1 /* 2 * Copyright (c) 1997, 2011, 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 } 237 238 // The following two overloads are used in connection with the 239 // ByteSize type (see sizes.hpp). They simplify the use of 240 // ByteSize'd arguments in assembly code. Note that their equivalent 241 // for the optimized build are the member functions with int disp 242 // argument since ByteSize is mapped to an int type in that case. 243 // 244 // Note: DO NOT introduce similar overloaded functions for WordSize 245 // arguments as in the optimized mode, both ByteSize and WordSize 246 // are mapped to the same type and thus the compiler cannot make a 247 // distinction anymore (=> compiler errors). 248 249 #ifdef ASSERT 250 Address(Register base, ByteSize disp) 251 : _base(base), 252 _index(noreg), 253 _scale(no_scale), 254 _disp(in_bytes(disp)) { 255 } 256 257 Address(Register base, Register index, ScaleFactor scale, ByteSize disp) 258 : _base(base), 259 _index(index), 260 _scale(scale), 261 _disp(in_bytes(disp)) { 262 assert(!index->is_valid() == (scale == Address::no_scale), 263 "inconsistent address"); 264 } 265 266 Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp) 267 : _base (base), 268 _index(index.register_or_noreg()), 269 _scale(scale), 270 _disp (in_bytes(disp) + (index.constant_or_zero() * scale_size(scale))) { 271 if (!index.is_register()) scale = Address::no_scale; 272 assert(!_index->is_valid() == (scale == Address::no_scale), 273 "inconsistent address"); 274 } 275 276 #endif // ASSERT 277 278 // accessors 279 bool uses(Register reg) const { return _base == reg || _index == reg; } 280 Register base() const { return _base; } 281 Register index() const { return _index; } 282 ScaleFactor scale() const { return _scale; } 283 int disp() const { return _disp; } 284 285 // Convert the raw encoding form into the form expected by the constructor for 286 // Address. An index of 4 (rsp) corresponds to having no index, so convert 287 // that to noreg for the Address constructor. 288 static Address make_raw(int base, int index, int scale, int disp, bool disp_is_oop); 289 290 static Address make_array(ArrayAddress); 291 292 private: 293 bool base_needs_rex() const { 294 return _base != noreg && _base->encoding() >= 8; 295 } 296 297 bool index_needs_rex() const { 298 return _index != noreg &&_index->encoding() >= 8; 299 } 300 301 relocInfo::relocType reloc() const { return _rspec.type(); } 302 303 friend class Assembler; 304 friend class MacroAssembler; 305 friend class LIR_Assembler; // base/index/scale/disp 306 }; 307 308 // 309 // AddressLiteral has been split out from Address because operands of this type 310 // need to be treated specially on 32bit vs. 64bit platforms. By splitting it out 311 // the few instructions that need to deal with address literals are unique and the 312 // MacroAssembler does not have to implement every instruction in the Assembler 313 // in order to search for address literals that may need special handling depending 314 // on the instruction and the platform. As small step on the way to merging i486/amd64 315 // directories. 316 // 317 class AddressLiteral VALUE_OBJ_CLASS_SPEC { 318 friend class ArrayAddress; 319 RelocationHolder _rspec; 320 // Typically we use AddressLiterals we want to use their rval 321 // However in some situations we want the lval (effect address) of the item. 322 // We provide a special factory for making those lvals. 323 bool _is_lval; 324 325 // If the target is far we'll need to load the ea of this to 326 // a register to reach it. Otherwise if near we can do rip 327 // relative addressing. 328 329 address _target; 330 331 protected: 332 // creation 333 AddressLiteral() 334 : _is_lval(false), 335 _target(NULL) 336 {} 337 338 public: 339 340 341 AddressLiteral(address target, relocInfo::relocType rtype); 342 343 AddressLiteral(address target, RelocationHolder const& rspec) 344 : _rspec(rspec), 345 _is_lval(false), 346 _target(target) 347 {} 348 349 AddressLiteral addr() { 350 AddressLiteral ret = *this; 351 ret._is_lval = true; 352 return ret; 353 } 354 355 356 private: 357 358 address target() { return _target; } 359 bool is_lval() { return _is_lval; } 360 361 relocInfo::relocType reloc() const { return _rspec.type(); } 362 const RelocationHolder& rspec() const { return _rspec; } 363 364 friend class Assembler; 365 friend class MacroAssembler; 366 friend class Address; 367 friend class LIR_Assembler; 368 }; 369 370 // Convience classes 371 class RuntimeAddress: public AddressLiteral { 372 373 public: 374 375 RuntimeAddress(address target) : AddressLiteral(target, relocInfo::runtime_call_type) {} 376 377 }; 378 379 class OopAddress: public AddressLiteral { 380 381 public: 382 383 OopAddress(address target) : AddressLiteral(target, relocInfo::oop_type){} 384 385 }; 386 387 class ExternalAddress: public AddressLiteral { 388 private: 389 static relocInfo::relocType reloc_for_target(address target) { 390 // Sometimes ExternalAddress is used for values which aren't 391 // exactly addresses, like the card table base. 392 // external_word_type can't be used for values in the first page 393 // so just skip the reloc in that case. 394 return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none; 395 } 396 397 public: 398 399 ExternalAddress(address target) : AddressLiteral(target, reloc_for_target(target)) {} 400 401 }; 402 403 class InternalAddress: public AddressLiteral { 404 405 public: 406 407 InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {} 408 409 }; 410 411 // x86 can do array addressing as a single operation since disp can be an absolute 412 // address amd64 can't. We create a class that expresses the concept but does extra 413 // magic on amd64 to get the final result 414 415 class ArrayAddress VALUE_OBJ_CLASS_SPEC { 416 private: 417 418 AddressLiteral _base; 419 Address _index; 420 421 public: 422 423 ArrayAddress() {}; 424 ArrayAddress(AddressLiteral base, Address index): _base(base), _index(index) {}; 425 AddressLiteral base() { return _base; } 426 Address index() { return _index; } 427 428 }; 429 430 const int FPUStateSizeInWords = NOT_LP64(27) LP64_ONLY( 512 / wordSize); 431 432 // The Intel x86/Amd64 Assembler: Pure assembler doing NO optimizations on the instruction 433 // level (e.g. mov rax, 0 is not translated into xor rax, rax!); i.e., what you write 434 // is what you get. The Assembler is generating code into a CodeBuffer. 435 436 class Assembler : public AbstractAssembler { 437 friend class AbstractAssembler; // for the non-virtual hack 438 friend class LIR_Assembler; // as_Address() 439 friend class StubGenerator; 440 441 public: 442 enum Condition { // The x86 condition codes used for conditional jumps/moves. 443 zero = 0x4, 444 notZero = 0x5, 445 equal = 0x4, 446 notEqual = 0x5, 447 less = 0xc, 448 lessEqual = 0xe, 449 greater = 0xf, 450 greaterEqual = 0xd, 451 below = 0x2, 452 belowEqual = 0x6, 453 above = 0x7, 454 aboveEqual = 0x3, 455 overflow = 0x0, 456 noOverflow = 0x1, 457 carrySet = 0x2, 458 carryClear = 0x3, 459 negative = 0x8, 460 positive = 0x9, 461 parity = 0xa, 462 noParity = 0xb 463 }; 464 465 enum Prefix { 466 // segment overrides 467 CS_segment = 0x2e, 468 SS_segment = 0x36, 469 DS_segment = 0x3e, 470 ES_segment = 0x26, 471 FS_segment = 0x64, 472 GS_segment = 0x65, 473 474 REX = 0x40, 475 476 REX_B = 0x41, 477 REX_X = 0x42, 478 REX_XB = 0x43, 479 REX_R = 0x44, 480 REX_RB = 0x45, 481 REX_RX = 0x46, 482 REX_RXB = 0x47, 483 484 REX_W = 0x48, 485 486 REX_WB = 0x49, 487 REX_WX = 0x4A, 488 REX_WXB = 0x4B, 489 REX_WR = 0x4C, 490 REX_WRB = 0x4D, 491 REX_WRX = 0x4E, 492 REX_WRXB = 0x4F 493 }; 494 495 enum WhichOperand { 496 // input to locate_operand, and format code for relocations 497 imm_operand = 0, // embedded 32-bit|64-bit immediate operand 498 disp32_operand = 1, // embedded 32-bit displacement or address 499 call32_operand = 2, // embedded 32-bit self-relative displacement 500 #ifndef _LP64 501 _WhichOperand_limit = 3 502 #else 503 narrow_oop_operand = 3, // embedded 32-bit immediate narrow oop 504 _WhichOperand_limit = 4 505 #endif 506 }; 507 508 509 510 // NOTE: The general philopsophy of the declarations here is that 64bit versions 511 // of instructions are freely declared without the need for wrapping them an ifdef. 512 // (Some dangerous instructions are ifdef's out of inappropriate jvm's.) 513 // In the .cpp file the implementations are wrapped so that they are dropped out 514 // of the resulting jvm. This is done mostly to keep the footprint of KERNEL 515 // to the size it was prior to merging up the 32bit and 64bit assemblers. 516 // 517 // This does mean you'll get a linker/runtime error if you use a 64bit only instruction 518 // in a 32bit vm. This is somewhat unfortunate but keeps the ifdef noise down. 519 520 private: 521 522 523 // 64bit prefixes 524 int prefix_and_encode(int reg_enc, bool byteinst = false); 525 int prefixq_and_encode(int reg_enc); 526 527 int prefix_and_encode(int dst_enc, int src_enc, bool byteinst = false); 528 int prefixq_and_encode(int dst_enc, int src_enc); 529 530 void prefix(Register reg); 531 void prefix(Address adr); 532 void prefixq(Address adr); 533 534 void prefix(Address adr, Register reg, bool byteinst = false); 535 void prefixq(Address adr, Register reg); 536 537 void prefix(Address adr, XMMRegister reg); 538 539 void prefetch_prefix(Address src); 540 541 // Helper functions for groups of instructions 542 void emit_arith_b(int op1, int op2, Register dst, int imm8); 543 544 void emit_arith(int op1, int op2, Register dst, int32_t imm32); 545 // only 32bit?? 546 void emit_arith(int op1, int op2, Register dst, jobject obj); 547 void emit_arith(int op1, int op2, Register dst, Register src); 548 549 void emit_operand(Register reg, 550 Register base, Register index, Address::ScaleFactor scale, 551 int disp, 552 RelocationHolder const& rspec, 553 int rip_relative_correction = 0); 554 555 void emit_operand(Register reg, Address adr, int rip_relative_correction = 0); 556 557 // operands that only take the original 32bit registers 558 void emit_operand32(Register reg, Address adr); 559 560 void emit_operand(XMMRegister reg, 561 Register base, Register index, Address::ScaleFactor scale, 562 int disp, 563 RelocationHolder const& rspec); 564 565 void emit_operand(XMMRegister reg, Address adr); 566 567 void emit_operand(MMXRegister reg, Address adr); 568 569 // workaround gcc (3.2.1-7) bug 570 void emit_operand(Address adr, MMXRegister reg); 571 572 573 // Immediate-to-memory forms 574 void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32); 575 576 void emit_farith(int b1, int b2, int i); 577 578 579 protected: 580 #ifdef ASSERT 581 void check_relocation(RelocationHolder const& rspec, int format); 582 #endif 583 584 inline void emit_long64(jlong x); 585 586 void emit_data(jint data, relocInfo::relocType rtype, int format); 587 void emit_data(jint data, RelocationHolder const& rspec, int format); 588 void emit_data64(jlong data, relocInfo::relocType rtype, int format = 0); 589 void emit_data64(jlong data, RelocationHolder const& rspec, int format = 0); 590 591 bool reachable(AddressLiteral adr) NOT_LP64({ return true;}); 592 593 // These are all easily abused and hence protected 594 595 // 32BIT ONLY SECTION 596 #ifndef _LP64 597 // Make these disappear in 64bit mode since they would never be correct 598 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 599 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 600 601 void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 602 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 603 604 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 605 #else 606 // 64BIT ONLY SECTION 607 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 608 609 void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 610 void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 611 612 void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 613 void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 614 #endif // _LP64 615 616 // These are unique in that we are ensured by the caller that the 32bit 617 // relative in these instructions will always be able to reach the potentially 618 // 64bit address described by entry. Since they can take a 64bit address they 619 // don't have the 32 suffix like the other instructions in this class. 620 621 void call_literal(address entry, RelocationHolder const& rspec); 622 void jmp_literal(address entry, RelocationHolder const& rspec); 623 624 // Avoid using directly section 625 // Instructions in this section are actually usable by anyone without danger 626 // of failure but have performance issues that are addressed my enhanced 627 // instructions which will do the proper thing base on the particular cpu. 628 // We protect them because we don't trust you... 629 630 // Don't use next inc() and dec() methods directly. INC & DEC instructions 631 // could cause a partial flag stall since they don't set CF flag. 632 // Use MacroAssembler::decrement() & MacroAssembler::increment() methods 633 // which call inc() & dec() or add() & sub() in accordance with 634 // the product flag UseIncDec value. 635 636 void decl(Register dst); 637 void decl(Address dst); 638 void decq(Register dst); 639 void decq(Address dst); 640 641 void incl(Register dst); 642 void incl(Address dst); 643 void incq(Register dst); 644 void incq(Address dst); 645 646 // New cpus require use of movsd and movss to avoid partial register stall 647 // when loading from memory. But for old Opteron use movlpd instead of movsd. 648 // The selection is done in MacroAssembler::movdbl() and movflt(). 649 650 // Move Scalar Single-Precision Floating-Point Values 651 void movss(XMMRegister dst, Address src); 652 void movss(XMMRegister dst, XMMRegister src); 653 void movss(Address dst, XMMRegister src); 654 655 // Move Scalar Double-Precision Floating-Point Values 656 void movsd(XMMRegister dst, Address src); 657 void movsd(XMMRegister dst, XMMRegister src); 658 void movsd(Address dst, XMMRegister src); 659 void movlpd(XMMRegister dst, Address src); 660 661 // New cpus require use of movaps and movapd to avoid partial register stall 662 // when moving between registers. 663 void movaps(XMMRegister dst, XMMRegister src); 664 void movapd(XMMRegister dst, XMMRegister src); 665 666 // End avoid using directly 667 668 669 // Instruction prefixes 670 void prefix(Prefix p); 671 672 public: 673 674 // Creation 675 Assembler(CodeBuffer* code) : AbstractAssembler(code) {} 676 677 // Decoding 678 static address locate_operand(address inst, WhichOperand which); 679 static address locate_next_instruction(address inst); 680 681 // Utilities 682 683 #ifdef _LP64 684 static bool is_simm(int64_t x, int nbits) { return -(CONST64(1) << (nbits-1)) <= x && 685 x < (CONST64(1) << (nbits-1)); } 686 static bool is_simm32(int64_t x) { return x == (int64_t)(int32_t)x; } 687 #else 688 static bool is_simm(int32_t x, int nbits) { return -(1 << (nbits-1)) <= x && 689 x < (1 << (nbits-1)); } 690 static bool is_simm32(int32_t x) { return true; } 691 #endif // _LP64 692 693 static bool is_polling_page_far() NOT_LP64({ return false;}); 694 695 // Generic instructions 696 // Does 32bit or 64bit as needed for the platform. In some sense these 697 // belong in macro assembler but there is no need for both varieties to exist 698 699 void lea(Register dst, Address src); 700 701 void mov(Register dst, Register src); 702 703 void pusha(); 704 void popa(); 705 706 void pushf(); 707 void popf(); 708 709 void push(int32_t imm32); 710 711 void push(Register src); 712 713 void pop(Register dst); 714 715 // These are dummies to prevent surprise implicit conversions to Register 716 void push(void* v); 717 void pop(void* v); 718 719 // These do register sized moves/scans 720 void rep_mov(); 721 void rep_set(); 722 void repne_scan(); 723 #ifdef _LP64 724 void repne_scanl(); 725 #endif 726 727 // Vanilla instructions in lexical order 728 729 void adcl(Address dst, int32_t imm32); 730 void adcl(Address dst, Register src); 731 void adcl(Register dst, int32_t imm32); 732 void adcl(Register dst, Address src); 733 void adcl(Register dst, Register src); 734 735 void adcq(Register dst, int32_t imm32); 736 void adcq(Register dst, Address src); 737 void adcq(Register dst, Register src); 738 739 void addl(Address dst, int32_t imm32); 740 void addl(Address dst, Register src); 741 void addl(Register dst, int32_t imm32); 742 void addl(Register dst, Address src); 743 void addl(Register dst, Register src); 744 745 void addq(Address dst, int32_t imm32); 746 void addq(Address dst, Register src); 747 void addq(Register dst, int32_t imm32); 748 void addq(Register dst, Address src); 749 void addq(Register dst, Register src); 750 751 void addr_nop_4(); 752 void addr_nop_5(); 753 void addr_nop_7(); 754 void addr_nop_8(); 755 756 // Add Scalar Double-Precision Floating-Point Values 757 void addsd(XMMRegister dst, Address src); 758 void addsd(XMMRegister dst, XMMRegister src); 759 760 // Add Scalar Single-Precision Floating-Point Values 761 void addss(XMMRegister dst, Address src); 762 void addss(XMMRegister dst, XMMRegister src); 763 764 void andl(Register dst, int32_t imm32); 765 void andl(Register dst, Address src); 766 void andl(Register dst, Register src); 767 768 void andq(Register dst, int32_t imm32); 769 void andq(Register dst, Address src); 770 void andq(Register dst, Register src); 771 772 // Bitwise Logical AND of Packed Double-Precision Floating-Point Values 773 void andpd(XMMRegister dst, Address src); 774 void andpd(XMMRegister dst, XMMRegister src); 775 776 void bsfl(Register dst, Register src); 777 void bsrl(Register dst, Register src); 778 779 #ifdef _LP64 780 void bsfq(Register dst, Register src); 781 void bsrq(Register dst, Register src); 782 #endif 783 784 void bswapl(Register reg); 785 786 void bswapq(Register reg); 787 788 void call(Label& L, relocInfo::relocType rtype); 789 void call(Register reg); // push pc; pc <- reg 790 void call(Address adr); // push pc; pc <- adr 791 792 void cdql(); 793 794 void cdqq(); 795 796 void cld() { emit_byte(0xfc); } 797 798 void clflush(Address adr); 799 800 void cmovl(Condition cc, Register dst, Register src); 801 void cmovl(Condition cc, Register dst, Address src); 802 803 void cmovq(Condition cc, Register dst, Register src); 804 void cmovq(Condition cc, Register dst, Address src); 805 806 807 void cmpb(Address dst, int imm8); 808 809 void cmpl(Address dst, int32_t imm32); 810 811 void cmpl(Register dst, int32_t imm32); 812 void cmpl(Register dst, Register src); 813 void cmpl(Register dst, Address src); 814 815 void cmpq(Address dst, int32_t imm32); 816 void cmpq(Address dst, Register src); 817 818 void cmpq(Register dst, int32_t imm32); 819 void cmpq(Register dst, Register src); 820 void cmpq(Register dst, Address src); 821 822 // these are dummies used to catch attempting to convert NULL to Register 823 void cmpl(Register dst, void* junk); // dummy 824 void cmpq(Register dst, void* junk); // dummy 825 826 void cmpw(Address dst, int imm16); 827 828 void cmpxchg8 (Address adr); 829 830 void cmpxchgl(Register reg, Address adr); 831 832 void cmpxchgq(Register reg, Address adr); 833 834 // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 835 void comisd(XMMRegister dst, Address src); 836 837 // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 838 void comiss(XMMRegister dst, Address src); 839 840 // Identify processor type and features 841 void cpuid() { 842 emit_byte(0x0F); 843 emit_byte(0xA2); 844 } 845 846 // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value 847 void cvtsd2ss(XMMRegister dst, XMMRegister src); 848 849 // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value 850 void cvtsi2sdl(XMMRegister dst, Register src); 851 void cvtsi2sdq(XMMRegister dst, Register src); 852 853 // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value 854 void cvtsi2ssl(XMMRegister dst, Register src); 855 void cvtsi2ssq(XMMRegister dst, Register src); 856 857 // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value 858 void cvtdq2pd(XMMRegister dst, XMMRegister src); 859 860 // Convert Packed Signed Doubleword Integers to Packed Single-Precision Floating-Point Value 861 void cvtdq2ps(XMMRegister dst, XMMRegister src); 862 863 // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value 864 void cvtss2sd(XMMRegister dst, XMMRegister src); 865 866 // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer 867 void cvttsd2sil(Register dst, Address src); 868 void cvttsd2sil(Register dst, XMMRegister src); 869 void cvttsd2siq(Register dst, XMMRegister src); 870 871 // Convert with Truncation Scalar Single-Precision Floating-Point Value to Doubleword Integer 872 void cvttss2sil(Register dst, XMMRegister src); 873 void cvttss2siq(Register dst, XMMRegister src); 874 875 // Divide Scalar Double-Precision Floating-Point Values 876 void divsd(XMMRegister dst, Address src); 877 void divsd(XMMRegister dst, XMMRegister src); 878 879 // Divide Scalar Single-Precision Floating-Point Values 880 void divss(XMMRegister dst, Address src); 881 void divss(XMMRegister dst, XMMRegister src); 882 883 void emms(); 884 885 void fabs(); 886 887 void fadd(int i); 888 889 void fadd_d(Address src); 890 void fadd_s(Address src); 891 892 // "Alternate" versions of x87 instructions place result down in FPU 893 // stack instead of on TOS 894 895 void fadda(int i); // "alternate" fadd 896 void faddp(int i = 1); 897 898 void fchs(); 899 900 void fcom(int i); 901 902 void fcomp(int i = 1); 903 void fcomp_d(Address src); 904 void fcomp_s(Address src); 905 906 void fcompp(); 907 908 void fcos(); 909 910 void fdecstp(); 911 912 void fdiv(int i); 913 void fdiv_d(Address src); 914 void fdivr_s(Address src); 915 void fdiva(int i); // "alternate" fdiv 916 void fdivp(int i = 1); 917 918 void fdivr(int i); 919 void fdivr_d(Address src); 920 void fdiv_s(Address src); 921 922 void fdivra(int i); // "alternate" reversed fdiv 923 924 void fdivrp(int i = 1); 925 926 void ffree(int i = 0); 927 928 void fild_d(Address adr); 929 void fild_s(Address adr); 930 931 void fincstp(); 932 933 void finit(); 934 935 void fist_s (Address adr); 936 void fistp_d(Address adr); 937 void fistp_s(Address adr); 938 939 void fld1(); 940 941 void fld_d(Address adr); 942 void fld_s(Address adr); 943 void fld_s(int index); 944 void fld_x(Address adr); // extended-precision (80-bit) format 945 946 void fldcw(Address src); 947 948 void fldenv(Address src); 949 950 void fldlg2(); 951 952 void fldln2(); 953 954 void fldz(); 955 956 void flog(); 957 void flog10(); 958 959 void fmul(int i); 960 961 void fmul_d(Address src); 962 void fmul_s(Address src); 963 964 void fmula(int i); // "alternate" fmul 965 966 void fmulp(int i = 1); 967 968 void fnsave(Address dst); 969 970 void fnstcw(Address src); 971 972 void fnstsw_ax(); 973 974 void fprem(); 975 void fprem1(); 976 977 void frstor(Address src); 978 979 void fsin(); 980 981 void fsqrt(); 982 983 void fst_d(Address adr); 984 void fst_s(Address adr); 985 986 void fstp_d(Address adr); 987 void fstp_d(int index); 988 void fstp_s(Address adr); 989 void fstp_x(Address adr); // extended-precision (80-bit) format 990 991 void fsub(int i); 992 void fsub_d(Address src); 993 void fsub_s(Address src); 994 995 void fsuba(int i); // "alternate" fsub 996 997 void fsubp(int i = 1); 998 999 void fsubr(int i); 1000 void fsubr_d(Address src); 1001 void fsubr_s(Address src); 1002 1003 void fsubra(int i); // "alternate" reversed fsub 1004 1005 void fsubrp(int i = 1); 1006 1007 void ftan(); 1008 1009 void ftst(); 1010 1011 void fucomi(int i = 1); 1012 void fucomip(int i = 1); 1013 1014 void fwait(); 1015 1016 void fxch(int i = 1); 1017 1018 void fxrstor(Address src); 1019 1020 void fxsave(Address dst); 1021 1022 void fyl2x(); 1023 1024 void hlt(); 1025 1026 void idivl(Register src); 1027 void divl(Register src); // Unsigned division 1028 1029 void idivq(Register src); 1030 1031 void imull(Register dst, Register src); 1032 void imull(Register dst, Register src, int value); 1033 1034 void imulq(Register dst, Register src); 1035 void imulq(Register dst, Register src, int value); 1036 1037 1038 // jcc is the generic conditional branch generator to run- 1039 // time routines, jcc is used for branches to labels. jcc 1040 // takes a branch opcode (cc) and a label (L) and generates 1041 // either a backward branch or a forward branch and links it 1042 // to the label fixup chain. Usage: 1043 // 1044 // Label L; // unbound label 1045 // jcc(cc, L); // forward branch to unbound label 1046 // bind(L); // bind label to the current pc 1047 // jcc(cc, L); // backward branch to bound label 1048 // bind(L); // illegal: a label may be bound only once 1049 // 1050 // Note: The same Label can be used for forward and backward branches 1051 // but it may be bound only once. 1052 1053 void jcc(Condition cc, Label& L, 1054 relocInfo::relocType rtype = relocInfo::none); 1055 1056 // Conditional jump to a 8-bit offset 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 jccb(Condition cc, Label& L); 1061 1062 void jmp(Address entry); // pc <- entry 1063 1064 // Label operations & relative jumps (PPUM Appendix D) 1065 void jmp(Label& L, relocInfo::relocType rtype = relocInfo::none); // unconditional jump to L 1066 1067 void jmp(Register entry); // pc <- entry 1068 1069 // Unconditional 8-bit offset jump to L. 1070 // WARNING: be very careful using this for forward jumps. If the label is 1071 // not bound within an 8-bit offset of this instruction, a run-time error 1072 // will occur. 1073 void jmpb(Label& L); 1074 1075 void ldmxcsr( Address src ); 1076 1077 void leal(Register dst, Address src); 1078 1079 void leaq(Register dst, Address src); 1080 1081 void lfence() { 1082 emit_byte(0x0F); 1083 emit_byte(0xAE); 1084 emit_byte(0xE8); 1085 } 1086 1087 void lock(); 1088 1089 void lzcntl(Register dst, Register src); 1090 1091 #ifdef _LP64 1092 void lzcntq(Register dst, Register src); 1093 #endif 1094 1095 enum Membar_mask_bits { 1096 StoreStore = 1 << 3, 1097 LoadStore = 1 << 2, 1098 StoreLoad = 1 << 1, 1099 LoadLoad = 1 << 0 1100 }; 1101 1102 // Serializes memory and blows flags 1103 void membar(Membar_mask_bits order_constraint) { 1104 if (os::is_MP()) { 1105 // We only have to handle StoreLoad 1106 if (order_constraint & StoreLoad) { 1107 // All usable chips support "locked" instructions which suffice 1108 // as barriers, and are much faster than the alternative of 1109 // using cpuid instruction. We use here a locked add [esp],0. 1110 // This is conveniently otherwise a no-op except for blowing 1111 // flags. 1112 // Any change to this code may need to revisit other places in 1113 // the code where this idiom is used, in particular the 1114 // orderAccess code. 1115 lock(); 1116 addl(Address(rsp, 0), 0);// Assert the lock# signal here 1117 } 1118 } 1119 } 1120 1121 void mfence(); 1122 1123 // Moves 1124 1125 void mov64(Register dst, int64_t imm64); 1126 1127 void movb(Address dst, Register src); 1128 void movb(Address dst, int imm8); 1129 void movb(Register dst, Address src); 1130 1131 void movdl(XMMRegister dst, Register src); 1132 void movdl(Register dst, XMMRegister src); 1133 void movdl(XMMRegister dst, Address src); 1134 1135 // Move Double Quadword 1136 void movdq(XMMRegister dst, Register src); 1137 void movdq(Register dst, XMMRegister src); 1138 1139 // Move Aligned Double Quadword 1140 void movdqa(Address dst, XMMRegister src); 1141 void movdqa(XMMRegister dst, Address src); 1142 void movdqa(XMMRegister dst, XMMRegister src); 1143 1144 // Move Unaligned Double Quadword 1145 void movdqu(Address dst, XMMRegister src); 1146 void movdqu(XMMRegister dst, Address src); 1147 void movdqu(XMMRegister dst, XMMRegister src); 1148 1149 void movl(Register dst, int32_t imm32); 1150 void movl(Address dst, int32_t imm32); 1151 void movl(Register dst, Register src); 1152 void movl(Register dst, Address src); 1153 void movl(Address dst, Register src); 1154 1155 // These dummies prevent using movl from converting a zero (like NULL) into Register 1156 // by giving the compiler two choices it can't resolve 1157 1158 void movl(Address dst, void* junk); 1159 void movl(Register dst, void* junk); 1160 1161 #ifdef _LP64 1162 void movq(Register dst, Register src); 1163 void movq(Register dst, Address src); 1164 void movq(Address dst, Register src); 1165 #endif 1166 1167 void movq(Address dst, MMXRegister src ); 1168 void movq(MMXRegister dst, Address src ); 1169 1170 #ifdef _LP64 1171 // These dummies prevent using movq from converting a zero (like NULL) into Register 1172 // by giving the compiler two choices it can't resolve 1173 1174 void movq(Address dst, void* dummy); 1175 void movq(Register dst, void* dummy); 1176 #endif 1177 1178 // Move Quadword 1179 void movq(Address dst, XMMRegister src); 1180 void movq(XMMRegister dst, Address src); 1181 1182 void movsbl(Register dst, Address src); 1183 void movsbl(Register dst, Register src); 1184 1185 #ifdef _LP64 1186 void movsbq(Register dst, Address src); 1187 void movsbq(Register dst, Register src); 1188 1189 // Move signed 32bit immediate to 64bit extending sign 1190 void movslq(Address dst, int32_t imm64); 1191 void movslq(Register dst, int32_t imm64); 1192 1193 void movslq(Register dst, Address src); 1194 void movslq(Register dst, Register src); 1195 void movslq(Register dst, void* src); // Dummy declaration to cause NULL to be ambiguous 1196 #endif 1197 1198 void movswl(Register dst, Address src); 1199 void movswl(Register dst, Register src); 1200 1201 #ifdef _LP64 1202 void movswq(Register dst, Address src); 1203 void movswq(Register dst, Register src); 1204 #endif 1205 1206 void movw(Address dst, int imm16); 1207 void movw(Register dst, Address src); 1208 void movw(Address dst, Register src); 1209 1210 void movzbl(Register dst, Address src); 1211 void movzbl(Register dst, Register src); 1212 1213 #ifdef _LP64 1214 void movzbq(Register dst, Address src); 1215 void movzbq(Register dst, Register src); 1216 #endif 1217 1218 void movzwl(Register dst, Address src); 1219 void movzwl(Register dst, Register src); 1220 1221 #ifdef _LP64 1222 void movzwq(Register dst, Address src); 1223 void movzwq(Register dst, Register src); 1224 #endif 1225 1226 void mull(Address src); 1227 void mull(Register src); 1228 1229 // Multiply Scalar Double-Precision Floating-Point Values 1230 void mulsd(XMMRegister dst, Address src); 1231 void mulsd(XMMRegister dst, XMMRegister src); 1232 1233 // Multiply Scalar Single-Precision Floating-Point Values 1234 void mulss(XMMRegister dst, Address src); 1235 void mulss(XMMRegister dst, XMMRegister src); 1236 1237 void negl(Register dst); 1238 1239 #ifdef _LP64 1240 void negq(Register dst); 1241 #endif 1242 1243 void nop(int i = 1); 1244 1245 void notl(Register dst); 1246 1247 #ifdef _LP64 1248 void notq(Register dst); 1249 #endif 1250 1251 void orl(Address dst, int32_t imm32); 1252 void orl(Register dst, int32_t imm32); 1253 void orl(Register dst, Address src); 1254 void orl(Register dst, Register src); 1255 1256 void orq(Address dst, int32_t imm32); 1257 void orq(Register dst, int32_t imm32); 1258 void orq(Register dst, Address src); 1259 void orq(Register dst, Register src); 1260 1261 // SSE4.2 string instructions 1262 void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8); 1263 void pcmpestri(XMMRegister xmm1, Address src, int imm8); 1264 1265 #ifndef _LP64 // no 32bit push/pop on amd64 1266 void popl(Address dst); 1267 #endif 1268 1269 #ifdef _LP64 1270 void popq(Address dst); 1271 #endif 1272 1273 void popcntl(Register dst, Address src); 1274 void popcntl(Register dst, Register src); 1275 1276 #ifdef _LP64 1277 void popcntq(Register dst, Address src); 1278 void popcntq(Register dst, Register src); 1279 #endif 1280 1281 // Prefetches (SSE, SSE2, 3DNOW only) 1282 1283 void prefetchnta(Address src); 1284 void prefetchr(Address src); 1285 void prefetcht0(Address src); 1286 void prefetcht1(Address src); 1287 void prefetcht2(Address src); 1288 void prefetchw(Address src); 1289 1290 // POR - Bitwise logical OR 1291 void por(XMMRegister dst, XMMRegister src); 1292 1293 // Shuffle Packed Doublewords 1294 void pshufd(XMMRegister dst, XMMRegister src, int mode); 1295 void pshufd(XMMRegister dst, Address src, int mode); 1296 1297 // Shuffle Packed Low Words 1298 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1299 void pshuflw(XMMRegister dst, Address src, int mode); 1300 1301 // Shift Right by bits Logical Quadword Immediate 1302 void psrlq(XMMRegister dst, int shift); 1303 1304 // Shift Right by bytes Logical DoubleQuadword Immediate 1305 void psrldq(XMMRegister dst, int shift); 1306 1307 // Logical Compare Double Quadword 1308 void ptest(XMMRegister dst, XMMRegister src); 1309 void ptest(XMMRegister dst, Address src); 1310 1311 // Interleave Low Bytes 1312 void punpcklbw(XMMRegister dst, XMMRegister src); 1313 1314 #ifndef _LP64 // no 32bit push/pop on amd64 1315 void pushl(Address src); 1316 #endif 1317 1318 void pushq(Address src); 1319 1320 // Xor Packed Byte Integer Values 1321 void pxor(XMMRegister dst, Address src); 1322 void pxor(XMMRegister dst, XMMRegister src); 1323 1324 void rcll(Register dst, int imm8); 1325 1326 void rclq(Register dst, int imm8); 1327 1328 void ret(int imm16); 1329 1330 void sahf(); 1331 1332 void sarl(Register dst, int imm8); 1333 void sarl(Register dst); 1334 1335 void sarq(Register dst, int imm8); 1336 void sarq(Register dst); 1337 1338 void sbbl(Address dst, int32_t imm32); 1339 void sbbl(Register dst, int32_t imm32); 1340 void sbbl(Register dst, Address src); 1341 void sbbl(Register dst, Register src); 1342 1343 void sbbq(Address dst, int32_t imm32); 1344 void sbbq(Register dst, int32_t imm32); 1345 void sbbq(Register dst, Address src); 1346 void sbbq(Register dst, Register src); 1347 1348 void setb(Condition cc, Register dst); 1349 1350 void shldl(Register dst, Register src); 1351 1352 void shll(Register dst, int imm8); 1353 void shll(Register dst); 1354 1355 void shlq(Register dst, int imm8); 1356 void shlq(Register dst); 1357 1358 void shrdl(Register dst, Register src); 1359 1360 void shrl(Register dst, int imm8); 1361 void shrl(Register dst); 1362 1363 void shrq(Register dst, int imm8); 1364 void shrq(Register dst); 1365 1366 void smovl(); // QQQ generic? 1367 1368 // Compute Square Root of Scalar Double-Precision Floating-Point Value 1369 void sqrtsd(XMMRegister dst, Address src); 1370 void sqrtsd(XMMRegister dst, XMMRegister src); 1371 1372 // Compute Square Root of Scalar Single-Precision Floating-Point Value 1373 void sqrtss(XMMRegister dst, Address src); 1374 void sqrtss(XMMRegister dst, XMMRegister src); 1375 1376 void std() { emit_byte(0xfd); } 1377 1378 void stmxcsr( Address dst ); 1379 1380 void subl(Address dst, int32_t imm32); 1381 void subl(Address dst, Register src); 1382 void subl(Register dst, int32_t imm32); 1383 void subl(Register dst, Address src); 1384 void subl(Register dst, Register src); 1385 1386 void subq(Address dst, int32_t imm32); 1387 void subq(Address dst, Register src); 1388 void subq(Register dst, int32_t imm32); 1389 void subq(Register dst, Address src); 1390 void subq(Register dst, Register src); 1391 1392 1393 // Subtract Scalar Double-Precision Floating-Point Values 1394 void subsd(XMMRegister dst, Address src); 1395 void subsd(XMMRegister dst, XMMRegister src); 1396 1397 // Subtract Scalar Single-Precision Floating-Point Values 1398 void subss(XMMRegister dst, Address src); 1399 void subss(XMMRegister dst, XMMRegister src); 1400 1401 void testb(Register dst, int imm8); 1402 1403 void testl(Register dst, int32_t imm32); 1404 void testl(Register dst, Register src); 1405 void testl(Register dst, Address src); 1406 1407 void testq(Register dst, int32_t imm32); 1408 void testq(Register dst, Register src); 1409 1410 1411 // Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS 1412 void ucomisd(XMMRegister dst, Address src); 1413 void ucomisd(XMMRegister dst, XMMRegister src); 1414 1415 // Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS 1416 void ucomiss(XMMRegister dst, Address src); 1417 void ucomiss(XMMRegister dst, XMMRegister src); 1418 1419 void xaddl(Address dst, Register src); 1420 1421 void xaddq(Address dst, Register src); 1422 1423 void xchgl(Register reg, Address adr); 1424 void xchgl(Register dst, Register src); 1425 1426 void xchgq(Register reg, Address adr); 1427 void xchgq(Register dst, Register src); 1428 1429 void xorl(Register dst, int32_t imm32); 1430 void xorl(Register dst, Address src); 1431 void xorl(Register dst, Register src); 1432 1433 void xorq(Register dst, Address src); 1434 void xorq(Register dst, Register src); 1435 1436 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 1437 void xorpd(XMMRegister dst, Address src); 1438 void xorpd(XMMRegister dst, XMMRegister src); 1439 1440 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 1441 void xorps(XMMRegister dst, Address src); 1442 void xorps(XMMRegister dst, XMMRegister src); 1443 1444 void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0 1445 }; 1446 1447 1448 // MacroAssembler extends Assembler by frequently used macros. 1449 // 1450 // Instructions for which a 'better' code sequence exists depending 1451 // on arguments should also go in here. 1452 1453 class MacroAssembler: public Assembler { 1454 friend class LIR_Assembler; 1455 friend class Runtime1; // as_Address() 1456 protected: 1457 1458 Address as_Address(AddressLiteral adr); 1459 Address as_Address(ArrayAddress adr); 1460 1461 // Support for VM calls 1462 // 1463 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 1464 // may customize this version by overriding it for its purposes (e.g., to save/restore 1465 // additional registers when doing a VM call). 1466 #ifdef CC_INTERP 1467 // c++ interpreter never wants to use interp_masm version of call_VM 1468 #define VIRTUAL 1469 #else 1470 #define VIRTUAL virtual 1471 #endif 1472 1473 VIRTUAL void call_VM_leaf_base( 1474 address entry_point, // the entry point 1475 int number_of_arguments // the number of arguments to pop after the call 1476 ); 1477 1478 // This is the base routine called by the different versions of call_VM. The interpreter 1479 // may customize this version by overriding it for its purposes (e.g., to save/restore 1480 // additional registers when doing a VM call). 1481 // 1482 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base 1483 // returns the register which contains the thread upon return. If a thread register has been 1484 // specified, the return value will correspond to that register. If no last_java_sp is specified 1485 // (noreg) than rsp will be used instead. 1486 VIRTUAL void call_VM_base( // returns the register containing the thread upon return 1487 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 1488 Register java_thread, // the thread if computed before ; use noreg otherwise 1489 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 1490 address entry_point, // the entry point 1491 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 1492 bool check_exceptions // whether to check for pending exceptions after return 1493 ); 1494 1495 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 1496 // The implementation is only non-empty for the InterpreterMacroAssembler, 1497 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 1498 virtual void check_and_handle_popframe(Register java_thread); 1499 virtual void check_and_handle_earlyret(Register java_thread); 1500 1501 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 1502 1503 // helpers for FPU flag access 1504 // tmp is a temporary register, if none is available use noreg 1505 void save_rax (Register tmp); 1506 void restore_rax(Register tmp); 1507 1508 public: 1509 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 1510 1511 // Support for NULL-checks 1512 // 1513 // Generates code that causes a NULL OS exception if the content of reg is NULL. 1514 // If the accessed location is M[reg + offset] and the offset is known, provide the 1515 // offset. No explicit code generation is needed if the offset is within a certain 1516 // range (0 <= offset <= page_size). 1517 1518 void null_check(Register reg, int offset = -1); 1519 static bool needs_explicit_null_check(intptr_t offset); 1520 1521 // Required platform-specific helpers for Label::patch_instructions. 1522 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 1523 void pd_patch_instruction(address branch, address target); 1524 #ifndef PRODUCT 1525 static void pd_print_patched_instruction(address branch); 1526 #endif 1527 1528 // The following 4 methods return the offset of the appropriate move instruction 1529 1530 // Support for fast byte/short loading with zero extension (depending on particular CPU) 1531 int load_unsigned_byte(Register dst, Address src); 1532 int load_unsigned_short(Register dst, Address src); 1533 1534 // Support for fast byte/short loading with sign extension (depending on particular CPU) 1535 int load_signed_byte(Register dst, Address src); 1536 int load_signed_short(Register dst, Address src); 1537 1538 // Support for sign-extension (hi:lo = extend_sign(lo)) 1539 void extend_sign(Register hi, Register lo); 1540 1541 // Load and store values by size and signed-ness 1542 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); 1543 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); 1544 1545 // Support for inc/dec with optimal instruction selection depending on value 1546 1547 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } 1548 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } 1549 1550 void decrementl(Address dst, int value = 1); 1551 void decrementl(Register reg, int value = 1); 1552 1553 void decrementq(Register reg, int value = 1); 1554 void decrementq(Address dst, int value = 1); 1555 1556 void incrementl(Address dst, int value = 1); 1557 void incrementl(Register reg, int value = 1); 1558 1559 void incrementq(Register reg, int value = 1); 1560 void incrementq(Address dst, int value = 1); 1561 1562 1563 // Support optimal SSE move instructions. 1564 void movflt(XMMRegister dst, XMMRegister src) { 1565 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } 1566 else { movss (dst, src); return; } 1567 } 1568 void movflt(XMMRegister dst, Address src) { movss(dst, src); } 1569 void movflt(XMMRegister dst, AddressLiteral src); 1570 void movflt(Address dst, XMMRegister src) { movss(dst, src); } 1571 1572 void movdbl(XMMRegister dst, XMMRegister src) { 1573 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } 1574 else { movsd (dst, src); return; } 1575 } 1576 1577 void movdbl(XMMRegister dst, AddressLiteral src); 1578 1579 void movdbl(XMMRegister dst, Address src) { 1580 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } 1581 else { movlpd(dst, src); return; } 1582 } 1583 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } 1584 1585 void incrementl(AddressLiteral dst); 1586 void incrementl(ArrayAddress dst); 1587 1588 // Alignment 1589 void align(int modulus); 1590 1591 // Misc 1592 void fat_nop(); // 5 byte nop 1593 1594 // Stack frame creation/removal 1595 void enter(); 1596 void leave(); 1597 1598 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 1599 // The pointer will be loaded into the thread register. 1600 void get_thread(Register thread); 1601 1602 1603 // Support for VM calls 1604 // 1605 // It is imperative that all calls into the VM are handled via the call_VM macros. 1606 // They make sure that the stack linkage is setup correctly. call_VM's correspond 1607 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 1608 1609 1610 void call_VM(Register oop_result, 1611 address entry_point, 1612 bool check_exceptions = true); 1613 void call_VM(Register oop_result, 1614 address entry_point, 1615 Register arg_1, 1616 bool check_exceptions = true); 1617 void call_VM(Register oop_result, 1618 address entry_point, 1619 Register arg_1, Register arg_2, 1620 bool check_exceptions = true); 1621 void call_VM(Register oop_result, 1622 address entry_point, 1623 Register arg_1, Register arg_2, Register arg_3, 1624 bool check_exceptions = true); 1625 1626 // Overloadings with last_Java_sp 1627 void call_VM(Register oop_result, 1628 Register last_java_sp, 1629 address entry_point, 1630 int number_of_arguments = 0, 1631 bool check_exceptions = true); 1632 void call_VM(Register oop_result, 1633 Register last_java_sp, 1634 address entry_point, 1635 Register arg_1, bool 1636 check_exceptions = true); 1637 void call_VM(Register oop_result, 1638 Register last_java_sp, 1639 address entry_point, 1640 Register arg_1, Register arg_2, 1641 bool check_exceptions = true); 1642 void call_VM(Register oop_result, 1643 Register last_java_sp, 1644 address entry_point, 1645 Register arg_1, Register arg_2, Register arg_3, 1646 bool check_exceptions = true); 1647 1648 void call_VM_leaf(address entry_point, 1649 int number_of_arguments = 0); 1650 void call_VM_leaf(address entry_point, 1651 Register arg_1); 1652 void call_VM_leaf(address entry_point, 1653 Register arg_1, Register arg_2); 1654 void call_VM_leaf(address entry_point, 1655 Register arg_1, Register arg_2, Register arg_3); 1656 1657 // last Java Frame (fills frame anchor) 1658 void set_last_Java_frame(Register thread, 1659 Register last_java_sp, 1660 Register last_java_fp, 1661 address last_java_pc); 1662 1663 // thread in the default location (r15_thread on 64bit) 1664 void set_last_Java_frame(Register last_java_sp, 1665 Register last_java_fp, 1666 address last_java_pc); 1667 1668 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); 1669 1670 // thread in the default location (r15_thread on 64bit) 1671 void reset_last_Java_frame(bool clear_fp, bool clear_pc); 1672 1673 // Stores 1674 void store_check(Register obj); // store check for obj - register is destroyed afterwards 1675 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 1676 1677 void g1_write_barrier_pre(Register obj, 1678 #ifndef _LP64 1679 Register thread, 1680 #endif 1681 Register tmp, 1682 Register tmp2, 1683 bool tosca_live); 1684 void g1_write_barrier_post(Register store_addr, 1685 Register new_val, 1686 #ifndef _LP64 1687 Register thread, 1688 #endif 1689 Register tmp, 1690 Register tmp2); 1691 1692 1693 // split store_check(Register obj) to enhance instruction interleaving 1694 void store_check_part_1(Register obj); 1695 void store_check_part_2(Register obj); 1696 1697 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 1698 void c2bool(Register x); 1699 1700 // C++ bool manipulation 1701 1702 void movbool(Register dst, Address src); 1703 void movbool(Address dst, bool boolconst); 1704 void movbool(Address dst, Register src); 1705 void testbool(Register dst); 1706 1707 // oop manipulations 1708 void load_klass(Register dst, Register src); 1709 void store_klass(Register dst, Register src); 1710 1711 void load_heap_oop(Register dst, Address src); 1712 void load_heap_oop_not_null(Register dst, Address src); 1713 void store_heap_oop(Address dst, Register src); 1714 1715 // Used for storing NULL. All other oop constants should be 1716 // stored using routines that take a jobject. 1717 void store_heap_oop_null(Address dst); 1718 1719 void load_prototype_header(Register dst, Register src); 1720 1721 #ifdef _LP64 1722 void store_klass_gap(Register dst, Register src); 1723 1724 // This dummy is to prevent a call to store_heap_oop from 1725 // converting a zero (like NULL) into a Register by giving 1726 // the compiler two choices it can't resolve 1727 1728 void store_heap_oop(Address dst, void* dummy); 1729 1730 void encode_heap_oop(Register r); 1731 void decode_heap_oop(Register r); 1732 void encode_heap_oop_not_null(Register r); 1733 void decode_heap_oop_not_null(Register r); 1734 void encode_heap_oop_not_null(Register dst, Register src); 1735 void decode_heap_oop_not_null(Register dst, Register src); 1736 1737 void set_narrow_oop(Register dst, jobject obj); 1738 void set_narrow_oop(Address dst, jobject obj); 1739 void cmp_narrow_oop(Register dst, jobject obj); 1740 void cmp_narrow_oop(Address dst, jobject obj); 1741 1742 // if heap base register is used - reinit it with the correct value 1743 void reinit_heapbase(); 1744 1745 DEBUG_ONLY(void verify_heapbase(const char* msg);) 1746 1747 #endif // _LP64 1748 1749 // Int division/remainder for Java 1750 // (as idivl, but checks for special case as described in JVM spec.) 1751 // returns idivl instruction offset for implicit exception handling 1752 int corrected_idivl(Register reg); 1753 1754 // Long division/remainder for Java 1755 // (as idivq, but checks for special case as described in JVM spec.) 1756 // returns idivq instruction offset for implicit exception handling 1757 int corrected_idivq(Register reg); 1758 1759 void int3(); 1760 1761 // Long operation macros for a 32bit cpu 1762 // Long negation for Java 1763 void lneg(Register hi, Register lo); 1764 1765 // Long multiplication for Java 1766 // (destroys contents of eax, ebx, ecx and edx) 1767 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y 1768 1769 // Long shifts for Java 1770 // (semantics as described in JVM spec.) 1771 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) 1772 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) 1773 1774 // Long compare for Java 1775 // (semantics as described in JVM spec.) 1776 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) 1777 1778 1779 // misc 1780 1781 // Sign extension 1782 void sign_extend_short(Register reg); 1783 void sign_extend_byte(Register reg); 1784 1785 // Division by power of 2, rounding towards 0 1786 void division_with_shift(Register reg, int shift_value); 1787 1788 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: 1789 // 1790 // CF (corresponds to C0) if x < y 1791 // PF (corresponds to C2) if unordered 1792 // ZF (corresponds to C3) if x = y 1793 // 1794 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 1795 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) 1796 void fcmp(Register tmp); 1797 // Variant of the above which allows y to be further down the stack 1798 // and which only pops x and y if specified. If pop_right is 1799 // specified then pop_left must also be specified. 1800 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); 1801 1802 // Floating-point comparison for Java 1803 // Compares the top-most stack entries on the FPU stack and stores the result in dst. 1804 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 1805 // (semantics as described in JVM spec.) 1806 void fcmp2int(Register dst, bool unordered_is_less); 1807 // Variant of the above which allows y to be further down the stack 1808 // and which only pops x and y if specified. If pop_right is 1809 // specified then pop_left must also be specified. 1810 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); 1811 1812 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) 1813 // tmp is a temporary register, if none is available use noreg 1814 void fremr(Register tmp); 1815 1816 1817 // same as fcmp2int, but using SSE2 1818 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 1819 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 1820 1821 // Inlined sin/cos generator for Java; must not use CPU instruction 1822 // directly on Intel as it does not have high enough precision 1823 // outside of the range [-pi/4, pi/4]. Extra argument indicate the 1824 // number of FPU stack slots in use; all but the topmost will 1825 // require saving if a slow case is necessary. Assumes argument is 1826 // on FP TOS; result is on FP TOS. No cpu registers are changed by 1827 // this code. 1828 void trigfunc(char trig, int num_fpu_regs_in_use = 1); 1829 1830 // branch to L if FPU flag C2 is set/not set 1831 // tmp is a temporary register, if none is available use noreg 1832 void jC2 (Register tmp, Label& L); 1833 void jnC2(Register tmp, Label& L); 1834 1835 // Pop ST (ffree & fincstp combined) 1836 void fpop(); 1837 1838 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack 1839 void push_fTOS(); 1840 1841 // pops double TOS element from CPU stack and pushes on FPU stack 1842 void pop_fTOS(); 1843 1844 void empty_FPU_stack(); 1845 1846 void push_IU_state(); 1847 void pop_IU_state(); 1848 1849 void push_FPU_state(); 1850 void pop_FPU_state(); 1851 1852 void push_CPU_state(); 1853 void pop_CPU_state(); 1854 1855 // Round up to a power of two 1856 void round_to(Register reg, int modulus); 1857 1858 // Callee saved registers handling 1859 void push_callee_saved_registers(); 1860 void pop_callee_saved_registers(); 1861 1862 // allocation 1863 void eden_allocate( 1864 Register obj, // result: pointer to object after successful allocation 1865 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 1866 int con_size_in_bytes, // object size in bytes if known at compile time 1867 Register t1, // temp register 1868 Label& slow_case // continuation point if fast allocation fails 1869 ); 1870 void tlab_allocate( 1871 Register obj, // result: pointer to object after successful allocation 1872 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 1873 int con_size_in_bytes, // object size in bytes if known at compile time 1874 Register t1, // temp register 1875 Register t2, // temp register 1876 Label& slow_case // continuation point if fast allocation fails 1877 ); 1878 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address 1879 void incr_allocated_bytes(Register thread, 1880 Register var_size_in_bytes, int con_size_in_bytes, 1881 Register t1 = noreg); 1882 1883 // interface method calling 1884 void lookup_interface_method(Register recv_klass, 1885 Register intf_klass, 1886 RegisterOrConstant itable_index, 1887 Register method_result, 1888 Register scan_temp, 1889 Label& no_such_interface); 1890 1891 // Test sub_klass against super_klass, with fast and slow paths. 1892 1893 // The fast path produces a tri-state answer: yes / no / maybe-slow. 1894 // One of the three labels can be NULL, meaning take the fall-through. 1895 // If super_check_offset is -1, the value is loaded up from super_klass. 1896 // No registers are killed, except temp_reg. 1897 void check_klass_subtype_fast_path(Register sub_klass, 1898 Register super_klass, 1899 Register temp_reg, 1900 Label* L_success, 1901 Label* L_failure, 1902 Label* L_slow_path, 1903 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 1904 1905 // The rest of the type check; must be wired to a corresponding fast path. 1906 // It does not repeat the fast path logic, so don't use it standalone. 1907 // The temp_reg and temp2_reg can be noreg, if no temps are available. 1908 // Updates the sub's secondary super cache as necessary. 1909 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 1910 void check_klass_subtype_slow_path(Register sub_klass, 1911 Register super_klass, 1912 Register temp_reg, 1913 Register temp2_reg, 1914 Label* L_success, 1915 Label* L_failure, 1916 bool set_cond_codes = false); 1917 1918 // Simplified, combined version, good for typical uses. 1919 // Falls through on failure. 1920 void check_klass_subtype(Register sub_klass, 1921 Register super_klass, 1922 Register temp_reg, 1923 Label& L_success); 1924 1925 // method handles (JSR 292) 1926 void check_method_handle_type(Register mtype_reg, Register mh_reg, 1927 Register temp_reg, 1928 Label& wrong_method_type); 1929 void load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, 1930 Register temp_reg); 1931 void jump_to_method_handle_entry(Register mh_reg, Register temp_reg); 1932 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 1933 1934 1935 //---- 1936 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 1937 1938 // Debugging 1939 1940 // only if +VerifyOops 1941 void verify_oop(Register reg, const char* s = "broken oop"); 1942 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 1943 1944 // only if +VerifyFPU 1945 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 1946 1947 // prints msg, dumps registers and stops execution 1948 void stop(const char* msg); 1949 1950 // prints msg and continues 1951 void warn(const char* msg); 1952 1953 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); 1954 static void debug64(char* msg, int64_t pc, int64_t regs[]); 1955 1956 void os_breakpoint(); 1957 1958 void untested() { stop("untested"); } 1959 1960 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } 1961 1962 void should_not_reach_here() { stop("should not reach here"); } 1963 1964 void print_CPU_state(); 1965 1966 // Stack overflow checking 1967 void bang_stack_with_offset(int offset) { 1968 // stack grows down, caller passes positive offset 1969 assert(offset > 0, "must bang with negative offset"); 1970 movl(Address(rsp, (-offset)), rax); 1971 } 1972 1973 // Writes to stack successive pages until offset reached to check for 1974 // stack overflow + shadow pages. Also, clobbers tmp 1975 void bang_stack_size(Register size, Register tmp); 1976 1977 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 1978 Register tmp, 1979 int offset); 1980 1981 // Support for serializing memory accesses between threads 1982 void serialize_memory(Register thread, Register tmp); 1983 1984 void verify_tlab(); 1985 1986 // Biased locking support 1987 // lock_reg and obj_reg must be loaded up with the appropriate values. 1988 // swap_reg must be rax, and is killed. 1989 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will 1990 // be killed; if not supplied, push/pop will be used internally to 1991 // allocate a temporary (inefficient, avoid if possible). 1992 // Optional slow case is for implementations (interpreter and C1) which branch to 1993 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 1994 // Returns offset of first potentially-faulting instruction for null 1995 // check info (currently consumed only by C1). If 1996 // swap_reg_contains_mark is true then returns -1 as it is assumed 1997 // the calling code has already passed any potential faults. 1998 int biased_locking_enter(Register lock_reg, Register obj_reg, 1999 Register swap_reg, Register tmp_reg, 2000 bool swap_reg_contains_mark, 2001 Label& done, Label* slow_case = NULL, 2002 BiasedLockingCounters* counters = NULL); 2003 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 2004 2005 2006 Condition negate_condition(Condition cond); 2007 2008 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit 2009 // operands. In general the names are modified to avoid hiding the instruction in Assembler 2010 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers 2011 // here in MacroAssembler. The major exception to this rule is call 2012 2013 // Arithmetics 2014 2015 2016 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } 2017 void addptr(Address dst, Register src); 2018 2019 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } 2020 void addptr(Register dst, int32_t src); 2021 void addptr(Register dst, Register src); 2022 2023 void andptr(Register dst, int32_t src); 2024 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } 2025 2026 void cmp8(AddressLiteral src1, int imm); 2027 2028 // renamed to drag out the casting of address to int32_t/intptr_t 2029 void cmp32(Register src1, int32_t imm); 2030 2031 void cmp32(AddressLiteral src1, int32_t imm); 2032 // compare reg - mem, or reg - &mem 2033 void cmp32(Register src1, AddressLiteral src2); 2034 2035 void cmp32(Register src1, Address src2); 2036 2037 #ifndef _LP64 2038 void cmpoop(Address dst, jobject obj); 2039 void cmpoop(Register dst, jobject obj); 2040 #endif // _LP64 2041 2042 // NOTE src2 must be the lval. This is NOT an mem-mem compare 2043 void cmpptr(Address src1, AddressLiteral src2); 2044 2045 void cmpptr(Register src1, AddressLiteral src2); 2046 2047 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2048 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2049 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2050 2051 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2052 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 2053 2054 // cmp64 to avoild hiding cmpq 2055 void cmp64(Register src1, AddressLiteral src); 2056 2057 void cmpxchgptr(Register reg, Address adr); 2058 2059 void locked_cmpxchgptr(Register reg, AddressLiteral adr); 2060 2061 2062 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } 2063 2064 2065 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } 2066 2067 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } 2068 2069 void shlptr(Register dst, int32_t shift); 2070 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } 2071 2072 void shrptr(Register dst, int32_t shift); 2073 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } 2074 2075 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } 2076 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } 2077 2078 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2079 2080 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 2081 void subptr(Register dst, int32_t src); 2082 void subptr(Register dst, Register src); 2083 2084 2085 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2086 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 2087 2088 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2089 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 2090 2091 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } 2092 2093 2094 2095 // Helper functions for statistics gathering. 2096 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. 2097 void cond_inc32(Condition cond, AddressLiteral counter_addr); 2098 // Unconditional atomic increment. 2099 void atomic_incl(AddressLiteral counter_addr); 2100 2101 void lea(Register dst, AddressLiteral adr); 2102 void lea(Address dst, AddressLiteral adr); 2103 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } 2104 2105 void leal32(Register dst, Address src) { leal(dst, src); } 2106 2107 // Import other testl() methods from the parent class or else 2108 // they will be hidden by the following overriding declaration. 2109 using Assembler::testl; 2110 void testl(Register dst, AddressLiteral src); 2111 2112 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2113 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2114 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 2115 2116 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } 2117 void testptr(Register src1, Register src2); 2118 2119 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2120 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 2121 2122 // Calls 2123 2124 void call(Label& L, relocInfo::relocType rtype); 2125 void call(Register entry); 2126 2127 // NOTE: this call tranfers to the effective address of entry NOT 2128 // the address contained by entry. This is because this is more natural 2129 // for jumps/calls. 2130 void call(AddressLiteral entry); 2131 2132 // Jumps 2133 2134 // NOTE: these jumps tranfer to the effective address of dst NOT 2135 // the address contained by dst. This is because this is more natural 2136 // for jumps/calls. 2137 void jump(AddressLiteral dst); 2138 void jump_cc(Condition cc, AddressLiteral dst); 2139 2140 // 32bit can do a case table jump in one instruction but we no longer allow the base 2141 // to be installed in the Address class. This jump will tranfers to the address 2142 // contained in the location described by entry (not the address of entry) 2143 void jump(ArrayAddress entry); 2144 2145 // Floating 2146 2147 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } 2148 void andpd(XMMRegister dst, AddressLiteral src); 2149 2150 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } 2151 void comiss(XMMRegister dst, AddressLiteral src); 2152 2153 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } 2154 void comisd(XMMRegister dst, AddressLiteral src); 2155 2156 void fadd_s(Address src) { Assembler::fadd_s(src); } 2157 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); } 2158 2159 void fldcw(Address src) { Assembler::fldcw(src); } 2160 void fldcw(AddressLiteral src); 2161 2162 void fld_s(int index) { Assembler::fld_s(index); } 2163 void fld_s(Address src) { Assembler::fld_s(src); } 2164 void fld_s(AddressLiteral src); 2165 2166 void fld_d(Address src) { Assembler::fld_d(src); } 2167 void fld_d(AddressLiteral src); 2168 2169 void fld_x(Address src) { Assembler::fld_x(src); } 2170 void fld_x(AddressLiteral src); 2171 2172 void fmul_s(Address src) { Assembler::fmul_s(src); } 2173 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); } 2174 2175 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } 2176 void ldmxcsr(AddressLiteral src); 2177 2178 private: 2179 // these are private because users should be doing movflt/movdbl 2180 2181 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } 2182 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } 2183 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } 2184 void movss(XMMRegister dst, AddressLiteral src); 2185 2186 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } 2187 void movlpd(XMMRegister dst, AddressLiteral src); 2188 2189 public: 2190 2191 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); } 2192 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); } 2193 void addsd(XMMRegister dst, AddressLiteral src) { Assembler::addsd(dst, as_Address(src)); } 2194 2195 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); } 2196 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); } 2197 void addss(XMMRegister dst, AddressLiteral src) { Assembler::addss(dst, as_Address(src)); } 2198 2199 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); } 2200 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); } 2201 void divsd(XMMRegister dst, AddressLiteral src) { Assembler::divsd(dst, as_Address(src)); } 2202 2203 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); } 2204 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); } 2205 void divss(XMMRegister dst, AddressLiteral src) { Assembler::divss(dst, as_Address(src)); } 2206 2207 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } 2208 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } 2209 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } 2210 void movsd(XMMRegister dst, AddressLiteral src) { Assembler::movsd(dst, as_Address(src)); } 2211 2212 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); } 2213 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); } 2214 void mulsd(XMMRegister dst, AddressLiteral src) { Assembler::mulsd(dst, as_Address(src)); } 2215 2216 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); } 2217 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); } 2218 void mulss(XMMRegister dst, AddressLiteral src) { Assembler::mulss(dst, as_Address(src)); } 2219 2220 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); } 2221 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); } 2222 void sqrtsd(XMMRegister dst, AddressLiteral src) { Assembler::sqrtsd(dst, as_Address(src)); } 2223 2224 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); } 2225 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); } 2226 void sqrtss(XMMRegister dst, AddressLiteral src) { Assembler::sqrtss(dst, as_Address(src)); } 2227 2228 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); } 2229 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); } 2230 void subsd(XMMRegister dst, AddressLiteral src) { Assembler::subsd(dst, as_Address(src)); } 2231 2232 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); } 2233 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); } 2234 void subss(XMMRegister dst, AddressLiteral src) { Assembler::subss(dst, as_Address(src)); } 2235 2236 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } 2237 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } 2238 void ucomiss(XMMRegister dst, AddressLiteral src); 2239 2240 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } 2241 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } 2242 void ucomisd(XMMRegister dst, AddressLiteral src); 2243 2244 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 2245 void xorpd(XMMRegister dst, XMMRegister src) { Assembler::xorpd(dst, src); } 2246 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } 2247 void xorpd(XMMRegister dst, AddressLiteral src); 2248 2249 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 2250 void xorps(XMMRegister dst, XMMRegister src) { Assembler::xorps(dst, src); } 2251 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } 2252 void xorps(XMMRegister dst, AddressLiteral src); 2253 2254 // Data 2255 2256 void cmov32( Condition cc, Register dst, Address src); 2257 void cmov32( Condition cc, Register dst, Register src); 2258 2259 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); } 2260 2261 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 2262 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 2263 2264 void movoop(Register dst, jobject obj); 2265 void movoop(Address dst, jobject obj); 2266 2267 void movptr(ArrayAddress dst, Register src); 2268 // can this do an lea? 2269 void movptr(Register dst, ArrayAddress src); 2270 2271 void movptr(Register dst, Address src); 2272 2273 void movptr(Register dst, AddressLiteral src); 2274 2275 void movptr(Register dst, intptr_t src); 2276 void movptr(Register dst, Register src); 2277 void movptr(Address dst, intptr_t src); 2278 2279 void movptr(Address dst, Register src); 2280 2281 #ifdef _LP64 2282 // Generally the next two are only used for moving NULL 2283 // Although there are situations in initializing the mark word where 2284 // they could be used. They are dangerous. 2285 2286 // They only exist on LP64 so that int32_t and intptr_t are not the same 2287 // and we have ambiguous declarations. 2288 2289 void movptr(Address dst, int32_t imm32); 2290 void movptr(Register dst, int32_t imm32); 2291 #endif // _LP64 2292 2293 // to avoid hiding movl 2294 void mov32(AddressLiteral dst, Register src); 2295 void mov32(Register dst, AddressLiteral src); 2296 2297 // to avoid hiding movb 2298 void movbyte(ArrayAddress dst, int src); 2299 2300 // Can push value or effective address 2301 void pushptr(AddressLiteral src); 2302 2303 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } 2304 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } 2305 2306 void pushoop(jobject obj); 2307 2308 // sign extend as need a l to ptr sized element 2309 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } 2310 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } 2311 2312 // IndexOf strings. 2313 // Small strings are loaded through stack if they cross page boundary. 2314 void string_indexof(Register str1, Register str2, 2315 Register cnt1, Register cnt2, 2316 int int_cnt2, Register result, 2317 XMMRegister vec, Register tmp); 2318 2319 // IndexOf for constant substrings with size >= 8 elements 2320 // which don't need to be loaded through stack. 2321 void string_indexofC8(Register str1, Register str2, 2322 Register cnt1, Register cnt2, 2323 int int_cnt2, Register result, 2324 XMMRegister vec, Register tmp); 2325 2326 // Smallest code: we don't need to load through stack, 2327 // check string tail. 2328 2329 // Compare strings. 2330 void string_compare(Register str1, Register str2, 2331 Register cnt1, Register cnt2, Register result, 2332 XMMRegister vec1); 2333 2334 // Compare char[] arrays. 2335 void char_arrays_equals(bool is_array_equ, Register ary1, Register ary2, 2336 Register limit, Register result, Register chr, 2337 XMMRegister vec1, XMMRegister vec2); 2338 2339 // Fill primitive arrays 2340 void generate_fill(BasicType t, bool aligned, 2341 Register to, Register value, Register count, 2342 Register rtmp, XMMRegister xtmp); 2343 2344 #undef VIRTUAL 2345 2346 }; 2347 2348 /** 2349 * class SkipIfEqual: 2350 * 2351 * Instantiating this class will result in assembly code being output that will 2352 * jump around any code emitted between the creation of the instance and it's 2353 * automatic destruction at the end of a scope block, depending on the value of 2354 * the flag passed to the constructor, which will be checked at run-time. 2355 */ 2356 class SkipIfEqual { 2357 private: 2358 MacroAssembler* _masm; 2359 Label _label; 2360 2361 public: 2362 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 2363 ~SkipIfEqual(); 2364 }; 2365 2366 #ifdef ASSERT 2367 inline bool AbstractAssembler::pd_check_instruction_mark() { return true; } 2368 #endif 2369 2370 #endif // CPU_X86_VM_ASSEMBLER_X86_HPP