistp_s(Address adr); 1266 1267 void fld1(); 1268 1269 void fld_d(Address adr); 1270 void fld_s(Address adr); 1271 void fld_s(int index); 1272 void fld_x(Address adr); // extended-precision (80-bit) format 1273 1274 void fldcw(Address src); 1275 1276 void fldenv(Address src); 1277 1278 void fldlg2(); 1279 1280 void fldln2(); 1281 1282 void fldz(); 1283 1284 void flog(); 1285 void flog10(); 1286 1287 void fmul(int i); 1288 1289 void fmul_d(Address src); 1290 void fmul_s(Address src); 1291 1292 void fmula(int i); // "alternate" fmul 1293 1294 void fmulp(int i = 1); 1295 1296 void fnsave(Address dst); 1297 1298 void fnstcw(Address src); 1299 1300 void fnstsw_ax(); 1301 1302 void fprem(); 1303 void fprem1(); 1304 1305 void frstor(Address src); 1306 1307 void fsin(); 1308 1309 void fsqrt(); 1310 1311 void fst_d(Address adr); 1312 void fst_s(Address adr); 1313 1314 void fstp_d(Address adr); 1315 void fstp_d(int index); 1316 void fstp_s(Address adr); 1317 void fstp_x(Address adr); // extended-precision (80-bit) format 1318 1319 void fsub(int i); 1320 void fsub_d(Address src); 1321 void fsub_s(Address src); 1322 1323 void fsuba(int i); // "alternate" fsub 1324 1325 void fsubp(int i = 1); 1326 1327 void fsubr(int i); 1328 void fsubr_d(Address src); 1329 void fsubr_s(Address src); 1330 1331 void fsubra(int i); // "alternate" reversed fsub 1332 1333 void fsubrp(int i = 1); 1334 1335 void ftan(); 1336 1337 void ftst(); 1338 1339 void fucomi(int i = 1); 1340 void fucomip(int i = 1); 1341 1342 void fwait(); 1343 1344 void fxch(int i = 1); 1345 1346 void fxrstor(Address src); 1347 1348 void fxsave(Address dst); 1349 1350 void fyl2x(); 1351 void frndint(); 1352 void f2xm1(); 1353 void fldl2e(); 1354 1355 void hlt(); 1356 1357 void idivl(Register src); 1358 void divl(Register src); // Unsigned division 1359 1360 #ifdef _LP64 1361 void idivq(Register src); 1362 #endif 1363 1364 void imull(Register dst, Register src); 1365 void imull(Register dst, Register src, int value); 1366 void imull(Register dst, Address src); 1367 1368 #ifdef _LP64 1369 void imulq(Register dst, Register src); 1370 void imulq(Register dst, Register src, int value); 1371 void imulq(Register dst, Address src); 1372 #endif 1373 1374 // jcc is the generic conditional branch generator to run- 1375 // time routines, jcc is used for branches to labels. jcc 1376 // takes a branch opcode (cc) and a label (L) and generates 1377 // either a backward branch or a forward branch and links it 1378 // to the label fixup chain. Usage: 1379 // 1380 // Label L; // unbound label 1381 // jcc(cc, L); // forward branch to unbound label 1382 // bind(L); // bind label to the current pc 1383 // jcc(cc, L); // backward branch to bound label 1384 // bind(L); // illegal: a label may be bound only once 1385 // 1386 // Note: The same Label can be used for forward and backward branches 1387 // but it may be bound only once. 1388 1389 void jcc(Condition cc, Label& L, bool maybe_short = true); 1390 1391 // Conditional jump to a 8-bit offset to L. 1392 // WARNING: be very careful using this for forward jumps. If the label is 1393 // not bound within an 8-bit offset of this instruction, a run-time error 1394 // will occur. 1395 void jccb(Condition cc, Label& L); 1396 1397 void jmp(Address entry); // pc <- entry 1398 1399 // Label operations & relative jumps (PPUM Appendix D) 1400 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L 1401 1402 void jmp(Register entry); // pc <- entry 1403 1404 // Unconditional 8-bit offset jump to L. 1405 // WARNING: be very careful using this for forward jumps. If the label is 1406 // not bound within an 8-bit offset of this instruction, a run-time error 1407 // will occur. 1408 void jmpb(Label& L); 1409 1410 void ldmxcsr( Address src ); 1411 1412 void leal(Register dst, Address src); 1413 1414 void leaq(Register dst, Address src); 1415 1416 void lfence(); 1417 1418 void lock(); 1419 1420 void lzcntl(Register dst, Register src); 1421 1422 #ifdef _LP64 1423 void lzcntq(Register dst, Register src); 1424 #endif 1425 1426 enum Membar_mask_bits { 1427 StoreStore = 1 << 3, 1428 LoadStore = 1 << 2, 1429 StoreLoad = 1 << 1, 1430 LoadLoad = 1 << 0 1431 }; 1432 1433 // Serializes memory and blows flags 1434 void membar(Membar_mask_bits order_constraint) { 1435 if (os::is_MP()) { 1436 // We only have to handle StoreLoad 1437 if (order_constraint & StoreLoad) { 1438 // All usable chips support "locked" instructions which suffice 1439 // as barriers, and are much faster than the alternative of 1440 // using cpuid instruction. We use here a locked add [esp-C],0. 