1 /* 2 * Copyright (c) 1997, 2012, 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 class BiasedLockingCounters; 29 30 // Contains all the definitions needed for x86 assembly code generation. 31 32 // Calling convention 33 class Argument VALUE_OBJ_CLASS_SPEC { 34 public: 35 enum { 36 #ifdef _LP64 37 #ifdef _WIN64 38 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) 39 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) 40 #else 41 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) 42 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) 43 #endif // _WIN64 44 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... 45 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... 46 #else 47 n_register_parameters = 0 // 0 registers used to pass arguments 48 #endif // _LP64 49 }; 50 }; 51 52 53 #ifdef _LP64 54 // Symbolically name the register arguments used by the c calling convention. 55 // Windows is different from linux/solaris. So much for standards... 56 57 #ifdef _WIN64 58 59 REGISTER_DECLARATION(Register, c_rarg0, rcx); 60 REGISTER_DECLARATION(Register, c_rarg1, rdx); 61 REGISTER_DECLARATION(Register, c_rarg2, r8); 62 REGISTER_DECLARATION(Register, c_rarg3, r9); 63 64 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 65 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 66 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 67 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 68 69 #else 70 71 REGISTER_DECLARATION(Register, c_rarg0, rdi); 72 REGISTER_DECLARATION(Register, c_rarg1, rsi); 73 REGISTER_DECLARATION(Register, c_rarg2, rdx); 74 REGISTER_DECLARATION(Register, c_rarg3, rcx); 75 REGISTER_DECLARATION(Register, c_rarg4, r8); 76 REGISTER_DECLARATION(Register, c_rarg5, r9); 77 78 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 79 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 80 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 81 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 82 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); 83 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); 84 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); 85 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); 86 87 #endif // _WIN64 88 89 // Symbolically name the register arguments used by the Java calling convention. 90 // We have control over the convention for java so we can do what we please. 91 // What pleases us is to offset the java calling convention so that when 92 // we call a suitable jni method the arguments are lined up and we don't 93 // have to do little shuffling. A suitable jni method is non-static and a 94 // small number of arguments (two fewer args on windows) 95 // 96 // |-------------------------------------------------------| 97 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | 98 // |-------------------------------------------------------| 99 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) 100 // | rdi rsi rdx rcx r8 r9 | solaris/linux 101 // |-------------------------------------------------------| 102 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | 103 // |-------------------------------------------------------| 104 105 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); 106 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); 107 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); 108 // Windows runs out of register args here 109 #ifdef _WIN64 110 REGISTER_DECLARATION(Register, j_rarg3, rdi); 111 REGISTER_DECLARATION(Register, j_rarg4, rsi); 112 #else 113 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); 114 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); 115 #endif /* _WIN64 */ 116 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); 117 118 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); 119 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); 120 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); 121 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); 122 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); 123 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); 124 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); 125 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); 126 127 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile 128 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile 129 130 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved 131 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved 132 133 #else 134 // rscratch1 will apear in 32bit code that is dead but of course must compile 135 // Using noreg ensures if the dead code is incorrectly live and executed it 136 // will cause an assertion failure 137 #define rscratch1 noreg 138 #define rscratch2 noreg 139 140 #endif // _LP64 141 142 // JSR 292 fixed register usages: 143 REGISTER_DECLARATION(Register, rbp_mh_SP_save, rbp); 144 145 // Address is an abstraction used to represent a memory location 146 // using any of the amd64 addressing modes with one object. 147 // 148 // Note: A register location is represented via a Register, not 149 // via an address for efficiency & simplicity reasons. 150 151 class ArrayAddress; 152 153 class Address VALUE_OBJ_CLASS_SPEC { 154 public: 155 enum ScaleFactor { 156 no_scale = -1, 157 times_1 = 0, 158 times_2 = 1, 159 times_4 = 2, 160 times_8 = 3, 161 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) 162 }; 163 static ScaleFactor times(int size) { 164 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size"); 165 if (size == 8) return times_8; 166 if (size == 4) return times_4; 167 if (size == 2) return times_2; 168 return times_1; 169 } 170 static int scale_size(ScaleFactor scale) { 171 assert(scale != no_scale, ""); 172 assert(((1 << (int)times_1) == 1 && 173 (1 << (int)times_2) == 2 && 174 (1 << (int)times_4) == 4 && 175 (1 << (int)times_8) == 8), ""); 176 return (1 << (int)scale); 177 } 178 179 private: 180 Register _base; 181 Register _index; 182 ScaleFactor _scale; 183 int _disp; 184 RelocationHolder _rspec; 185 186 // Easily misused constructors make them private 187 // %%% can we make these go away? 188 NOT_LP64(Address(address loc, RelocationHolder spec);) 189 Address(int disp, address loc, relocInfo::relocType rtype); 190 Address(int disp, address loc, RelocationHolder spec); 191 192 public: 193 194 int disp() { return _disp; } 195 // creation 196 Address() 197 : _base(noreg), 198 _index(noreg), 199 _scale(no_scale), 200 _disp(0) { 201 } 202 203 // No default displacement otherwise Register can be implicitly 204 // converted to 0(Register) which is quite a different animal. 205 206 Address(Register base, int disp) 207 : _base(base), 208 _index(noreg), 209 _scale(no_scale), 210 _disp(disp) { 211 } 212 213 Address(Register base, Register index, ScaleFactor scale, int disp = 0) 214 : _base (base), 215 _index(index), 216 _scale(scale), 217 _disp (disp) { 218 assert(!index->is_valid() == (scale == Address::no_scale), 219 "inconsistent address"); 220 } 221 222 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) 223 : _base (base), 224 _index(index.register_or_noreg()), 225 _scale(scale), 226 _disp (disp + (index.constant_or_zero() * scale_size(scale))) { 227 if (!index.is_register()) scale = Address::no_scale; 228 assert(!_index->is_valid() == (scale == Address::no_scale), 229 "inconsistent address"); 230 } 231 232 Address plus_disp(int disp) const { 233 Address a = (*this); 234 a._disp += disp; 235 return a; 236 } 237 Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const { 238 Address a = (*this); 239 a._disp += disp.constant_or_zero() * scale_size(scale); 240 if (disp.is_register()) { 241 assert(!a.index()->is_valid(), "competing indexes"); 242 a._index = disp.as_register(); 243 a._scale = scale; 244 } 245 return a; 246 } 247 bool is_same_address(Address a) const { 248 // disregard _rspec 249 return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale; 250 } 251 252 // The following two overloads are used in connection with the 253 // ByteSize type (see sizes.hpp). They simplify the use of 254 // ByteSize'd arguments in assembly code. Note that their equivalent 255 // for the optimized build are the member functions with int disp 256 // argument since ByteSize is mapped to an int type in that case. 257 // 258 // Note: DO NOT introduce similar overloaded functions for WordSize 259 // arguments as in the optimized mode, both ByteSize and WordSize 260 // are mapped to the same type and thus the compiler cannot make a 261 // distinction anymore (=> compiler errors). 262 263 #ifdef ASSERT 264 Address(Register base, ByteSize disp) 265 : _base(base), 266 _index(noreg), 267 _scale(no_scale), 268 _disp(in_bytes(disp)) { 269 } 270 271 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 272 : _base(base), 273 _index(index), 274 _scale(scale), 275 _disp(in_bytes(disp)) { 276 assert(!index->is_valid() == (scale == Address::no_scale), 277 "inconsistent address"); 278 } 279 280 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 281 : _base (base), 282 _index(index.register_or_noreg()), 283 _scale(scale), 284 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { 285 if (!index.is_register()) scale = Address::no_scale; 286 assert(!_index->is_valid() == (scale == Address::no_scale), 287 "inconsistent address"); 288 } 289 290 #endif // ASSERT 291 292 // accessors 293 bool uses(Register reg) const { return _base == reg || _index == reg; } 294 Register base() const { return _base; } 295 Register index() const { return _index; } 296 ScaleFactor scale() const { return _scale; } 297 int disp() const { return _disp; } 298 299 // Convert the raw encoding form into the form expected by the constructor for 300 // Address. An index of 4 (rsp) corresponds to having no index, so convert 301 // that to noreg for the Address constructor. 302 static Address make_raw(int base, int index, int scale, int disp, relocInfo::relocType disp_reloc); 303 304 static Address make_array(ArrayAddress); 305 306 private: 307 bool base_needs_rex() const { 308 return _base != noreg && _base->encoding() >= 8; 309 } 310 311 bool index_needs_rex() const { 312 return _index != noreg &&_index->encoding() >= 8; 313 } 314 315 relocInfo::relocType reloc() const { return _rspec.type(); } 316 317 friend class Assembler; 318 friend class MacroAssembler; 319 friend class LIR_Assembler; // base/index/scale/disp 320 }; 321 322 // 323 // AddressLiteral has been split out from Address because operands of this type 324 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 325 // the few instructions that need to deal with address literals are unique and the 326 // MacroAssembler does not have to implement every instruction in the Assembler 327 // in order to search for address literals that may need special handling depending 328 // on the instruction and the platform. As small step on the way to merging i486/amd64 329 // directories. 330 // 331 class AddressLiteral VALUE_OBJ_CLASS_SPEC { 332 friend class ArrayAddress; 333 RelocationHolder _rspec; 334 // Typically we use AddressLiterals we want to use their rval 335 // However in some situations we want the lval (effect address) of the item. 336 // We provide a special factory for making those lvals. 337 bool _is_lval; 338 339 // If the target is far we'll need to load the ea of this to 340 // a register to reach it. Otherwise if near we can do rip 341 // relative addressing. 342 343 address _target; 344 345 protected: 346 // creation 347 AddressLiteral() 348 : _is_lval(false), 349 _target(NULL) 350 {} 351 352 public: 353 354 355 AddressLiteral(address target, relocInfo::relocType rtype); 356 357 AddressLiteral(address target, RelocationHolder const& rspec) 358 : _rspec(rspec), 359 _is_lval(false), 360 _target(target) 361 {} 362 363 AddressLiteral addr() { 364 AddressLiteral ret = *this; 365 ret._is_lval = true; 366 return ret; 367 } 368 369 370 private: 371 372 address target() { return _target; } 373 bool is_lval() { return _is_lval; } 374 375 relocInfo::relocType reloc() const { return _rspec.type(); } 376 const RelocationHolder& rspec() const { return _rspec; } 377 378 friend class Assembler; 379 friend class MacroAssembler; 380 friend class Address; 381 friend class LIR_Assembler; 382 }; 383 384 // Convience classes 385 class RuntimeAddress: public AddressLiteral { 386 387 public: 388 389 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 390 391 }; 392 393 class ExternalAddress: public AddressLiteral { 394 private: 395 static relocInfo::relocType reloc_for_target(address target) { 396 // Sometimes ExternalAddress is used for values which aren't 397 // exactly addresses, like the card table base. 398 // external_word_type can't be used for values in the first page 399 // so just skip the reloc in that case. 400 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 401 } 402 403 public: 404 405 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} 406 407 }; 408 409 class InternalAddress: public AddressLiteral { 410 411 public: 412 413 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 414 415 }; 416 417 // x86 can do array addressing as a single operation since disp can be an absolute 418 // address amd64 can't. We create a class that expresses the concept but does extra 419 // magic on amd64 to get the final result 420 421 class ArrayAddress VALUE_OBJ_CLASS_SPEC { 422 private: 423 424 AddressLiteral _base; 425 Address _index; 426 427 public: 428 429 ArrayAddress() {}; 430 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 431 AddressLiteral base() { return _base; } 432 Address index() { return _index; } 433 434 }; 435 436 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); 437 438 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 439 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 440 // is what you get. The Assembler is generating code into a CodeBuffer. 441 442 class Assembler : public AbstractAssembler { 443 friend class AbstractAssembler; // for the non-virtual hack 444 friend class LIR_Assembler; // as_Address() 445 friend class StubGenerator; 446 447 public: 448 enum Condition { // The x86 condition codes used for conditional jumps/moves. 449 zero = 0x4, 450 notZero = 0x5, 451 equal = 0x4, 452 notEqual = 0x5, 453 less = 0xc, 454 lessEqual = 0xe, 455 greater = 0xf, 456 greaterEqual = 0xd, 457 below = 0x2, 458 belowEqual = 0x6, 459 above = 0x7, 460 aboveEqual = 0x3, 461 overflow = 0x0, 462 noOverflow = 0x1, 463 carrySet = 0x2, 464 carryClear = 0x3, 465 negative = 0x8, 466 positive = 0x9, 467 parity = 0xa, 468 noParity = 0xb 469 }; 470 471 enum Prefix { 472 // segment overrides 473 CS_segment = 0x2e, 474 SS_segment = 0x36, 475 DS_segment = 0x3e, 476 ES_segment = 0x26, 477 FS_segment = 0x64, 478 GS_segment = 0x65, 479 480 REX = 0x40, 481 482 REX_B = 0x41, 483 REX_X = 0x42, 484 REX_XB = 0x43, 485 REX_R = 0x44, 486 REX_RB = 0x45, 487 REX_RX = 0x46, 488 REX_RXB = 0x47, 489 490 REX_W = 0x48, 491 492 REX_WB = 0x49, 493 REX_WX = 0x4A, 494 REX_WXB = 0x4B, 495 REX_WR = 0x4C, 496 REX_WRB = 0x4D, 497 REX_WRX = 0x4E, 498 REX_WRXB = 0x4F, 499 500 VEX_3bytes = 0xC4, 501 VEX_2bytes = 0xC5 502 }; 503 504 enum VexPrefix { 505 VEX_B = 0x20, 506 VEX_X = 0x40, 507 VEX_R = 0x80, 508 VEX_W = 0x80 509 }; 510 511 enum VexSimdPrefix { 512 VEX_SIMD_NONE = 0x0, 513 VEX_SIMD_66 = 0x1, 514 VEX_SIMD_F3 = 0x2, 515 VEX_SIMD_F2 = 0x3 516 }; 517 518 enum VexOpcode { 519 VEX_OPCODE_NONE = 0x0, 520 VEX_OPCODE_0F = 0x1, 521 VEX_OPCODE_0F_38 = 0x2, 522 VEX_OPCODE_0F_3A = 0x3 523 }; 524 525 enum WhichOperand { 526 // input to locate_operand, and format code for relocations 527 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 528 disp32_operand = 1, // embedded 32-bit displacement or address 529 call32_operand = 2, // embedded 32-bit self-relative displacement 530 #ifndef _LP64 531 _WhichOperand_limit = 3 532 #else 533 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 534 _WhichOperand_limit = 4 535 #endif 536 }; 537 538 539 540 // NOTE: The general philopsophy of the declarations here is that 64bit versions 541 // of instructions are freely declared without the need for wrapping them an ifdef. 542 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 543 // In the .cpp file the implementations are wrapped so that they are dropped out 544 // of the resulting jvm. This is done mostly to keep the footprint of KERNEL 545 // to the size it was prior to merging up the 32bit and 64bit assemblers. 546 // 547 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 548 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 549 550 private: 551 552 553 // 64bit prefixes 554 int prefix_and_encode(int reg_enc, bool byteinst = false); 555 int prefixq_and_encode(int reg_enc); 556 557 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); 558 int prefixq_and_encode(int dst_enc, int src_enc); 559 560 void prefix(Register reg); 561 void prefix(Address adr); 562 void prefixq(Address adr); 563 564 void prefix(Address adr, Register reg, bool byteinst = false); 565 void prefix(Address adr, XMMRegister reg); 566 void prefixq(Address adr, Register reg); 567 void prefixq(Address adr, XMMRegister reg); 568 569 void prefetch_prefix(Address src); 570 571 void rex_prefix(Address adr, XMMRegister xreg, 572 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 573 int rex_prefix_and_encode(int dst_enc, int src_enc, 574 VexSimdPrefix pre, VexOpcode opc, bool rex_w); 575 576 void vex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w, 577 int nds_enc, VexSimdPrefix pre, VexOpcode opc, 578 bool vector256); 579 580 void vex_prefix(Address adr, int nds_enc, int xreg_enc, 581 VexSimdPrefix pre, VexOpcode opc, 582 bool vex_w, bool vector256); 583 584 void vex_prefix(XMMRegister dst, XMMRegister nds, Address src, 585 VexSimdPrefix pre, bool vector256 = false) { 586 int dst_enc = dst->encoding(); 587 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 588 vex_prefix(src, nds_enc, dst_enc, pre, VEX_OPCODE_0F, false, vector256); 589 } 590 591 int vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, 592 VexSimdPrefix pre, VexOpcode opc, 593 bool vex_w, bool vector256); 594 595 int vex_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 596 VexSimdPrefix pre, bool vector256 = false, 597 VexOpcode opc = VEX_OPCODE_0F) { 598 int src_enc = src->encoding(); 599 int dst_enc = dst->encoding(); 600 int nds_enc = nds->is_valid() ? nds->encoding() : 0; 601 return vex_prefix_and_encode(dst_enc, nds_enc, src_enc, pre, opc, false, vector256); 602 } 603 604 void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr, 605 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 606 bool rex_w = false, bool vector256 = false); 607 608 void simd_prefix(XMMRegister dst, Address src, 609 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { 610 simd_prefix(dst, xnoreg, src, pre, opc); 611 } 612 613 void simd_prefix(Address dst, XMMRegister src, VexSimdPrefix pre) { 614 simd_prefix(src, dst, pre); 615 } 616 void simd_prefix_q(XMMRegister dst, XMMRegister nds, Address src, 617 VexSimdPrefix pre) { 618 bool rex_w = true; 619 simd_prefix(dst, nds, src, pre, VEX_OPCODE_0F, rex_w); 620 } 621 622 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src, 623 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F, 624 bool rex_w = false, bool vector256 = false); 625 626 // Move/convert 32-bit integer value. 627 int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, Register src, 628 VexSimdPrefix pre) { 629 // It is OK to cast from Register to XMMRegister to pass argument here 630 // since only encoding is used in simd_prefix_and_encode() and number of 631 // Gen and Xmm registers are the same. 632 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre); 633 } 634 int simd_prefix_and_encode(XMMRegister dst, Register src, VexSimdPrefix pre) { 635 return simd_prefix_and_encode(dst, xnoreg, src, pre); 636 } 637 int simd_prefix_and_encode(Register dst, XMMRegister src, 638 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { 639 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc); 640 } 641 642 // Move/convert 64-bit integer value. 643 int simd_prefix_and_encode_q(XMMRegister dst, XMMRegister nds, Register src, 644 VexSimdPrefix pre) { 645 bool rex_w = true; 646 return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, VEX_OPCODE_0F, rex_w); 647 } 648 int simd_prefix_and_encode_q(XMMRegister dst, Register src, VexSimdPrefix pre) { 649 return simd_prefix_and_encode_q(dst, xnoreg, src, pre); 650 } 651 int simd_prefix_and_encode_q(Register dst, XMMRegister src, 652 VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) { 653 bool rex_w = true; 654 return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc, rex_w); 655 } 656 657 // Helper functions for groups of instructions 658 void emit_arith_b(int op1, int op2, Register dst, int imm8); 659 660 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 661 // Force generation of a 4 byte immediate value even if it fits into 8bit 662 void emit_arith_imm32(int op1, int op2, Register dst, int32_t imm32); 663 void emit_arith(int op1, int op2, Register dst, Register src); 664 665 void emit_simd_arith(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre); 666 void emit_simd_arith(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre); 667 void emit_simd_arith_nonds(int opcode, XMMRegister dst, Address src, VexSimdPrefix pre); 668 void emit_simd_arith_nonds(int opcode, XMMRegister dst, XMMRegister src, VexSimdPrefix pre); 669 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 670 Address src, VexSimdPrefix pre, bool vector256); 671 void emit_vex_arith(int opcode, XMMRegister dst, XMMRegister nds, 672 XMMRegister src, VexSimdPrefix pre, bool vector256); 673 674 void emit_operand(Register reg, 675 Register base, Register index, Address::ScaleFactor scale, 676 int disp, 677 RelocationHolder const& rspec, 678 int rip_relative_correction = 0); 679 680 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 681 682 // operands that only take the original 32bit registers 683 void emit_operand32(Register reg, Address adr); 684 685 void emit_operand(XMMRegister reg, 686 Register base, Register index, Address::ScaleFactor scale, 687 int disp, 688 RelocationHolder const& rspec); 689 690 void emit_operand(XMMRegister reg, Address adr); 691 692 void emit_operand(MMXRegister reg, Address adr); 693 694 // workaround gcc (3.2.1-7) bug 695 void emit_operand(Address adr, MMXRegister reg); 696 697 698 // Immediate-to-memory forms 699 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 700 701 void emit_farith(int b1, int b2, int i); 702 703 704 protected: 705 #ifdef ASSERT 706 void check_relocation(RelocationHolder const& rspec, int format); 707 #endif 708 709 inline void emit_long64(jlong x); 710 711 void emit_data(jint data, relocInfo::relocType rtype, int format); 712 void emit_data(jint data, RelocationHolder const& rspec, int format); 713 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 714 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 715 716 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 717 718 // These are all easily abused and hence protected 719 720 // 32BIT ONLY SECTION 721 #ifndef _LP64 722 // Make these disappear in 64bit mode since they would never be correct 723 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 724 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 725 726 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 727 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 728 729 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 730 #else 731 // 64BIT ONLY SECTION 732 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 733 734 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 735 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 736 737 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 738 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 739 #endif // _LP64 740 741 // These are unique in that we are ensured by the caller that the 32bit 742 // relative in these instructions will always be able to reach the potentially 743 // 64bit address described by entry. Since they can take a 64bit address they 744 // don't have the 32 suffix like the other instructions in this class. 745 746 void call_literal(address entry, RelocationHolder const& rspec); 747 void jmp_literal(address entry, RelocationHolder const& rspec); 748 749 // Avoid using directly section 750 // Instructions in this section are actually usable by anyone without danger 751 // of failure but have performance issues that are addressed my enhanced 752 // instructions which will do the proper thing base on the particular cpu. 753 // We protect them because we don't trust you... 754 755 // Don't use next inc() and dec() methods directly. INC & DEC instructions 756 // could cause a partial flag stall since they don't set CF flag. 757 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 758 // which call inc() & dec() or add() & sub() in accordance with 759 // the product flag UseIncDec value. 760 761 void decl(Register dst); 762 void decl(Address dst); 763 void decq(Register dst); 764 void decq(Address dst); 765 766 void incl(Register dst); 767 void incl(Address dst); 768 void incq(Register dst); 769 void incq(Address dst); 770 771 // New cpus require use of movsd and movss to avoid partial register stall 772 // when loading from memory. But for old Opteron use movlpd instead of movsd. 773 // The selection is done in MacroAssembler::movdbl() and movflt(). 774 775 // Move Scalar Single-Precision Floating-Point Values 776 void movss(XMMRegister dst, Address src); 777 void movss(XMMRegister dst, XMMRegister src); 778 void movss(Address dst, XMMRegister src); 779 780 // Move Scalar Double-Precision Floating-Point Values 781 void movsd(XMMRegister dst, Address src); 782 void movsd(XMMRegister dst, XMMRegister src); 783 void movsd(Address dst, XMMRegister src); 784 void movlpd(XMMRegister dst, Address src); 785 786 // New cpus require use of movaps and movapd to avoid partial register stall 787 // when moving between registers. 788 void movaps(XMMRegister dst, XMMRegister src); 789 void movapd(XMMRegister dst, XMMRegister src); 790 791 // End avoid using directly 792 793 794 // Instruction prefixes 795 void prefix(Prefix p); 796 797 public: 798 799 // Creation 800 Assembler(CodeBuffer* code) : AbstractAssembler(code) {} 801 802 // Decoding 803 static address locate_operand(address inst, WhichOperand which); 804 static address locate_next_instruction(address inst); 805 806 // Utilities 807 static bool is_polling_page_far() NOT_LP64({ return false;}); 808 809 // Generic instructions 810 // Does 32bit or 64bit as needed for the platform. In some sense these 811 // belong in macro assembler but there is no need for both varieties to exist 812 813 void lea(Register dst, Address src); 814 815 void mov(Register dst, Register src); 816 817 void pusha(); 818 void popa(); 819 820 void pushf(); 821 void popf(); 822 823 void push(int32_t imm32); 824 825 void push(Register src); 826 827 void pop(Register dst); 828 829 // These are dummies to prevent surprise implicit conversions to Register 830 void push(void* v); 831 void pop(void* v); 832 833 // These do register sized moves/scans 834 void rep_mov(); 835 void rep_set(); 836 void repne_scan(); 837 #ifdef _LP64 838 void repne_scanl(); 839 #endif 840 841 // Vanilla instructions in lexical order 842 843 void adcl(Address dst, int32_t imm32); 844 void adcl(Address dst, Register src); 845 void adcl(Register dst, int32_t imm32); 846 void adcl(Register dst, Address src); 847 void adcl(Register dst, Register src); 848 849 void adcq(Register dst, int32_t imm32); 850 void adcq(Register dst, Address src); 851 void adcq(Register dst, Register src); 852 853 void addl(Address dst, int32_t imm32); 854 void addl(Address dst, Register src); 855 void addl(Register dst, int32_t imm32); 856 void addl(Register dst, Address src); 857 void addl(Register dst, Register src); 858 859 void addq(Address dst, int32_t imm32); 860 void addq(Address dst, Register src); 861 void addq(Register dst, int32_t imm32); 862 void addq(Register dst, Address src); 863 void addq(Register dst, Register src); 864 865 void addr_nop_4(); 866 void addr_nop_5(); 867 void addr_nop_7(); 868 void addr_nop_8(); 869 870 // Add Scalar Double-Precision Floating-Point Values 871 void addsd(XMMRegister dst, Address src); 872 void addsd(XMMRegister dst, XMMRegister src); 873 874 // Add Scalar Single-Precision Floating-Point Values 875 void addss(XMMRegister dst, Address src); 876 void addss(XMMRegister dst, XMMRegister src); 877 878 void andl(Address dst, int32_t imm32); 879 void andl(Register dst, int32_t imm32); 880 void andl(Register dst, Address src); 881 void andl(Register dst, Register src); 882 883 void andq(Address dst, int32_t imm32); 884 void andq(Register dst, int32_t imm32); 885 void andq(Register dst, Address src); 886 void andq(Register dst, Register src); 887 888 void bsfl(Register dst, Register src); 889 void bsrl(Register dst, Register src); 890 891 #ifdef _LP64 892 void bsfq(Register dst, Register src); 893 void bsrq(Register dst, Register src); 894 #endif 895 896 void bswapl(Register reg); 897 898 void bswapq(Register reg); 899 900 void call(Label& L, relocInfo::relocType rtype); 901 void call(Register reg); // push pc; pc <- reg 902 void call(Address adr); // push pc; pc <- adr 903 904 void cdql(); 905 906 void cdqq(); 907 908 void cld() { emit_byte(0xfc); } 909 910 void clflush(Address adr); 911 912 void cmovl(Condition cc, Register dst, Register src); 913 void cmovl(Condition cc, Register dst, Address src); 914 915 void cmovq(Condition cc, Register dst, Register src); 916 void cmovq(Condition cc, Register dst, Address src); 917 918 919 void cmpb(Address dst, int imm8); 920 921 void cmpl(Address dst, int32_t imm32); 922 923 void cmpl(Register dst, int32_t imm32); 924 void cmpl(Register dst, Register src); 925 void cmpl(Register dst, Address src); 926 927 void cmpq(Address dst, int32_t imm32); 928 void cmpq(Address dst, Register src); 929 930 void cmpq(Register dst, int32_t imm32); 931 void cmpq(Register dst, Register src); 932 void cmpq(Register dst, Address src); 933 934 // these are dummies used to catch attempting to convert NULL to Register 935 void cmpl(Register dst, void* junk); // dummy 936 void cmpq(Register dst, void* junk); // dummy 937 938 void cmpw(Address dst, int imm16); 939 940 void cmpxchg8 (Address adr); 941 942 void cmpxchgl(Register reg, Address adr); 943 944 void cmpxchgq(Register reg, Address adr); 945 946 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 947 void comisd(XMMRegister dst, Address src); 948 void comisd(XMMRegister dst, XMMRegister src); 949 950 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 951 void comiss(XMMRegister dst, Address src); 952 void comiss(XMMRegister dst, XMMRegister src); 953 954 // Identify processor type and features 955 void cpuid() { 956 emit_byte(0x0F); 957 emit_byte(0xA2); 958 } 959 960 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 961 void cvtsd2ss(XMMRegister dst, XMMRegister src); 962 void cvtsd2ss(XMMRegister dst, Address src); 963 964 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 965 void cvtsi2sdl(XMMRegister dst, Register src); 966 void cvtsi2sdl(XMMRegister dst, Address src); 967 void cvtsi2sdq(XMMRegister dst, Register src); 968 void cvtsi2sdq(XMMRegister dst, Address src); 969 970 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 971 void cvtsi2ssl(XMMRegister dst, Register src); 972 void cvtsi2ssl(XMMRegister dst, Address src); 973 void cvtsi2ssq(XMMRegister dst, Register src); 974 void cvtsi2ssq(XMMRegister dst, Address src); 975 976 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 977 void cvtdq2pd(XMMRegister dst, XMMRegister src); 978 979 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 980 void cvtdq2ps(XMMRegister dst, XMMRegister src); 981 982 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 983 void cvtss2sd(XMMRegister dst, XMMRegister src); 984 void cvtss2sd(XMMRegister dst, Address src); 985 986 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 987 void cvttsd2sil(Register dst, Address src); 988 void cvttsd2sil(Register dst, XMMRegister src); 989 void cvttsd2siq(Register dst, XMMRegister src); 990 991 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 992 void cvttss2sil(Register dst, XMMRegister src); 993 void cvttss2siq(Register dst, XMMRegister src); 994 995 // Divide Scalar Double-Precision Floating-Point Values 996 void divsd(XMMRegister dst, Address src); 997 void divsd(XMMRegister dst, XMMRegister src); 998 999 // Divide Scalar Single-Precision Floating-Point Values 1000 void divss(XMMRegister dst, Address src); 1001 void divss(XMMRegister dst, XMMRegister src); 1002 1003 void emms(); 1004 1005 void fabs(); 1006 1007 void fadd(int i); 1008 1009 void fadd_d(Address src); 1010 void fadd_s(Address src); 1011 1012 // "Alternate" versions of x87 instructions place result down in FPU 1013 // stack instead of on TOS 1014 1015 void fadda(int i); // "alternate" fadd 1016 void faddp(int i = 1); 1017 1018 void fchs(); 1019 1020 void fcom(int i); 1021 1022 void fcomp(int i = 1); 1023 void fcomp_d(Address src); 1024 void fcomp_s(Address src); 1025 1026 void fcompp(); 1027 1028 void fcos(); 1029 1030 void fdecstp(); 1031 1032 void fdiv(int i); 1033 void fdiv_d(Address src); 1034 void fdivr_s(Address src); 1035 void fdiva(int i); // "alternate" fdiv 1036 void fdivp(int i = 1); 1037 1038 void fdivr(int i); 1039 void fdivr_d(Address src); 1040 void fdiv_s(Address src); 1041 1042 void fdivra(int i); // "alternate" reversed fdiv 1043 1044 void fdivrp(int i = 1); 1045 1046 void ffree(int i = 0); 1047 1048 void fild_d(Address adr); 1049 void fild_s(Address adr); 1050 1051 void fincstp(); 1052 1053 void finit(); 1054 1055 void fist_s (Address adr); 1056 void fistp_d(Address adr); 1057 void fistp_s(Address adr); 1058 1059 void fld1(); 1060 1061 void fld_d(Address adr); 1062 void fld_s(Address adr); 1063 void fld_s(int index); 1064 void fld_x(Address adr); // extended-precision (80-bit) format 1065 1066 void fldcw(Address src); 1067 1068 void fldenv(Address src); 1069 1070 void fldlg2(); 1071 1072 void fldln2(); 1073 1074 void fldz(); 1075 1076 void flog(); 1077 void flog10(); 1078 1079 void fmul(int i); 1080 1081 void fmul_d(Address src); 1082 void fmul_s(Address src); 1083 1084 void fmula(int i); // "alternate" fmul 1085 1086 void fmulp(int i = 1); 1087 1088 void fnsave(Address dst); 1089 1090 void fnstcw(Address src); 1091 1092 void fnstsw_ax(); 1093 1094 void fprem(); 1095 void fprem1(); 1096 1097 void frstor(Address src); 1098 1099 void fsin(); 1100 1101 void fsqrt(); 1102 1103 void fst_d(Address adr); 1104 void fst_s(Address adr); 1105 1106 void fstp_d(Address adr); 1107 void fstp_d(int index); 1108 void fstp_s(Address adr); 1109 void fstp_x(Address adr); // extended-precision (80-bit) format 1110 1111 void fsub(int i); 1112 void fsub_d(Address src); 1113 void fsub_s(Address src); 1114 1115 void fsuba(int i); // "alternate" fsub 1116 1117 void fsubp(int i = 1); 1118 1119 void fsubr(int i); 1120 void fsubr_d(Address src); 1121 void fsubr_s(Address src); 1122 1123 void fsubra(int i); // "alternate" reversed fsub 1124 1125 void fsubrp(int i = 1); 1126 1127 void ftan(); 1128 1129 void ftst(); 1130 1131 void fucomi(int i = 1); 1132 void fucomip(int i = 1); 1133 1134 void fwait(); 1135 1136 void fxch(int i = 1); 1137 1138 void fxrstor(Address src); 1139 1140 void fxsave(Address dst); 1141 1142 void fyl2x(); 1143 void frndint(); 1144 void f2xm1(); 1145 void fldl2e(); 1146 1147 void hlt(); 1148 1149 void idivl(Register src); 1150 void divl(Register src); // Unsigned division 1151 1152 void idivq(Register src); 1153 1154 void imull(Register dst, Register src); 1155 void imull(Register dst, Register src, int value); 1156 1157 void imulq(Register dst, Register src); 1158 void imulq(Register dst, Register src, int value); 1159 1160 1161 // jcc is the generic conditional branch generator to run- 1162 // time routines, jcc is used for branches to labels. jcc 1163 // takes a branch opcode (cc) and a label (L) and generates 1164 // either a backward branch or a forward branch and links it 1165 // to the label fixup chain. Usage: 1166 // 1167 // Label L; // unbound label 1168 // jcc(cc, L); // forward branch to unbound label 1169 // bind(L); // bind label to the current pc 1170 // jcc(cc, L); // backward branch to bound label 1171 // bind(L); // illegal: a label may be bound only once 1172 // 1173 // Note: The same Label can be used for forward and backward branches 1174 // but it may be bound only once. 1175 1176 void jcc(Condition cc, Label& L, bool maybe_short = true); 1177 1178 // Conditional jump to a 8-bit offset to L. 1179 // WARNING: be very careful using this for forward jumps. If the label is 1180 // not bound within an 8-bit offset of this instruction, a run-time error 1181 // will occur. 1182 void jccb(Condition cc, Label& L); 1183 1184 void jmp(Address entry); // pc <- entry 1185 1186 // Label operations & relative jumps (PPUM Appendix D) 1187 void jmp(Label& L, bool maybe_short = true); // unconditional jump to L 1188 1189 void jmp(Register entry); // pc <- entry 1190 1191 // Unconditional 8-bit offset jump to L. 1192 // WARNING: be very careful using this for forward jumps. If the label is 1193 // not bound within an 8-bit offset of this instruction, a run-time error 1194 // will occur. 1195 void jmpb(Label& L); 1196 1197 void ldmxcsr( Address src ); 1198 1199 void leal(Register dst, Address src); 1200 1201 void leaq(Register dst, Address src); 1202 1203 void lfence() { 1204 emit_byte(0x0F); 1205 emit_byte(0xAE); 1206 emit_byte(0xE8); 1207 } 1208 1209 void lock(); 1210 1211 void lzcntl(Register dst, Register src); 1212 1213 #ifdef _LP64 1214 void lzcntq(Register dst, Register src); 1215 #endif 1216 1217 enum Membar_mask_bits { 1218 StoreStore = 1 << 3, 1219 LoadStore = 1 << 2, 1220 StoreLoad = 1 << 1, 1221 LoadLoad = 1 << 0 1222 }; 1223 1224 // Serializes memory and blows flags 1225 void membar(Membar_mask_bits order_constraint) { 1226 if (os::is_MP()) { 1227 // We only have to handle StoreLoad 1228 if (order_constraint & StoreLoad) { 1229 // All usable chips support "locked" instructions which suffice 1230 // as barriers, and are much faster than the alternative of 1231 // using cpuid instruction. We use here a locked add [esp],0. 1232 // This is conveniently otherwise a no-op except for blowing 1233 // flags. 1234 // Any change to this code may need to revisit other places in 1235 // the code where this idiom is used, in particular the 1236 // orderAccess code. 1237 lock(); 1238 addl(Address(rsp, 0), 0);// Assert the lock# signal here 1239 } 1240 } 1241 } 1242 1243 void mfence(); 1244 1245 // Moves 1246 1247 void mov64(Register dst, int64_t imm64); 1248 1249 void movb(Address dst, Register src); 1250 void movb(Address dst, int imm8); 1251 void movb(Register dst, Address src); 1252 1253 void movdl(XMMRegister dst, Register src); 1254 void movdl(Register dst, XMMRegister src); 1255 void movdl(XMMRegister dst, Address src); 1256 void movdl(Address dst, XMMRegister src); 1257 1258 // Move Double Quadword 1259 void movdq(XMMRegister dst, Register src); 1260 void movdq(Register dst, XMMRegister src); 1261 1262 // Move Aligned Double Quadword 1263 void movdqa(XMMRegister dst, XMMRegister src); 1264 1265 // Move Unaligned Double Quadword 1266 void movdqu(Address dst, XMMRegister src); 1267 void movdqu(XMMRegister dst, Address src); 1268 void movdqu(XMMRegister dst, XMMRegister src); 1269 1270 // Move Unaligned 256bit Vector 1271 void vmovdqu(Address dst, XMMRegister src); 1272 void vmovdqu(XMMRegister dst, Address src); 1273 void vmovdqu(XMMRegister dst, XMMRegister src); 1274 1275 // Move lower 64bit to high 64bit in 128bit register 1276 void movlhps(XMMRegister dst, XMMRegister src); 1277 1278 void movl(Register dst, int32_t imm32); 1279 void movl(Address dst, int32_t imm32); 1280 void movl(Register dst, Register src); 1281 void movl(Register dst, Address src); 1282 void movl(Address dst, Register src); 1283 1284 // These dummies prevent using movl from converting a zero (like NULL) into Register 1285 // by giving the compiler two choices it can't resolve 1286 1287 void movl(Address dst, void* junk); 1288 void movl(Register dst, void* junk); 1289 1290 #ifdef _LP64 1291 void movq(Register dst, Register src); 1292 void movq(Register dst, Address src); 1293 void movq(Address dst, Register src); 1294 #endif 1295 1296 void movq(Address dst, MMXRegister src ); 1297 void movq(MMXRegister dst, Address src ); 1298 1299 #ifdef _LP64 1300 // These dummies prevent using movq from converting a zero (like NULL) into Register 1301 // by giving the compiler two choices it can't resolve 1302 1303 void movq(Address dst, void* dummy); 1304 void movq(Register dst, void* dummy); 1305 #endif 1306 1307 // Move Quadword 1308 void movq(Address dst, XMMRegister src); 1309 void movq(XMMRegister dst, Address src); 1310 1311 void movsbl(Register dst, Address src); 1312 void movsbl(Register dst, Register src); 1313 1314 #ifdef _LP64 1315 void movsbq(Register dst, Address src); 1316 void movsbq(Register dst, Register src); 1317 1318 // Move signed 32bit immediate to 64bit extending sign 1319 void movslq(Address dst, int32_t imm64); 1320 void movslq(Register dst, int32_t imm64); 1321 1322 void movslq(Register dst, Address src); 1323 void movslq(Register dst, Register src); 1324 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1325 #endif 1326 1327 void movswl(Register dst, Address src); 1328 void movswl(Register dst, Register src); 1329 1330 #ifdef _LP64 1331 void movswq(Register dst, Address src); 1332 void movswq(Register dst, Register src); 1333 #endif 1334 1335 void movw(Address dst, int imm16); 1336 void movw(Register dst, Address src); 1337 void movw(Address dst, Register src); 1338 1339 void movzbl(Register dst, Address src); 1340 void movzbl(Register dst, Register src); 1341 1342 #ifdef _LP64 1343 void movzbq(Register dst, Address src); 1344 void movzbq(Register dst, Register src); 1345 #endif 1346 1347 void movzwl(Register dst, Address src); 1348 void movzwl(Register dst, Register src); 1349 1350 #ifdef _LP64 1351 void movzwq(Register dst, Address src); 1352 void movzwq(Register dst, Register src); 1353 #endif 1354 1355 void mull(Address src); 1356 void mull(Register src); 1357 1358 // Multiply Scalar Double-Precision Floating-Point Values 1359 void mulsd(XMMRegister dst, Address src); 1360 void mulsd(XMMRegister dst, XMMRegister src); 1361 1362 // Multiply Scalar Single-Precision Floating-Point Values 1363 void mulss(XMMRegister dst, Address src); 1364 void mulss(XMMRegister dst, XMMRegister src); 1365 1366 void negl(Register dst); 1367 1368 #ifdef _LP64 1369 void negq(Register dst); 1370 #endif 1371 1372 void nop(int i = 1); 1373 1374 void notl(Register dst); 1375 1376 #ifdef _LP64 1377 void notq(Register dst); 1378 #endif 1379 1380 void orl(Address dst, int32_t imm32); 1381 void orl(Register dst, int32_t imm32); 1382 void orl(Register dst, Address src); 1383 void orl(Register dst, Register src); 1384 1385 void orq(Address dst, int32_t imm32); 1386 void orq(Register dst, int32_t imm32); 1387 void orq(Register dst, Address src); 1388 void orq(Register dst, Register src); 1389 1390 // Pack with unsigned saturation 1391 void packuswb(XMMRegister dst, XMMRegister src); 1392 void packuswb(XMMRegister dst, Address src); 1393 1394 // SSE4.2 string instructions 1395 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1396 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1397 1398 // SSE4.1 packed move 1399 void pmovzxbw(XMMRegister dst, XMMRegister src); 1400 void pmovzxbw(XMMRegister dst, Address src); 1401 1402 #ifndef _LP64 // no 32bit push/pop on amd64 1403 void popl(Address dst); 1404 #endif 1405 1406 #ifdef _LP64 1407 void popq(Address dst); 1408 #endif 1409 1410 void popcntl(Register dst, Address src); 1411 void popcntl(Register dst, Register src); 1412 1413 #ifdef _LP64 1414 void popcntq(Register dst, Address src); 1415 void popcntq(Register dst, Register src); 1416 #endif 1417 1418 // Prefetches (SSE, SSE2, 3DNOW only) 1419 1420 void prefetchnta(Address src); 1421 void prefetchr(Address src); 1422 void prefetcht0(Address src); 1423 void prefetcht1(Address src); 1424 void prefetcht2(Address src); 1425 void prefetchw(Address src); 1426 1427 // Shuffle Packed Doublewords 1428 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1429 void pshufd(XMMRegister dst, Address src, int mode); 1430 1431 // Shuffle Packed Low Words 1432 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1433 void pshuflw(XMMRegister dst, Address src, int mode); 1434 1435 // Shift Right by bytes Logical DoubleQuadword Immediate 1436 void psrldq(XMMRegister dst, int shift); 1437 1438 // Logical Compare Double Quadword 1439 void ptest(XMMRegister dst, XMMRegister src); 1440 void ptest(XMMRegister dst, Address src); 1441 1442 // Interleave Low Bytes 1443 void punpcklbw(XMMRegister dst, XMMRegister src); 1444 void punpcklbw(XMMRegister dst, Address src); 1445 1446 // Interleave Low Doublewords 1447 void punpckldq(XMMRegister dst, XMMRegister src); 1448 void punpckldq(XMMRegister dst, Address src); 1449 1450 // Interleave Low Quadwords 1451 void punpcklqdq(XMMRegister dst, XMMRegister src); 1452 1453 #ifndef _LP64 // no 32bit push/pop on amd64 1454 void pushl(Address src); 1455 #endif 1456 1457 void pushq(Address src); 1458 1459 void rcll(Register dst, int imm8); 1460 1461 void rclq(Register dst, int imm8); 1462 1463 void ret(int imm16); 1464 1465 void sahf(); 1466 1467 void sarl(Register dst, int imm8); 1468 void sarl(Register dst); 1469 1470 void sarq(Register dst, int imm8); 1471 void sarq(Register dst); 1472 1473 void sbbl(Address dst, int32_t imm32); 1474 void sbbl(Register dst, int32_t imm32); 1475 void sbbl(Register dst, Address src); 1476 void sbbl(Register dst, Register src); 1477 1478 void sbbq(Address dst, int32_t imm32); 1479 void sbbq(Register dst, int32_t imm32); 1480 void sbbq(Register dst, Address src); 1481 void sbbq(Register dst, Register src); 1482 1483 void setb(Condition cc, Register dst); 1484 1485 void shldl(Register dst, Register src); 1486 1487 void shll(Register dst, int imm8); 1488 void shll(Register dst); 1489 1490 void shlq(Register dst, int imm8); 1491 void shlq(Register dst); 1492 1493 void shrdl(Register dst, Register src); 1494 1495 void shrl(Register dst, int imm8); 1496 void shrl(Register dst); 1497 1498 void shrq(Register dst, int imm8); 1499 void shrq(Register dst); 1500 1501 void smovl(); // QQQ generic? 1502 1503 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1504 void sqrtsd(XMMRegister dst, Address src); 1505 void sqrtsd(XMMRegister dst, XMMRegister src); 1506 1507 // Compute Square Root of Scalar Single-Precision Floating-Point Value 1508 void sqrtss(XMMRegister dst, Address src); 1509 void sqrtss(XMMRegister dst, XMMRegister src); 1510 1511 void std() { emit_byte(0xfd); } 1512 1513 void stmxcsr( Address dst ); 1514 1515 void subl(Address dst, int32_t imm32); 1516 void subl(Address dst, Register src); 1517 void subl(Register dst, int32_t imm32); 1518 void subl(Register dst, Address src); 1519 void subl(Register dst, Register src); 1520 1521 void subq(Address dst, int32_t imm32); 1522 void subq(Address dst, Register src); 1523 void subq(Register dst, int32_t imm32); 1524 void subq(Register dst, Address src); 1525 void subq(Register dst, Register src); 1526 1527 // Force generation of a 4 byte immediate value even if it fits into 8bit 1528 void subl_imm32(Register dst, int32_t imm32); 1529 void subq_imm32(Register dst, int32_t imm32); 1530 1531 // Subtract Scalar Double-Precision Floating-Point Values 1532 void subsd(XMMRegister dst, Address src); 1533 void subsd(XMMRegister dst, XMMRegister src); 1534 1535 // Subtract Scalar Single-Precision Floating-Point Values 1536 void subss(XMMRegister dst, Address src); 1537 void subss(XMMRegister dst, XMMRegister src); 1538 1539 void testb(Register dst, int imm8); 1540 1541 void testl(Register dst, int32_t imm32); 1542 void testl(Register dst, Register src); 1543 void testl(Register dst, Address src); 1544 1545 void testq(Register dst, int32_t imm32); 1546 void testq(Register dst, Register src); 1547 1548 1549 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1550 void ucomisd(XMMRegister dst, Address src); 1551 void ucomisd(XMMRegister dst, XMMRegister src); 1552 1553 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1554 void ucomiss(XMMRegister dst, Address src); 1555 void ucomiss(XMMRegister dst, XMMRegister src); 1556 1557 void xaddl(Address dst, Register src); 1558 1559 void xaddq(Address dst, Register src); 1560 1561 void xchgl(Register reg, Address adr); 1562 void xchgl(Register dst, Register src); 1563 1564 void xchgq(Register reg, Address adr); 1565 void xchgq(Register dst, Register src); 1566 1567 // Get Value of Extended Control Register 1568 void xgetbv() { 1569 emit_byte(0x0F); 1570 emit_byte(0x01); 1571 emit_byte(0xD0); 1572 } 1573 1574 void xorl(Register dst, int32_t imm32); 1575 void xorl(Register dst, Address src); 1576 void xorl(Register dst, Register src); 1577 1578 void xorq(Register dst, Address src); 1579 void xorq(Register dst, Register src); 1580 1581 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1582 1583 // AVX 3-operands scalar instructions (encoded with VEX prefix) 1584 1585 void vaddsd(XMMRegister dst, XMMRegister nds, Address src); 1586 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1587 void vaddss(XMMRegister dst, XMMRegister nds, Address src); 1588 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1589 void vdivsd(XMMRegister dst, XMMRegister nds, Address src); 1590 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1591 void vdivss(XMMRegister dst, XMMRegister nds, Address src); 1592 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1593 void vmulsd(XMMRegister dst, XMMRegister nds, Address src); 1594 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1595 void vmulss(XMMRegister dst, XMMRegister nds, Address src); 1596 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1597 void vsubsd(XMMRegister dst, XMMRegister nds, Address src); 1598 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src); 1599 void vsubss(XMMRegister dst, XMMRegister nds, Address src); 1600 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src); 1601 1602 1603 //====================VECTOR ARITHMETIC===================================== 1604 1605 // Add Packed Floating-Point Values 1606 void addpd(XMMRegister dst, XMMRegister src); 1607 void addps(XMMRegister dst, XMMRegister src); 1608 void vaddpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1609 void vaddps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1610 void vaddpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1611 void vaddps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1612 1613 // Subtract Packed Floating-Point Values 1614 void subpd(XMMRegister dst, XMMRegister src); 1615 void subps(XMMRegister dst, XMMRegister src); 1616 void vsubpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1617 void vsubps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1618 void vsubpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1619 void vsubps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1620 1621 // Multiply Packed Floating-Point Values 1622 void mulpd(XMMRegister dst, XMMRegister src); 1623 void mulps(XMMRegister dst, XMMRegister src); 1624 void vmulpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1625 void vmulps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1626 void vmulpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1627 void vmulps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1628 1629 // Divide Packed Floating-Point Values 1630 void divpd(XMMRegister dst, XMMRegister src); 1631 void divps(XMMRegister dst, XMMRegister src); 1632 void vdivpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1633 void vdivps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1634 void vdivpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1635 void vdivps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1636 1637 // Bitwise Logical AND of Packed Floating-Point Values 1638 void andpd(XMMRegister dst, XMMRegister src); 1639 void andps(XMMRegister dst, XMMRegister src); 1640 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1641 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1642 void vandpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1643 void vandps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1644 1645 // Bitwise Logical XOR of Packed Floating-Point Values 1646 void xorpd(XMMRegister dst, XMMRegister src); 1647 void xorps(XMMRegister dst, XMMRegister src); 1648 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1649 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1650 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1651 void vxorps(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1652 1653 // Add packed integers 1654 void paddb(XMMRegister dst, XMMRegister src); 1655 void paddw(XMMRegister dst, XMMRegister src); 1656 void paddd(XMMRegister dst, XMMRegister src); 1657 void paddq(XMMRegister dst, XMMRegister src); 1658 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1659 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1660 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1661 void vpaddq(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1662 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1663 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1664 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1665 void vpaddq(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1666 1667 // Sub packed integers 1668 void psubb(XMMRegister dst, XMMRegister src); 1669 void psubw(XMMRegister dst, XMMRegister src); 1670 void psubd(XMMRegister dst, XMMRegister src); 1671 void psubq(XMMRegister dst, XMMRegister src); 1672 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1673 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1674 void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1675 void vpsubq(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1676 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1677 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1678 void vpsubd(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1679 void vpsubq(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1680 1681 // Multiply packed integers (only shorts and ints) 1682 void pmullw(XMMRegister dst, XMMRegister src); 1683 void pmulld(XMMRegister dst, XMMRegister src); 1684 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1685 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1686 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1687 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1688 1689 // Shift left packed integers 1690 void psllw(XMMRegister dst, int shift); 1691 void pslld(XMMRegister dst, int shift); 1692 void psllq(XMMRegister dst, int shift); 1693 void psllw(XMMRegister dst, XMMRegister shift); 1694 void pslld(XMMRegister dst, XMMRegister shift); 1695 void psllq(XMMRegister dst, XMMRegister shift); 1696 void vpsllw(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1697 void vpslld(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1698 void vpsllq(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1699 void vpsllw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1700 void vpslld(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1701 void vpsllq(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1702 1703 // Logical shift right packed integers 1704 void psrlw(XMMRegister dst, int shift); 1705 void psrld(XMMRegister dst, int shift); 1706 void psrlq(XMMRegister dst, int shift); 1707 void psrlw(XMMRegister dst, XMMRegister shift); 1708 void psrld(XMMRegister dst, XMMRegister shift); 1709 void psrlq(XMMRegister dst, XMMRegister shift); 1710 void vpsrlw(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1711 void vpsrld(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1712 void vpsrlq(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1713 void vpsrlw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1714 void vpsrld(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1715 void vpsrlq(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1716 1717 // Arithmetic shift right packed integers (only shorts and ints, no instructions for longs) 1718 void psraw(XMMRegister dst, int shift); 1719 void psrad(XMMRegister dst, int shift); 1720 void psraw(XMMRegister dst, XMMRegister shift); 1721 void psrad(XMMRegister dst, XMMRegister shift); 1722 void vpsraw(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1723 void vpsrad(XMMRegister dst, XMMRegister src, int shift, bool vector256); 1724 void vpsraw(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1725 void vpsrad(XMMRegister dst, XMMRegister src, XMMRegister shift, bool vector256); 1726 1727 // And packed integers 1728 void pand(XMMRegister dst, XMMRegister src); 1729 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1730 void vpand(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1731 1732 // Or packed integers 1733 void por(XMMRegister dst, XMMRegister src); 1734 void vpor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1735 void vpor(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1736 1737 // Xor packed integers 1738 void pxor(XMMRegister dst, XMMRegister src); 1739 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256); 1740 void vpxor(XMMRegister dst, XMMRegister nds, Address src, bool vector256); 1741 1742 // Copy low 128bit into high 128bit of YMM registers. 1743 void vinsertf128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 1744 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src); 1745 1746 // Load/store high 128bit of YMM registers which does not destroy other half. 1747 void vinsertf128h(XMMRegister dst, Address src); 1748 void vinserti128h(XMMRegister dst, Address src); 1749 void vextractf128h(Address dst, XMMRegister src); 1750 void vextracti128h(Address dst, XMMRegister src); 1751 1752 // AVX instruction which is used to clear upper 128 bits of YMM registers and 1753 // to avoid transaction penalty between AVX and SSE states. There is no 1754 // penalty if legacy SSE instructions are encoded using VEX prefix because 1755 // they always clear upper 128 bits. It should be used before calling 1756 // runtime code and native libraries. 1757 void vzeroupper(); 1758 1759 protected: 1760 // Next instructions require address alignment 16 bytes SSE mode. 1761 // They should be called only from corresponding MacroAssembler instructions. 1762 void andpd(XMMRegister dst, Address src); 1763 void andps(XMMRegister dst, Address src); 1764 void xorpd(XMMRegister dst, Address src); 1765 void xorps(XMMRegister dst, Address src); 1766 1767 }; 1768 1769 1770 // MacroAssembler extends Assembler by frequently used macros. 1771 // 1772 // Instructions for which a 'better' code sequence exists depending 1773 // on arguments should also go in here. 1774 1775 class MacroAssembler: public Assembler { 1776 friend class LIR_Assembler; 1777 friend class Runtime1; // as_Address() 1778 1779 protected: 1780 1781 Address as_Address(AddressLiteral adr); 1782 Address as_Address(ArrayAddress adr); 1783 1784 // Support for VM calls 1785 // 1786 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 1787 // may customize this version by overriding it for its purposes (e.g., to save/restore 1788 // additional registers when doing a VM call). 1789 #ifdef CC_INTERP 1790 // c++ interpreter never wants to use interp_masm version of call_VM 1791 #define VIRTUAL 1792 #else 1793 #define VIRTUAL virtual 1794 #endif 1795 1796 VIRTUAL void call_VM_leaf_base( 1797 address entry_point, // the entry point 1798 int number_of_arguments // the number of arguments to pop after the call 1799 ); 1800 1801 // This is the base routine called by the different versions of call_VM. The interpreter 1802 // may customize this version by overriding it for its purposes (e.g., to save/restore 1803 // additional registers when doing a VM call). 1804 // 1805 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base 1806 // returns the register which contains the thread upon return. If a thread register has been 1807 // specified, the return value will correspond to that register. If no last_java_sp is specified 1808 // (noreg) than rsp will be used instead. 1809 VIRTUAL void call_VM_base( // returns the register containing the thread upon return 1810 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 1811 Register java_thread, // the thread if computed before ; use noreg otherwise 1812 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 1813 address entry_point, // the entry point 1814 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 1815 bool check_exceptions // whether to check for pending exceptions after return 1816 ); 1817 1818 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 1819 // The implementation is only non-empty for the InterpreterMacroAssembler, 1820 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 1821 virtual void check_and_handle_popframe(Register java_thread); 1822 virtual void check_and_handle_earlyret(Register java_thread); 1823 1824 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 1825 1826 // helpers for FPU flag access 1827 // tmp is a temporary register, if none is available use noreg 1828 void save_rax (Register tmp); 1829 void restore_rax(Register tmp); 1830 1831 public: 1832 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 1833 1834 // Support for NULL-checks 1835 // 1836 // Generates code that causes a NULL OS exception if the content of reg is NULL. 1837 // If the accessed location is M[reg + offset] and the offset is known, provide the 1838 // offset. No explicit code generation is needed if the offset is within a certain 1839 // range (0 <= offset <= page_size). 1840 1841 void null_check(Register reg, int offset = -1); 1842 static bool needs_explicit_null_check(intptr_t offset); 1843 1844 // Required platform-specific helpers for Label::patch_instructions. 1845 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 1846 void pd_patch_instruction(address branch, address target); 1847 #ifndef PRODUCT 1848 static void pd_print_patched_instruction(address branch); 1849 #endif 1850 1851 // The following 4 methods return the offset of the appropriate move instruction 1852 1853 // Support for fast byte/short loading with zero extension (depending on particular CPU) 1854 int load_unsigned_byte(Register dst, Address src); 1855 int load_unsigned_short(Register dst, Address src); 1856 1857 // Support for fast byte/short loading with sign extension (depending on particular CPU) 1858 int load_signed_byte(Register dst, Address src); 1859 int load_signed_short(Register dst, Address src); 1860 1861 // Support for sign-extension (hi:lo = extend_sign(lo)) 1862 void extend_sign(Register hi, Register lo); 1863 1864 // Load and store values by size and signed-ness 1865 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); 1866 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); 1867 1868 // Support for inc/dec with optimal instruction selection depending on value 1869 1870 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } 1871 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } 1872 1873 void decrementl(Address dst, int value = 1); 1874 void decrementl(Register reg, int value = 1); 1875 1876 void decrementq(Register reg, int value = 1); 1877 void decrementq(Address dst, int value = 1); 1878 1879 void incrementl(Address dst, int value = 1); 1880 void incrementl(Register reg, int value = 1); 1881 1882 void incrementq(Register reg, int value = 1); 1883 void incrementq(Address dst, int value = 1); 1884 1885 1886 // Support optimal SSE move instructions. 1887 void movflt(XMMRegister dst, XMMRegister src) { 1888 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } 1889 else { movss (dst, src); return; } 1890 } 1891 void movflt(XMMRegister dst, Address src) { movss(dst, src); } 1892 void movflt(XMMRegister dst, AddressLiteral src); 1893 void movflt(Address dst, XMMRegister src) { movss(dst, src); } 1894 1895 void movdbl(XMMRegister dst, XMMRegister src) { 1896 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } 1897 else { movsd (dst, src); return; } 1898 } 1899 1900 void movdbl(XMMRegister dst, AddressLiteral src); 1901 1902 void movdbl(XMMRegister dst, Address src) { 1903 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } 1904 else { movlpd(dst, src); return; } 1905 } 1906 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } 1907 1908 void incrementl(AddressLiteral dst); 1909 void incrementl(ArrayAddress dst); 1910 1911 // Alignment 1912 void align(int modulus); 1913 1914 // A 5 byte nop that is safe for patching (see patch_verified_entry) 1915 void fat_nop(); 1916 1917 // Stack frame creation/removal 1918 void enter(); 1919 void leave(); 1920 1921 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 1922 // The pointer will be loaded into the thread register. 1923 void get_thread(Register thread); 1924 1925 1926 // Support for VM calls 1927 // 1928 // It is imperative that all calls into the VM are handled via the call_VM macros. 1929 // They make sure that the stack linkage is setup correctly. call_VM's correspond 1930 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 1931 1932 1933 void call_VM(Register oop_result, 1934 address entry_point, 1935 bool check_exceptions = true); 1936 void call_VM(Register oop_result, 1937 address entry_point, 1938 Register arg_1, 1939 bool check_exceptions = true); 1940 void call_VM(Register oop_result, 1941 address entry_point, 1942 Register arg_1, Register arg_2, 1943 bool check_exceptions = true); 1944 void call_VM(Register oop_result, 1945 address entry_point, 1946 Register arg_1, Register arg_2, Register arg_3, 1947 bool check_exceptions = true); 1948 1949 // Overloadings with last_Java_sp 1950 void call_VM(Register oop_result, 1951 Register last_java_sp, 1952 address entry_point, 1953 int number_of_arguments = 0, 1954 bool check_exceptions = true); 1955 void call_VM(Register oop_result, 1956 Register last_java_sp, 1957 address entry_point, 1958 Register arg_1, bool 1959 check_exceptions = true); 1960 void call_VM(Register oop_result, 1961 Register last_java_sp, 1962 address entry_point, 1963 Register arg_1, Register arg_2, 1964 bool check_exceptions = true); 1965 void call_VM(Register oop_result, 1966 Register last_java_sp, 1967 address entry_point, 1968 Register arg_1, Register arg_2, Register arg_3, 1969 bool check_exceptions = true); 1970 1971 void get_vm_result (Register oop_result, Register thread); 1972 void get_vm_result_2(Register metadata_result, Register thread); 1973 1974 // These always tightly bind to MacroAssembler::call_VM_base 1975 // bypassing the virtual implementation 1976 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); 1977 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); 1978 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); 1979 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true); 1980 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true); 1981 1982 void call_VM_leaf(address entry_point, 1983 int number_of_arguments = 0); 1984 void call_VM_leaf(address entry_point, 1985 Register arg_1); 1986 void call_VM_leaf(address entry_point, 1987 Register arg_1, Register arg_2); 1988 void call_VM_leaf(address entry_point, 1989 Register arg_1, Register arg_2, Register arg_3); 1990 1991 // These always tightly bind to MacroAssembler::call_VM_leaf_base 1992 // bypassing the virtual implementation 1993 void super_call_VM_leaf(address entry_point); 1994 void super_call_VM_leaf(address entry_point, Register arg_1); 1995 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); 1996 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); 1997 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); 1998 1999 // last Java Frame (fills frame anchor) 2000 void set_last_Java_frame(Register thread, 2001 Register last_java_sp, 2002 Register last_java_fp, 2003 address last_java_pc); 2004 2005 // thread in the default location (r15_thread on 64bit) 2006 void set_last_Java_frame(Register last_java_sp, 2007 Register last_java_fp, 2008 address last_java_pc); 2009 2010 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); 2011 2012 // thread in the default location (r15_thread on 64bit) 2013 void reset_last_Java_frame(bool clear_fp, bool clear_pc); 2014 2015 // Stores 2016 void store_check(Register obj); // store check for obj - register is destroyed afterwards 2017 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 2018 2019 #ifndef SERIALGC 2020 2021 void g1_write_barrier_pre(Register obj, 2022 Register pre_val, 2023 Register thread, 2024 Register tmp, 2025 bool tosca_live, 2026 bool expand_call); 2027 2028 void g1_write_barrier_post(Register store_addr, 2029 Register new_val, 2030 Register thread, 2031 Register tmp, 2032 Register tmp2); 2033 2034 #endif // SERIALGC 2035 2036 // split store_check(Register obj) to enhance instruction interleaving 2037 void store_check_part_1(Register obj); 2038 void store_check_part_2(Register obj); 2039 2040 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 2041 void c2bool(Register x); 2042 2043 // C++ bool manipulation 2044 2045 void movbool(Register dst, Address src); 2046 void movbool(Address dst, bool boolconst); 2047 void movbool(Address dst, Register src); 2048 void testbool(Register dst); 2049 2050 // oop manipulations 2051 void load_klass(Register dst, Register src); 2052 void store_klass(Register dst, Register src); 2053 2054 void load_heap_oop(Register dst, Address src); 2055 void load_heap_oop_not_null(Register dst, Address src); 2056 void store_heap_oop(Address dst, Register src); 2057 void cmp_heap_oop(Register src1, Address src2, Register tmp = noreg); 2058 2059 // Used for storing NULL. All other oop constants should be 2060 // stored using routines that take a jobject. 2061 void store_heap_oop_null(Address dst); 2062 2063 void load_prototype_header(Register dst, Register src); 2064 2065 #ifdef _LP64 2066 void store_klass_gap(Register dst, Register src); 2067 2068 // This dummy is to prevent a call to store_heap_oop from 2069 // converting a zero (like NULL) into a Register by giving 2070 // the compiler two choices it can't resolve 2071 2072 void store_heap_oop(Address dst, void* dummy); 2073 2074 void encode_heap_oop(Register r); 2075 void decode_heap_oop(Register r); 2076 void encode_heap_oop_not_null(Register r); 2077 void decode_heap_oop_not_null(Register r); 2078 void encode_heap_oop_not_null(Register dst, Register src); 2079 void decode_heap_oop_not_null(Register dst, Register src); 2080 2081 void set_narrow_oop(Register dst, jobject obj); 2082 void set_narrow_oop(Address dst, jobject obj); 2083 void cmp_narrow_oop(Register dst, jobject obj); 2084 void cmp_narrow_oop(Address dst, jobject obj); 2085 2086 // if heap base register is used - reinit it with the correct value 2087 void reinit_heapbase(); 2088 2089 DEBUG_ONLY(void verify_heapbase(const char* msg);) 2090 2091 #endif // _LP64 2092 2093 // Int division/remainder for Java 2094 // (as idivl, but checks for special case as described in JVM spec.) 2095 // returns idivl instruction offset for implicit exception handling 2096 int corrected_idivl(Register reg); 2097 2098 // Long division/remainder for Java 2099 // (as idivq, but checks for special case as described in JVM spec.) 2100 // returns idivq instruction offset for implicit exception handling 2101 int corrected_idivq(Register reg); 2102 2103 void int3(); 2104 2105 // Long operation macros for a 32bit cpu 2106 // Long negation for Java 2107 void lneg(Register hi, Register lo); 2108 2109 // Long multiplication for Java 2110 // (destroys contents of eax, ebx, ecx and edx) 2111 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y 2112 2113 // Long shifts for Java 2114 // (semantics as described in JVM spec.) 2115 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) 2116 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) 2117 2118 // Long compare for Java 2119 // (semantics as described in JVM spec.) 2120 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) 2121 2122 2123 // misc 2124 2125 // Sign extension 2126 void sign_extend_short(Register reg); 2127 void sign_extend_byte(Register reg); 2128 2129 // Division by power of 2, rounding towards 0 2130 void division_with_shift(Register reg, int shift_value); 2131 2132 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: 2133 // 2134 // CF (corresponds to C0) if x < y 2135 // PF (corresponds to C2) if unordered 2136 // ZF (corresponds to C3) if x = y 2137 // 2138 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 2139 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) 2140 void fcmp(Register tmp); 2141 // Variant of the above which allows y to be further down the stack 2142 // and which only pops x and y if specified. If pop_right is 2143 // specified then pop_left must also be specified. 2144 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); 2145 2146 // Floating-point comparison for Java 2147 // Compares the top-most stack entries on the FPU stack and stores the result in dst. 2148 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 2149 // (semantics as described in JVM spec.) 2150 void fcmp2int(Register dst, bool unordered_is_less); 2151 // Variant of the above which allows y to be further down the stack 2152 // and which only pops x and y if specified. If pop_right is 2153 // specified then pop_left must also be specified. 2154 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); 2155 2156 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) 2157 // tmp is a temporary register, if none is available use noreg 2158 void fremr(Register tmp); 2159 2160 2161 // same as fcmp2int, but using SSE2 2162 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 2163 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 2164 2165 // Inlined sin/cos generator for Java; must not use CPU instruction 2166 // directly on Intel as it does not have high enough precision 2167 // outside of the range [-pi/4, pi/4]. Extra argument indicate the 2168 // number of FPU stack slots in use; all but the topmost will 2169 // require saving if a slow case is necessary. Assumes argument is 2170 // on FP TOS; result is on FP TOS. No cpu registers are changed by 2171 // this code. 2172 void trigfunc(char trig, int num_fpu_regs_in_use = 1); 2173 2174 // branch to L if FPU flag C2 is set/not set 2175 // tmp is a temporary register, if none is available use noreg 2176 void jC2 (Register tmp, Label& L); 2177 void jnC2(Register tmp, Label& L); 2178 2179 // Pop ST (ffree & fincstp combined) 2180 void fpop(); 2181 2182 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack 2183 void push_fTOS(); 2184 2185 // pops double TOS element from CPU stack and pushes on FPU stack 2186 void pop_fTOS(); 2187 2188 void empty_FPU_stack(); 2189 2190 void push_IU_state(); 2191 void pop_IU_state(); 2192 2193 void push_FPU_state(); 2194 void pop_FPU_state(); 2195 2196 void push_CPU_state(); 2197 void pop_CPU_state(); 2198 2199 // Round up to a power of two 2200 void round_to(Register reg, int modulus); 2201 2202 // Callee saved registers handling 2203 void push_callee_saved_registers(); 2204 void pop_callee_saved_registers(); 2205 2206 // allocation 2207 void eden_allocate( 2208 Register obj, // result: pointer to object after successful allocation 2209 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 2210 int con_size_in_bytes, // object size in bytes if known at compile time 2211 Register t1, // temp register 2212 Label& slow_case // continuation point if fast allocation fails 2213 ); 2214 void tlab_allocate( 2215 Register obj, // result: pointer to object after successful allocation 2216 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 2217 int con_size_in_bytes, // object size in bytes if known at compile time 2218 Register t1, // temp register 2219 Register t2, // temp register 2220 Label& slow_case // continuation point if fast allocation fails 2221 ); 2222 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address 2223 void incr_allocated_bytes(Register thread, 2224 Register var_size_in_bytes, int con_size_in_bytes, 2225 Register t1 = noreg); 2226 2227 // interface method calling 2228 void lookup_interface_method(Register recv_klass, 2229 Register intf_klass, 2230 RegisterOrConstant itable_index, 2231 Register method_result, 2232 Register scan_temp, 2233 Label& no_such_interface); 2234 2235 // virtual method calling 2236 void lookup_virtual_method(Register recv_klass, 2237 RegisterOrConstant vtable_index, 2238 Register method_result); 2239 2240 // Test sub_klass against super_klass, with fast and slow paths. 2241 2242 // The fast path produces a tri-state answer: yes / no / maybe-slow. 2243 // One of the three labels can be NULL, meaning take the fall-through. 2244 // If super_check_offset is -1, the value is loaded up from super_klass. 2245 // No registers are killed, except temp_reg. 2246 void check_klass_subtype_fast_path(Register sub_klass, 2247 Register super_klass, 2248 Register temp_reg, 2249 Label* L_success, 2250 Label* L_failure, 2251 Label* L_slow_path, 2252 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 2253 2254 // The rest of the type check; must be wired to a corresponding fast path. 2255 // It does not repeat the fast path logic, so don't use it standalone. 2256 // The temp_reg and temp2_reg can be noreg, if no temps are available. 2257 // Updates the sub's secondary super cache as necessary. 2258 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 2259 void check_klass_subtype_slow_path(Register sub_klass, 2260 Register super_klass, 2261 Register temp_reg, 2262 Register temp2_reg, 2263 Label* L_success, 2264 Label* L_failure, 2265 bool set_cond_codes = false); 2266 2267 // Simplified, combined version, good for typical uses. 2268 // Falls through on failure. 2269 void check_klass_subtype(Register sub_klass, 2270 Register super_klass, 2271 Register temp_reg, 2272 Label& L_success); 2273 2274 // method handles (JSR 292) 2275 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 2276 2277 //---- 2278 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 2279 2280 // Debugging 2281 2282 // only if +VerifyOops 2283 // TODO: Make these macros with file and line like sparc version! 2284 void verify_oop(Register reg, const char* s = "broken oop"); 2285 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 2286 2287 // TODO: verify method and klass metadata (compare against vptr?) 2288 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {} 2289 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){} 2290 2291 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__) 2292 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__) 2293 2294 // only if +VerifyFPU 2295 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 2296 2297 // prints msg, dumps registers and stops execution 2298 void stop(const char* msg); 2299 2300 // prints msg and continues 2301 void warn(const char* msg); 2302 2303 // dumps registers and other state 2304 void print_state(); 2305 2306 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); 2307 static void debug64(char* msg, int64_t pc, int64_t regs[]); 2308 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip); 2309 static void print_state64(int64_t pc, int64_t regs[]); 2310 2311 void os_breakpoint(); 2312 2313 void untested() { stop("untested"); } 2314 2315 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } 2316 2317 void should_not_reach_here() { stop("should not reach here"); } 2318 2319 void print_CPU_state(); 2320 2321 // Stack overflow checking 2322 void bang_stack_with_offset(int offset) { 2323 // stack grows down, caller passes positive offset 2324 assert(offset > 0, "must bang with negative offset"); 2325 movl(Address(rsp, (-offset)), rax); 2326 } 2327 2328 // Writes to stack successive pages until offset reached to check for 2329 // stack overflow + shadow pages. Also, clobbers tmp 2330 void bang_stack_size(Register size, Register tmp); 2331 2332 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 2333 Register tmp, 2334 int offset); 2335 2336 // Support for serializing memory accesses between threads 2337 void serialize_memory(Register thread, Register tmp); 2338 2339 void verify_tlab(); 2340 2341 // Biased locking support 2342 // lock_reg and obj_reg must be loaded up with the appropriate values. 2343 // swap_reg must be rax, and is killed. 2344 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will 2345 // be killed; if not supplied, push/pop will be used internally to 2346 // allocate a temporary (inefficient, avoid if possible). 2347 // Optional slow case is for implementations (interpreter and C1) which branch to 2348 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 2349 // Returns offset of first potentially-faulting instruction for null 2350 // check info (currently consumed only by C1). If 2351 // swap_reg_contains_mark is true then returns -1 as it is assumed 2352 // the calling code has already passed any potential faults. 2353 int biased_locking_enter(Register lock_reg, Register obj_reg, 2354 Register swap_reg, Register tmp_reg, 2355 bool swap_reg_contains_mark, 2356 Label& done, Label* slow_case = NULL, 2357 BiasedLockingCounters* counters = NULL); 2358 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 2359 2360 2361 Condition negate_condition(Condition cond); 2362 2363 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit 2364 // operands. In general the names are modified to avoid hiding the instruction in Assembler 2365 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers 2366 // here in MacroAssembler. The major exception to this rule is call 2367 2368 // Arithmetics 2369 2370 2371 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } 2372 void addptr(Address dst, Register src); 2373 2374 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } 2375 void addptr(Register dst, int32_t src); 2376 void addptr(Register dst, Register src); 2377 void addptr(Register dst, RegisterOrConstant src) { 2378 if (src.is_constant()) addptr(dst, (int) src.as_constant()); 2379 else addptr(dst, src.as_register()); 2380 } 2381 2382 void andptr(Register dst, int32_t src); 2383 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } 2384 2385 void cmp8(AddressLiteral src1, int imm); 2386 2387 // renamed to drag out the casting of address to int32_t/intptr_t 2388 void cmp32(Register src1, int32_t imm); 2389 2390 void cmp32(AddressLiteral src1, int32_t imm); 2391 // compare reg - mem, or reg - &mem 2392 void cmp32(Register src1, AddressLiteral src2); 2393 2394 void cmp32(Register src1, Address src2); 2395 2396 #ifndef _LP64 2397 void cmpklass(Address dst, Metadata* obj); 2398 void cmpklass(Register dst, Metadata* obj); 2399 void cmpoop(Address dst, jobject obj); 2400 void cmpoop(Register dst, jobject obj); 2401 #endif // _LP64 2402 2403 // NOTE src2 must be the lval. This is NOT an mem-mem compare 2404 void cmpptr(Address src1, AddressLiteral src2); 2405 2406 void cmpptr(Register src1, AddressLiteral src2); 2407 2408 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2409 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2410 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2411 2412 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2413 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2414 2415 // cmp64 to avoild hiding cmpq 2416 void cmp64(Register src1, AddressLiteral src); 2417 2418 void cmpxchgptr(Register reg, Address adr); 2419 2420 void locked_cmpxchgptr(Register reg, AddressLiteral adr); 2421 2422 2423 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } 2424 2425 2426 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } 2427 2428 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } 2429 2430 void shlptr(Register dst, int32_t shift); 2431 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } 2432 2433 void shrptr(Register dst, int32_t shift); 2434 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } 2435 2436 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } 2437 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } 2438 2439 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2440 2441 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2442 void subptr(Register dst, int32_t src); 2443 // Force generation of a 4 byte immediate value even if it fits into 8bit 2444 void subptr_imm32(Register dst, int32_t src); 2445 void subptr(Register dst, Register src); 2446 void subptr(Register dst, RegisterOrConstant src) { 2447 if (src.is_constant()) subptr(dst, (int) src.as_constant()); 2448 else subptr(dst, src.as_register()); 2449 } 2450 2451 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2452 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2453 2454 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2455 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2456 2457 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } 2458 2459 2460 2461 // Helper functions for statistics gathering. 2462 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. 2463 void cond_inc32(Condition cond, AddressLiteral counter_addr); 2464 // Unconditional atomic increment. 2465 void atomic_incl(AddressLiteral counter_addr); 2466 2467 void lea(Register dst, AddressLiteral adr); 2468 void lea(Address dst, AddressLiteral adr); 2469 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } 2470 2471 void leal32(Register dst, Address src) { leal(dst, src); } 2472 2473 // Import other testl() methods from the parent class or else 2474 // they will be hidden by the following overriding declaration. 2475 using Assembler::testl; 2476 void testl(Register dst, AddressLiteral src); 2477 2478 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2479 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2480 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2481 2482 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } 2483 void testptr(Register src1, Register src2); 2484 2485 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2486 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2487 2488 // Calls 2489 2490 void call(Label& L, relocInfo::relocType rtype); 2491 void call(Register entry); 2492 2493 // NOTE: this call tranfers to the effective address of entry NOT 2494 // the address contained by entry. This is because this is more natural 2495 // for jumps/calls. 2496 void call(AddressLiteral entry); 2497 2498 // Emit the CompiledIC call idiom 2499 void ic_call(address entry); 2500 2501 // Jumps 2502 2503 // NOTE: these jumps tranfer to the effective address of dst NOT 2504 // the address contained by dst. This is because this is more natural 2505 // for jumps/calls. 2506 void jump(AddressLiteral dst); 2507 void jump_cc(Condition cc, AddressLiteral dst); 2508 2509 // 32bit can do a case table jump in one instruction but we no longer allow the base 2510 // to be installed in the Address class. This jump will tranfers to the address 2511 // contained in the location described by entry (not the address of entry) 2512 void jump(ArrayAddress entry); 2513 2514 // Floating 2515 2516 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } 2517 void andpd(XMMRegister dst, AddressLiteral src); 2518 2519 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); } 2520 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); } 2521 void andps(XMMRegister dst, AddressLiteral src); 2522 2523 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); } 2524 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } 2525 void comiss(XMMRegister dst, AddressLiteral src); 2526 2527 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); } 2528 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } 2529 void comisd(XMMRegister dst, AddressLiteral src); 2530 2531 void fadd_s(Address src) { Assembler::fadd_s(src); } 2532 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); } 2533 2534 void fldcw(Address src) { Assembler::fldcw(src); } 2535 void fldcw(AddressLiteral src); 2536 2537 void fld_s(int index) { Assembler::fld_s(index); } 2538 void fld_s(Address src) { Assembler::fld_s(src); } 2539 void fld_s(AddressLiteral src); 2540 2541 void fld_d(Address src) { Assembler::fld_d(src); } 2542 void fld_d(AddressLiteral src); 2543 2544 void fld_x(Address src) { Assembler::fld_x(src); } 2545 void fld_x(AddressLiteral src); 2546 2547 void fmul_s(Address src) { Assembler::fmul_s(src); } 2548 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); } 2549 2550 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } 2551 void ldmxcsr(AddressLiteral src); 2552 2553 // compute pow(x,y) and exp(x) with x86 instructions. Don't cover 2554 // all corner cases and may result in NaN and require fallback to a 2555 // runtime call. 2556 void fast_pow(); 2557 void fast_exp(); 2558 void increase_precision(); 2559 void restore_precision(); 2560 2561 // computes exp(x). Fallback to runtime call included. 2562 void exp_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(true, num_fpu_regs_in_use); } 2563 // computes pow(x,y). Fallback to runtime call included. 2564 void pow_with_fallback(int num_fpu_regs_in_use) { pow_or_exp(false, num_fpu_regs_in_use); } 2565 2566 private: 2567 2568 // call runtime as a fallback for trig functions and pow/exp. 2569 void fp_runtime_fallback(address runtime_entry, int nb_args, int num_fpu_regs_in_use); 2570 2571 // computes 2^(Ylog2X); Ylog2X in ST(0) 2572 void pow_exp_core_encoding(); 2573 2574 // computes pow(x,y) or exp(x). Fallback to runtime call included. 2575 void pow_or_exp(bool is_exp, int num_fpu_regs_in_use); 2576 2577 // these are private because users should be doing movflt/movdbl 2578 2579 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } 2580 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } 2581 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } 2582 void movss(XMMRegister dst, AddressLiteral src); 2583 2584 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } 2585 void movlpd(XMMRegister dst, AddressLiteral src); 2586 2587 public: 2588 2589 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); } 2590 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); } 2591 void addsd(XMMRegister dst, AddressLiteral src); 2592 2593 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); } 2594 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); } 2595 void addss(XMMRegister dst, AddressLiteral src); 2596 2597 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); } 2598 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); } 2599 void divsd(XMMRegister dst, AddressLiteral src); 2600 2601 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); } 2602 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); } 2603 void divss(XMMRegister dst, AddressLiteral src); 2604 2605 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } 2606 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } 2607 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } 2608 void movsd(XMMRegister dst, AddressLiteral src); 2609 2610 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); } 2611 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); } 2612 void mulsd(XMMRegister dst, AddressLiteral src); 2613 2614 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); } 2615 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); } 2616 void mulss(XMMRegister dst, AddressLiteral src); 2617 2618 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); } 2619 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); } 2620 void sqrtsd(XMMRegister dst, AddressLiteral src); 2621 2622 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); } 2623 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); } 2624 void sqrtss(XMMRegister dst, AddressLiteral src); 2625 2626 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); } 2627 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); } 2628 void subsd(XMMRegister dst, AddressLiteral src); 2629 2630 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); } 2631 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); } 2632 void subss(XMMRegister dst, AddressLiteral src); 2633 2634 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } 2635 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } 2636 void ucomiss(XMMRegister dst, AddressLiteral src); 2637 2638 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } 2639 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } 2640 void ucomisd(XMMRegister dst, AddressLiteral src); 2641 2642 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 2643 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); } 2644 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } 2645 void xorpd(XMMRegister dst, AddressLiteral src); 2646 2647 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 2648 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); } 2649 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } 2650 void xorps(XMMRegister dst, AddressLiteral src); 2651 2652 // AVX 3-operands instructions 2653 2654 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); } 2655 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); } 2656 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2657 2658 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); } 2659 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); } 2660 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2661 2662 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); } 2663 void vandpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandpd(dst, nds, src, vector256); } 2664 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); 2665 2666 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); } 2667 void vandps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vandps(dst, nds, src, vector256); } 2668 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); 2669 2670 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); } 2671 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); } 2672 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2673 2674 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); } 2675 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); } 2676 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2677 2678 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); } 2679 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); } 2680 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2681 2682 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); } 2683 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); } 2684 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2685 2686 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); } 2687 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); } 2688 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2689 2690 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); } 2691 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); } 2692 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 2693 2694 // AVX Vector instructions 2695 2696 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); } 2697 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorpd(dst, nds, src, vector256); } 2698 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); 2699 2700 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); } 2701 void vxorps(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { Assembler::vxorps(dst, nds, src, vector256); } 2702 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, bool vector256); 2703 2704 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, bool vector256) { 2705 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2 2706 Assembler::vpxor(dst, nds, src, vector256); 2707 else 2708 Assembler::vxorpd(dst, nds, src, vector256); 2709 } 2710 void vpxor(XMMRegister dst, XMMRegister nds, Address src, bool vector256) { 2711 if (UseAVX > 1 || !vector256) // vpxor 256 bit is available only in AVX2 2712 Assembler::vpxor(dst, nds, src, vector256); 2713 else 2714 Assembler::vxorpd(dst, nds, src, vector256); 2715 } 2716 2717 // Move packed integer values from low 128 bit to hign 128 bit in 256 bit vector. 2718 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src) { 2719 if (UseAVX > 1) // vinserti128h is available only in AVX2 2720 Assembler::vinserti128h(dst, nds, src); 2721 else 2722 Assembler::vinsertf128h(dst, nds, src); 2723 } 2724 2725 // Data 2726 2727 void cmov32( Condition cc, Register dst, Address src); 2728 void cmov32( Condition cc, Register dst, Register src); 2729 2730 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); } 2731 2732 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 2733 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 2734 2735 void movoop(Register dst, jobject obj); 2736 void movoop(Address dst, jobject obj); 2737 2738 void mov_metadata(Register dst, Metadata* obj); 2739 void mov_metadata(Address dst, Metadata* obj); 2740 2741 void movptr(ArrayAddress dst, Register src); 2742 // can this do an lea? 2743 void movptr(Register dst, ArrayAddress src); 2744 2745 void movptr(Register dst, Address src); 2746 2747 void movptr(Register dst, AddressLiteral src); 2748 2749 void movptr(Register dst, intptr_t src); 2750 void movptr(Register dst, Register src); 2751 void movptr(Address dst, intptr_t src); 2752 2753 void movptr(Address dst, Register src); 2754 2755 void movptr(Register dst, RegisterOrConstant src) { 2756 if (src.is_constant()) movptr(dst, src.as_constant()); 2757 else movptr(dst, src.as_register()); 2758 } 2759 2760 #ifdef _LP64 2761 // Generally the next two are only used for moving NULL 2762 // Although there are situations in initializing the mark word where 2763 // they could be used. They are dangerous. 2764 2765 // They only exist on LP64 so that int32_t and intptr_t are not the same 2766 // and we have ambiguous declarations. 2767 2768 void movptr(Address dst, int32_t imm32); 2769 void movptr(Register dst, int32_t imm32); 2770 #endif // _LP64 2771 2772 // to avoid hiding movl 2773 void mov32(AddressLiteral dst, Register src); 2774 void mov32(Register dst, AddressLiteral src); 2775 2776 // to avoid hiding movb 2777 void movbyte(ArrayAddress dst, int src); 2778 2779 // Import other mov() methods from the parent class or else 2780 // they will be hidden by the following overriding declaration. 2781 using Assembler::movdl; 2782 using Assembler::movq; 2783 void movdl(XMMRegister dst, AddressLiteral src); 2784 void movq(XMMRegister dst, AddressLiteral src); 2785 2786 // Can push value or effective address 2787 void pushptr(AddressLiteral src); 2788 2789 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } 2790 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } 2791 2792 void pushoop(jobject obj); 2793 void pushklass(Metadata* obj); 2794 2795 // sign extend as need a l to ptr sized element 2796 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } 2797 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } 2798 2799 // C2 compiled method's prolog code. 2800 void verified_entry(int framesize, bool stack_bang, bool fp_mode_24b); 2801 2802 // IndexOf strings. 2803 // Small strings are loaded through stack if they cross page boundary. 2804 void string_indexof(Register str1, Register str2, 2805 Register cnt1, Register cnt2, 2806 int int_cnt2, Register result, 2807 XMMRegister vec, Register tmp); 2808 2809 // IndexOf for constant substrings with size >= 8 elements 2810 // which don't need to be loaded through stack. 2811 void string_indexofC8(Register str1, Register str2, 2812 Register cnt1, Register cnt2, 2813 int int_cnt2, Register result, 2814 XMMRegister vec, Register tmp); 2815 2816 // Smallest code: we don't need to load through stack, 2817 // check string tail. 2818 2819 // Compare strings. 2820 void string_compare(Register str1, Register str2, 2821 Register cnt1, Register cnt2, Register result, 2822 XMMRegister vec1); 2823 2824 // Compare char[] arrays. 2825 void char_arrays_equals(bool is_array_equ, Register ary1, Register ary2, 2826 Register limit, Register result, Register chr, 2827 XMMRegister vec1, XMMRegister vec2); 2828 2829 // Fill primitive arrays 2830 void generate_fill(BasicType t, bool aligned, 2831 Register to, Register value, Register count, 2832 Register rtmp, XMMRegister xtmp); 2833 2834 #undef VIRTUAL 2835 2836 }; 2837 2838 /** 2839 * class SkipIfEqual: 2840 * 2841 * Instantiating this class will result in assembly code being output that will 2842 * jump around any code emitted between the creation of the instance and it's 2843 * automatic destruction at the end of a scope block, depending on the value of 2844 * the flag passed to the constructor, which will be checked at run-time. 2845 */ 2846 class SkipIfEqual { 2847 private: 2848 MacroAssembler* _masm; 2849 Label _label; 2850 2851 public: 2852 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 2853 ~SkipIfEqual(); 2854 }; 2855 2856 #ifdef ASSERT 2857 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } 2858 #endif 2859 2860 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP