1 /* 2 * Copyright 1997-2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20 * CA 95054 USA or visit www.sun.com if you need additional information or 21 * have any questions. 22 * 23 */ 24 25 class BiasedLockingCounters; 26 27 // Contains all the definitions needed for x86 assembly code generation. 28 29 // Calling convention 30 class Argument VALUE_OBJ_CLASS_SPEC { 31 public: 32 enum { 33 #ifdef _LP64 34 #ifdef _WIN64 35 n_int_register_parameters_c = 4, // rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...) 36 n_float_register_parameters_c = 4, // xmm0 - xmm3 (c_farg0, c_farg1, ... ) 37 #else 38 n_int_register_parameters_c = 6, // rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...) 39 n_float_register_parameters_c = 8, // xmm0 - xmm7 (c_farg0, c_farg1, ... ) 40 #endif // _WIN64 41 n_int_register_parameters_j = 6, // j_rarg0, j_rarg1, ... 42 n_float_register_parameters_j = 8 // j_farg0, j_farg1, ... 43 #else 44 n_register_parameters = 0 // 0 registers used to pass arguments 45 #endif // _LP64 46 }; 47 }; 48 49 50 #ifdef _LP64 51 // Symbolically name the register arguments used by the c calling convention. 52 // Windows is different from linux/solaris. So much for standards... 53 54 #ifdef _WIN64 55 56 REGISTER_DECLARATION(Register, c_rarg0, rcx); 57 REGISTER_DECLARATION(Register, c_rarg1, rdx); 58 REGISTER_DECLARATION(Register, c_rarg2, r8); 59 REGISTER_DECLARATION(Register, c_rarg3, r9); 60 61 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 62 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 63 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 64 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 65 66 #else 67 68 REGISTER_DECLARATION(Register, c_rarg0, rdi); 69 REGISTER_DECLARATION(Register, c_rarg1, rsi); 70 REGISTER_DECLARATION(Register, c_rarg2, rdx); 71 REGISTER_DECLARATION(Register, c_rarg3, rcx); 72 REGISTER_DECLARATION(Register, c_rarg4, r8); 73 REGISTER_DECLARATION(Register, c_rarg5, r9); 74 75 REGISTER_DECLARATION(XMMRegister, c_farg0, xmm0); 76 REGISTER_DECLARATION(XMMRegister, c_farg1, xmm1); 77 REGISTER_DECLARATION(XMMRegister, c_farg2, xmm2); 78 REGISTER_DECLARATION(XMMRegister, c_farg3, xmm3); 79 REGISTER_DECLARATION(XMMRegister, c_farg4, xmm4); 80 REGISTER_DECLARATION(XMMRegister, c_farg5, xmm5); 81 REGISTER_DECLARATION(XMMRegister, c_farg6, xmm6); 82 REGISTER_DECLARATION(XMMRegister, c_farg7, xmm7); 83 84 #endif // _WIN64 85 86 // Symbolically name the register arguments used by the Java calling convention. 87 // We have control over the convention for java so we can do what we please. 88 // What pleases us is to offset the java calling convention so that when 89 // we call a suitable jni method the arguments are lined up and we don't 90 // have to do little shuffling. A suitable jni method is non-static and a 91 // small number of arguments (two fewer args on windows) 92 // 93 // |-------------------------------------------------------| 94 // | c_rarg0 c_rarg1 c_rarg2 c_rarg3 c_rarg4 c_rarg5 | 95 // |-------------------------------------------------------| 96 // | rcx rdx r8 r9 rdi* rsi* | windows (* not a c_rarg) 97 // | rdi rsi rdx rcx r8 r9 | solaris/linux 98 // |-------------------------------------------------------| 99 // | j_rarg5 j_rarg0 j_rarg1 j_rarg2 j_rarg3 j_rarg4 | 100 // |-------------------------------------------------------| 101 102 REGISTER_DECLARATION(Register, j_rarg0, c_rarg1); 103 REGISTER_DECLARATION(Register, j_rarg1, c_rarg2); 104 REGISTER_DECLARATION(Register, j_rarg2, c_rarg3); 105 // Windows runs out of register args here 106 #ifdef _WIN64 107 REGISTER_DECLARATION(Register, j_rarg3, rdi); 108 REGISTER_DECLARATION(Register, j_rarg4, rsi); 109 #else 110 REGISTER_DECLARATION(Register, j_rarg3, c_rarg4); 111 REGISTER_DECLARATION(Register, j_rarg4, c_rarg5); 112 #endif /* _WIN64 */ 113 REGISTER_DECLARATION(Register, j_rarg5, c_rarg0); 114 115 REGISTER_DECLARATION(XMMRegister, j_farg0, xmm0); 116 REGISTER_DECLARATION(XMMRegister, j_farg1, xmm1); 117 REGISTER_DECLARATION(XMMRegister, j_farg2, xmm2); 118 REGISTER_DECLARATION(XMMRegister, j_farg3, xmm3); 119 REGISTER_DECLARATION(XMMRegister, j_farg4, xmm4); 120 REGISTER_DECLARATION(XMMRegister, j_farg5, xmm5); 121 REGISTER_DECLARATION(XMMRegister, j_farg6, xmm6); 122 REGISTER_DECLARATION(XMMRegister, j_farg7, xmm7); 123 124 REGISTER_DECLARATION(Register, rscratch1, r10); // volatile 125 REGISTER_DECLARATION(Register, rscratch2, r11); // volatile 126 127 REGISTER_DECLARATION(Register, r12_heapbase, r12); // callee-saved 128 REGISTER_DECLARATION(Register, r15_thread, r15); // callee-saved 129 130 #else 131 // rscratch1 will apear in 32bit code that is dead but of course must compile 132 // Using noreg ensures if the dead code is incorrectly live and executed it 133 // will cause an assertion failure 134 #define rscratch1 noreg 135 136 #endif // _LP64 137 138 // Address is an abstraction used to represent a memory location 139 // using any of the amd64 addressing modes with one object. 140 // 141 // Note: A register location is represented via a Register, not 142 // via an address for efficiency & simplicity reasons. 143 144 class ArrayAddress; 145 146 class Address VALUE_OBJ_CLASS_SPEC { 147 public: 148 enum ScaleFactor { 149 no_scale = -1, 150 times_1 = 0, 151 times_2 = 1, 152 times_4 = 2, 153 times_8 = 3, 154 times_ptr = LP64_ONLY(times_8) NOT_LP64(times_4) 155 }; 156 static ScaleFactor times(int size) { 157 assert(size >= 1 && size <= 8 && is_power_of_2(size), "bad scale size"); 158 if (size == 8) return times_8; 159 if (size == 4) return times_4; 160 if (size == 2) return times_2; 161 return times_1; 162 } 163 static int scale_size(ScaleFactor scale) { 164 assert(scale != no_scale, ""); 165 assert(((1 << (int)times_1) == 1 && 166 (1 << (int)times_2) == 2 && 167 (1 << (int)times_4) == 4 && 168 (1 << (int)times_8) == 8), ""); 169 return (1 << (int)scale); 170 } 171 172 private: 173 Register _base; 174 Register _index; 175 ScaleFactor _scale; 176 int _disp; 177 RelocationHolder _rspec; 178 179 // Easily misused constructors make them private 180 // %%% can we make these go away? 181 NOT_LP64(Address(address loc, RelocationHolder spec);) 182 Address(int disp, address loc, relocInfo::relocType rtype); 183 Address(int disp, address loc, RelocationHolder spec); 184 185 public: 186 187 int disp() { return _disp; } 188 // creation 189 Address() 190 : _base(noreg), 191 _index(noreg), 192 _scale(no_scale), 193 _disp(0) { 194 } 195 196 // No default displacement otherwise Register can be implicitly 197 // converted to 0(Register) which is quite a different animal. 198 199 Address(Register base, int disp) 200 : _base(base), 201 _index(noreg), 202 _scale(no_scale), 203 _disp(disp) { 204 } 205 206 Address(Register base, Register index, ScaleFactor scale, int disp = 0) 207 : _base (base), 208 _index(index), 209 _scale(scale), 210 _disp (disp) { 211 assert(!index->is_valid() == (scale == Address::no_scale), 212 "inconsistent address"); 213 } 214 215 Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0) 216 : _base (base), 217 _index(index.register_or_noreg()), 218 _scale(scale), 219 _disp (disp + (index.constant_or_zero() * scale_size(scale))) { 220 if (!index.is_register()) scale = Address::no_scale; 221 assert(!_index->is_valid() == (scale == Address::no_scale), 222 "inconsistent address"); 223 } 224 225 Address plus_disp(int disp) const { 226 Address a = (*this); 227 a._disp += disp; 228 return a; 229 } 230 231 // The following two overloads are used in connection with the 232 // ByteSize type (see sizes.hpp). They simplify the use of 233 // ByteSize'd arguments in assembly code. Note that their equivalent 234 // for the optimized build are the member functions with int disp 235 // argument since ByteSize is mapped to an int type in that case. 236 // 237 // Note: DO NOT introduce similar overloaded functions for WordSize 238 // arguments as in the optimized mode, both ByteSize and WordSize 239 // are mapped to the same type and thus the compiler cannot make a 240 // distinction anymore (=> compiler errors). 241 242 #ifdef ASSERT 243 Address(Register base, ByteSize disp) 244 : _base(base), 245 _index(noreg), 246 _scale(no_scale), 247 _disp(in_bytes(disp)) { 248 } 249 250 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 251 : _base(base), 252 _index(index), 253 _scale(scale), 254 _disp(in_bytes(disp)) { 255 assert(!index->is_valid() == (scale == Address::no_scale), 256 "inconsistent address"); 257 } 258 259 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 260 : _base (base), 261 _index(index.register_or_noreg()), 262 _scale(scale), 263 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { 264 if (!index.is_register()) scale = Address::no_scale; 265 assert(!_index->is_valid() == (scale == Address::no_scale), 266 "inconsistent address"); 267 } 268 269 #endif // ASSERT 270 271 // accessors 272 bool uses(Register reg) const { return _base == reg || _index == reg; } 273 Register base() const { return _base; } 274 Register index() const { return _index; } 275 ScaleFactor scale() const { return _scale; } 276 int disp() const { return _disp; } 277 278 // Convert the raw encoding form into the form expected by the constructor for 279 // Address. An index of 4 (rsp) corresponds to having no index, so convert 280 // that to noreg for the Address constructor. 281 static Address make_raw(int base, int index, int scale, int disp, bool disp_is_oop); 282 283 static Address make_array(ArrayAddress); 284 285 private: 286 bool base_needs_rex() const { 287 return _base != noreg && _base->encoding() >= 8; 288 } 289 290 bool index_needs_rex() const { 291 return _index != noreg &&_index->encoding() >= 8; 292 } 293 294 relocInfo::relocType reloc() const { return _rspec.type(); } 295 296 friend class Assembler; 297 friend class MacroAssembler; 298 friend class LIR_Assembler; // base/index/scale/disp 299 }; 300 301 // 302 // AddressLiteral has been split out from Address because operands of this type 303 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 304 // the few instructions that need to deal with address literals are unique and the 305 // MacroAssembler does not have to implement every instruction in the Assembler 306 // in order to search for address literals that may need special handling depending 307 // on the instruction and the platform. As small step on the way to merging i486/amd64 308 // directories. 309 // 310 class AddressLiteral VALUE_OBJ_CLASS_SPEC { 311 friend class ArrayAddress; 312 RelocationHolder _rspec; 313 // Typically we use AddressLiterals we want to use their rval 314 // However in some situations we want the lval (effect address) of the item. 315 // We provide a special factory for making those lvals. 316 bool _is_lval; 317 318 // If the target is far we'll need to load the ea of this to 319 // a register to reach it. Otherwise if near we can do rip 320 // relative addressing. 321 322 address _target; 323 324 protected: 325 // creation 326 AddressLiteral() 327 : _is_lval(false), 328 _target(NULL) 329 {} 330 331 public: 332 333 334 AddressLiteral(address target, relocInfo::relocType rtype); 335 336 AddressLiteral(address target, RelocationHolder const& rspec) 337 : _rspec(rspec), 338 _is_lval(false), 339 _target(target) 340 {} 341 342 AddressLiteral addr() { 343 AddressLiteral ret = *this; 344 ret._is_lval = true; 345 return ret; 346 } 347 348 349 private: 350 351 address target() { return _target; } 352 bool is_lval() { return _is_lval; } 353 354 relocInfo::relocType reloc() const { return _rspec.type(); } 355 const RelocationHolder& rspec() const { return _rspec; } 356 357 friend class Assembler; 358 friend class MacroAssembler; 359 friend class Address; 360 friend class LIR_Assembler; 361 }; 362 363 // Convience classes 364 class RuntimeAddress: public AddressLiteral { 365 366 public: 367 368 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 369 370 }; 371 372 class OopAddress: public AddressLiteral { 373 374 public: 375 376 OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){} 377 378 }; 379 380 class ExternalAddress: public AddressLiteral { 381 382 public: 383 384 ExternalAddress(address target) : AddressLiteral(target, relocInfo::external_word_type){} 385 386 }; 387 388 class InternalAddress: public AddressLiteral { 389 390 public: 391 392 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 393 394 }; 395 396 // x86 can do array addressing as a single operation since disp can be an absolute 397 // address amd64 can't. We create a class that expresses the concept but does extra 398 // magic on amd64 to get the final result 399 400 class ArrayAddress VALUE_OBJ_CLASS_SPEC { 401 private: 402 403 AddressLiteral _base; 404 Address _index; 405 406 public: 407 408 ArrayAddress() {}; 409 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 410 AddressLiteral base() { return _base; } 411 Address index() { return _index; } 412 413 }; 414 415 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); 416 417 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 418 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 419 // is what you get. The Assembler is generating code into a CodeBuffer. 420 421 class Assembler : public AbstractAssembler { 422 friend class AbstractAssembler; // for the non-virtual hack 423 friend class LIR_Assembler; // as_Address() 424 friend class StubGenerator; 425 426 public: 427 enum Condition { // The x86 condition codes used for conditional jumps/moves. 428 zero = 0x4, 429 notZero = 0x5, 430 equal = 0x4, 431 notEqual = 0x5, 432 less = 0xc, 433 lessEqual = 0xe, 434 greater = 0xf, 435 greaterEqual = 0xd, 436 below = 0x2, 437 belowEqual = 0x6, 438 above = 0x7, 439 aboveEqual = 0x3, 440 overflow = 0x0, 441 noOverflow = 0x1, 442 carrySet = 0x2, 443 carryClear = 0x3, 444 negative = 0x8, 445 positive = 0x9, 446 parity = 0xa, 447 noParity = 0xb 448 }; 449 450 enum Prefix { 451 // segment overrides 452 CS_segment = 0x2e, 453 SS_segment = 0x36, 454 DS_segment = 0x3e, 455 ES_segment = 0x26, 456 FS_segment = 0x64, 457 GS_segment = 0x65, 458 459 REX = 0x40, 460 461 REX_B = 0x41, 462 REX_X = 0x42, 463 REX_XB = 0x43, 464 REX_R = 0x44, 465 REX_RB = 0x45, 466 REX_RX = 0x46, 467 REX_RXB = 0x47, 468 469 REX_W = 0x48, 470 471 REX_WB = 0x49, 472 REX_WX = 0x4A, 473 REX_WXB = 0x4B, 474 REX_WR = 0x4C, 475 REX_WRB = 0x4D, 476 REX_WRX = 0x4E, 477 REX_WRXB = 0x4F 478 }; 479 480 enum WhichOperand { 481 // input to locate_operand, and format code for relocations 482 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 483 disp32_operand = 1, // embedded 32-bit displacement or address 484 call32_operand = 2, // embedded 32-bit self-relative displacement 485 #ifndef _LP64 486 _WhichOperand_limit = 3 487 #else 488 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 489 _WhichOperand_limit = 4 490 #endif 491 }; 492 493 494 495 // NOTE: The general philopsophy of the declarations here is that 64bit versions 496 // of instructions are freely declared without the need for wrapping them an ifdef. 497 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 498 // In the .cpp file the implementations are wrapped so that they are dropped out 499 // of the resulting jvm. This is done mostly to keep the footprint of KERNEL 500 // to the size it was prior to merging up the 32bit and 64bit assemblers. 501 // 502 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 503 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 504 505 private: 506 507 508 // 64bit prefixes 509 int prefix_and_encode(int reg_enc, bool byteinst = false); 510 int prefixq_and_encode(int reg_enc); 511 512 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); 513 int prefixq_and_encode(int dst_enc, int src_enc); 514 515 void prefix(Register reg); 516 void prefix(Address adr); 517 void prefixq(Address adr); 518 519 void prefix(Address adr, Register reg, bool byteinst = false); 520 void prefixq(Address adr, Register reg); 521 522 void prefix(Address adr, XMMRegister reg); 523 524 void prefetch_prefix(Address src); 525 526 // Helper functions for groups of instructions 527 void emit_arith_b(int op1, int op2, Register dst, int imm8); 528 529 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 530 // only 32bit?? 531 void emit_arith(int op1, int op2, Register dst, jobject obj); 532 void emit_arith(int op1, int op2, Register dst, Register src); 533 534 void emit_operand(Register reg, 535 Register base, Register index, Address::ScaleFactor scale, 536 int disp, 537 RelocationHolder const& rspec, 538 int rip_relative_correction = 0); 539 540 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 541 542 // operands that only take the original 32bit registers 543 void emit_operand32(Register reg, Address adr); 544 545 void emit_operand(XMMRegister reg, 546 Register base, Register index, Address::ScaleFactor scale, 547 int disp, 548 RelocationHolder const& rspec); 549 550 void emit_operand(XMMRegister reg, Address adr); 551 552 void emit_operand(MMXRegister reg, Address adr); 553 554 // workaround gcc (3.2.1-7) bug 555 void emit_operand(Address adr, MMXRegister reg); 556 557 558 // Immediate-to-memory forms 559 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 560 561 void emit_farith(int b1, int b2, int i); 562 563 564 protected: 565 #ifdef ASSERT 566 void check_relocation(RelocationHolder const& rspec, int format); 567 #endif 568 569 inline void emit_long64(jlong x); 570 571 void emit_data(jint data, relocInfo::relocType rtype, int format); 572 void emit_data(jint data, RelocationHolder const& rspec, int format); 573 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 574 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 575 576 577 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 578 579 // These are all easily abused and hence protected 580 581 // 32BIT ONLY SECTION 582 #ifndef _LP64 583 // Make these disappear in 64bit mode since they would never be correct 584 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 585 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 586 587 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 588 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 589 590 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 591 #else 592 // 64BIT ONLY SECTION 593 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 594 595 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 596 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 597 598 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 599 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 600 #endif // _LP64 601 602 // These are unique in that we are ensured by the caller that the 32bit 603 // relative in these instructions will always be able to reach the potentially 604 // 64bit address described by entry. Since they can take a 64bit address they 605 // don't have the 32 suffix like the other instructions in this class. 606 607 void call_literal(address entry, RelocationHolder const& rspec); 608 void jmp_literal(address entry, RelocationHolder const& rspec); 609 610 // Avoid using directly section 611 // Instructions in this section are actually usable by anyone without danger 612 // of failure but have performance issues that are addressed my enhanced 613 // instructions which will do the proper thing base on the particular cpu. 614 // We protect them because we don't trust you... 615 616 // Don't use next inc() and dec() methods directly. INC & DEC instructions 617 // could cause a partial flag stall since they don't set CF flag. 618 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 619 // which call inc() & dec() or add() & sub() in accordance with 620 // the product flag UseIncDec value. 621 622 void decl(Register dst); 623 void decl(Address dst); 624 void decq(Register dst); 625 void decq(Address dst); 626 627 void incl(Register dst); 628 void incl(Address dst); 629 void incq(Register dst); 630 void incq(Address dst); 631 632 // New cpus require use of movsd and movss to avoid partial register stall 633 // when loading from memory. But for old Opteron use movlpd instead of movsd. 634 // The selection is done in MacroAssembler::movdbl() and movflt(). 635 636 // Move Scalar Single-Precision Floating-Point Values 637 void movss(XMMRegister dst, Address src); 638 void movss(XMMRegister dst, XMMRegister src); 639 void movss(Address dst, XMMRegister src); 640 641 // Move Scalar Double-Precision Floating-Point Values 642 void movsd(XMMRegister dst, Address src); 643 void movsd(XMMRegister dst, XMMRegister src); 644 void movsd(Address dst, XMMRegister src); 645 void movlpd(XMMRegister dst, Address src); 646 647 // New cpus require use of movaps and movapd to avoid partial register stall 648 // when moving between registers. 649 void movaps(XMMRegister dst, XMMRegister src); 650 void movapd(XMMRegister dst, XMMRegister src); 651 652 // End avoid using directly 653 654 655 // Instruction prefixes 656 void prefix(Prefix p); 657 658 public: 659 660 // Creation 661 Assembler(CodeBuffer* code) : AbstractAssembler(code) {} 662 663 // Decoding 664 static address locate_operand(address inst, WhichOperand which); 665 static address locate_next_instruction(address inst); 666 667 // Utilities 668 669 #ifdef _LP64 670 static bool is_simm(int64_t x, int nbits) { return -( CONST64(1) << (nbits-1) ) <= x && x < ( CONST64(1) << (nbits-1) ); } 671 static bool is_simm32(int64_t x) { return x == (int64_t)(int32_t)x; } 672 #else 673 static bool is_simm(int32_t x, int nbits) { return -( 1 << (nbits-1) ) <= x && x < ( 1 << (nbits-1) ); } 674 static bool is_simm32(int32_t x) { return true; } 675 #endif // LP64 676 677 // Generic instructions 678 // Does 32bit or 64bit as needed for the platform. In some sense these 679 // belong in macro assembler but there is no need for both varieties to exist 680 681 void lea(Register dst, Address src); 682 683 void mov(Register dst, Register src); 684 685 void pusha(); 686 void popa(); 687 688 void pushf(); 689 void popf(); 690 691 void push(int32_t imm32); 692 693 void push(Register src); 694 695 void pop(Register dst); 696 697 // These are dummies to prevent surprise implicit conversions to Register 698 void push(void* v); 699 void pop(void* v); 700 701 702 // These do register sized moves/scans 703 void rep_mov(); 704 void rep_set(); 705 void repne_scan(); 706 #ifdef _LP64 707 void repne_scanl(); 708 #endif 709 710 // Vanilla instructions in lexical order 711 712 void adcl(Register dst, int32_t imm32); 713 void adcl(Register dst, Address src); 714 void adcl(Register dst, Register src); 715 716 void adcq(Register dst, int32_t imm32); 717 void adcq(Register dst, Address src); 718 void adcq(Register dst, Register src); 719 720 721 void addl(Address dst, int32_t imm32); 722 void addl(Address dst, Register src); 723 void addl(Register dst, int32_t imm32); 724 void addl(Register dst, Address src); 725 void addl(Register dst, Register src); 726 727 void addq(Address dst, int32_t imm32); 728 void addq(Address dst, Register src); 729 void addq(Register dst, int32_t imm32); 730 void addq(Register dst, Address src); 731 void addq(Register dst, Register src); 732 733 734 void addr_nop_4(); 735 void addr_nop_5(); 736 void addr_nop_7(); 737 void addr_nop_8(); 738 739 // Add Scalar Double-Precision Floating-Point Values 740 void addsd(XMMRegister dst, Address src); 741 void addsd(XMMRegister dst, XMMRegister src); 742 743 // Add Scalar Single-Precision Floating-Point Values 744 void addss(XMMRegister dst, Address src); 745 void addss(XMMRegister dst, XMMRegister src); 746 747 void andl(Register dst, int32_t imm32); 748 void andl(Register dst, Address src); 749 void andl(Register dst, Register src); 750 751 void andq(Register dst, int32_t imm32); 752 void andq(Register dst, Address src); 753 void andq(Register dst, Register src); 754 755 756 // Bitwise Logical AND of Packed Double-Precision Floating-Point Values 757 void andpd(XMMRegister dst, Address src); 758 void andpd(XMMRegister dst, XMMRegister src); 759 760 void bsfl(Register dst, Register src); 761 762 #ifdef _LP64 763 void bsfq(Register dst, Register src); 764 #endif 765 766 void bsrl(Register dst, Register src); 767 768 #ifdef _LP64 769 void bsrq(Register dst, Register src); 770 #endif 771 772 void bswapl(Register reg); 773 774 void bswapq(Register reg); 775 776 void call(Label& L, relocInfo::relocType rtype); 777 void call(Register reg); // push pc; pc <- reg 778 void call(Address adr); // push pc; pc <- adr 779 780 void cdql(); 781 782 void cdqq(); 783 784 void cld() { emit_byte(0xfc); } 785 786 void clflush(Address adr); 787 788 void cmovl(Condition cc, Register dst, Register src); 789 void cmovl(Condition cc, Register dst, Address src); 790 791 void cmovq(Condition cc, Register dst, Register src); 792 void cmovq(Condition cc, Register dst, Address src); 793 794 795 void cmpb(Address dst, int imm8); 796 797 void cmpl(Address dst, int32_t imm32); 798 799 void cmpl(Register dst, int32_t imm32); 800 void cmpl(Register dst, Register src); 801 void cmpl(Register dst, Address src); 802 803 void cmpq(Address dst, int32_t imm32); 804 void cmpq(Address dst, Register src); 805 806 void cmpq(Register dst, int32_t imm32); 807 void cmpq(Register dst, Register src); 808 void cmpq(Register dst, Address src); 809 810 // these are dummies used to catch attempting to convert NULL to Register 811 void cmpl(Register dst, void* junk); // dummy 812 void cmpq(Register dst, void* junk); // dummy 813 814 void cmpw(Address dst, int imm16); 815 816 void cmpxchg8 (Address adr); 817 818 void cmpxchgl(Register reg, Address adr); 819 820 void cmpxchgq(Register reg, Address adr); 821 822 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 823 void comisd(XMMRegister dst, Address src); 824 825 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 826 void comiss(XMMRegister dst, Address src); 827 828 // Identify processor type and features 829 void cpuid() { 830 emit_byte(0x0F); 831 emit_byte(0xA2); 832 } 833 834 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 835 void cvtsd2ss(XMMRegister dst, XMMRegister src); 836 837 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 838 void cvtsi2sdl(XMMRegister dst, Register src); 839 void cvtsi2sdq(XMMRegister dst, Register src); 840 841 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 842 void cvtsi2ssl(XMMRegister dst, Register src); 843 void cvtsi2ssq(XMMRegister dst, Register src); 844 845 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 846 void cvtdq2pd(XMMRegister dst, XMMRegister src); 847 848 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 849 void cvtdq2ps(XMMRegister dst, XMMRegister src); 850 851 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 852 void cvtss2sd(XMMRegister dst, XMMRegister src); 853 854 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 855 void cvttsd2sil(Register dst, Address src); 856 void cvttsd2sil(Register dst, XMMRegister src); 857 void cvttsd2siq(Register dst, XMMRegister src); 858 859 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 860 void cvttss2sil(Register dst, XMMRegister src); 861 void cvttss2siq(Register dst, XMMRegister src); 862 863 // Divide Scalar Double-Precision Floating-Point Values 864 void divsd(XMMRegister dst, Address src); 865 void divsd(XMMRegister dst, XMMRegister src); 866 867 // Divide Scalar Single-Precision Floating-Point Values 868 void divss(XMMRegister dst, Address src); 869 void divss(XMMRegister dst, XMMRegister src); 870 871 void emms(); 872 873 void fabs(); 874 875 void fadd(int i); 876 877 void fadd_d(Address src); 878 void fadd_s(Address src); 879 880 // "Alternate" versions of x87 instructions place result down in FPU 881 // stack instead of on TOS 882 883 void fadda(int i); // "alternate" fadd 884 void faddp(int i = 1); 885 886 void fchs(); 887 888 void fcom(int i); 889 890 void fcomp(int i = 1); 891 void fcomp_d(Address src); 892 void fcomp_s(Address src); 893 894 void fcompp(); 895 896 void fcos(); 897 898 void fdecstp(); 899 900 void fdiv(int i); 901 void fdiv_d(Address src); 902 void fdivr_s(Address src); 903 void fdiva(int i); // "alternate" fdiv 904 void fdivp(int i = 1); 905 906 void fdivr(int i); 907 void fdivr_d(Address src); 908 void fdiv_s(Address src); 909 910 void fdivra(int i); // "alternate" reversed fdiv 911 912 void fdivrp(int i = 1); 913 914 void ffree(int i = 0); 915 916 void fild_d(Address adr); 917 void fild_s(Address adr); 918 919 void fincstp(); 920 921 void finit(); 922 923 void fist_s (Address adr); 924 void fistp_d(Address adr); 925 void fistp_s(Address adr); 926 927 void fld1(); 928 929 void fld_d(Address adr); 930 void fld_s(Address adr); 931 void fld_s(int index); 932 void fld_x(Address adr); // extended-precision (80-bit) format 933 934 void fldcw(Address src); 935 936 void fldenv(Address src); 937 938 void fldlg2(); 939 940 void fldln2(); 941 942 void fldz(); 943 944 void flog(); 945 void flog10(); 946 947 void fmul(int i); 948 949 void fmul_d(Address src); 950 void fmul_s(Address src); 951 952 void fmula(int i); // "alternate" fmul 953 954 void fmulp(int i = 1); 955 956 void fnsave(Address dst); 957 958 void fnstcw(Address src); 959 960 void fnstsw_ax(); 961 962 void fprem(); 963 void fprem1(); 964 965 void frstor(Address src); 966 967 void fsin(); 968 969 void fsqrt(); 970 971 void fst_d(Address adr); 972 void fst_s(Address adr); 973 974 void fstp_d(Address adr); 975 void fstp_d(int index); 976 void fstp_s(Address adr); 977 void fstp_x(Address adr); // extended-precision (80-bit) format 978 979 void fsub(int i); 980 void fsub_d(Address src); 981 void fsub_s(Address src); 982 983 void fsuba(int i); // "alternate" fsub 984 985 void fsubp(int i = 1); 986 987 void fsubr(int i); 988 void fsubr_d(Address src); 989 void fsubr_s(Address src); 990 991 void fsubra(int i); // "alternate" reversed fsub 992 993 void fsubrp(int i = 1); 994 995 void ftan(); 996 997 void ftst(); 998 999 void fucomi(int i = 1); 1000 void fucomip(int i = 1); 1001 1002 void fwait(); 1003 1004 void fxch(int i = 1); 1005 1006 void fxrstor(Address src); 1007 1008 void fxsave(Address dst); 1009 1010 void fyl2x(); 1011 1012 void hlt(); 1013 1014 void idivl(Register src); 1015 1016 void idivq(Register src); 1017 1018 void imull(Register dst, Register src); 1019 void imull(Register dst, Register src, int value); 1020 1021 void imulq(Register dst, Register src); 1022 void imulq(Register dst, Register src, int value); 1023 1024 1025 // jcc is the generic conditional branch generator to run- 1026 // time routines, jcc is used for branches to labels. jcc 1027 // takes a branch opcode (cc) and a label (L) and generates 1028 // either a backward branch or a forward branch and links it 1029 // to the label fixup chain. Usage: 1030 // 1031 // Label L; // unbound label 1032 // jcc(cc, L); // forward branch to unbound label 1033 // bind(L); // bind label to the current pc 1034 // jcc(cc, L); // backward branch to bound label 1035 // bind(L); // illegal: a label may be bound only once 1036 // 1037 // Note: The same Label can be used for forward and backward branches 1038 // but it may be bound only once. 1039 1040 void jcc(Condition cc, Label& L, 1041 relocInfo::relocType rtype = relocInfo::none); 1042 1043 // Conditional jump to a 8-bit offset to L. 1044 // WARNING: be very careful using this for forward jumps. If the label is 1045 // not bound within an 8-bit offset of this instruction, a run-time error 1046 // will occur. 1047 void jccb(Condition cc, Label& L); 1048 1049 void jmp(Address entry); // pc <- entry 1050 1051 // Label operations & relative jumps (PPUM Appendix D) 1052 void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none); // unconditional jump to L 1053 1054 void jmp(Register entry); // pc <- entry 1055 1056 // Unconditional 8-bit offset jump to L. 1057 // WARNING: be very careful using this for forward jumps. If the label is 1058 // not bound within an 8-bit offset of this instruction, a run-time error 1059 // will occur. 1060 void jmpb(Label& L); 1061 1062 void ldmxcsr( Address src ); 1063 1064 void leal(Register dst, Address src); 1065 1066 void leaq(Register dst, Address src); 1067 1068 void lfence() { 1069 emit_byte(0x0F); 1070 emit_byte(0xAE); 1071 emit_byte(0xE8); 1072 } 1073 1074 void lock(); 1075 1076 void lzcntl(Register dst, Register src); 1077 1078 #ifdef _LP64 1079 void lzcntq(Register dst, Register src); 1080 #endif 1081 1082 enum Membar_mask_bits { 1083 StoreStore = 1 << 3, 1084 LoadStore = 1 << 2, 1085 StoreLoad = 1 << 1, 1086 LoadLoad = 1 << 0 1087 }; 1088 1089 // Serializes memory and blows flags 1090 void membar(Membar_mask_bits order_constraint) { 1091 if (os::is_MP()) { 1092 // We only have to handle StoreLoad 1093 if (order_constraint & StoreLoad) { 1094 // All usable chips support "locked" instructions which suffice 1095 // as barriers, and are much faster than the alternative of 1096 // using cpuid instruction. We use here a locked add [esp],0. 1097 // This is conveniently otherwise a no-op except for blowing 1098 // flags. 1099 // Any change to this code may need to revisit other places in 1100 // the code where this idiom is used, in particular the 1101 // orderAccess code. 1102 lock(); 1103 addl(Address(rsp, 0), 0);// Assert the lock# signal here 1104 } 1105 } 1106 } 1107 1108 void mfence(); 1109 1110 // Moves 1111 1112 void mov64(Register dst, int64_t imm64); 1113 1114 void movb(Address dst, Register src); 1115 void movb(Address dst, int imm8); 1116 void movb(Register dst, Address src); 1117 1118 void movdl(XMMRegister dst, Register src); 1119 void movdl(Register dst, XMMRegister src); 1120 1121 // Move Double Quadword 1122 void movdq(XMMRegister dst, Register src); 1123 void movdq(Register dst, XMMRegister src); 1124 1125 // Move Aligned Double Quadword 1126 void movdqa(Address dst, XMMRegister src); 1127 void movdqa(XMMRegister dst, Address src); 1128 void movdqa(XMMRegister dst, XMMRegister src); 1129 1130 // Move Unaligned Double Quadword 1131 void movdqu(Address dst, XMMRegister src); 1132 void movdqu(XMMRegister dst, Address src); 1133 void movdqu(XMMRegister dst, XMMRegister src); 1134 1135 void movl(Register dst, int32_t imm32); 1136 void movl(Address dst, int32_t imm32); 1137 void movl(Register dst, Register src); 1138 void movl(Register dst, Address src); 1139 void movl(Address dst, Register src); 1140 1141 // These dummies prevent using movl from converting a zero (like NULL) into Register 1142 // by giving the compiler two choices it can't resolve 1143 1144 void movl(Address dst, void* junk); 1145 void movl(Register dst, void* junk); 1146 1147 #ifdef _LP64 1148 void movq(Register dst, Register src); 1149 void movq(Register dst, Address src); 1150 void movq(Address dst, Register src); 1151 #endif 1152 1153 void movq(Address dst, MMXRegister src ); 1154 void movq(MMXRegister dst, Address src ); 1155 1156 #ifdef _LP64 1157 // These dummies prevent using movq from converting a zero (like NULL) into Register 1158 // by giving the compiler two choices it can't resolve 1159 1160 void movq(Address dst, void* dummy); 1161 void movq(Register dst, void* dummy); 1162 #endif 1163 1164 // Move Quadword 1165 void movq(Address dst, XMMRegister src); 1166 void movq(XMMRegister dst, Address src); 1167 1168 void movsbl(Register dst, Address src); 1169 void movsbl(Register dst, Register src); 1170 1171 #ifdef _LP64 1172 void movsbq(Register dst, Address src); 1173 void movsbq(Register dst, Register src); 1174 1175 // Move signed 32bit immediate to 64bit extending sign 1176 void movslq(Address dst, int32_t imm64); 1177 void movslq(Register dst, int32_t imm64); 1178 1179 void movslq(Register dst, Address src); 1180 void movslq(Register dst, Register src); 1181 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1182 #endif 1183 1184 void movswl(Register dst, Address src); 1185 void movswl(Register dst, Register src); 1186 1187 #ifdef _LP64 1188 void movswq(Register dst, Address src); 1189 void movswq(Register dst, Register src); 1190 #endif 1191 1192 void movw(Address dst, int imm16); 1193 void movw(Register dst, Address src); 1194 void movw(Address dst, Register src); 1195 1196 void movzbl(Register dst, Address src); 1197 void movzbl(Register dst, Register src); 1198 1199 #ifdef _LP64 1200 void movzbq(Register dst, Address src); 1201 void movzbq(Register dst, Register src); 1202 #endif 1203 1204 void movzwl(Register dst, Address src); 1205 void movzwl(Register dst, Register src); 1206 1207 #ifdef _LP64 1208 void movzwq(Register dst, Address src); 1209 void movzwq(Register dst, Register src); 1210 #endif 1211 1212 void mull(Address src); 1213 void mull(Register src); 1214 1215 // Multiply Scalar Double-Precision Floating-Point Values 1216 void mulsd(XMMRegister dst, Address src); 1217 void mulsd(XMMRegister dst, XMMRegister src); 1218 1219 // Multiply Scalar Single-Precision Floating-Point Values 1220 void mulss(XMMRegister dst, Address src); 1221 void mulss(XMMRegister dst, XMMRegister src); 1222 1223 void negl(Register dst); 1224 1225 #ifdef _LP64 1226 void negq(Register dst); 1227 #endif 1228 1229 void nop(int i = 1); 1230 1231 void notl(Register dst); 1232 1233 #ifdef _LP64 1234 void notq(Register dst); 1235 #endif 1236 1237 void orl(Address dst, int32_t imm32); 1238 void orl(Register dst, int32_t imm32); 1239 void orl(Register dst, Address src); 1240 void orl(Register dst, Register src); 1241 1242 void orq(Address dst, int32_t imm32); 1243 void orq(Register dst, int32_t imm32); 1244 void orq(Register dst, Address src); 1245 void orq(Register dst, Register src); 1246 1247 // SSE4.2 string instructions 1248 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1249 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1250 1251 void popl(Address dst); 1252 1253 #ifdef _LP64 1254 void popq(Address dst); 1255 #endif 1256 1257 void popcntl(Register dst, Address src); 1258 void popcntl(Register dst, Register src); 1259 1260 #ifdef _LP64 1261 void popcntq(Register dst, Address src); 1262 void popcntq(Register dst, Register src); 1263 #endif 1264 1265 // Prefetches (SSE, SSE2, 3DNOW only) 1266 1267 void prefetchnta(Address src); 1268 void prefetchr(Address src); 1269 void prefetcht0(Address src); 1270 void prefetcht1(Address src); 1271 void prefetcht2(Address src); 1272 void prefetchw(Address src); 1273 1274 // Shuffle Packed Doublewords 1275 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1276 void pshufd(XMMRegister dst, Address src, int mode); 1277 1278 // Shuffle Packed Low Words 1279 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1280 void pshuflw(XMMRegister dst, Address src, int mode); 1281 1282 // Shift Right Logical Quadword Immediate 1283 void psrlq(XMMRegister dst, int shift); 1284 1285 // Logical Compare Double Quadword 1286 void ptest(XMMRegister dst, XMMRegister src); 1287 void ptest(XMMRegister dst, Address src); 1288 1289 // Interleave Low Bytes 1290 void punpcklbw(XMMRegister dst, XMMRegister src); 1291 1292 void pushl(Address src); 1293 1294 void pushq(Address src); 1295 1296 // Xor Packed Byte Integer Values 1297 void pxor(XMMRegister dst, Address src); 1298 void pxor(XMMRegister dst, XMMRegister src); 1299 1300 void rcll(Register dst, int imm8); 1301 1302 void rclq(Register dst, int imm8); 1303 1304 void ret(int imm16); 1305 1306 void sahf(); 1307 1308 void sarl(Register dst, int imm8); 1309 void sarl(Register dst); 1310 1311 void sarq(Register dst, int imm8); 1312 void sarq(Register dst); 1313 1314 void sbbl(Address dst, int32_t imm32); 1315 void sbbl(Register dst, int32_t imm32); 1316 void sbbl(Register dst, Address src); 1317 void sbbl(Register dst, Register src); 1318 1319 void sbbq(Address dst, int32_t imm32); 1320 void sbbq(Register dst, int32_t imm32); 1321 void sbbq(Register dst, Address src); 1322 void sbbq(Register dst, Register src); 1323 1324 void setb(Condition cc, Register dst); 1325 1326 void shldl(Register dst, Register src); 1327 1328 void shll(Register dst, int imm8); 1329 void shll(Register dst); 1330 1331 void shlq(Register dst, int imm8); 1332 void shlq(Register dst); 1333 1334 void shrdl(Register dst, Register src); 1335 1336 void shrl(Register dst, int imm8); 1337 void shrl(Register dst); 1338 1339 void shrq(Register dst, int imm8); 1340 void shrq(Register dst); 1341 1342 void smovl(); // QQQ generic? 1343 1344 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1345 void sqrtsd(XMMRegister dst, Address src); 1346 void sqrtsd(XMMRegister dst, XMMRegister src); 1347 1348 void std() { emit_byte(0xfd); } 1349 1350 void stmxcsr( Address dst ); 1351 1352 void subl(Address dst, int32_t imm32); 1353 void subl(Address dst, Register src); 1354 void subl(Register dst, int32_t imm32); 1355 void subl(Register dst, Address src); 1356 void subl(Register dst, Register src); 1357 1358 void subq(Address dst, int32_t imm32); 1359 void subq(Address dst, Register src); 1360 void subq(Register dst, int32_t imm32); 1361 void subq(Register dst, Address src); 1362 void subq(Register dst, Register src); 1363 1364 1365 // Subtract Scalar Double-Precision Floating-Point Values 1366 void subsd(XMMRegister dst, Address src); 1367 void subsd(XMMRegister dst, XMMRegister src); 1368 1369 // Subtract Scalar Single-Precision Floating-Point Values 1370 void subss(XMMRegister dst, Address src); 1371 void subss(XMMRegister dst, XMMRegister src); 1372 1373 void testb(Register dst, int imm8); 1374 1375 void testl(Register dst, int32_t imm32); 1376 void testl(Register dst, Register src); 1377 void testl(Register dst, Address src); 1378 1379 void testq(Register dst, int32_t imm32); 1380 void testq(Register dst, Register src); 1381 1382 1383 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1384 void ucomisd(XMMRegister dst, Address src); 1385 void ucomisd(XMMRegister dst, XMMRegister src); 1386 1387 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1388 void ucomiss(XMMRegister dst, Address src); 1389 void ucomiss(XMMRegister dst, XMMRegister src); 1390 1391 void xaddl(Address dst, Register src); 1392 1393 void xaddq(Address dst, Register src); 1394 1395 void xchgl(Register reg, Address adr); 1396 void xchgl(Register dst, Register src); 1397 1398 void xchgq(Register reg, Address adr); 1399 void xchgq(Register dst, Register src); 1400 1401 void xorl(Register dst, int32_t imm32); 1402 void xorl(Register dst, Address src); 1403 void xorl(Register dst, Register src); 1404 1405 void xorq(Register dst, Address src); 1406 void xorq(Register dst, Register src); 1407 1408 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 1409 void xorpd(XMMRegister dst, Address src); 1410 void xorpd(XMMRegister dst, XMMRegister src); 1411 1412 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 1413 void xorps(XMMRegister dst, Address src); 1414 void xorps(XMMRegister dst, XMMRegister src); 1415 1416 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1417 }; 1418 1419 1420 // MacroAssembler extends Assembler by frequently used macros. 1421 // 1422 // Instructions for which a 'better' code sequence exists depending 1423 // on arguments should also go in here. 1424 1425 class MacroAssembler: public Assembler { 1426 friend class LIR_Assembler; 1427 friend class Runtime1; // as_Address() 1428 protected: 1429 1430 Address as_Address(AddressLiteral adr); 1431 Address as_Address(ArrayAddress adr); 1432 1433 // Support for VM calls 1434 // 1435 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 1436 // may customize this version by overriding it for its purposes (e.g., to save/restore 1437 // additional registers when doing a VM call). 1438 #ifdef CC_INTERP 1439 // c++ interpreter never wants to use interp_masm version of call_VM 1440 #define VIRTUAL 1441 #else 1442 #define VIRTUAL virtual 1443 #endif 1444 1445 VIRTUAL void call_VM_leaf_base( 1446 address entry_point, // the entry point 1447 int number_of_arguments // the number of arguments to pop after the call 1448 ); 1449 1450 // This is the base routine called by the different versions of call_VM. The interpreter 1451 // may customize this version by overriding it for its purposes (e.g., to save/restore 1452 // additional registers when doing a VM call). 1453 // 1454 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base 1455 // returns the register which contains the thread upon return. If a thread register has been 1456 // specified, the return value will correspond to that register. If no last_java_sp is specified 1457 // (noreg) than rsp will be used instead. 1458 VIRTUAL void call_VM_base( // returns the register containing the thread upon return 1459 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 1460 Register java_thread, // the thread if computed before ; use noreg otherwise 1461 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 1462 address entry_point, // the entry point 1463 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 1464 bool check_exceptions // whether to check for pending exceptions after return 1465 ); 1466 1467 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 1468 // The implementation is only non-empty for the InterpreterMacroAssembler, 1469 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 1470 virtual void check_and_handle_popframe(Register java_thread); 1471 virtual void check_and_handle_earlyret(Register java_thread); 1472 1473 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 1474 1475 // helpers for FPU flag access 1476 // tmp is a temporary register, if none is available use noreg 1477 void save_rax (Register tmp); 1478 void restore_rax(Register tmp); 1479 1480 public: 1481 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 1482 1483 // Support for NULL-checks 1484 // 1485 // Generates code that causes a NULL OS exception if the content of reg is NULL. 1486 // If the accessed location is M[reg + offset] and the offset is known, provide the 1487 // offset. No explicit code generation is needed if the offset is within a certain 1488 // range (0 <= offset <= page_size). 1489 1490 void null_check(Register reg, int offset = -1); 1491 static bool needs_explicit_null_check(intptr_t offset); 1492 1493 // Required platform-specific helpers for Label::patch_instructions. 1494 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 1495 void pd_patch_instruction(address branch, address target); 1496 #ifndef PRODUCT 1497 static void pd_print_patched_instruction(address branch); 1498 #endif 1499 1500 // The following 4 methods return the offset of the appropriate move instruction 1501 1502 // Support for fast byte/short loading with zero extension (depending on particular CPU) 1503 int load_unsigned_byte(Register dst, Address src); 1504 int load_unsigned_short(Register dst, Address src); 1505 1506 // Support for fast byte/short loading with sign extension (depending on particular CPU) 1507 int load_signed_byte(Register dst, Address src); 1508 int load_signed_short(Register dst, Address src); 1509 1510 // Support for sign-extension (hi:lo = extend_sign(lo)) 1511 void extend_sign(Register hi, Register lo); 1512 1513 // Loading values by size and signed-ness 1514 void load_sized_value(Register dst, Address src, int size_in_bytes, bool is_signed); 1515 1516 // Support for inc/dec with optimal instruction selection depending on value 1517 1518 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } 1519 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } 1520 1521 void decrementl(Address dst, int value = 1); 1522 void decrementl(Register reg, int value = 1); 1523 1524 void decrementq(Register reg, int value = 1); 1525 void decrementq(Address dst, int value = 1); 1526 1527 void incrementl(Address dst, int value = 1); 1528 void incrementl(Register reg, int value = 1); 1529 1530 void incrementq(Register reg, int value = 1); 1531 void incrementq(Address dst, int value = 1); 1532 1533 1534 // Support optimal SSE move instructions. 1535 void movflt(XMMRegister dst, XMMRegister src) { 1536 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } 1537 else { movss (dst, src); return; } 1538 } 1539 void movflt(XMMRegister dst, Address src) { movss(dst, src); } 1540 void movflt(XMMRegister dst, AddressLiteral src); 1541 void movflt(Address dst, XMMRegister src) { movss(dst, src); } 1542 1543 void movdbl(XMMRegister dst, XMMRegister src) { 1544 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } 1545 else { movsd (dst, src); return; } 1546 } 1547 1548 void movdbl(XMMRegister dst, AddressLiteral src); 1549 1550 void movdbl(XMMRegister dst, Address src) { 1551 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } 1552 else { movlpd(dst, src); return; } 1553 } 1554 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } 1555 1556 void incrementl(AddressLiteral dst); 1557 void incrementl(ArrayAddress dst); 1558 1559 // Alignment 1560 void align(int modulus); 1561 1562 // Misc 1563 void fat_nop(); // 5 byte nop 1564 1565 // Stack frame creation/removal 1566 void enter(); 1567 void leave(); 1568 1569 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 1570 // The pointer will be loaded into the thread register. 1571 void get_thread(Register thread); 1572 1573 1574 // Support for VM calls 1575 // 1576 // It is imperative that all calls into the VM are handled via the call_VM macros. 1577 // They make sure that the stack linkage is setup correctly. call_VM's correspond 1578 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 1579 1580 1581 void call_VM(Register oop_result, 1582 address entry_point, 1583 bool check_exceptions = true); 1584 void call_VM(Register oop_result, 1585 address entry_point, 1586 Register arg_1, 1587 bool check_exceptions = true); 1588 void call_VM(Register oop_result, 1589 address entry_point, 1590 Register arg_1, Register arg_2, 1591 bool check_exceptions = true); 1592 void call_VM(Register oop_result, 1593 address entry_point, 1594 Register arg_1, Register arg_2, Register arg_3, 1595 bool check_exceptions = true); 1596 1597 // Overloadings with last_Java_sp 1598 void call_VM(Register oop_result, 1599 Register last_java_sp, 1600 address entry_point, 1601 int number_of_arguments = 0, 1602 bool check_exceptions = true); 1603 void call_VM(Register oop_result, 1604 Register last_java_sp, 1605 address entry_point, 1606 Register arg_1, bool 1607 check_exceptions = true); 1608 void call_VM(Register oop_result, 1609 Register last_java_sp, 1610 address entry_point, 1611 Register arg_1, Register arg_2, 1612 bool check_exceptions = true); 1613 void call_VM(Register oop_result, 1614 Register last_java_sp, 1615 address entry_point, 1616 Register arg_1, Register arg_2, Register arg_3, 1617 bool check_exceptions = true); 1618 1619 void call_VM_leaf(address entry_point, 1620 int number_of_arguments = 0); 1621 void call_VM_leaf(address entry_point, 1622 Register arg_1); 1623 void call_VM_leaf(address entry_point, 1624 Register arg_1, Register arg_2); 1625 void call_VM_leaf(address entry_point, 1626 Register arg_1, Register arg_2, Register arg_3); 1627 1628 // last Java Frame (fills frame anchor) 1629 void set_last_Java_frame(Register thread, 1630 Register last_java_sp, 1631 Register last_java_fp, 1632 address last_java_pc); 1633 1634 // thread in the default location (r15_thread on 64bit) 1635 void set_last_Java_frame(Register last_java_sp, 1636 Register last_java_fp, 1637 address last_java_pc); 1638 1639 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); 1640 1641 // thread in the default location (r15_thread on 64bit) 1642 void reset_last_Java_frame(bool clear_fp, bool clear_pc); 1643 1644 // Stores 1645 void store_check(Register obj); // store check for obj - register is destroyed afterwards 1646 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 1647 1648 void g1_write_barrier_pre(Register obj, 1649 #ifndef _LP64 1650 Register thread, 1651 #endif 1652 Register tmp, 1653 Register tmp2, 1654 bool tosca_live); 1655 void g1_write_barrier_post(Register store_addr, 1656 Register new_val, 1657 #ifndef _LP64 1658 Register thread, 1659 #endif 1660 Register tmp, 1661 Register tmp2); 1662 1663 1664 // split store_check(Register obj) to enhance instruction interleaving 1665 void store_check_part_1(Register obj); 1666 void store_check_part_2(Register obj); 1667 1668 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 1669 void c2bool(Register x); 1670 1671 // C++ bool manipulation 1672 1673 void movbool(Register dst, Address src); 1674 void movbool(Address dst, bool boolconst); 1675 void movbool(Address dst, Register src); 1676 void testbool(Register dst); 1677 1678 // oop manipulations 1679 void load_klass(Register dst, Register src); 1680 void store_klass(Register dst, Register src); 1681 1682 void load_prototype_header(Register dst, Register src); 1683 1684 #ifdef _LP64 1685 void store_klass_gap(Register dst, Register src); 1686 1687 void load_heap_oop(Register dst, Address src); 1688 void store_heap_oop(Address dst, Register src); 1689 void encode_heap_oop(Register r); 1690 void decode_heap_oop(Register r); 1691 void encode_heap_oop_not_null(Register r); 1692 void decode_heap_oop_not_null(Register r); 1693 void encode_heap_oop_not_null(Register dst, Register src); 1694 void decode_heap_oop_not_null(Register dst, Register src); 1695 1696 void set_narrow_oop(Register dst, jobject obj); 1697 void set_narrow_oop(Address dst, jobject obj); 1698 void cmp_narrow_oop(Register dst, jobject obj); 1699 void cmp_narrow_oop(Address dst, jobject obj); 1700 1701 // if heap base register is used - reinit it with the correct value 1702 void reinit_heapbase(); 1703 #endif // _LP64 1704 1705 // Int division/remainder for Java 1706 // (as idivl, but checks for special case as described in JVM spec.) 1707 // returns idivl instruction offset for implicit exception handling 1708 int corrected_idivl(Register reg); 1709 1710 // Long division/remainder for Java 1711 // (as idivq, but checks for special case as described in JVM spec.) 1712 // returns idivq instruction offset for implicit exception handling 1713 int corrected_idivq(Register reg); 1714 1715 void int3(); 1716 1717 // Long operation macros for a 32bit cpu 1718 // Long negation for Java 1719 void lneg(Register hi, Register lo); 1720 1721 // Long multiplication for Java 1722 // (destroys contents of eax, ebx, ecx and edx) 1723 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y 1724 1725 // Long shifts for Java 1726 // (semantics as described in JVM spec.) 1727 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) 1728 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) 1729 1730 // Long compare for Java 1731 // (semantics as described in JVM spec.) 1732 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) 1733 1734 1735 // misc 1736 1737 // Sign extension 1738 void sign_extend_short(Register reg); 1739 void sign_extend_byte(Register reg); 1740 1741 // Division by power of 2, rounding towards 0 1742 void division_with_shift(Register reg, int shift_value); 1743 1744 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: 1745 // 1746 // CF (corresponds to C0) if x < y 1747 // PF (corresponds to C2) if unordered 1748 // ZF (corresponds to C3) if x = y 1749 // 1750 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 1751 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) 1752 void fcmp(Register tmp); 1753 // Variant of the above which allows y to be further down the stack 1754 // and which only pops x and y if specified. If pop_right is 1755 // specified then pop_left must also be specified. 1756 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); 1757 1758 // Floating-point comparison for Java 1759 // Compares the top-most stack entries on the FPU stack and stores the result in dst. 1760 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 1761 // (semantics as described in JVM spec.) 1762 void fcmp2int(Register dst, bool unordered_is_less); 1763 // Variant of the above which allows y to be further down the stack 1764 // and which only pops x and y if specified. If pop_right is 1765 // specified then pop_left must also be specified. 1766 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); 1767 1768 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) 1769 // tmp is a temporary register, if none is available use noreg 1770 void fremr(Register tmp); 1771 1772 1773 // same as fcmp2int, but using SSE2 1774 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 1775 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 1776 1777 // Inlined sin/cos generator for Java; must not use CPU instruction 1778 // directly on Intel as it does not have high enough precision 1779 // outside of the range [-pi/4, pi/4]. Extra argument indicate the 1780 // number of FPU stack slots in use; all but the topmost will 1781 // require saving if a slow case is necessary. Assumes argument is 1782 // on FP TOS; result is on FP TOS. No cpu registers are changed by 1783 // this code. 1784 void trigfunc(char trig, int num_fpu_regs_in_use = 1); 1785 1786 // branch to L if FPU flag C2 is set/not set 1787 // tmp is a temporary register, if none is available use noreg 1788 void jC2 (Register tmp, Label& L); 1789 void jnC2(Register tmp, Label& L); 1790 1791 // Pop ST (ffree & fincstp combined) 1792 void fpop(); 1793 1794 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack 1795 void push_fTOS(); 1796 1797 // pops double TOS element from CPU stack and pushes on FPU stack 1798 void pop_fTOS(); 1799 1800 void empty_FPU_stack(); 1801 1802 void push_IU_state(); 1803 void pop_IU_state(); 1804 1805 void push_FPU_state(); 1806 void pop_FPU_state(); 1807 1808 void push_CPU_state(); 1809 void pop_CPU_state(); 1810 1811 // Round up to a power of two 1812 void round_to(Register reg, int modulus); 1813 1814 // Callee saved registers handling 1815 void push_callee_saved_registers(); 1816 void pop_callee_saved_registers(); 1817 1818 // allocation 1819 void eden_allocate( 1820 Register obj, // result: pointer to object after successful allocation 1821 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 1822 int con_size_in_bytes, // object size in bytes if known at compile time 1823 Register t1, // temp register 1824 Label& slow_case // continuation point if fast allocation fails 1825 ); 1826 void tlab_allocate( 1827 Register obj, // result: pointer to object after successful allocation 1828 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 1829 int con_size_in_bytes, // object size in bytes if known at compile time 1830 Register t1, // temp register 1831 Register t2, // temp register 1832 Label& slow_case // continuation point if fast allocation fails 1833 ); 1834 void tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); 1835 1836 // interface method calling 1837 void lookup_interface_method(Register recv_klass, 1838 Register intf_klass, 1839 RegisterOrConstant itable_index, 1840 Register method_result, 1841 Register scan_temp, 1842 Label& no_such_interface); 1843 1844 // Test sub_klass against super_klass, with fast and slow paths. 1845 1846 // The fast path produces a tri-state answer: yes / no / maybe-slow. 1847 // One of the three labels can be NULL, meaning take the fall-through. 1848 // If super_check_offset is -1, the value is loaded up from super_klass. 1849 // No registers are killed, except temp_reg. 1850 void check_klass_subtype_fast_path(Register sub_klass, 1851 Register super_klass, 1852 Register temp_reg, 1853 Label* L_success, 1854 Label* L_failure, 1855 Label* L_slow_path, 1856 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 1857 1858 // The rest of the type check; must be wired to a corresponding fast path. 1859 // It does not repeat the fast path logic, so don't use it standalone. 1860 // The temp_reg and temp2_reg can be noreg, if no temps are available. 1861 // Updates the sub's secondary super cache as necessary. 1862 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 1863 void check_klass_subtype_slow_path(Register sub_klass, 1864 Register super_klass, 1865 Register temp_reg, 1866 Register temp2_reg, 1867 Label* L_success, 1868 Label* L_failure, 1869 bool set_cond_codes = false); 1870 1871 // Simplified, combined version, good for typical uses. 1872 // Falls through on failure. 1873 void check_klass_subtype(Register sub_klass, 1874 Register super_klass, 1875 Register temp_reg, 1876 Label& L_success); 1877 1878 // method handles (JSR 292) 1879 void check_method_handle_type(Register mtype_reg, Register mh_reg, 1880 Register temp_reg, 1881 Label& wrong_method_type); 1882 void load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, 1883 Register temp_reg); 1884 void jump_to_method_handle_entry(Register mh_reg, Register temp_reg); 1885 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 1886 1887 1888 //---- 1889 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 1890 1891 // Debugging 1892 1893 // only if +VerifyOops 1894 void verify_oop(Register reg, const char* s = "broken oop"); 1895 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 1896 1897 // only if +VerifyFPU 1898 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 1899 1900 // prints msg, dumps registers and stops execution 1901 void stop(const char* msg); 1902 1903 // prints msg and continues 1904 void warn(const char* msg); 1905 1906 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); 1907 static void debug64(char* msg, int64_t pc, int64_t regs[]); 1908 1909 void os_breakpoint(); 1910 1911 void untested() { stop("untested"); } 1912 1913 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, sizeof(b), "unimplemented: %s", what); stop(b); } 1914 1915 void should_not_reach_here() { stop("should not reach here"); } 1916 1917 void print_CPU_state(); 1918 1919 // Stack overflow checking 1920 void bang_stack_with_offset(int offset) { 1921 // stack grows down, caller passes positive offset 1922 assert(offset > 0, "must bang with negative offset"); 1923 movl(Address(rsp, (-offset)), rax); 1924 } 1925 1926 // Writes to stack successive pages until offset reached to check for 1927 // stack overflow + shadow pages. Also, clobbers tmp 1928 void bang_stack_size(Register size, Register tmp); 1929 1930 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 1931 Register tmp, 1932 int offset); 1933 1934 // Support for serializing memory accesses between threads 1935 void serialize_memory(Register thread, Register tmp); 1936 1937 void verify_tlab(); 1938 1939 // Biased locking support 1940 // lock_reg and obj_reg must be loaded up with the appropriate values. 1941 // swap_reg must be rax, and is killed. 1942 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will 1943 // be killed; if not supplied, push/pop will be used internally to 1944 // allocate a temporary (inefficient, avoid if possible). 1945 // Optional slow case is for implementations (interpreter and C1) which branch to 1946 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 1947 // Returns offset of first potentially-faulting instruction for null 1948 // check info (currently consumed only by C1). If 1949 // swap_reg_contains_mark is true then returns -1 as it is assumed 1950 // the calling code has already passed any potential faults. 1951 int biased_locking_enter(Register lock_reg, Register obj_reg, 1952 Register swap_reg, Register tmp_reg, 1953 bool swap_reg_contains_mark, 1954 Label& done, Label* slow_case = NULL, 1955 BiasedLockingCounters* counters = NULL); 1956 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 1957 1958 1959 Condition negate_condition(Condition cond); 1960 1961 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit 1962 // operands. In general the names are modified to avoid hiding the instruction in Assembler 1963 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers 1964 // here in MacroAssembler. The major exception to this rule is call 1965 1966 // Arithmetics 1967 1968 1969 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } 1970 void addptr(Address dst, Register src); 1971 1972 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } 1973 void addptr(Register dst, int32_t src); 1974 void addptr(Register dst, Register src); 1975 1976 void andptr(Register dst, int32_t src); 1977 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } 1978 1979 void cmp8(AddressLiteral src1, int imm); 1980 1981 // renamed to drag out the casting of address to int32_t/intptr_t 1982 void cmp32(Register src1, int32_t imm); 1983 1984 void cmp32(AddressLiteral src1, int32_t imm); 1985 // compare reg - mem, or reg - &mem 1986 void cmp32(Register src1, AddressLiteral src2); 1987 1988 void cmp32(Register src1, Address src2); 1989 1990 #ifndef _LP64 1991 void cmpoop(Address dst, jobject obj); 1992 void cmpoop(Register dst, jobject obj); 1993 #endif // _LP64 1994 1995 // NOTE src2 must be the lval. This is NOT an mem-mem compare 1996 void cmpptr(Address src1, AddressLiteral src2); 1997 1998 void cmpptr(Register src1, AddressLiteral src2); 1999 2000 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2001 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2002 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2003 2004 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2005 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2006 2007 // cmp64 to avoild hiding cmpq 2008 void cmp64(Register src1, AddressLiteral src); 2009 2010 void cmpxchgptr(Register reg, Address adr); 2011 2012 void locked_cmpxchgptr(Register reg, AddressLiteral adr); 2013 2014 2015 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } 2016 2017 2018 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } 2019 2020 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } 2021 2022 void shlptr(Register dst, int32_t shift); 2023 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } 2024 2025 void shrptr(Register dst, int32_t shift); 2026 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } 2027 2028 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } 2029 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } 2030 2031 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2032 2033 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2034 void subptr(Register dst, int32_t src); 2035 void subptr(Register dst, Register src); 2036 2037 2038 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2039 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2040 2041 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2042 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2043 2044 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } 2045 2046 2047 2048 // Helper functions for statistics gathering. 2049 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. 2050 void cond_inc32(Condition cond, AddressLiteral counter_addr); 2051 // Unconditional atomic increment. 2052 void atomic_incl(AddressLiteral counter_addr); 2053 2054 void lea(Register dst, AddressLiteral adr); 2055 void lea(Address dst, AddressLiteral adr); 2056 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } 2057 2058 void leal32(Register dst, Address src) { leal(dst, src); } 2059 2060 void test32(Register src1, AddressLiteral src2); 2061 2062 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2063 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2064 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2065 2066 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } 2067 void testptr(Register src1, Register src2); 2068 2069 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2070 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2071 2072 // Calls 2073 2074 void call(Label& L, relocInfo::relocType rtype); 2075 void call(Register entry); 2076 2077 // NOTE: this call tranfers to the effective address of entry NOT 2078 // the address contained by entry. This is because this is more natural 2079 // for jumps/calls. 2080 void call(AddressLiteral entry); 2081 2082 // Jumps 2083 2084 // NOTE: these jumps tranfer to the effective address of dst NOT 2085 // the address contained by dst. This is because this is more natural 2086 // for jumps/calls. 2087 void jump(AddressLiteral dst); 2088 void jump_cc(Condition cc, AddressLiteral dst); 2089 2090 // 32bit can do a case table jump in one instruction but we no longer allow the base 2091 // to be installed in the Address class. This jump will tranfers to the address 2092 // contained in the location described by entry (not the address of entry) 2093 void jump(ArrayAddress entry); 2094 2095 // Floating 2096 2097 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } 2098 void andpd(XMMRegister dst, AddressLiteral src); 2099 2100 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } 2101 void comiss(XMMRegister dst, AddressLiteral src); 2102 2103 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } 2104 void comisd(XMMRegister dst, AddressLiteral src); 2105 2106 void fldcw(Address src) { Assembler::fldcw(src); } 2107 void fldcw(AddressLiteral src); 2108 2109 void fld_s(int index) { Assembler::fld_s(index); } 2110 void fld_s(Address src) { Assembler::fld_s(src); } 2111 void fld_s(AddressLiteral src); 2112 2113 void fld_d(Address src) { Assembler::fld_d(src); } 2114 void fld_d(AddressLiteral src); 2115 2116 void fld_x(Address src) { Assembler::fld_x(src); } 2117 void fld_x(AddressLiteral src); 2118 2119 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } 2120 void ldmxcsr(AddressLiteral src); 2121 2122 private: 2123 // these are private because users should be doing movflt/movdbl 2124 2125 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } 2126 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } 2127 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } 2128 void movss(XMMRegister dst, AddressLiteral src); 2129 2130 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } 2131 void movlpd(XMMRegister dst, AddressLiteral src); 2132 2133 public: 2134 2135 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } 2136 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } 2137 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } 2138 void movsd(XMMRegister dst, AddressLiteral src); 2139 2140 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } 2141 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } 2142 void ucomiss(XMMRegister dst, AddressLiteral src); 2143 2144 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } 2145 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } 2146 void ucomisd(XMMRegister dst, AddressLiteral src); 2147 2148 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 2149 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); } 2150 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } 2151 void xorpd(XMMRegister dst, AddressLiteral src); 2152 2153 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 2154 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); } 2155 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } 2156 void xorps(XMMRegister dst, AddressLiteral src); 2157 2158 // Data 2159 2160 void cmov(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } 2161 2162 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } 2163 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmovl(cc, dst, src)); } 2164 2165 void movoop(Register dst, jobject obj); 2166 void movoop(Address dst, jobject obj); 2167 2168 void movptr(ArrayAddress dst, Register src); 2169 // can this do an lea? 2170 void movptr(Register dst, ArrayAddress src); 2171 2172 void movptr(Register dst, Address src); 2173 2174 void movptr(Register dst, AddressLiteral src); 2175 2176 void movptr(Register dst, intptr_t src); 2177 void movptr(Register dst, Register src); 2178 void movptr(Address dst, intptr_t src); 2179 2180 void movptr(Address dst, Register src); 2181 2182 #ifdef _LP64 2183 // Generally the next two are only used for moving NULL 2184 // Although there are situations in initializing the mark word where 2185 // they could be used. They are dangerous. 2186 2187 // They only exist on LP64 so that int32_t and intptr_t are not the same 2188 // and we have ambiguous declarations. 2189 2190 void movptr(Address dst, int32_t imm32); 2191 void movptr(Register dst, int32_t imm32); 2192 #endif // _LP64 2193 2194 // to avoid hiding movl 2195 void mov32(AddressLiteral dst, Register src); 2196 void mov32(Register dst, AddressLiteral src); 2197 2198 // to avoid hiding movb 2199 void movbyte(ArrayAddress dst, int src); 2200 2201 // Can push value or effective address 2202 void pushptr(AddressLiteral src); 2203 2204 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } 2205 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } 2206 2207 void pushoop(jobject obj); 2208 2209 // sign extend as need a l to ptr sized element 2210 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } 2211 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } 2212 2213 2214 #undef VIRTUAL 2215 2216 }; 2217 2218 /** 2219 * class SkipIfEqual: 2220 * 2221 * Instantiating this class will result in assembly code being output that will 2222 * jump around any code emitted between the creation of the instance and it's 2223 * automatic destruction at the end of a scope block, depending on the value of 2224 * the flag passed to the constructor, which will be checked at run-time. 2225 */ 2226 class SkipIfEqual { 2227 private: 2228 MacroAssembler* _masm; 2229 Label _label; 2230 2231 public: 2232 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 2233 ~SkipIfEqual(); 2234 }; 2235 2236 #ifdef ASSERT 2237 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } 2238 #endif