1441 // This is conveniently otherwise a no-op except for blowing 1442 // flags, and introducing a false dependency on target memory 1443 // location. We can't do anything with flags, but we can avoid 1444 // memory dependencies in the current method by locked-adding 1445 // somewhere else on the stack. Doing [esp+C] will collide with 1446 // something on stack in current method, hence we go for [esp-C]. 1447 // It is convenient since it is almost always in data cache, for 1448 // any small C. We need to step back from SP to avoid data 1449 // dependencies with other things on below SP (callee-saves, for 1450 // example). Without a clear way to figure out the minimal safe 1451 // distance from SP, it makes sense to step back the complete 1452 // cache line, as this will also avoid possible second-order effects 1453 // with locked ops against the cache line. Our choice of offset 1454 // is bounded by x86 operand encoding, which should stay within 1455 // [-128; +127] to have the 8-byte displacement encoding. 1456 // 1457 // Any change to this code may need to revisit other places in 1458 // the code where this idiom is used, in particular the 1459 // orderAccess code. 1460 1461 int offset = -VM_Version::L1_line_size(); 1462 if (offset < -128) { 1463 offset = -128; 1464 } 1465 1466 lock(); 1467 addl(Address(rsp, offset), 0);// Assert the lock# signal here 1468 } 1469 } 1470 } 1471 1472 void mfence(); 1473 1474 // Moves 1475 1476 void mov64(Register dst, int64_t imm64); 1477 1478 void movb(Address dst, Register src); 1479 void movb(Address dst, int imm8); 1480 void movb(Register dst, Address src); 1481 1482 void kmovq(KRegister dst, KRegister src); 1483 void kmovql(KRegister dst, Register src); 1484 void kmovdl(KRegister dst, Register src); 1485 void kmovq(Address dst, KRegister src); 1486 void kmovq(KRegister dst, Address src); 1487 1488 void movdl(XMMRegister dst, Register src); 1489 void movdl(Register dst, XMMRegister src); 1490 void movdl(XMMRegister dst, Address src); 1491 void movdl(Address dst, XMMRegister src); 1492 1493 // Move Double Quadword 1494 void movdq(XMMRegister dst, Register src); 1495 void movdq(Register dst, XMMRegister src); 1496 1497 // Move Aligned Double Quadword 1498 void movdqa(XMMRegister dst, XMMRegister src); 1499 void movdqa(XMMRegister dst, Address src); 1500 1501 // Move Unaligned Double Quadword 1502 void movdqu(Address dst, XMMRegister src); 1503 void movdqu(XMMRegister dst, Address src); 1504 void movdqu(XMMRegister dst, XMMRegister src); 1505 1506 // Move Unaligned 256bit Vector 1507 void vmovdqu(Address dst, XMMRegister src); 1508 void vmovdqu(XMMRegister dst, Address src); 1509 void vmovdqu(XMMRegister dst, XMMRegister src); 1510 1511 // Move Unaligned 512bit Vector 1512 void evmovdqu(Address dst, XMMRegister src, int vector_len); 1513 void evmovdqu(XMMRegister dst, Address src, int vector_len); 1514 void evmovdqu(XMMRegister dst, XMMRegister src, int vector_len); 1515 1516 // Move lower 64bit to high 64bit in 128bit register 1517 void movlhps(XMMRegister dst, XMMRegister src); 1518 1519 void movl(Register dst, int32_t imm32); 1520 void movl(Address dst, int32_t imm32); 1521 void movl(Register dst, Register src); 1522 void movl(Register dst, Address src); 1523 void movl(Address dst, Register src); 1524 1525 // These dummies prevent using movl from converting a zero (like NULL) into Register 1526 // by giving the compiler two choices it can't resolve 1527 1528 void movl(Address dst, void* junk); 1529 void movl(Register dst, void* junk); 1530 1531 #ifdef _LP64 1532 void movq(Register dst, Register src); 1533 void movq(Register dst, Address src); 1534 void movq(Address dst, Register src); 1535 #endif 1536 1537 void movq(Address dst, MMXRegister src ); 1538 void movq(MMXRegister dst, Address src ); 1539 1540 #ifdef _LP64 1541 // These dummies prevent using movq from converting a zero (like NULL) into Register 1542 // by giving the compiler two choices it can't resolve 1543 1544 void movq(Address dst, void* dummy); 1545 void movq(Register dst, void* dummy); 1546 #endif 1547 1548 // Move Quadword 1549 void movq(Address dst, XMMRegister src); 1550 void movq(XMMRegister dst, Address src); 1551 1552 void movsbl(Register dst, Address src); 1553 void movsbl(Register dst, Register src); 1554 1555 #ifdef _LP64 1556 void movsbq(Register dst, Address src); 1557 void movsbq(Register dst, Register src); 1558 1559 // Move signed 32bit immediate to 64bit extending sign 1560 void movslq(Address dst, int32_t imm64); 1561 void movslq(Register dst, int32_t imm64); 1562 1563 void movslq(Register dst, Address src); 1564 void movslq(Register dst, Register src); 1565 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1566 #endif 1567 1568 void movswl(Register dst, Address src); 1569 void movswl(Register dst, Register src); 1570 1571 #ifdef _LP64 1572 void movswq(Register dst, Address src); 1573 void movswq(Register dst, Register src); 1574 #endif 1575 1576 void movw(Address dst, int imm16); 1577 void movw(Register dst, Address src); 1578 void movw(Address dst, Register src); 1579 1580 void movzbl(Register dst, Address src); 1581 void movzbl(Register dst, Register src); 1582 1583 #ifdef _LP64 1584 void movzbq(Register dst, Address src); 1585 void movzbq(Register dst, Register src); 1586 #endif 1587 1588 void movzwl(Register dst, Address src); 1589 void movzwl(Register dst, Register src); 1590 1591 #ifdef _LP64 1592 void movzwq(Register dst, Address src); 1593 void movzwq(Register dst, Register src); 1594 #endif 1595 1596 // Unsigned multiply with RAX destination register 1597 void mull(Address src); 1598 void mull(Register src); 1599 1600 #ifdef _LP64 1601 void mulq(Address src); 1602 void mulq(Register src); 1603 void mulxq(Register dst1, Register dst2, Register src); 1604 #endif 1605 1606 // Multiply Scalar Double-Precision Floating-Point Values 1607 void mulsd(XMMRegister dst, Address src); 1608 void mulsd(XMMRegister dst, XMMRegister src); 1609 1610 // Multiply Scalar Single-Precision Floating-Point Values 1611 void mulss(XMMRegister dst, Address src); 1612 void mulss(XMMRegister dst, XMMRegister src); 1613 1614 void negl(Register dst); 1615 1616 #ifdef _LP64 1617 void negq(Register dst); 1618 #endif 1619 1620 void nop(int i = 1); 1621 1622 void notl(Register dst); 1623 1624 #ifdef _LP64 1625 void notq(Register dst); 1626 #endif 1627 1628 void orl(Address dst, int32_t imm32); 1629 void orl(Register dst, int32_t imm32); 1630 void orl(Register dst, Address src); 1631 void orl(Register dst, Register src); 1632 void orl(Address dst, Register src); 1633 1634 void orq(Address dst, int32_t imm32); 1635 void orq(Register dst, int32_t imm32); 1636 void orq(Register dst, Address src); 1637 void orq(Register dst, Register src); 1638 1639 // Pack with unsigned saturation 1640 void packuswb(XMMRegister dst, XMMRegister src); 1641 void packuswb(XMMRegister dst, Address src); 1642 void vpackuswb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1643 1644 // Pemutation of 64bit words 1645 void vpermq(XMMRegister dst, XMMRegister src, int imm8, int vector_len); 1646 1647 void pause(); 1648 1649 // SSE4.2 string instructions 1650 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1651 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1652 1653 // SSE 4.1 extract 1654 void pextrd(Register dst, XMMRegister src, int imm8); 1655 void pextrq(Register dst, XMMRegister src, int imm8); 1656 1657 // SSE 4.1 insert 1658 void pinsrd(XMMRegister dst, Register src, int imm8); 1659 void pinsrq(XMMRegister dst, Register src, int imm8); 1660 1661 // SSE4.1 packed move 1662 void pmovzxbw(XMMRegister dst, XMMRegister src); 1663 void pmovzxbw(XMMRegister dst, Address src); 1664 1665 #ifndef _LP64 // no 32bit push/pop on amd64 1666 void popl(Address dst); 1667 #endif 1668 1669 #ifdef _LP64 1670 void popq(Address dst); 1671 #endif 1672 1673 void popcntl(Register dst, Address src); 1674 void popcntl(Register dst, Register src); 1675 1676 #ifdef _LP64 1677 void popcntq(Register dst, Address src); 1678 void popcntq(Register dst, Register src); 1679 #endif 1680 1681 // Prefetches (SSE, SSE2, 3DNOW only) 1682 1683 void prefetchnta(Address src); 1684 void prefetchr(Address src); 1685 void prefetcht0(Address src); 1686 void prefetcht1(Address src); 1687 void prefetcht2(Address src); 1688 void prefetchw(Address src); 1689 1690 // Shuffle Bytes 1691 void pshufb(XMMRegister dst, XMMRegister src); 1692 void pshufb(XMMRegister dst, Address src); 1693 1694 // Shuffle Packed Doublewords 1695 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1696 void pshufd(XMMRegister dst, Address src, int mode); 1697 1698 // Shuffle Packed Low Words 1699 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1700 void pshuflw(XMMRegister dst, Address src, int mode); 1701 1702 // Shift Right by bytes Logical DoubleQuadword Immediate 1703 void psrldq(XMMRegister dst, int shift); 1704 // Shift Left by bytes Logical DoubleQuadword Immediate 1705 void pslldq(XMMRegister dst, int shift); 1706 1707 // Logical Compare 128bit 1708 void ptest(XMMRegister dst, XMMRegister src); 1709 void ptest(XMMRegister dst, Address src); 1710 // Logical Compare 256bit 1711 void vptest(XMMRegister dst, XMMRegister src); 1712 void vptest(XMMRegister dst, Address src); 1713 1714 // Interleave Low Bytes 1715 void punpcklbw(XMMRegister dst, XMMRegister src); 1716 void punpcklbw(XMMRegister dst, Address src); 1717 1718 // Interleave Low Doublewords 1719 void punpckldq(XMMRegister dst, XMMRegister src); 1720 void punpckldq(XMMRegister dst, Address src); 1721 1722 // Interleave Low Quadwords 1723 void punpcklqdq(XMMRegister dst, XMMRegister src); 1724 1725 #ifndef _LP64 // no 32bit push/pop on amd64 1726 void pushl(Address src); 1727 #endif 1728 1729 void pushq(Address src); 1730 1731 void rcll(Register dst, int imm8); 1732 1733 void rclq(Register dst, int imm8); 1734 1735 void rcrq(Register dst, int imm8); 1736 1737 void rdtsc(); 1738 1739 void ret(int imm16); 1740 1741 #ifdef _LP64 1742 void rorq(Register dst, int imm8); 1743 void rorxq(Register dst, Register src, int imm8); 1744 #endif 1745 1746 void sahf(); 1747 1748 void sarl(Register dst, int imm8); 1749 void sarl(Register dst); 1750 1751 void sarq(Register dst, int imm8); 1752 void sarq(Register dst); 1753 1754 void sbbl(Address dst, int32_t imm32); 1755 void sbbl(Register dst, int32_t imm32); 1756 void sbbl(Register dst, Address src); 1757 void sbbl(Register dst, Register src); 1758 1759 void sbbq(Address dst, int32_t imm32); 1760 void sbbq(Register dst, int32_t imm32); 1761 void sbbq(Register dst, Address src); 1762 void sbbq(Register dst, Register src); 1763 1764 void setb(Condition cc, Register dst); 1765 1766 void shldl(Register dst, Register src); 1767 1768 void shll(Register dst, int imm8); 1769 void shll(Register dst); 1770 1771 void shlq(Register dst, int imm8); 1772 void shlq(Register dst); 1773 1774 void shrdl(Register dst, Register src); 1775 1776 void shrl(Register dst, int imm8); 1777 void shrl(Register dst); 1778 1779 void shrq(Register dst, int imm8); 1780 void shrq(Register dst); 1781 1782 void smovl(); // QQQ generic? 1783 1784 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1785 void sqrtsd(XMMRegister dst, Address src); 1786 void sqrtsd(XMMRegister dst, XMMRegister src); 1787 1788 // Compute Square Root of Scalar Single-Precision Floating-Point Value 1789 void sqrtss(XMMRegister dst, Address src); 1790 void sqrtss(XMMRegister dst, XMMRegister src); 1791 1792 void std(); 1793 1794 void stmxcsr( Address dst ); 1795 1796 void subl(Address dst, int32_t imm32); 1797 void subl(Address dst, Register src); 1798 void subl(Register dst, int32_t imm32); 1799 void subl(Register dst, Address src); 1800 void subl(Register dst, Register src); 1801 1802 void subq(Address dst, int32_t imm32); 1803 void subq(Address dst, Register src); 1804 void subq(Register dst, int32_t imm32); 1805 void subq(Register dst, Address src); 1806 void subq(Register dst, Register src); 1807 1808 // Force generation of a 4 byte immediate value even if it fits into 8bit 1809 void subl_imm32(Register dst, int32_t imm32); 1810 void subq_imm32(Register dst, int32_t imm32); 1811 1812 // Subtract Scalar Double-Precision Floating-Point Values 1813 void subsd(XMMRegister dst, Address src); 1814 void subsd(XMMRegister dst, XMMRegister src); 1815 1816 // Subtract Scalar Single-Precision Floating-Point Values 1817 void subss(XMMRegister dst, Address src); 1818 void subss(XMMRegister dst, XMMRegister src); 1819 1820 void testb(Register dst, int imm8); 1821 1822 void testl(Register dst, int32_t imm32); 1823 void testl(Register dst, Register src); 1824 void testl(Register dst, Address src); 1825 1826 void testq(Register dst, int32_t imm32); 1827 void testq(Register dst, Register src); 1828 1829 // BMI - count trailing zeros 1830 void tzcntl(Register dst, Register src); 1831 void tzcntq(Register dst, Register src); 1832 1833 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1834 void ucomisd(XMMRegister dst, Address src); 1835 void ucomisd(XMMRegister dst, XMMRegister src); 1836 1837 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1838 void ucomiss(XMMRegister dst, Address src); 1839 void ucomiss(XMMRegister dst, XMMRegister src); 1840 1841 void xabort(int8_t imm8); 1842 1843 void xaddl(Address dst, Register src); 1844 1845 void xaddq(Address dst, Register src); 1846 1847 void xbegin(Label& abort, relocInfo::relocType rtype = relocInfo::none); 1848 1849 void xchgl(Register reg, Address adr); 1850 void xchgl(Register dst, Register src); 1851 1852 void xchgq(Register reg, Address adr); 1853 void xchgq(Register dst, Register src); 1854 1855 void xend(); 1856 1857 // Get Value of Extended Control Register 1858 void xgetbv(); 1859 1860 void xorl(Register dst, int32_t imm32); 1861 void xorl(Register dst, Address src); 1862 void xorl(Register dst, Register src); 1863 1864 void xorq(Register dst, Address src); 1865 void xorq(Register dst, Register src); 1866 1867 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1868 1869 // AVX 3-operands scalar instructions (encoded with VEX prefix) 1870 1871 void vaddsd(XMMRegister dst, XMMRegister nds, Address src); 1872 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1873 void vaddss(XMMRegister dst, XMMRegister nds, Address src); 1874 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1875 void vdivsd(XMMRegister dst, XMMRegister nds, Address src); 1876 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1877 void vdivss(XMMRegister dst, XMMRegister nds, Address src); 1878 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1879 void vmulsd(XMMRegister dst, XMMRegister nds, Address src); 1880 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1881 void vmulss(XMMRegister dst, XMMRegister nds, Address src); 1882 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1883 void vsubsd(XMMRegister dst, XMMRegister nds, Address src); 1884 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1885 void vsubss(XMMRegister dst, XMMRegister nds, Address src); 1886 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1887 1888 1889 //====================VECTOR ARITHMETIC===================================== 1890 1891 // Add Packed Floating-Point Values 1892 void addpd(XMMRegister dst, XMMRegister src); 1893 void addps(XMMRegister dst, XMMRegister src); 1894 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1895 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1896 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1897 void vaddps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1898 1899 // Subtract Packed Floating-Point Values 1900 void subpd(XMMRegister dst, XMMRegister src); 1901 void subps(XMMRegister dst, XMMRegister src); 1902 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1903 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1904 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1905 void vsubps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1906 1907 // Multiply Packed Floating-Point Values 1908 void mulpd(XMMRegister dst, XMMRegister src); 1909 void mulps(XMMRegister dst, XMMRegister src); 1910 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1911 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1912 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1913 void vmulps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1914 1915 // Divide Packed Floating-Point Values 1916 void divpd(XMMRegister dst, XMMRegister src); 1917 void divps(XMMRegister dst, XMMRegister src); 1918 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1919 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1920 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1921 void vdivps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1922 1923 // Bitwise Logical AND of Packed Floating-Point Values 1924 void andpd(XMMRegister dst, XMMRegister src); 1925 void andps(XMMRegister dst, XMMRegister src); 1926 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1927 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1928 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1929 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1930 1931 // Bitwise Logical XOR of Packed Floating-Point Values 1932 void xorpd(XMMRegister dst, XMMRegister src); 1933 void xorps(XMMRegister dst, XMMRegister src); 1934 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1935 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1936 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1937 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1938 1939 // Add horizontal packed integers 1940 void vphaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1941 void vphaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1942 void phaddw(XMMRegister dst, XMMRegister src); 1943 void phaddd(XMMRegister dst, XMMRegister src); 1944 1945 // Add packed integers 1946 void paddb(XMMRegister dst, XMMRegister src); 1947 void paddw(XMMRegister dst, XMMRegister src); 1948 void paddd(XMMRegister dst, XMMRegister src); 1949 void paddq(XMMRegister dst, XMMRegister src); 1950 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1951 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1952 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1953 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1954 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1955 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1956 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1957 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1958 1959 // Sub packed integers 1960 void psubb(XMMRegister dst, XMMRegister src); 1961 void psubw(XMMRegister dst, XMMRegister src); 1962 void psubd(XMMRegister dst, XMMRegister src); 1963 void psubq(XMMRegister dst, XMMRegister src); 1964 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1965 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1966 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1967 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1968 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1969 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1970 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1971 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1972 1973 // Multiply packed integers (only shorts and ints) 1974 void pmullw(XMMRegister dst, XMMRegister src); 1975 void pmulld(XMMRegister dst, XMMRegister src); 1976 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1977 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1978 void vpmullq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1979 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1980 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1981 void vpmullq(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1982 1983 // Shift left packed integers 1984 void psllw(XMMRegister dst, int shift); 1985 void pslld(XMMRegister dst, int shift); 1986 void psllq(XMMRegister dst, int shift); 1987 void psllw(XMMRegister dst, XMMRegister shift); 1988 void pslld(XMMRegister dst, XMMRegister shift); 1989 void psllq(XMMRegister dst, XMMRegister shift); 1990 void vpsllw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1991 void vpslld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1992 void vpsllq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 1993 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1994 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1995 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 1996 1997 // Logical shift right packed integers 1998 void psrlw(XMMRegister dst, int shift); 1999 void psrld(XMMRegister dst, int shift); 2000 void psrlq(XMMRegister dst, int shift); 2001 void psrlw(XMMRegister dst, XMMRegister shift); 2002 void psrld(XMMRegister dst, XMMRegister shift); 2003 void psrlq(XMMRegister dst, XMMRegister shift); 2004 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2005 void vpsrld(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2006 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2007 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2008 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2009 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2010 2011 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs) 2012 void psraw(XMMRegister dst, int shift); 2013 void psrad(XMMRegister dst, int shift); 2014 void psraw(XMMRegister dst, XMMRegister shift); 2015 void psrad(XMMRegister dst, XMMRegister shift); 2016 void vpsraw(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2017 void vpsrad(XMMRegister dst, XMMRegister src, int shift, int vector_len); 2018 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2019 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, int vector_len); 2020 2021 // And packed integers 2022 void pand(XMMRegister dst, XMMRegister src); 2023 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2024 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2025 2026 // Or packed integers 2027 void por(XMMRegister dst, XMMRegister src); 2028 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2029 void vpor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2030 2031 // Xor packed integers 2032 void pxor(XMMRegister dst, XMMRegister src); 2033 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 2034 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 2035 2036 // Copy low 128bit into high 128bit of YMM registers. 2037 void vinsertf128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2038 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2039 void vextractf128h(XMMRegister dst, XMMRegister src); 2040 void vextracti128h(XMMRegister dst, XMMRegister src); 2041 2042 // Load/store high 128bit of YMM registers which does not destroy other half. 2043 void vinsertf128h(XMMRegister dst, Address src); 2044 void vinserti128h(XMMRegister dst, Address src); 2045 void vextractf128h(Address dst, XMMRegister src); 2046 void vextracti128h(Address dst, XMMRegister src); 2047 2048 // Copy low 256bit into high 256bit of ZMM registers. 2049 void vinserti64x4h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2050 void vinsertf64x4h(XMMRegister dst, XMMRegister nds, XMMRegister src); 2051 void vextracti64x4h(XMMRegister dst, XMMRegister src); 2052 void vextractf64x4h(XMMRegister dst, XMMRegister src); 2053 void vextractf64x4h(Address dst, XMMRegister src); 2054 void vinsertf64x4h(XMMRegister dst, Address src); 2055 2056 // Copy targeted 128bit segments of the ZMM registers 2057 void vextracti64x2h(XMMRegister dst, XMMRegister src, int value); 2058 void vextractf64x2h(XMMRegister dst, XMMRegister src, int value); 2059 void vextractf32x4h(XMMRegister dst, XMMRegister src, int value); 2060 2061 // duplicate 4-bytes integer data from src into 8 locations in dest 2062 void vpbroadcastd(XMMRegister dst, XMMRegister src); 2063 2064 // duplicate n-bytes integer data from src into vector_len locations in dest 2065 void evpbroadcastb(XMMRegister dst, XMMRegister src, int vector_len); 2066 void evpbroadcastb(XMMRegister dst, Address src, int vector_len); 2067 void evpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len); 2068 void evpbroadcastw(XMMRegister dst, Address src, int vector_len); 2069 void evpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len); 2070 void evpbroadcastd(XMMRegister dst, Address src, int vector_len); 2071 void evpbroadcastq(XMMRegister dst, XMMRegister src, int vector_len); 2072 void evpbroadcastq(XMMRegister dst, Address src, int vector_len); 2073 2074 void evpbroadcastss(XMMRegister dst, XMMRegister src, int vector_len); 2075 void evpbroadcastss(XMMRegister dst, Address src, int vector_len); 2076 void evpbroadcastsd(XMMRegister dst, XMMRegister src, int vector_len); 2077 void evpbroadcastsd(XMMRegister dst, Address src, int vector_len); 2078 2079 void evpbroadcastb(XMMRegister dst, Register src, int vector_len); 2080 void evpbroadcastw(XMMRegister dst, Register src, int vector_len); 2081 void evpbroadcastd(XMMRegister dst, Register src, int vector_len); 2082 void evpbroadcastq(XMMRegister dst, Register src, int vector_len); 2083 2084 // Carry-Less Multiplication Quadword 2085 void pclmulqdq(XMMRegister dst, XMMRegister src, int mask); 2086 void vpclmulqdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int mask); 2087 2088 // AVX instruction which is used to clear upper 128 bits of YMM registers and 2089 // to avoid transaction penalty between AVX and SSE states. There is no 2090 // penalty if legacy SSE instructions are encoded using VEX prefix because 2091 // they always clear upper 128 bits. It should be used before calling 2092 // runtime code and native libraries. 2093 void vzeroupper(); 2094 2095 protected: 2096 // Next instructions require address alignment 16 bytes SSE mode. 2097 // They should be called only from corresponding MacroAssembler instructions. 2098 void andpd(XMMRegister dst, Address src); 2099 void andps(XMMRegister dst, Address src); 2100 void xorpd(XMMRegister dst, Address src); 2101 void xorps(XMMRegister dst, Address src); 2102 2103 }; 2104 2105 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP