1 /* 2 * Copyright (c) 1997, 2010, 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 #include "incls/_precompiled.incl" 26 #include "incls/_assembler_x86.cpp.incl" 27 28 // Implementation of AddressLiteral 29 30 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) { 31 _is_lval = false; 32 _target = target; 33 switch (rtype) { 34 case relocInfo::oop_type: 35 // Oops are a special case. Normally they would be their own section 36 // but in cases like icBuffer they are literals in the code stream that 37 // we don't have a section for. We use none so that we get a literal address 38 // which is always patchable. 39 break; 40 case relocInfo::external_word_type: 41 _rspec = external_word_Relocation::spec(target); 42 break; 43 case relocInfo::internal_word_type: 44 _rspec = internal_word_Relocation::spec(target); 45 break; 46 case relocInfo::opt_virtual_call_type: 47 _rspec = opt_virtual_call_Relocation::spec(); 48 break; 49 case relocInfo::static_call_type: 50 _rspec = static_call_Relocation::spec(); 51 break; 52 case relocInfo::runtime_call_type: 53 _rspec = runtime_call_Relocation::spec(); 54 break; 55 case relocInfo::poll_type: 56 case relocInfo::poll_return_type: 57 _rspec = Relocation::spec_simple(rtype); 58 break; 59 case relocInfo::none: 60 break; 61 default: 62 ShouldNotReachHere(); 63 break; 64 } 65 } 66 67 // Implementation of Address 68 69 #ifdef _LP64 70 71 Address Address::make_array(ArrayAddress adr) { 72 // Not implementable on 64bit machines 73 // Should have been handled higher up the call chain. 74 ShouldNotReachHere(); 75 return Address(); 76 } 77 78 // exceedingly dangerous constructor 79 Address::Address(int disp, address loc, relocInfo::relocType rtype) { 80 _base = noreg; 81 _index = noreg; 82 _scale = no_scale; 83 _disp = disp; 84 switch (rtype) { 85 case relocInfo::external_word_type: 86 _rspec = external_word_Relocation::spec(loc); 87 break; 88 case relocInfo::internal_word_type: 89 _rspec = internal_word_Relocation::spec(loc); 90 break; 91 case relocInfo::runtime_call_type: 92 // HMM 93 _rspec = runtime_call_Relocation::spec(); 94 break; 95 case relocInfo::poll_type: 96 case relocInfo::poll_return_type: 97 _rspec = Relocation::spec_simple(rtype); 98 break; 99 case relocInfo::none: 100 break; 101 default: 102 ShouldNotReachHere(); 103 } 104 } 105 #else // LP64 106 107 Address Address::make_array(ArrayAddress adr) { 108 AddressLiteral base = adr.base(); 109 Address index = adr.index(); 110 assert(index._disp == 0, "must not have disp"); // maybe it can? 111 Address array(index._base, index._index, index._scale, (intptr_t) base.target()); 112 array._rspec = base._rspec; 113 return array; 114 } 115 116 // exceedingly dangerous constructor 117 Address::Address(address loc, RelocationHolder spec) { 118 _base = noreg; 119 _index = noreg; 120 _scale = no_scale; 121 _disp = (intptr_t) loc; 122 _rspec = spec; 123 } 124 125 #endif // _LP64 126 127 128 129 // Convert the raw encoding form into the form expected by the constructor for 130 // Address. An index of 4 (rsp) corresponds to having no index, so convert 131 // that to noreg for the Address constructor. 132 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) { 133 RelocationHolder rspec; 134 if (disp_is_oop) { 135 rspec = Relocation::spec_simple(relocInfo::oop_type); 136 } 137 bool valid_index = index != rsp->encoding(); 138 if (valid_index) { 139 Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp)); 140 madr._rspec = rspec; 141 return madr; 142 } else { 143 Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp)); 144 madr._rspec = rspec; 145 return madr; 146 } 147 } 148 149 // Implementation of Assembler 150 151 int AbstractAssembler::code_fill_byte() { 152 return (u_char)'\xF4'; // hlt 153 } 154 155 // make this go away someday 156 void Assembler::emit_data(jint data, relocInfo::relocType rtype, int format) { 157 if (rtype == relocInfo::none) 158 emit_long(data); 159 else emit_data(data, Relocation::spec_simple(rtype), format); 160 } 161 162 void Assembler::emit_data(jint data, RelocationHolder const& rspec, int format) { 163 assert(imm_operand == 0, "default format must be immediate in this file"); 164 assert(inst_mark() != NULL, "must be inside InstructionMark"); 165 if (rspec.type() != relocInfo::none) { 166 #ifdef ASSERT 167 check_relocation(rspec, format); 168 #endif 169 // Do not use AbstractAssembler::relocate, which is not intended for 170 // embedded words. Instead, relocate to the enclosing instruction. 171 172 // hack. call32 is too wide for mask so use disp32 173 if (format == call32_operand) 174 code_section()->relocate(inst_mark(), rspec, disp32_operand); 175 else 176 code_section()->relocate(inst_mark(), rspec, format); 177 } 178 emit_long(data); 179 } 180 181 static int encode(Register r) { 182 int enc = r->encoding(); 183 if (enc >= 8) { 184 enc -= 8; 185 } 186 return enc; 187 } 188 189 static int encode(XMMRegister r) { 190 int enc = r->encoding(); 191 if (enc >= 8) { 192 enc -= 8; 193 } 194 return enc; 195 } 196 197 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) { 198 assert(dst->has_byte_register(), "must have byte register"); 199 assert(isByte(op1) && isByte(op2), "wrong opcode"); 200 assert(isByte(imm8), "not a byte"); 201 assert((op1 & 0x01) == 0, "should be 8bit operation"); 202 emit_byte(op1); 203 emit_byte(op2 | encode(dst)); 204 emit_byte(imm8); 205 } 206 207 208 void Assembler::emit_arith(int op1, int op2, Register dst, int32_t imm32) { 209 assert(isByte(op1) && isByte(op2), "wrong opcode"); 210 assert((op1 & 0x01) == 1, "should be 32bit operation"); 211 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); 212 if (is8bit(imm32)) { 213 emit_byte(op1 | 0x02); // set sign bit 214 emit_byte(op2 | encode(dst)); 215 emit_byte(imm32 & 0xFF); 216 } else { 217 emit_byte(op1); 218 emit_byte(op2 | encode(dst)); 219 emit_long(imm32); 220 } 221 } 222 223 // immediate-to-memory forms 224 void Assembler::emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32) { 225 assert((op1 & 0x01) == 1, "should be 32bit operation"); 226 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); 227 if (is8bit(imm32)) { 228 emit_byte(op1 | 0x02); // set sign bit 229 emit_operand(rm, adr, 1); 230 emit_byte(imm32 & 0xFF); 231 } else { 232 emit_byte(op1); 233 emit_operand(rm, adr, 4); 234 emit_long(imm32); 235 } 236 } 237 238 void Assembler::emit_arith(int op1, int op2, Register dst, jobject obj) { 239 LP64_ONLY(ShouldNotReachHere()); 240 assert(isByte(op1) && isByte(op2), "wrong opcode"); 241 assert((op1 & 0x01) == 1, "should be 32bit operation"); 242 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); 243 InstructionMark im(this); 244 emit_byte(op1); 245 emit_byte(op2 | encode(dst)); 246 emit_data((intptr_t)obj, relocInfo::oop_type, 0); 247 } 248 249 250 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) { 251 assert(isByte(op1) && isByte(op2), "wrong opcode"); 252 emit_byte(op1); 253 emit_byte(op2 | encode(dst) << 3 | encode(src)); 254 } 255 256 257 void Assembler::emit_operand(Register reg, Register base, Register index, 258 Address::ScaleFactor scale, int disp, 259 RelocationHolder const& rspec, 260 int rip_relative_correction) { 261 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type(); 262 263 // Encode the registers as needed in the fields they are used in 264 265 int regenc = encode(reg) << 3; 266 int indexenc = index->is_valid() ? encode(index) << 3 : 0; 267 int baseenc = base->is_valid() ? encode(base) : 0; 268 269 if (base->is_valid()) { 270 if (index->is_valid()) { 271 assert(scale != Address::no_scale, "inconsistent address"); 272 // [base + index*scale + disp] 273 if (disp == 0 && rtype == relocInfo::none && 274 base != rbp LP64_ONLY(&& base != r13)) { 275 // [base + index*scale] 276 // [00 reg 100][ss index base] 277 assert(index != rsp, "illegal addressing mode"); 278 emit_byte(0x04 | regenc); 279 emit_byte(scale << 6 | indexenc | baseenc); 280 } else if (is8bit(disp) && rtype == relocInfo::none) { 281 // [base + index*scale + imm8] 282 // [01 reg 100][ss index base] imm8 283 assert(index != rsp, "illegal addressing mode"); 284 emit_byte(0x44 | regenc); 285 emit_byte(scale << 6 | indexenc | baseenc); 286 emit_byte(disp & 0xFF); 287 } else { 288 // [base + index*scale + disp32] 289 // [10 reg 100][ss index base] disp32 290 assert(index != rsp, "illegal addressing mode"); 291 emit_byte(0x84 | regenc); 292 emit_byte(scale << 6 | indexenc | baseenc); 293 emit_data(disp, rspec, disp32_operand); 294 } 295 } else if (base == rsp LP64_ONLY(|| base == r12)) { 296 // [rsp + disp] 297 if (disp == 0 && rtype == relocInfo::none) { 298 // [rsp] 299 // [00 reg 100][00 100 100] 300 emit_byte(0x04 | regenc); 301 emit_byte(0x24); 302 } else if (is8bit(disp) && rtype == relocInfo::none) { 303 // [rsp + imm8] 304 // [01 reg 100][00 100 100] disp8 305 emit_byte(0x44 | regenc); 306 emit_byte(0x24); 307 emit_byte(disp & 0xFF); 308 } else { 309 // [rsp + imm32] 310 // [10 reg 100][00 100 100] disp32 311 emit_byte(0x84 | regenc); 312 emit_byte(0x24); 313 emit_data(disp, rspec, disp32_operand); 314 } 315 } else { 316 // [base + disp] 317 assert(base != rsp LP64_ONLY(&& base != r12), "illegal addressing mode"); 318 if (disp == 0 && rtype == relocInfo::none && 319 base != rbp LP64_ONLY(&& base != r13)) { 320 // [base] 321 // [00 reg base] 322 emit_byte(0x00 | regenc | baseenc); 323 } else if (is8bit(disp) && rtype == relocInfo::none) { 324 // [base + disp8] 325 // [01 reg base] disp8 326 emit_byte(0x40 | regenc | baseenc); 327 emit_byte(disp & 0xFF); 328 } else { 329 // [base + disp32] 330 // [10 reg base] disp32 331 emit_byte(0x80 | regenc | baseenc); 332 emit_data(disp, rspec, disp32_operand); 333 } 334 } 335 } else { 336 if (index->is_valid()) { 337 assert(scale != Address::no_scale, "inconsistent address"); 338 // [index*scale + disp] 339 // [00 reg 100][ss index 101] disp32 340 assert(index != rsp, "illegal addressing mode"); 341 emit_byte(0x04 | regenc); 342 emit_byte(scale << 6 | indexenc | 0x05); 343 emit_data(disp, rspec, disp32_operand); 344 } else if (rtype != relocInfo::none ) { 345 // [disp] (64bit) RIP-RELATIVE (32bit) abs 346 // [00 000 101] disp32 347 348 emit_byte(0x05 | regenc); 349 // Note that the RIP-rel. correction applies to the generated 350 // disp field, but _not_ to the target address in the rspec. 351 352 // disp was created by converting the target address minus the pc 353 // at the start of the instruction. That needs more correction here. 354 // intptr_t disp = target - next_ip; 355 assert(inst_mark() != NULL, "must be inside InstructionMark"); 356 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction; 357 int64_t adjusted = disp; 358 // Do rip-rel adjustment for 64bit 359 LP64_ONLY(adjusted -= (next_ip - inst_mark())); 360 assert(is_simm32(adjusted), 361 "must be 32bit offset (RIP relative address)"); 362 emit_data((int32_t) adjusted, rspec, disp32_operand); 363 364 } else { 365 // 32bit never did this, did everything as the rip-rel/disp code above 366 // [disp] ABSOLUTE 367 // [00 reg 100][00 100 101] disp32 368 emit_byte(0x04 | regenc); 369 emit_byte(0x25); 370 emit_data(disp, rspec, disp32_operand); 371 } 372 } 373 } 374 375 void Assembler::emit_operand(XMMRegister reg, Register base, Register index, 376 Address::ScaleFactor scale, int disp, 377 RelocationHolder const& rspec) { 378 emit_operand((Register)reg, base, index, scale, disp, rspec); 379 } 380 381 // Secret local extension to Assembler::WhichOperand: 382 #define end_pc_operand (_WhichOperand_limit) 383 384 address Assembler::locate_operand(address inst, WhichOperand which) { 385 // Decode the given instruction, and return the address of 386 // an embedded 32-bit operand word. 387 388 // If "which" is disp32_operand, selects the displacement portion 389 // of an effective address specifier. 390 // If "which" is imm64_operand, selects the trailing immediate constant. 391 // If "which" is call32_operand, selects the displacement of a call or jump. 392 // Caller is responsible for ensuring that there is such an operand, 393 // and that it is 32/64 bits wide. 394 395 // If "which" is end_pc_operand, find the end of the instruction. 396 397 address ip = inst; 398 bool is_64bit = false; 399 400 debug_only(bool has_disp32 = false); 401 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn 402 403 again_after_prefix: 404 switch (0xFF & *ip++) { 405 406 // These convenience macros generate groups of "case" labels for the switch. 407 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3 408 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \ 409 case (x)+4: case (x)+5: case (x)+6: case (x)+7 410 #define REP16(x) REP8((x)+0): \ 411 case REP8((x)+8) 412 413 case CS_segment: 414 case SS_segment: 415 case DS_segment: 416 case ES_segment: 417 case FS_segment: 418 case GS_segment: 419 // Seems dubious 420 LP64_ONLY(assert(false, "shouldn't have that prefix")); 421 assert(ip == inst+1, "only one prefix allowed"); 422 goto again_after_prefix; 423 424 case 0x67: 425 case REX: 426 case REX_B: 427 case REX_X: 428 case REX_XB: 429 case REX_R: 430 case REX_RB: 431 case REX_RX: 432 case REX_RXB: 433 NOT_LP64(assert(false, "64bit prefixes")); 434 goto again_after_prefix; 435 436 case REX_W: 437 case REX_WB: 438 case REX_WX: 439 case REX_WXB: 440 case REX_WR: 441 case REX_WRB: 442 case REX_WRX: 443 case REX_WRXB: 444 NOT_LP64(assert(false, "64bit prefixes")); 445 is_64bit = true; 446 goto again_after_prefix; 447 448 case 0xFF: // pushq a; decl a; incl a; call a; jmp a 449 case 0x88: // movb a, r 450 case 0x89: // movl a, r 451 case 0x8A: // movb r, a 452 case 0x8B: // movl r, a 453 case 0x8F: // popl a 454 debug_only(has_disp32 = true); 455 break; 456 457 case 0x68: // pushq #32 458 if (which == end_pc_operand) { 459 return ip + 4; 460 } 461 assert(which == imm_operand && !is_64bit, "pushl has no disp32 or 64bit immediate"); 462 return ip; // not produced by emit_operand 463 464 case 0x66: // movw ... (size prefix) 465 again_after_size_prefix2: 466 switch (0xFF & *ip++) { 467 case REX: 468 case REX_B: 469 case REX_X: 470 case REX_XB: 471 case REX_R: 472 case REX_RB: 473 case REX_RX: 474 case REX_RXB: 475 case REX_W: 476 case REX_WB: 477 case REX_WX: 478 case REX_WXB: 479 case REX_WR: 480 case REX_WRB: 481 case REX_WRX: 482 case REX_WRXB: 483 NOT_LP64(assert(false, "64bit prefix found")); 484 goto again_after_size_prefix2; 485 case 0x8B: // movw r, a 486 case 0x89: // movw a, r 487 debug_only(has_disp32 = true); 488 break; 489 case 0xC7: // movw a, #16 490 debug_only(has_disp32 = true); 491 tail_size = 2; // the imm16 492 break; 493 case 0x0F: // several SSE/SSE2 variants 494 ip--; // reparse the 0x0F 495 goto again_after_prefix; 496 default: 497 ShouldNotReachHere(); 498 } 499 break; 500 501 case REP8(0xB8): // movl/q r, #32/#64(oop?) 502 if (which == end_pc_operand) return ip + (is_64bit ? 8 : 4); 503 // these asserts are somewhat nonsensical 504 #ifndef _LP64 505 assert(which == imm_operand || which == disp32_operand, ""); 506 #else 507 assert((which == call32_operand || which == imm_operand) && is_64bit || 508 which == narrow_oop_operand && !is_64bit, ""); 509 #endif // _LP64 510 return ip; 511 512 case 0x69: // imul r, a, #32 513 case 0xC7: // movl a, #32(oop?) 514 tail_size = 4; 515 debug_only(has_disp32 = true); // has both kinds of operands! 516 break; 517 518 case 0x0F: // movx..., etc. 519 switch (0xFF & *ip++) { 520 case 0x12: // movlps 521 case 0x28: // movaps 522 case 0x2E: // ucomiss 523 case 0x2F: // comiss 524 case 0x54: // andps 525 case 0x55: // andnps 526 case 0x56: // orps 527 case 0x57: // xorps 528 case 0x6E: // movd 529 case 0x7E: // movd 530 case 0xAE: // ldmxcsr a 531 // 64bit side says it these have both operands but that doesn't 532 // appear to be true 533 debug_only(has_disp32 = true); 534 break; 535 536 case 0xAD: // shrd r, a, %cl 537 case 0xAF: // imul r, a 538 case 0xBE: // movsbl r, a (movsxb) 539 case 0xBF: // movswl r, a (movsxw) 540 case 0xB6: // movzbl r, a (movzxb) 541 case 0xB7: // movzwl r, a (movzxw) 542 case REP16(0x40): // cmovl cc, r, a 543 case 0xB0: // cmpxchgb 544 case 0xB1: // cmpxchg 545 case 0xC1: // xaddl 546 case 0xC7: // cmpxchg8 547 case REP16(0x90): // setcc a 548 debug_only(has_disp32 = true); 549 // fall out of the switch to decode the address 550 break; 551 552 case 0xAC: // shrd r, a, #8 553 debug_only(has_disp32 = true); 554 tail_size = 1; // the imm8 555 break; 556 557 case REP16(0x80): // jcc rdisp32 558 if (which == end_pc_operand) return ip + 4; 559 assert(which == call32_operand, "jcc has no disp32 or imm"); 560 return ip; 561 default: 562 ShouldNotReachHere(); 563 } 564 break; 565 566 case 0x81: // addl a, #32; addl r, #32 567 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl 568 // on 32bit in the case of cmpl, the imm might be an oop 569 tail_size = 4; 570 debug_only(has_disp32 = true); // has both kinds of operands! 571 break; 572 573 case 0x83: // addl a, #8; addl r, #8 574 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl 575 debug_only(has_disp32 = true); // has both kinds of operands! 576 tail_size = 1; 577 break; 578 579 case 0x9B: 580 switch (0xFF & *ip++) { 581 case 0xD9: // fnstcw a 582 debug_only(has_disp32 = true); 583 break; 584 default: 585 ShouldNotReachHere(); 586 } 587 break; 588 589 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a 590 case REP4(0x10): // adc... 591 case REP4(0x20): // and... 592 case REP4(0x30): // xor... 593 case REP4(0x08): // or... 594 case REP4(0x18): // sbb... 595 case REP4(0x28): // sub... 596 case 0xF7: // mull a 597 case 0x8D: // lea r, a 598 case 0x87: // xchg r, a 599 case REP4(0x38): // cmp... 600 case 0x85: // test r, a 601 debug_only(has_disp32 = true); // has both kinds of operands! 602 break; 603 604 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8 605 case 0xC6: // movb a, #8 606 case 0x80: // cmpb a, #8 607 case 0x6B: // imul r, a, #8 608 debug_only(has_disp32 = true); // has both kinds of operands! 609 tail_size = 1; // the imm8 610 break; 611 612 case 0xE8: // call rdisp32 613 case 0xE9: // jmp rdisp32 614 if (which == end_pc_operand) return ip + 4; 615 assert(which == call32_operand, "call has no disp32 or imm"); 616 return ip; 617 618 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1 619 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl 620 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a 621 case 0xDD: // fld_d a; fst_d a; fstp_d a 622 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a 623 case 0xDF: // fild_d a; fistp_d a 624 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a 625 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a 626 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a 627 debug_only(has_disp32 = true); 628 break; 629 630 case 0xF0: // Lock 631 assert(os::is_MP(), "only on MP"); 632 goto again_after_prefix; 633 634 case 0xF3: // For SSE 635 case 0xF2: // For SSE2 636 switch (0xFF & *ip++) { 637 case REX: 638 case REX_B: 639 case REX_X: 640 case REX_XB: 641 case REX_R: 642 case REX_RB: 643 case REX_RX: 644 case REX_RXB: 645 case REX_W: 646 case REX_WB: 647 case REX_WX: 648 case REX_WXB: 649 case REX_WR: 650 case REX_WRB: 651 case REX_WRX: 652 case REX_WRXB: 653 NOT_LP64(assert(false, "found 64bit prefix")); 654 ip++; 655 default: 656 ip++; 657 } 658 debug_only(has_disp32 = true); // has both kinds of operands! 659 break; 660 661 default: 662 ShouldNotReachHere(); 663 664 #undef REP8 665 #undef REP16 666 } 667 668 assert(which != call32_operand, "instruction is not a call, jmp, or jcc"); 669 #ifdef _LP64 670 assert(which != imm_operand, "instruction is not a movq reg, imm64"); 671 #else 672 // assert(which != imm_operand || has_imm32, "instruction has no imm32 field"); 673 assert(which != imm_operand || has_disp32, "instruction has no imm32 field"); 674 #endif // LP64 675 assert(which != disp32_operand || has_disp32, "instruction has no disp32 field"); 676 677 // parse the output of emit_operand 678 int op2 = 0xFF & *ip++; 679 int base = op2 & 0x07; 680 int op3 = -1; 681 const int b100 = 4; 682 const int b101 = 5; 683 if (base == b100 && (op2 >> 6) != 3) { 684 op3 = 0xFF & *ip++; 685 base = op3 & 0x07; // refetch the base 686 } 687 // now ip points at the disp (if any) 688 689 switch (op2 >> 6) { 690 case 0: 691 // [00 reg 100][ss index base] 692 // [00 reg 100][00 100 esp] 693 // [00 reg base] 694 // [00 reg 100][ss index 101][disp32] 695 // [00 reg 101] [disp32] 696 697 if (base == b101) { 698 if (which == disp32_operand) 699 return ip; // caller wants the disp32 700 ip += 4; // skip the disp32 701 } 702 break; 703 704 case 1: 705 // [01 reg 100][ss index base][disp8] 706 // [01 reg 100][00 100 esp][disp8] 707 // [01 reg base] [disp8] 708 ip += 1; // skip the disp8 709 break; 710 711 case 2: 712 // [10 reg 100][ss index base][disp32] 713 // [10 reg 100][00 100 esp][disp32] 714 // [10 reg base] [disp32] 715 if (which == disp32_operand) 716 return ip; // caller wants the disp32 717 ip += 4; // skip the disp32 718 break; 719 720 case 3: 721 // [11 reg base] (not a memory addressing mode) 722 break; 723 } 724 725 if (which == end_pc_operand) { 726 return ip + tail_size; 727 } 728 729 #ifdef _LP64 730 assert(which == narrow_oop_operand && !is_64bit, "instruction is not a movl adr, imm32"); 731 #else 732 assert(which == imm_operand, "instruction has only an imm field"); 733 #endif // LP64 734 return ip; 735 } 736 737 address Assembler::locate_next_instruction(address inst) { 738 // Secretly share code with locate_operand: 739 return locate_operand(inst, end_pc_operand); 740 } 741 742 743 #ifdef ASSERT 744 void Assembler::check_relocation(RelocationHolder const& rspec, int format) { 745 address inst = inst_mark(); 746 assert(inst != NULL && inst < pc(), "must point to beginning of instruction"); 747 address opnd; 748 749 Relocation* r = rspec.reloc(); 750 if (r->type() == relocInfo::none) { 751 return; 752 } else if (r->is_call() || format == call32_operand) { 753 // assert(format == imm32_operand, "cannot specify a nonzero format"); 754 opnd = locate_operand(inst, call32_operand); 755 } else if (r->is_data()) { 756 assert(format == imm_operand || format == disp32_operand 757 LP64_ONLY(|| format == narrow_oop_operand), "format ok"); 758 opnd = locate_operand(inst, (WhichOperand)format); 759 } else { 760 assert(format == imm_operand, "cannot specify a format"); 761 return; 762 } 763 assert(opnd == pc(), "must put operand where relocs can find it"); 764 } 765 #endif // ASSERT 766 767 void Assembler::emit_operand32(Register reg, Address adr) { 768 assert(reg->encoding() < 8, "no extended registers"); 769 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); 770 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, 771 adr._rspec); 772 } 773 774 void Assembler::emit_operand(Register reg, Address adr, 775 int rip_relative_correction) { 776 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, 777 adr._rspec, 778 rip_relative_correction); 779 } 780 781 void Assembler::emit_operand(XMMRegister reg, Address adr) { 782 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, 783 adr._rspec); 784 } 785 786 // MMX operations 787 void Assembler::emit_operand(MMXRegister reg, Address adr) { 788 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); 789 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); 790 } 791 792 // work around gcc (3.2.1-7a) bug 793 void Assembler::emit_operand(Address adr, MMXRegister reg) { 794 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); 795 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); 796 } 797 798 799 void Assembler::emit_farith(int b1, int b2, int i) { 800 assert(isByte(b1) && isByte(b2), "wrong opcode"); 801 assert(0 <= i && i < 8, "illegal stack offset"); 802 emit_byte(b1); 803 emit_byte(b2 + i); 804 } 805 806 807 // Now the Assembler instruction (identical for 32/64 bits) 808 809 void Assembler::adcl(Register dst, int32_t imm32) { 810 prefix(dst); 811 emit_arith(0x81, 0xD0, dst, imm32); 812 } 813 814 void Assembler::adcl(Register dst, Address src) { 815 InstructionMark im(this); 816 prefix(src, dst); 817 emit_byte(0x13); 818 emit_operand(dst, src); 819 } 820 821 void Assembler::adcl(Register dst, Register src) { 822 (void) prefix_and_encode(dst->encoding(), src->encoding()); 823 emit_arith(0x13, 0xC0, dst, src); 824 } 825 826 void Assembler::addl(Address dst, int32_t imm32) { 827 InstructionMark im(this); 828 prefix(dst); 829 emit_arith_operand(0x81, rax, dst, imm32); 830 } 831 832 void Assembler::addl(Address dst, Register src) { 833 InstructionMark im(this); 834 prefix(dst, src); 835 emit_byte(0x01); 836 emit_operand(src, dst); 837 } 838 839 void Assembler::addl(Register dst, int32_t imm32) { 840 prefix(dst); 841 emit_arith(0x81, 0xC0, dst, imm32); 842 } 843 844 void Assembler::addl(Register dst, Address src) { 845 InstructionMark im(this); 846 prefix(src, dst); 847 emit_byte(0x03); 848 emit_operand(dst, src); 849 } 850 851 void Assembler::addl(Register dst, Register src) { 852 (void) prefix_and_encode(dst->encoding(), src->encoding()); 853 emit_arith(0x03, 0xC0, dst, src); 854 } 855 856 void Assembler::addr_nop_4() { 857 // 4 bytes: NOP DWORD PTR [EAX+0] 858 emit_byte(0x0F); 859 emit_byte(0x1F); 860 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc); 861 emit_byte(0); // 8-bits offset (1 byte) 862 } 863 864 void Assembler::addr_nop_5() { 865 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset 866 emit_byte(0x0F); 867 emit_byte(0x1F); 868 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4); 869 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); 870 emit_byte(0); // 8-bits offset (1 byte) 871 } 872 873 void Assembler::addr_nop_7() { 874 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset 875 emit_byte(0x0F); 876 emit_byte(0x1F); 877 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc); 878 emit_long(0); // 32-bits offset (4 bytes) 879 } 880 881 void Assembler::addr_nop_8() { 882 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset 883 emit_byte(0x0F); 884 emit_byte(0x1F); 885 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4); 886 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); 887 emit_long(0); // 32-bits offset (4 bytes) 888 } 889 890 void Assembler::addsd(XMMRegister dst, XMMRegister src) { 891 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 892 emit_byte(0xF2); 893 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 894 emit_byte(0x0F); 895 emit_byte(0x58); 896 emit_byte(0xC0 | encode); 897 } 898 899 void Assembler::addsd(XMMRegister dst, Address src) { 900 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 901 InstructionMark im(this); 902 emit_byte(0xF2); 903 prefix(src, dst); 904 emit_byte(0x0F); 905 emit_byte(0x58); 906 emit_operand(dst, src); 907 } 908 909 void Assembler::addss(XMMRegister dst, XMMRegister src) { 910 NOT_LP64(assert(VM_Version::supports_sse(), "")); 911 emit_byte(0xF3); 912 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 913 emit_byte(0x0F); 914 emit_byte(0x58); 915 emit_byte(0xC0 | encode); 916 } 917 918 void Assembler::addss(XMMRegister dst, Address src) { 919 NOT_LP64(assert(VM_Version::supports_sse(), "")); 920 InstructionMark im(this); 921 emit_byte(0xF3); 922 prefix(src, dst); 923 emit_byte(0x0F); 924 emit_byte(0x58); 925 emit_operand(dst, src); 926 } 927 928 void Assembler::andl(Register dst, int32_t imm32) { 929 prefix(dst); 930 emit_arith(0x81, 0xE0, dst, imm32); 931 } 932 933 void Assembler::andl(Register dst, Address src) { 934 InstructionMark im(this); 935 prefix(src, dst); 936 emit_byte(0x23); 937 emit_operand(dst, src); 938 } 939 940 void Assembler::andl(Register dst, Register src) { 941 (void) prefix_and_encode(dst->encoding(), src->encoding()); 942 emit_arith(0x23, 0xC0, dst, src); 943 } 944 945 void Assembler::andpd(XMMRegister dst, Address src) { 946 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 947 InstructionMark im(this); 948 emit_byte(0x66); 949 prefix(src, dst); 950 emit_byte(0x0F); 951 emit_byte(0x54); 952 emit_operand(dst, src); 953 } 954 955 void Assembler::bsfl(Register dst, Register src) { 956 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 957 emit_byte(0x0F); 958 emit_byte(0xBC); 959 emit_byte(0xC0 | encode); 960 } 961 962 void Assembler::bsrl(Register dst, Register src) { 963 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT"); 964 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 965 emit_byte(0x0F); 966 emit_byte(0xBD); 967 emit_byte(0xC0 | encode); 968 } 969 970 void Assembler::bswapl(Register reg) { // bswap 971 int encode = prefix_and_encode(reg->encoding()); 972 emit_byte(0x0F); 973 emit_byte(0xC8 | encode); 974 } 975 976 void Assembler::call(Label& L, relocInfo::relocType rtype) { 977 // suspect disp32 is always good 978 int operand = LP64_ONLY(disp32_operand) NOT_LP64(imm_operand); 979 980 if (L.is_bound()) { 981 const int long_size = 5; 982 int offs = (int)( target(L) - pc() ); 983 assert(offs <= 0, "assembler error"); 984 InstructionMark im(this); 985 // 1110 1000 #32-bit disp 986 emit_byte(0xE8); 987 emit_data(offs - long_size, rtype, operand); 988 } else { 989 InstructionMark im(this); 990 // 1110 1000 #32-bit disp 991 L.add_patch_at(code(), locator()); 992 993 emit_byte(0xE8); 994 emit_data(int(0), rtype, operand); 995 } 996 } 997 998 void Assembler::call(Register dst) { 999 // This was originally using a 32bit register encoding 1000 // and surely we want 64bit! 1001 // this is a 32bit encoding but in 64bit mode the default 1002 // operand size is 64bit so there is no need for the 1003 // wide prefix. So prefix only happens if we use the 1004 // new registers. Much like push/pop. 1005 int x = offset(); 1006 // this may be true but dbx disassembles it as if it 1007 // were 32bits... 1008 // int encode = prefix_and_encode(dst->encoding()); 1009 // if (offset() != x) assert(dst->encoding() >= 8, "what?"); 1010 int encode = prefixq_and_encode(dst->encoding()); 1011 1012 emit_byte(0xFF); 1013 emit_byte(0xD0 | encode); 1014 } 1015 1016 1017 void Assembler::call(Address adr) { 1018 InstructionMark im(this); 1019 prefix(adr); 1020 emit_byte(0xFF); 1021 emit_operand(rdx, adr); 1022 } 1023 1024 void Assembler::call_literal(address entry, RelocationHolder const& rspec) { 1025 assert(entry != NULL, "call most probably wrong"); 1026 InstructionMark im(this); 1027 emit_byte(0xE8); 1028 intptr_t disp = entry - (_code_pos + sizeof(int32_t)); 1029 assert(is_simm32(disp), "must be 32bit offset (call2)"); 1030 // Technically, should use call32_operand, but this format is 1031 // implied by the fact that we're emitting a call instruction. 1032 1033 int operand = LP64_ONLY(disp32_operand) NOT_LP64(call32_operand); 1034 emit_data((int) disp, rspec, operand); 1035 } 1036 1037 void Assembler::cdql() { 1038 emit_byte(0x99); 1039 } 1040 1041 void Assembler::cmovl(Condition cc, Register dst, Register src) { 1042 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction")); 1043 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1044 emit_byte(0x0F); 1045 emit_byte(0x40 | cc); 1046 emit_byte(0xC0 | encode); 1047 } 1048 1049 1050 void Assembler::cmovl(Condition cc, Register dst, Address src) { 1051 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction")); 1052 prefix(src, dst); 1053 emit_byte(0x0F); 1054 emit_byte(0x40 | cc); 1055 emit_operand(dst, src); 1056 } 1057 1058 void Assembler::cmpb(Address dst, int imm8) { 1059 InstructionMark im(this); 1060 prefix(dst); 1061 emit_byte(0x80); 1062 emit_operand(rdi, dst, 1); 1063 emit_byte(imm8); 1064 } 1065 1066 void Assembler::cmpl(Address dst, int32_t imm32) { 1067 InstructionMark im(this); 1068 prefix(dst); 1069 emit_byte(0x81); 1070 emit_operand(rdi, dst, 4); 1071 emit_long(imm32); 1072 } 1073 1074 void Assembler::cmpl(Register dst, int32_t imm32) { 1075 prefix(dst); 1076 emit_arith(0x81, 0xF8, dst, imm32); 1077 } 1078 1079 void Assembler::cmpl(Register dst, Register src) { 1080 (void) prefix_and_encode(dst->encoding(), src->encoding()); 1081 emit_arith(0x3B, 0xC0, dst, src); 1082 } 1083 1084 1085 void Assembler::cmpl(Register dst, Address src) { 1086 InstructionMark im(this); 1087 prefix(src, dst); 1088 emit_byte(0x3B); 1089 emit_operand(dst, src); 1090 } 1091 1092 void Assembler::cmpw(Address dst, int imm16) { 1093 InstructionMark im(this); 1094 assert(!dst.base_needs_rex() && !dst.index_needs_rex(), "no extended registers"); 1095 emit_byte(0x66); 1096 emit_byte(0x81); 1097 emit_operand(rdi, dst, 2); 1098 emit_word(imm16); 1099 } 1100 1101 // The 32-bit cmpxchg compares the value at adr with the contents of rax, 1102 // and stores reg into adr if so; otherwise, the value at adr is loaded into rax,. 1103 // The ZF is set if the compared values were equal, and cleared otherwise. 1104 void Assembler::cmpxchgl(Register reg, Address adr) { // cmpxchg 1105 if (Atomics & 2) { 1106 // caveat: no instructionmark, so this isn't relocatable. 1107 // Emit a synthetic, non-atomic, CAS equivalent. 1108 // Beware. The synthetic form sets all ICCs, not just ZF. 1109 // cmpxchg r,[m] is equivalent to rax, = CAS (m, rax, r) 1110 cmpl(rax, adr); 1111 movl(rax, adr); 1112 if (reg != rax) { 1113 Label L ; 1114 jcc(Assembler::notEqual, L); 1115 movl(adr, reg); 1116 bind(L); 1117 } 1118 } else { 1119 InstructionMark im(this); 1120 prefix(adr, reg); 1121 emit_byte(0x0F); 1122 emit_byte(0xB1); 1123 emit_operand(reg, adr); 1124 } 1125 } 1126 1127 void Assembler::comisd(XMMRegister dst, Address src) { 1128 // NOTE: dbx seems to decode this as comiss even though the 1129 // 0x66 is there. Strangly ucomisd comes out correct 1130 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1131 emit_byte(0x66); 1132 comiss(dst, src); 1133 } 1134 1135 void Assembler::comiss(XMMRegister dst, Address src) { 1136 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1137 1138 InstructionMark im(this); 1139 prefix(src, dst); 1140 emit_byte(0x0F); 1141 emit_byte(0x2F); 1142 emit_operand(dst, src); 1143 } 1144 1145 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) { 1146 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1147 emit_byte(0xF3); 1148 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1149 emit_byte(0x0F); 1150 emit_byte(0xE6); 1151 emit_byte(0xC0 | encode); 1152 } 1153 1154 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) { 1155 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1156 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1157 emit_byte(0x0F); 1158 emit_byte(0x5B); 1159 emit_byte(0xC0 | encode); 1160 } 1161 1162 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) { 1163 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1164 emit_byte(0xF2); 1165 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1166 emit_byte(0x0F); 1167 emit_byte(0x5A); 1168 emit_byte(0xC0 | encode); 1169 } 1170 1171 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) { 1172 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1173 emit_byte(0xF2); 1174 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1175 emit_byte(0x0F); 1176 emit_byte(0x2A); 1177 emit_byte(0xC0 | encode); 1178 } 1179 1180 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) { 1181 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1182 emit_byte(0xF3); 1183 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1184 emit_byte(0x0F); 1185 emit_byte(0x2A); 1186 emit_byte(0xC0 | encode); 1187 } 1188 1189 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) { 1190 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1191 emit_byte(0xF3); 1192 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1193 emit_byte(0x0F); 1194 emit_byte(0x5A); 1195 emit_byte(0xC0 | encode); 1196 } 1197 1198 void Assembler::cvttsd2sil(Register dst, XMMRegister src) { 1199 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1200 emit_byte(0xF2); 1201 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1202 emit_byte(0x0F); 1203 emit_byte(0x2C); 1204 emit_byte(0xC0 | encode); 1205 } 1206 1207 void Assembler::cvttss2sil(Register dst, XMMRegister src) { 1208 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1209 emit_byte(0xF3); 1210 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1211 emit_byte(0x0F); 1212 emit_byte(0x2C); 1213 emit_byte(0xC0 | encode); 1214 } 1215 1216 void Assembler::decl(Address dst) { 1217 // Don't use it directly. Use MacroAssembler::decrement() instead. 1218 InstructionMark im(this); 1219 prefix(dst); 1220 emit_byte(0xFF); 1221 emit_operand(rcx, dst); 1222 } 1223 1224 void Assembler::divsd(XMMRegister dst, Address src) { 1225 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1226 InstructionMark im(this); 1227 emit_byte(0xF2); 1228 prefix(src, dst); 1229 emit_byte(0x0F); 1230 emit_byte(0x5E); 1231 emit_operand(dst, src); 1232 } 1233 1234 void Assembler::divsd(XMMRegister dst, XMMRegister src) { 1235 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1236 emit_byte(0xF2); 1237 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1238 emit_byte(0x0F); 1239 emit_byte(0x5E); 1240 emit_byte(0xC0 | encode); 1241 } 1242 1243 void Assembler::divss(XMMRegister dst, Address src) { 1244 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1245 InstructionMark im(this); 1246 emit_byte(0xF3); 1247 prefix(src, dst); 1248 emit_byte(0x0F); 1249 emit_byte(0x5E); 1250 emit_operand(dst, src); 1251 } 1252 1253 void Assembler::divss(XMMRegister dst, XMMRegister src) { 1254 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1255 emit_byte(0xF3); 1256 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1257 emit_byte(0x0F); 1258 emit_byte(0x5E); 1259 emit_byte(0xC0 | encode); 1260 } 1261 1262 void Assembler::emms() { 1263 NOT_LP64(assert(VM_Version::supports_mmx(), "")); 1264 emit_byte(0x0F); 1265 emit_byte(0x77); 1266 } 1267 1268 void Assembler::hlt() { 1269 emit_byte(0xF4); 1270 } 1271 1272 void Assembler::idivl(Register src) { 1273 int encode = prefix_and_encode(src->encoding()); 1274 emit_byte(0xF7); 1275 emit_byte(0xF8 | encode); 1276 } 1277 1278 void Assembler::divl(Register src) { // Unsigned 1279 int encode = prefix_and_encode(src->encoding()); 1280 emit_byte(0xF7); 1281 emit_byte(0xF0 | encode); 1282 } 1283 1284 void Assembler::imull(Register dst, Register src) { 1285 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1286 emit_byte(0x0F); 1287 emit_byte(0xAF); 1288 emit_byte(0xC0 | encode); 1289 } 1290 1291 1292 void Assembler::imull(Register dst, Register src, int value) { 1293 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1294 if (is8bit(value)) { 1295 emit_byte(0x6B); 1296 emit_byte(0xC0 | encode); 1297 emit_byte(value & 0xFF); 1298 } else { 1299 emit_byte(0x69); 1300 emit_byte(0xC0 | encode); 1301 emit_long(value); 1302 } 1303 } 1304 1305 void Assembler::incl(Address dst) { 1306 // Don't use it directly. Use MacroAssembler::increment() instead. 1307 InstructionMark im(this); 1308 prefix(dst); 1309 emit_byte(0xFF); 1310 emit_operand(rax, dst); 1311 } 1312 1313 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) { 1314 InstructionMark im(this); 1315 relocate(rtype); 1316 assert((0 <= cc) && (cc < 16), "illegal cc"); 1317 if (L.is_bound()) { 1318 address dst = target(L); 1319 assert(dst != NULL, "jcc most probably wrong"); 1320 1321 const int short_size = 2; 1322 const int long_size = 6; 1323 intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos; 1324 if (rtype == relocInfo::none && is8bit(offs - short_size)) { 1325 // 0111 tttn #8-bit disp 1326 emit_byte(0x70 | cc); 1327 emit_byte((offs - short_size) & 0xFF); 1328 } else { 1329 // 0000 1111 1000 tttn #32-bit disp 1330 assert(is_simm32(offs - long_size), 1331 "must be 32bit offset (call4)"); 1332 emit_byte(0x0F); 1333 emit_byte(0x80 | cc); 1334 emit_long(offs - long_size); 1335 } 1336 } else { 1337 // Note: could eliminate cond. jumps to this jump if condition 1338 // is the same however, seems to be rather unlikely case. 1339 // Note: use jccb() if label to be bound is very close to get 1340 // an 8-bit displacement 1341 L.add_patch_at(code(), locator()); 1342 emit_byte(0x0F); 1343 emit_byte(0x80 | cc); 1344 emit_long(0); 1345 } 1346 } 1347 1348 void Assembler::jccb(Condition cc, Label& L) { 1349 if (L.is_bound()) { 1350 const int short_size = 2; 1351 address entry = target(L); 1352 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)), 1353 "Dispacement too large for a short jmp"); 1354 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos; 1355 // 0111 tttn #8-bit disp 1356 emit_byte(0x70 | cc); 1357 emit_byte((offs - short_size) & 0xFF); 1358 } else { 1359 InstructionMark im(this); 1360 L.add_patch_at(code(), locator()); 1361 emit_byte(0x70 | cc); 1362 emit_byte(0); 1363 } 1364 } 1365 1366 void Assembler::jmp(Address adr) { 1367 InstructionMark im(this); 1368 prefix(adr); 1369 emit_byte(0xFF); 1370 emit_operand(rsp, adr); 1371 } 1372 1373 void Assembler::jmp(Label& L, relocInfo::relocType rtype) { 1374 if (L.is_bound()) { 1375 address entry = target(L); 1376 assert(entry != NULL, "jmp most probably wrong"); 1377 InstructionMark im(this); 1378 const int short_size = 2; 1379 const int long_size = 5; 1380 intptr_t offs = entry - _code_pos; 1381 if (rtype == relocInfo::none && is8bit(offs - short_size)) { 1382 emit_byte(0xEB); 1383 emit_byte((offs - short_size) & 0xFF); 1384 } else { 1385 emit_byte(0xE9); 1386 emit_long(offs - long_size); 1387 } 1388 } else { 1389 // By default, forward jumps are always 32-bit displacements, since 1390 // we can't yet know where the label will be bound. If you're sure that 1391 // the forward jump will not run beyond 256 bytes, use jmpb to 1392 // force an 8-bit displacement. 1393 InstructionMark im(this); 1394 relocate(rtype); 1395 L.add_patch_at(code(), locator()); 1396 emit_byte(0xE9); 1397 emit_long(0); 1398 } 1399 } 1400 1401 void Assembler::jmp(Register entry) { 1402 int encode = prefix_and_encode(entry->encoding()); 1403 emit_byte(0xFF); 1404 emit_byte(0xE0 | encode); 1405 } 1406 1407 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) { 1408 InstructionMark im(this); 1409 emit_byte(0xE9); 1410 assert(dest != NULL, "must have a target"); 1411 intptr_t disp = dest - (_code_pos + sizeof(int32_t)); 1412 assert(is_simm32(disp), "must be 32bit offset (jmp)"); 1413 emit_data(disp, rspec.reloc(), call32_operand); 1414 } 1415 1416 void Assembler::jmpb(Label& L) { 1417 if (L.is_bound()) { 1418 const int short_size = 2; 1419 address entry = target(L); 1420 assert(is8bit((entry - _code_pos) + short_size), 1421 "Dispacement too large for a short jmp"); 1422 assert(entry != NULL, "jmp most probably wrong"); 1423 intptr_t offs = entry - _code_pos; 1424 emit_byte(0xEB); 1425 emit_byte((offs - short_size) & 0xFF); 1426 } else { 1427 InstructionMark im(this); 1428 L.add_patch_at(code(), locator()); 1429 emit_byte(0xEB); 1430 emit_byte(0); 1431 } 1432 } 1433 1434 void Assembler::ldmxcsr( Address src) { 1435 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1436 InstructionMark im(this); 1437 prefix(src); 1438 emit_byte(0x0F); 1439 emit_byte(0xAE); 1440 emit_operand(as_Register(2), src); 1441 } 1442 1443 void Assembler::leal(Register dst, Address src) { 1444 InstructionMark im(this); 1445 #ifdef _LP64 1446 emit_byte(0x67); // addr32 1447 prefix(src, dst); 1448 #endif // LP64 1449 emit_byte(0x8D); 1450 emit_operand(dst, src); 1451 } 1452 1453 void Assembler::lock() { 1454 if (Atomics & 1) { 1455 // Emit either nothing, a NOP, or a NOP: prefix 1456 emit_byte(0x90) ; 1457 } else { 1458 emit_byte(0xF0); 1459 } 1460 } 1461 1462 void Assembler::lzcntl(Register dst, Register src) { 1463 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR"); 1464 emit_byte(0xF3); 1465 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1466 emit_byte(0x0F); 1467 emit_byte(0xBD); 1468 emit_byte(0xC0 | encode); 1469 } 1470 1471 // Emit mfence instruction 1472 void Assembler::mfence() { 1473 NOT_LP64(assert(VM_Version::supports_sse2(), "unsupported");) 1474 emit_byte( 0x0F ); 1475 emit_byte( 0xAE ); 1476 emit_byte( 0xF0 ); 1477 } 1478 1479 void Assembler::mov(Register dst, Register src) { 1480 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); 1481 } 1482 1483 void Assembler::movapd(XMMRegister dst, XMMRegister src) { 1484 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1485 int dstenc = dst->encoding(); 1486 int srcenc = src->encoding(); 1487 emit_byte(0x66); 1488 if (dstenc < 8) { 1489 if (srcenc >= 8) { 1490 prefix(REX_B); 1491 srcenc -= 8; 1492 } 1493 } else { 1494 if (srcenc < 8) { 1495 prefix(REX_R); 1496 } else { 1497 prefix(REX_RB); 1498 srcenc -= 8; 1499 } 1500 dstenc -= 8; 1501 } 1502 emit_byte(0x0F); 1503 emit_byte(0x28); 1504 emit_byte(0xC0 | dstenc << 3 | srcenc); 1505 } 1506 1507 void Assembler::movaps(XMMRegister dst, XMMRegister src) { 1508 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1509 int dstenc = dst->encoding(); 1510 int srcenc = src->encoding(); 1511 if (dstenc < 8) { 1512 if (srcenc >= 8) { 1513 prefix(REX_B); 1514 srcenc -= 8; 1515 } 1516 } else { 1517 if (srcenc < 8) { 1518 prefix(REX_R); 1519 } else { 1520 prefix(REX_RB); 1521 srcenc -= 8; 1522 } 1523 dstenc -= 8; 1524 } 1525 emit_byte(0x0F); 1526 emit_byte(0x28); 1527 emit_byte(0xC0 | dstenc << 3 | srcenc); 1528 } 1529 1530 void Assembler::movb(Register dst, Address src) { 1531 NOT_LP64(assert(dst->has_byte_register(), "must have byte register")); 1532 InstructionMark im(this); 1533 prefix(src, dst, true); 1534 emit_byte(0x8A); 1535 emit_operand(dst, src); 1536 } 1537 1538 1539 void Assembler::movb(Address dst, int imm8) { 1540 InstructionMark im(this); 1541 prefix(dst); 1542 emit_byte(0xC6); 1543 emit_operand(rax, dst, 1); 1544 emit_byte(imm8); 1545 } 1546 1547 1548 void Assembler::movb(Address dst, Register src) { 1549 assert(src->has_byte_register(), "must have byte register"); 1550 InstructionMark im(this); 1551 prefix(dst, src, true); 1552 emit_byte(0x88); 1553 emit_operand(src, dst); 1554 } 1555 1556 void Assembler::movdl(XMMRegister dst, Register src) { 1557 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1558 emit_byte(0x66); 1559 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1560 emit_byte(0x0F); 1561 emit_byte(0x6E); 1562 emit_byte(0xC0 | encode); 1563 } 1564 1565 void Assembler::movdl(Register dst, XMMRegister src) { 1566 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1567 emit_byte(0x66); 1568 // swap src/dst to get correct prefix 1569 int encode = prefix_and_encode(src->encoding(), dst->encoding()); 1570 emit_byte(0x0F); 1571 emit_byte(0x7E); 1572 emit_byte(0xC0 | encode); 1573 } 1574 1575 void Assembler::movdqa(XMMRegister dst, Address src) { 1576 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1577 InstructionMark im(this); 1578 emit_byte(0x66); 1579 prefix(src, dst); 1580 emit_byte(0x0F); 1581 emit_byte(0x6F); 1582 emit_operand(dst, src); 1583 } 1584 1585 void Assembler::movdqa(XMMRegister dst, XMMRegister src) { 1586 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1587 emit_byte(0x66); 1588 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 1589 emit_byte(0x0F); 1590 emit_byte(0x6F); 1591 emit_byte(0xC0 | encode); 1592 } 1593 1594 void Assembler::movdqa(Address dst, XMMRegister src) { 1595 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1596 InstructionMark im(this); 1597 emit_byte(0x66); 1598 prefix(dst, src); 1599 emit_byte(0x0F); 1600 emit_byte(0x7F); 1601 emit_operand(src, dst); 1602 } 1603 1604 void Assembler::movdqu(XMMRegister dst, Address src) { 1605 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1606 InstructionMark im(this); 1607 emit_byte(0xF3); 1608 prefix(src, dst); 1609 emit_byte(0x0F); 1610 emit_byte(0x6F); 1611 emit_operand(dst, src); 1612 } 1613 1614 void Assembler::movdqu(XMMRegister dst, XMMRegister src) { 1615 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1616 emit_byte(0xF3); 1617 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 1618 emit_byte(0x0F); 1619 emit_byte(0x6F); 1620 emit_byte(0xC0 | encode); 1621 } 1622 1623 void Assembler::movdqu(Address dst, XMMRegister src) { 1624 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1625 InstructionMark im(this); 1626 emit_byte(0xF3); 1627 prefix(dst, src); 1628 emit_byte(0x0F); 1629 emit_byte(0x7F); 1630 emit_operand(src, dst); 1631 } 1632 1633 // Uses zero extension on 64bit 1634 1635 void Assembler::movl(Register dst, int32_t imm32) { 1636 int encode = prefix_and_encode(dst->encoding()); 1637 emit_byte(0xB8 | encode); 1638 emit_long(imm32); 1639 } 1640 1641 void Assembler::movl(Register dst, Register src) { 1642 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1643 emit_byte(0x8B); 1644 emit_byte(0xC0 | encode); 1645 } 1646 1647 void Assembler::movl(Register dst, Address src) { 1648 InstructionMark im(this); 1649 prefix(src, dst); 1650 emit_byte(0x8B); 1651 emit_operand(dst, src); 1652 } 1653 1654 void Assembler::movl(Address dst, int32_t imm32) { 1655 InstructionMark im(this); 1656 prefix(dst); 1657 emit_byte(0xC7); 1658 emit_operand(rax, dst, 4); 1659 emit_long(imm32); 1660 } 1661 1662 void Assembler::movl(Address dst, Register src) { 1663 InstructionMark im(this); 1664 prefix(dst, src); 1665 emit_byte(0x89); 1666 emit_operand(src, dst); 1667 } 1668 1669 // New cpus require to use movsd and movss to avoid partial register stall 1670 // when loading from memory. But for old Opteron use movlpd instead of movsd. 1671 // The selection is done in MacroAssembler::movdbl() and movflt(). 1672 void Assembler::movlpd(XMMRegister dst, Address src) { 1673 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1674 InstructionMark im(this); 1675 emit_byte(0x66); 1676 prefix(src, dst); 1677 emit_byte(0x0F); 1678 emit_byte(0x12); 1679 emit_operand(dst, src); 1680 } 1681 1682 void Assembler::movq( MMXRegister dst, Address src ) { 1683 assert( VM_Version::supports_mmx(), "" ); 1684 emit_byte(0x0F); 1685 emit_byte(0x6F); 1686 emit_operand(dst, src); 1687 } 1688 1689 void Assembler::movq( Address dst, MMXRegister src ) { 1690 assert( VM_Version::supports_mmx(), "" ); 1691 emit_byte(0x0F); 1692 emit_byte(0x7F); 1693 // workaround gcc (3.2.1-7a) bug 1694 // In that version of gcc with only an emit_operand(MMX, Address) 1695 // gcc will tail jump and try and reverse the parameters completely 1696 // obliterating dst in the process. By having a version available 1697 // that doesn't need to swap the args at the tail jump the bug is 1698 // avoided. 1699 emit_operand(dst, src); 1700 } 1701 1702 void Assembler::movq(XMMRegister dst, Address src) { 1703 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1704 InstructionMark im(this); 1705 emit_byte(0xF3); 1706 prefix(src, dst); 1707 emit_byte(0x0F); 1708 emit_byte(0x7E); 1709 emit_operand(dst, src); 1710 } 1711 1712 void Assembler::movq(Address dst, XMMRegister src) { 1713 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1714 InstructionMark im(this); 1715 emit_byte(0x66); 1716 prefix(dst, src); 1717 emit_byte(0x0F); 1718 emit_byte(0xD6); 1719 emit_operand(src, dst); 1720 } 1721 1722 void Assembler::movsbl(Register dst, Address src) { // movsxb 1723 InstructionMark im(this); 1724 prefix(src, dst); 1725 emit_byte(0x0F); 1726 emit_byte(0xBE); 1727 emit_operand(dst, src); 1728 } 1729 1730 void Assembler::movsbl(Register dst, Register src) { // movsxb 1731 NOT_LP64(assert(src->has_byte_register(), "must have byte register")); 1732 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); 1733 emit_byte(0x0F); 1734 emit_byte(0xBE); 1735 emit_byte(0xC0 | encode); 1736 } 1737 1738 void Assembler::movsd(XMMRegister dst, XMMRegister src) { 1739 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1740 emit_byte(0xF2); 1741 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1742 emit_byte(0x0F); 1743 emit_byte(0x10); 1744 emit_byte(0xC0 | encode); 1745 } 1746 1747 void Assembler::movsd(XMMRegister dst, Address src) { 1748 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1749 InstructionMark im(this); 1750 emit_byte(0xF2); 1751 prefix(src, dst); 1752 emit_byte(0x0F); 1753 emit_byte(0x10); 1754 emit_operand(dst, src); 1755 } 1756 1757 void Assembler::movsd(Address dst, XMMRegister src) { 1758 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1759 InstructionMark im(this); 1760 emit_byte(0xF2); 1761 prefix(dst, src); 1762 emit_byte(0x0F); 1763 emit_byte(0x11); 1764 emit_operand(src, dst); 1765 } 1766 1767 void Assembler::movss(XMMRegister dst, XMMRegister src) { 1768 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1769 emit_byte(0xF3); 1770 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1771 emit_byte(0x0F); 1772 emit_byte(0x10); 1773 emit_byte(0xC0 | encode); 1774 } 1775 1776 void Assembler::movss(XMMRegister dst, Address src) { 1777 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1778 InstructionMark im(this); 1779 emit_byte(0xF3); 1780 prefix(src, dst); 1781 emit_byte(0x0F); 1782 emit_byte(0x10); 1783 emit_operand(dst, src); 1784 } 1785 1786 void Assembler::movss(Address dst, XMMRegister src) { 1787 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1788 InstructionMark im(this); 1789 emit_byte(0xF3); 1790 prefix(dst, src); 1791 emit_byte(0x0F); 1792 emit_byte(0x11); 1793 emit_operand(src, dst); 1794 } 1795 1796 void Assembler::movswl(Register dst, Address src) { // movsxw 1797 InstructionMark im(this); 1798 prefix(src, dst); 1799 emit_byte(0x0F); 1800 emit_byte(0xBF); 1801 emit_operand(dst, src); 1802 } 1803 1804 void Assembler::movswl(Register dst, Register src) { // movsxw 1805 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1806 emit_byte(0x0F); 1807 emit_byte(0xBF); 1808 emit_byte(0xC0 | encode); 1809 } 1810 1811 void Assembler::movw(Address dst, int imm16) { 1812 InstructionMark im(this); 1813 1814 emit_byte(0x66); // switch to 16-bit mode 1815 prefix(dst); 1816 emit_byte(0xC7); 1817 emit_operand(rax, dst, 2); 1818 emit_word(imm16); 1819 } 1820 1821 void Assembler::movw(Register dst, Address src) { 1822 InstructionMark im(this); 1823 emit_byte(0x66); 1824 prefix(src, dst); 1825 emit_byte(0x8B); 1826 emit_operand(dst, src); 1827 } 1828 1829 void Assembler::movw(Address dst, Register src) { 1830 InstructionMark im(this); 1831 emit_byte(0x66); 1832 prefix(dst, src); 1833 emit_byte(0x89); 1834 emit_operand(src, dst); 1835 } 1836 1837 void Assembler::movzbl(Register dst, Address src) { // movzxb 1838 InstructionMark im(this); 1839 prefix(src, dst); 1840 emit_byte(0x0F); 1841 emit_byte(0xB6); 1842 emit_operand(dst, src); 1843 } 1844 1845 void Assembler::movzbl(Register dst, Register src) { // movzxb 1846 NOT_LP64(assert(src->has_byte_register(), "must have byte register")); 1847 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); 1848 emit_byte(0x0F); 1849 emit_byte(0xB6); 1850 emit_byte(0xC0 | encode); 1851 } 1852 1853 void Assembler::movzwl(Register dst, Address src) { // movzxw 1854 InstructionMark im(this); 1855 prefix(src, dst); 1856 emit_byte(0x0F); 1857 emit_byte(0xB7); 1858 emit_operand(dst, src); 1859 } 1860 1861 void Assembler::movzwl(Register dst, Register src) { // movzxw 1862 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1863 emit_byte(0x0F); 1864 emit_byte(0xB7); 1865 emit_byte(0xC0 | encode); 1866 } 1867 1868 void Assembler::mull(Address src) { 1869 InstructionMark im(this); 1870 prefix(src); 1871 emit_byte(0xF7); 1872 emit_operand(rsp, src); 1873 } 1874 1875 void Assembler::mull(Register src) { 1876 int encode = prefix_and_encode(src->encoding()); 1877 emit_byte(0xF7); 1878 emit_byte(0xE0 | encode); 1879 } 1880 1881 void Assembler::mulsd(XMMRegister dst, Address src) { 1882 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1883 InstructionMark im(this); 1884 emit_byte(0xF2); 1885 prefix(src, dst); 1886 emit_byte(0x0F); 1887 emit_byte(0x59); 1888 emit_operand(dst, src); 1889 } 1890 1891 void Assembler::mulsd(XMMRegister dst, XMMRegister src) { 1892 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 1893 emit_byte(0xF2); 1894 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1895 emit_byte(0x0F); 1896 emit_byte(0x59); 1897 emit_byte(0xC0 | encode); 1898 } 1899 1900 void Assembler::mulss(XMMRegister dst, Address src) { 1901 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1902 InstructionMark im(this); 1903 emit_byte(0xF3); 1904 prefix(src, dst); 1905 emit_byte(0x0F); 1906 emit_byte(0x59); 1907 emit_operand(dst, src); 1908 } 1909 1910 void Assembler::mulss(XMMRegister dst, XMMRegister src) { 1911 NOT_LP64(assert(VM_Version::supports_sse(), "")); 1912 emit_byte(0xF3); 1913 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 1914 emit_byte(0x0F); 1915 emit_byte(0x59); 1916 emit_byte(0xC0 | encode); 1917 } 1918 1919 void Assembler::negl(Register dst) { 1920 int encode = prefix_and_encode(dst->encoding()); 1921 emit_byte(0xF7); 1922 emit_byte(0xD8 | encode); 1923 } 1924 1925 void Assembler::nop(int i) { 1926 #ifdef ASSERT 1927 assert(i > 0, " "); 1928 // The fancy nops aren't currently recognized by debuggers making it a 1929 // pain to disassemble code while debugging. If asserts are on clearly 1930 // speed is not an issue so simply use the single byte traditional nop 1931 // to do alignment. 1932 1933 for (; i > 0 ; i--) emit_byte(0x90); 1934 return; 1935 1936 #endif // ASSERT 1937 1938 if (UseAddressNop && VM_Version::is_intel()) { 1939 // 1940 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel 1941 // 1: 0x90 1942 // 2: 0x66 0x90 1943 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) 1944 // 4: 0x0F 0x1F 0x40 0x00 1945 // 5: 0x0F 0x1F 0x44 0x00 0x00 1946 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 1947 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 1948 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 1949 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 1950 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 1951 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 1952 1953 // The rest coding is Intel specific - don't use consecutive address nops 1954 1955 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 1956 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 1957 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 1958 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 1959 1960 while(i >= 15) { 1961 // For Intel don't generate consecutive addess nops (mix with regular nops) 1962 i -= 15; 1963 emit_byte(0x66); // size prefix 1964 emit_byte(0x66); // size prefix 1965 emit_byte(0x66); // size prefix 1966 addr_nop_8(); 1967 emit_byte(0x66); // size prefix 1968 emit_byte(0x66); // size prefix 1969 emit_byte(0x66); // size prefix 1970 emit_byte(0x90); // nop 1971 } 1972 switch (i) { 1973 case 14: 1974 emit_byte(0x66); // size prefix 1975 case 13: 1976 emit_byte(0x66); // size prefix 1977 case 12: 1978 addr_nop_8(); 1979 emit_byte(0x66); // size prefix 1980 emit_byte(0x66); // size prefix 1981 emit_byte(0x66); // size prefix 1982 emit_byte(0x90); // nop 1983 break; 1984 case 11: 1985 emit_byte(0x66); // size prefix 1986 case 10: 1987 emit_byte(0x66); // size prefix 1988 case 9: 1989 emit_byte(0x66); // size prefix 1990 case 8: 1991 addr_nop_8(); 1992 break; 1993 case 7: 1994 addr_nop_7(); 1995 break; 1996 case 6: 1997 emit_byte(0x66); // size prefix 1998 case 5: 1999 addr_nop_5(); 2000 break; 2001 case 4: 2002 addr_nop_4(); 2003 break; 2004 case 3: 2005 // Don't use "0x0F 0x1F 0x00" - need patching safe padding 2006 emit_byte(0x66); // size prefix 2007 case 2: 2008 emit_byte(0x66); // size prefix 2009 case 1: 2010 emit_byte(0x90); // nop 2011 break; 2012 default: 2013 assert(i == 0, " "); 2014 } 2015 return; 2016 } 2017 if (UseAddressNop && VM_Version::is_amd()) { 2018 // 2019 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD. 2020 // 1: 0x90 2021 // 2: 0x66 0x90 2022 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) 2023 // 4: 0x0F 0x1F 0x40 0x00 2024 // 5: 0x0F 0x1F 0x44 0x00 0x00 2025 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 2026 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 2027 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 2028 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 2029 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 2030 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 2031 2032 // The rest coding is AMD specific - use consecutive address nops 2033 2034 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 2035 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 2036 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 2037 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 2038 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 2039 // Size prefixes (0x66) are added for larger sizes 2040 2041 while(i >= 22) { 2042 i -= 11; 2043 emit_byte(0x66); // size prefix 2044 emit_byte(0x66); // size prefix 2045 emit_byte(0x66); // size prefix 2046 addr_nop_8(); 2047 } 2048 // Generate first nop for size between 21-12 2049 switch (i) { 2050 case 21: 2051 i -= 1; 2052 emit_byte(0x66); // size prefix 2053 case 20: 2054 case 19: 2055 i -= 1; 2056 emit_byte(0x66); // size prefix 2057 case 18: 2058 case 17: 2059 i -= 1; 2060 emit_byte(0x66); // size prefix 2061 case 16: 2062 case 15: 2063 i -= 8; 2064 addr_nop_8(); 2065 break; 2066 case 14: 2067 case 13: 2068 i -= 7; 2069 addr_nop_7(); 2070 break; 2071 case 12: 2072 i -= 6; 2073 emit_byte(0x66); // size prefix 2074 addr_nop_5(); 2075 break; 2076 default: 2077 assert(i < 12, " "); 2078 } 2079 2080 // Generate second nop for size between 11-1 2081 switch (i) { 2082 case 11: 2083 emit_byte(0x66); // size prefix 2084 case 10: 2085 emit_byte(0x66); // size prefix 2086 case 9: 2087 emit_byte(0x66); // size prefix 2088 case 8: 2089 addr_nop_8(); 2090 break; 2091 case 7: 2092 addr_nop_7(); 2093 break; 2094 case 6: 2095 emit_byte(0x66); // size prefix 2096 case 5: 2097 addr_nop_5(); 2098 break; 2099 case 4: 2100 addr_nop_4(); 2101 break; 2102 case 3: 2103 // Don't use "0x0F 0x1F 0x00" - need patching safe padding 2104 emit_byte(0x66); // size prefix 2105 case 2: 2106 emit_byte(0x66); // size prefix 2107 case 1: 2108 emit_byte(0x90); // nop 2109 break; 2110 default: 2111 assert(i == 0, " "); 2112 } 2113 return; 2114 } 2115 2116 // Using nops with size prefixes "0x66 0x90". 2117 // From AMD Optimization Guide: 2118 // 1: 0x90 2119 // 2: 0x66 0x90 2120 // 3: 0x66 0x66 0x90 2121 // 4: 0x66 0x66 0x66 0x90 2122 // 5: 0x66 0x66 0x90 0x66 0x90 2123 // 6: 0x66 0x66 0x90 0x66 0x66 0x90 2124 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 2125 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90 2126 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 2127 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 2128 // 2129 while(i > 12) { 2130 i -= 4; 2131 emit_byte(0x66); // size prefix 2132 emit_byte(0x66); 2133 emit_byte(0x66); 2134 emit_byte(0x90); // nop 2135 } 2136 // 1 - 12 nops 2137 if(i > 8) { 2138 if(i > 9) { 2139 i -= 1; 2140 emit_byte(0x66); 2141 } 2142 i -= 3; 2143 emit_byte(0x66); 2144 emit_byte(0x66); 2145 emit_byte(0x90); 2146 } 2147 // 1 - 8 nops 2148 if(i > 4) { 2149 if(i > 6) { 2150 i -= 1; 2151 emit_byte(0x66); 2152 } 2153 i -= 3; 2154 emit_byte(0x66); 2155 emit_byte(0x66); 2156 emit_byte(0x90); 2157 } 2158 switch (i) { 2159 case 4: 2160 emit_byte(0x66); 2161 case 3: 2162 emit_byte(0x66); 2163 case 2: 2164 emit_byte(0x66); 2165 case 1: 2166 emit_byte(0x90); 2167 break; 2168 default: 2169 assert(i == 0, " "); 2170 } 2171 } 2172 2173 void Assembler::notl(Register dst) { 2174 int encode = prefix_and_encode(dst->encoding()); 2175 emit_byte(0xF7); 2176 emit_byte(0xD0 | encode ); 2177 } 2178 2179 void Assembler::orl(Address dst, int32_t imm32) { 2180 InstructionMark im(this); 2181 prefix(dst); 2182 emit_byte(0x81); 2183 emit_operand(rcx, dst, 4); 2184 emit_long(imm32); 2185 } 2186 2187 void Assembler::orl(Register dst, int32_t imm32) { 2188 prefix(dst); 2189 emit_arith(0x81, 0xC8, dst, imm32); 2190 } 2191 2192 2193 void Assembler::orl(Register dst, Address src) { 2194 InstructionMark im(this); 2195 prefix(src, dst); 2196 emit_byte(0x0B); 2197 emit_operand(dst, src); 2198 } 2199 2200 2201 void Assembler::orl(Register dst, Register src) { 2202 (void) prefix_and_encode(dst->encoding(), src->encoding()); 2203 emit_arith(0x0B, 0xC0, dst, src); 2204 } 2205 2206 void Assembler::pcmpestri(XMMRegister dst, Address src, int imm8) { 2207 assert(VM_Version::supports_sse4_2(), ""); 2208 2209 InstructionMark im(this); 2210 emit_byte(0x66); 2211 prefix(src, dst); 2212 emit_byte(0x0F); 2213 emit_byte(0x3A); 2214 emit_byte(0x61); 2215 emit_operand(dst, src); 2216 emit_byte(imm8); 2217 } 2218 2219 void Assembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) { 2220 assert(VM_Version::supports_sse4_2(), ""); 2221 2222 emit_byte(0x66); 2223 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 2224 emit_byte(0x0F); 2225 emit_byte(0x3A); 2226 emit_byte(0x61); 2227 emit_byte(0xC0 | encode); 2228 emit_byte(imm8); 2229 } 2230 2231 // generic 2232 void Assembler::pop(Register dst) { 2233 int encode = prefix_and_encode(dst->encoding()); 2234 emit_byte(0x58 | encode); 2235 } 2236 2237 void Assembler::popcntl(Register dst, Address src) { 2238 assert(VM_Version::supports_popcnt(), "must support"); 2239 InstructionMark im(this); 2240 emit_byte(0xF3); 2241 prefix(src, dst); 2242 emit_byte(0x0F); 2243 emit_byte(0xB8); 2244 emit_operand(dst, src); 2245 } 2246 2247 void Assembler::popcntl(Register dst, Register src) { 2248 assert(VM_Version::supports_popcnt(), "must support"); 2249 emit_byte(0xF3); 2250 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2251 emit_byte(0x0F); 2252 emit_byte(0xB8); 2253 emit_byte(0xC0 | encode); 2254 } 2255 2256 void Assembler::popf() { 2257 emit_byte(0x9D); 2258 } 2259 2260 #ifndef _LP64 // no 32bit push/pop on amd64 2261 void Assembler::popl(Address dst) { 2262 // NOTE: this will adjust stack by 8byte on 64bits 2263 InstructionMark im(this); 2264 prefix(dst); 2265 emit_byte(0x8F); 2266 emit_operand(rax, dst); 2267 } 2268 #endif 2269 2270 void Assembler::prefetch_prefix(Address src) { 2271 prefix(src); 2272 emit_byte(0x0F); 2273 } 2274 2275 void Assembler::prefetchnta(Address src) { 2276 NOT_LP64(assert(VM_Version::supports_sse2(), "must support")); 2277 InstructionMark im(this); 2278 prefetch_prefix(src); 2279 emit_byte(0x18); 2280 emit_operand(rax, src); // 0, src 2281 } 2282 2283 void Assembler::prefetchr(Address src) { 2284 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support")); 2285 InstructionMark im(this); 2286 prefetch_prefix(src); 2287 emit_byte(0x0D); 2288 emit_operand(rax, src); // 0, src 2289 } 2290 2291 void Assembler::prefetcht0(Address src) { 2292 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); 2293 InstructionMark im(this); 2294 prefetch_prefix(src); 2295 emit_byte(0x18); 2296 emit_operand(rcx, src); // 1, src 2297 } 2298 2299 void Assembler::prefetcht1(Address src) { 2300 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); 2301 InstructionMark im(this); 2302 prefetch_prefix(src); 2303 emit_byte(0x18); 2304 emit_operand(rdx, src); // 2, src 2305 } 2306 2307 void Assembler::prefetcht2(Address src) { 2308 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); 2309 InstructionMark im(this); 2310 prefetch_prefix(src); 2311 emit_byte(0x18); 2312 emit_operand(rbx, src); // 3, src 2313 } 2314 2315 void Assembler::prefetchw(Address src) { 2316 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support")); 2317 InstructionMark im(this); 2318 prefetch_prefix(src); 2319 emit_byte(0x0D); 2320 emit_operand(rcx, src); // 1, src 2321 } 2322 2323 void Assembler::prefix(Prefix p) { 2324 a_byte(p); 2325 } 2326 2327 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) { 2328 assert(isByte(mode), "invalid value"); 2329 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2330 2331 emit_byte(0x66); 2332 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2333 emit_byte(0x0F); 2334 emit_byte(0x70); 2335 emit_byte(0xC0 | encode); 2336 emit_byte(mode & 0xFF); 2337 2338 } 2339 2340 void Assembler::pshufd(XMMRegister dst, Address src, int mode) { 2341 assert(isByte(mode), "invalid value"); 2342 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2343 2344 InstructionMark im(this); 2345 emit_byte(0x66); 2346 prefix(src, dst); 2347 emit_byte(0x0F); 2348 emit_byte(0x70); 2349 emit_operand(dst, src); 2350 emit_byte(mode & 0xFF); 2351 } 2352 2353 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) { 2354 assert(isByte(mode), "invalid value"); 2355 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2356 2357 emit_byte(0xF2); 2358 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2359 emit_byte(0x0F); 2360 emit_byte(0x70); 2361 emit_byte(0xC0 | encode); 2362 emit_byte(mode & 0xFF); 2363 } 2364 2365 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) { 2366 assert(isByte(mode), "invalid value"); 2367 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2368 2369 InstructionMark im(this); 2370 emit_byte(0xF2); 2371 prefix(src, dst); // QQ new 2372 emit_byte(0x0F); 2373 emit_byte(0x70); 2374 emit_operand(dst, src); 2375 emit_byte(mode & 0xFF); 2376 } 2377 2378 void Assembler::psrlq(XMMRegister dst, int shift) { 2379 // HMM Table D-1 says sse2 or mmx 2380 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2381 2382 int encode = prefixq_and_encode(xmm2->encoding(), dst->encoding()); 2383 emit_byte(0x66); 2384 emit_byte(0x0F); 2385 emit_byte(0x73); 2386 emit_byte(0xC0 | encode); 2387 emit_byte(shift); 2388 } 2389 2390 void Assembler::ptest(XMMRegister dst, Address src) { 2391 assert(VM_Version::supports_sse4_1(), ""); 2392 2393 InstructionMark im(this); 2394 emit_byte(0x66); 2395 prefix(src, dst); 2396 emit_byte(0x0F); 2397 emit_byte(0x38); 2398 emit_byte(0x17); 2399 emit_operand(dst, src); 2400 } 2401 2402 void Assembler::ptest(XMMRegister dst, XMMRegister src) { 2403 assert(VM_Version::supports_sse4_1(), ""); 2404 2405 emit_byte(0x66); 2406 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 2407 emit_byte(0x0F); 2408 emit_byte(0x38); 2409 emit_byte(0x17); 2410 emit_byte(0xC0 | encode); 2411 } 2412 2413 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) { 2414 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2415 emit_byte(0x66); 2416 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2417 emit_byte(0x0F); 2418 emit_byte(0x60); 2419 emit_byte(0xC0 | encode); 2420 } 2421 2422 void Assembler::push(int32_t imm32) { 2423 // in 64bits we push 64bits onto the stack but only 2424 // take a 32bit immediate 2425 emit_byte(0x68); 2426 emit_long(imm32); 2427 } 2428 2429 void Assembler::push(Register src) { 2430 int encode = prefix_and_encode(src->encoding()); 2431 2432 emit_byte(0x50 | encode); 2433 } 2434 2435 void Assembler::pushf() { 2436 emit_byte(0x9C); 2437 } 2438 2439 #ifndef _LP64 // no 32bit push/pop on amd64 2440 void Assembler::pushl(Address src) { 2441 // Note this will push 64bit on 64bit 2442 InstructionMark im(this); 2443 prefix(src); 2444 emit_byte(0xFF); 2445 emit_operand(rsi, src); 2446 } 2447 #endif 2448 2449 void Assembler::pxor(XMMRegister dst, Address src) { 2450 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2451 InstructionMark im(this); 2452 emit_byte(0x66); 2453 prefix(src, dst); 2454 emit_byte(0x0F); 2455 emit_byte(0xEF); 2456 emit_operand(dst, src); 2457 } 2458 2459 void Assembler::pxor(XMMRegister dst, XMMRegister src) { 2460 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2461 InstructionMark im(this); 2462 emit_byte(0x66); 2463 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2464 emit_byte(0x0F); 2465 emit_byte(0xEF); 2466 emit_byte(0xC0 | encode); 2467 } 2468 2469 void Assembler::rcll(Register dst, int imm8) { 2470 assert(isShiftCount(imm8), "illegal shift count"); 2471 int encode = prefix_and_encode(dst->encoding()); 2472 if (imm8 == 1) { 2473 emit_byte(0xD1); 2474 emit_byte(0xD0 | encode); 2475 } else { 2476 emit_byte(0xC1); 2477 emit_byte(0xD0 | encode); 2478 emit_byte(imm8); 2479 } 2480 } 2481 2482 // copies data from [esi] to [edi] using rcx pointer sized words 2483 // generic 2484 void Assembler::rep_mov() { 2485 emit_byte(0xF3); 2486 // MOVSQ 2487 LP64_ONLY(prefix(REX_W)); 2488 emit_byte(0xA5); 2489 } 2490 2491 // sets rcx pointer sized words with rax, value at [edi] 2492 // generic 2493 void Assembler::rep_set() { // rep_set 2494 emit_byte(0xF3); 2495 // STOSQ 2496 LP64_ONLY(prefix(REX_W)); 2497 emit_byte(0xAB); 2498 } 2499 2500 // scans rcx pointer sized words at [edi] for occurance of rax, 2501 // generic 2502 void Assembler::repne_scan() { // repne_scan 2503 emit_byte(0xF2); 2504 // SCASQ 2505 LP64_ONLY(prefix(REX_W)); 2506 emit_byte(0xAF); 2507 } 2508 2509 #ifdef _LP64 2510 // scans rcx 4 byte words at [edi] for occurance of rax, 2511 // generic 2512 void Assembler::repne_scanl() { // repne_scan 2513 emit_byte(0xF2); 2514 // SCASL 2515 emit_byte(0xAF); 2516 } 2517 #endif 2518 2519 void Assembler::ret(int imm16) { 2520 if (imm16 == 0) { 2521 emit_byte(0xC3); 2522 } else { 2523 emit_byte(0xC2); 2524 emit_word(imm16); 2525 } 2526 } 2527 2528 void Assembler::sahf() { 2529 #ifdef _LP64 2530 // Not supported in 64bit mode 2531 ShouldNotReachHere(); 2532 #endif 2533 emit_byte(0x9E); 2534 } 2535 2536 void Assembler::sarl(Register dst, int imm8) { 2537 int encode = prefix_and_encode(dst->encoding()); 2538 assert(isShiftCount(imm8), "illegal shift count"); 2539 if (imm8 == 1) { 2540 emit_byte(0xD1); 2541 emit_byte(0xF8 | encode); 2542 } else { 2543 emit_byte(0xC1); 2544 emit_byte(0xF8 | encode); 2545 emit_byte(imm8); 2546 } 2547 } 2548 2549 void Assembler::sarl(Register dst) { 2550 int encode = prefix_and_encode(dst->encoding()); 2551 emit_byte(0xD3); 2552 emit_byte(0xF8 | encode); 2553 } 2554 2555 void Assembler::sbbl(Address dst, int32_t imm32) { 2556 InstructionMark im(this); 2557 prefix(dst); 2558 emit_arith_operand(0x81, rbx, dst, imm32); 2559 } 2560 2561 void Assembler::sbbl(Register dst, int32_t imm32) { 2562 prefix(dst); 2563 emit_arith(0x81, 0xD8, dst, imm32); 2564 } 2565 2566 2567 void Assembler::sbbl(Register dst, Address src) { 2568 InstructionMark im(this); 2569 prefix(src, dst); 2570 emit_byte(0x1B); 2571 emit_operand(dst, src); 2572 } 2573 2574 void Assembler::sbbl(Register dst, Register src) { 2575 (void) prefix_and_encode(dst->encoding(), src->encoding()); 2576 emit_arith(0x1B, 0xC0, dst, src); 2577 } 2578 2579 void Assembler::setb(Condition cc, Register dst) { 2580 assert(0 <= cc && cc < 16, "illegal cc"); 2581 int encode = prefix_and_encode(dst->encoding(), true); 2582 emit_byte(0x0F); 2583 emit_byte(0x90 | cc); 2584 emit_byte(0xC0 | encode); 2585 } 2586 2587 void Assembler::shll(Register dst, int imm8) { 2588 assert(isShiftCount(imm8), "illegal shift count"); 2589 int encode = prefix_and_encode(dst->encoding()); 2590 if (imm8 == 1 ) { 2591 emit_byte(0xD1); 2592 emit_byte(0xE0 | encode); 2593 } else { 2594 emit_byte(0xC1); 2595 emit_byte(0xE0 | encode); 2596 emit_byte(imm8); 2597 } 2598 } 2599 2600 void Assembler::shll(Register dst) { 2601 int encode = prefix_and_encode(dst->encoding()); 2602 emit_byte(0xD3); 2603 emit_byte(0xE0 | encode); 2604 } 2605 2606 void Assembler::shrl(Register dst, int imm8) { 2607 assert(isShiftCount(imm8), "illegal shift count"); 2608 int encode = prefix_and_encode(dst->encoding()); 2609 emit_byte(0xC1); 2610 emit_byte(0xE8 | encode); 2611 emit_byte(imm8); 2612 } 2613 2614 void Assembler::shrl(Register dst) { 2615 int encode = prefix_and_encode(dst->encoding()); 2616 emit_byte(0xD3); 2617 emit_byte(0xE8 | encode); 2618 } 2619 2620 // copies a single word from [esi] to [edi] 2621 void Assembler::smovl() { 2622 emit_byte(0xA5); 2623 } 2624 2625 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) { 2626 // HMM Table D-1 says sse2 2627 // NOT_LP64(assert(VM_Version::supports_sse(), "")); 2628 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2629 emit_byte(0xF2); 2630 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2631 emit_byte(0x0F); 2632 emit_byte(0x51); 2633 emit_byte(0xC0 | encode); 2634 } 2635 2636 void Assembler::sqrtsd(XMMRegister dst, Address src) { 2637 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2638 InstructionMark im(this); 2639 emit_byte(0xF2); 2640 prefix(src, dst); 2641 emit_byte(0x0F); 2642 emit_byte(0x51); 2643 emit_operand(dst, src); 2644 } 2645 2646 void Assembler::sqrtss(XMMRegister dst, XMMRegister src) { 2647 // HMM Table D-1 says sse2 2648 // NOT_LP64(assert(VM_Version::supports_sse(), "")); 2649 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2650 emit_byte(0xF3); 2651 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2652 emit_byte(0x0F); 2653 emit_byte(0x51); 2654 emit_byte(0xC0 | encode); 2655 } 2656 2657 void Assembler::sqrtss(XMMRegister dst, Address src) { 2658 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2659 InstructionMark im(this); 2660 emit_byte(0xF3); 2661 prefix(src, dst); 2662 emit_byte(0x0F); 2663 emit_byte(0x51); 2664 emit_operand(dst, src); 2665 } 2666 2667 void Assembler::stmxcsr( Address dst) { 2668 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2669 InstructionMark im(this); 2670 prefix(dst); 2671 emit_byte(0x0F); 2672 emit_byte(0xAE); 2673 emit_operand(as_Register(3), dst); 2674 } 2675 2676 void Assembler::subl(Address dst, int32_t imm32) { 2677 InstructionMark im(this); 2678 prefix(dst); 2679 if (is8bit(imm32)) { 2680 emit_byte(0x83); 2681 emit_operand(rbp, dst, 1); 2682 emit_byte(imm32 & 0xFF); 2683 } else { 2684 emit_byte(0x81); 2685 emit_operand(rbp, dst, 4); 2686 emit_long(imm32); 2687 } 2688 } 2689 2690 void Assembler::subl(Register dst, int32_t imm32) { 2691 prefix(dst); 2692 emit_arith(0x81, 0xE8, dst, imm32); 2693 } 2694 2695 void Assembler::subl(Address dst, Register src) { 2696 InstructionMark im(this); 2697 prefix(dst, src); 2698 emit_byte(0x29); 2699 emit_operand(src, dst); 2700 } 2701 2702 void Assembler::subl(Register dst, Address src) { 2703 InstructionMark im(this); 2704 prefix(src, dst); 2705 emit_byte(0x2B); 2706 emit_operand(dst, src); 2707 } 2708 2709 void Assembler::subl(Register dst, Register src) { 2710 (void) prefix_and_encode(dst->encoding(), src->encoding()); 2711 emit_arith(0x2B, 0xC0, dst, src); 2712 } 2713 2714 void Assembler::subsd(XMMRegister dst, XMMRegister src) { 2715 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2716 emit_byte(0xF2); 2717 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2718 emit_byte(0x0F); 2719 emit_byte(0x5C); 2720 emit_byte(0xC0 | encode); 2721 } 2722 2723 void Assembler::subsd(XMMRegister dst, Address src) { 2724 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2725 InstructionMark im(this); 2726 emit_byte(0xF2); 2727 prefix(src, dst); 2728 emit_byte(0x0F); 2729 emit_byte(0x5C); 2730 emit_operand(dst, src); 2731 } 2732 2733 void Assembler::subss(XMMRegister dst, XMMRegister src) { 2734 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2735 emit_byte(0xF3); 2736 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2737 emit_byte(0x0F); 2738 emit_byte(0x5C); 2739 emit_byte(0xC0 | encode); 2740 } 2741 2742 void Assembler::subss(XMMRegister dst, Address src) { 2743 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2744 InstructionMark im(this); 2745 emit_byte(0xF3); 2746 prefix(src, dst); 2747 emit_byte(0x0F); 2748 emit_byte(0x5C); 2749 emit_operand(dst, src); 2750 } 2751 2752 void Assembler::testb(Register dst, int imm8) { 2753 NOT_LP64(assert(dst->has_byte_register(), "must have byte register")); 2754 (void) prefix_and_encode(dst->encoding(), true); 2755 emit_arith_b(0xF6, 0xC0, dst, imm8); 2756 } 2757 2758 void Assembler::testl(Register dst, int32_t imm32) { 2759 // not using emit_arith because test 2760 // doesn't support sign-extension of 2761 // 8bit operands 2762 int encode = dst->encoding(); 2763 if (encode == 0) { 2764 emit_byte(0xA9); 2765 } else { 2766 encode = prefix_and_encode(encode); 2767 emit_byte(0xF7); 2768 emit_byte(0xC0 | encode); 2769 } 2770 emit_long(imm32); 2771 } 2772 2773 void Assembler::testl(Register dst, Register src) { 2774 (void) prefix_and_encode(dst->encoding(), src->encoding()); 2775 emit_arith(0x85, 0xC0, dst, src); 2776 } 2777 2778 void Assembler::testl(Register dst, Address src) { 2779 InstructionMark im(this); 2780 prefix(src, dst); 2781 emit_byte(0x85); 2782 emit_operand(dst, src); 2783 } 2784 2785 void Assembler::ucomisd(XMMRegister dst, Address src) { 2786 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2787 emit_byte(0x66); 2788 ucomiss(dst, src); 2789 } 2790 2791 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) { 2792 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2793 emit_byte(0x66); 2794 ucomiss(dst, src); 2795 } 2796 2797 void Assembler::ucomiss(XMMRegister dst, Address src) { 2798 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2799 2800 InstructionMark im(this); 2801 prefix(src, dst); 2802 emit_byte(0x0F); 2803 emit_byte(0x2E); 2804 emit_operand(dst, src); 2805 } 2806 2807 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) { 2808 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2809 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2810 emit_byte(0x0F); 2811 emit_byte(0x2E); 2812 emit_byte(0xC0 | encode); 2813 } 2814 2815 2816 void Assembler::xaddl(Address dst, Register src) { 2817 InstructionMark im(this); 2818 prefix(dst, src); 2819 emit_byte(0x0F); 2820 emit_byte(0xC1); 2821 emit_operand(src, dst); 2822 } 2823 2824 void Assembler::xchgl(Register dst, Address src) { // xchg 2825 InstructionMark im(this); 2826 prefix(src, dst); 2827 emit_byte(0x87); 2828 emit_operand(dst, src); 2829 } 2830 2831 void Assembler::xchgl(Register dst, Register src) { 2832 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2833 emit_byte(0x87); 2834 emit_byte(0xc0 | encode); 2835 } 2836 2837 void Assembler::xorl(Register dst, int32_t imm32) { 2838 prefix(dst); 2839 emit_arith(0x81, 0xF0, dst, imm32); 2840 } 2841 2842 void Assembler::xorl(Register dst, Address src) { 2843 InstructionMark im(this); 2844 prefix(src, dst); 2845 emit_byte(0x33); 2846 emit_operand(dst, src); 2847 } 2848 2849 void Assembler::xorl(Register dst, Register src) { 2850 (void) prefix_and_encode(dst->encoding(), src->encoding()); 2851 emit_arith(0x33, 0xC0, dst, src); 2852 } 2853 2854 void Assembler::xorpd(XMMRegister dst, XMMRegister src) { 2855 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2856 emit_byte(0x66); 2857 xorps(dst, src); 2858 } 2859 2860 void Assembler::xorpd(XMMRegister dst, Address src) { 2861 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 2862 InstructionMark im(this); 2863 emit_byte(0x66); 2864 prefix(src, dst); 2865 emit_byte(0x0F); 2866 emit_byte(0x57); 2867 emit_operand(dst, src); 2868 } 2869 2870 2871 void Assembler::xorps(XMMRegister dst, XMMRegister src) { 2872 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2873 int encode = prefix_and_encode(dst->encoding(), src->encoding()); 2874 emit_byte(0x0F); 2875 emit_byte(0x57); 2876 emit_byte(0xC0 | encode); 2877 } 2878 2879 void Assembler::xorps(XMMRegister dst, Address src) { 2880 NOT_LP64(assert(VM_Version::supports_sse(), "")); 2881 InstructionMark im(this); 2882 prefix(src, dst); 2883 emit_byte(0x0F); 2884 emit_byte(0x57); 2885 emit_operand(dst, src); 2886 } 2887 2888 #ifndef _LP64 2889 // 32bit only pieces of the assembler 2890 2891 void Assembler::cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec) { 2892 // NO PREFIX AS NEVER 64BIT 2893 InstructionMark im(this); 2894 emit_byte(0x81); 2895 emit_byte(0xF8 | src1->encoding()); 2896 emit_data(imm32, rspec, 0); 2897 } 2898 2899 void Assembler::cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec) { 2900 // NO PREFIX AS NEVER 64BIT (not even 32bit versions of 64bit regs 2901 InstructionMark im(this); 2902 emit_byte(0x81); 2903 emit_operand(rdi, src1); 2904 emit_data(imm32, rspec, 0); 2905 } 2906 2907 // The 64-bit (32bit platform) cmpxchg compares the value at adr with the contents of rdx:rax, 2908 // and stores rcx:rbx into adr if so; otherwise, the value at adr is loaded 2909 // into rdx:rax. The ZF is set if the compared values were equal, and cleared otherwise. 2910 void Assembler::cmpxchg8(Address adr) { 2911 InstructionMark im(this); 2912 emit_byte(0x0F); 2913 emit_byte(0xc7); 2914 emit_operand(rcx, adr); 2915 } 2916 2917 void Assembler::decl(Register dst) { 2918 // Don't use it directly. Use MacroAssembler::decrementl() instead. 2919 emit_byte(0x48 | dst->encoding()); 2920 } 2921 2922 #endif // _LP64 2923 2924 // 64bit typically doesn't use the x87 but needs to for the trig funcs 2925 2926 void Assembler::fabs() { 2927 emit_byte(0xD9); 2928 emit_byte(0xE1); 2929 } 2930 2931 void Assembler::fadd(int i) { 2932 emit_farith(0xD8, 0xC0, i); 2933 } 2934 2935 void Assembler::fadd_d(Address src) { 2936 InstructionMark im(this); 2937 emit_byte(0xDC); 2938 emit_operand32(rax, src); 2939 } 2940 2941 void Assembler::fadd_s(Address src) { 2942 InstructionMark im(this); 2943 emit_byte(0xD8); 2944 emit_operand32(rax, src); 2945 } 2946 2947 void Assembler::fadda(int i) { 2948 emit_farith(0xDC, 0xC0, i); 2949 } 2950 2951 void Assembler::faddp(int i) { 2952 emit_farith(0xDE, 0xC0, i); 2953 } 2954 2955 void Assembler::fchs() { 2956 emit_byte(0xD9); 2957 emit_byte(0xE0); 2958 } 2959 2960 void Assembler::fcom(int i) { 2961 emit_farith(0xD8, 0xD0, i); 2962 } 2963 2964 void Assembler::fcomp(int i) { 2965 emit_farith(0xD8, 0xD8, i); 2966 } 2967 2968 void Assembler::fcomp_d(Address src) { 2969 InstructionMark im(this); 2970 emit_byte(0xDC); 2971 emit_operand32(rbx, src); 2972 } 2973 2974 void Assembler::fcomp_s(Address src) { 2975 InstructionMark im(this); 2976 emit_byte(0xD8); 2977 emit_operand32(rbx, src); 2978 } 2979 2980 void Assembler::fcompp() { 2981 emit_byte(0xDE); 2982 emit_byte(0xD9); 2983 } 2984 2985 void Assembler::fcos() { 2986 emit_byte(0xD9); 2987 emit_byte(0xFF); 2988 } 2989 2990 void Assembler::fdecstp() { 2991 emit_byte(0xD9); 2992 emit_byte(0xF6); 2993 } 2994 2995 void Assembler::fdiv(int i) { 2996 emit_farith(0xD8, 0xF0, i); 2997 } 2998 2999 void Assembler::fdiv_d(Address src) { 3000 InstructionMark im(this); 3001 emit_byte(0xDC); 3002 emit_operand32(rsi, src); 3003 } 3004 3005 void Assembler::fdiv_s(Address src) { 3006 InstructionMark im(this); 3007 emit_byte(0xD8); 3008 emit_operand32(rsi, src); 3009 } 3010 3011 void Assembler::fdiva(int i) { 3012 emit_farith(0xDC, 0xF8, i); 3013 } 3014 3015 // Note: The Intel manual (Pentium Processor User's Manual, Vol.3, 1994) 3016 // is erroneous for some of the floating-point instructions below. 3017 3018 void Assembler::fdivp(int i) { 3019 emit_farith(0xDE, 0xF8, i); // ST(0) <- ST(0) / ST(1) and pop (Intel manual wrong) 3020 } 3021 3022 void Assembler::fdivr(int i) { 3023 emit_farith(0xD8, 0xF8, i); 3024 } 3025 3026 void Assembler::fdivr_d(Address src) { 3027 InstructionMark im(this); 3028 emit_byte(0xDC); 3029 emit_operand32(rdi, src); 3030 } 3031 3032 void Assembler::fdivr_s(Address src) { 3033 InstructionMark im(this); 3034 emit_byte(0xD8); 3035 emit_operand32(rdi, src); 3036 } 3037 3038 void Assembler::fdivra(int i) { 3039 emit_farith(0xDC, 0xF0, i); 3040 } 3041 3042 void Assembler::fdivrp(int i) { 3043 emit_farith(0xDE, 0xF0, i); // ST(0) <- ST(1) / ST(0) and pop (Intel manual wrong) 3044 } 3045 3046 void Assembler::ffree(int i) { 3047 emit_farith(0xDD, 0xC0, i); 3048 } 3049 3050 void Assembler::fild_d(Address adr) { 3051 InstructionMark im(this); 3052 emit_byte(0xDF); 3053 emit_operand32(rbp, adr); 3054 } 3055 3056 void Assembler::fild_s(Address adr) { 3057 InstructionMark im(this); 3058 emit_byte(0xDB); 3059 emit_operand32(rax, adr); 3060 } 3061 3062 void Assembler::fincstp() { 3063 emit_byte(0xD9); 3064 emit_byte(0xF7); 3065 } 3066 3067 void Assembler::finit() { 3068 emit_byte(0x9B); 3069 emit_byte(0xDB); 3070 emit_byte(0xE3); 3071 } 3072 3073 void Assembler::fist_s(Address adr) { 3074 InstructionMark im(this); 3075 emit_byte(0xDB); 3076 emit_operand32(rdx, adr); 3077 } 3078 3079 void Assembler::fistp_d(Address adr) { 3080 InstructionMark im(this); 3081 emit_byte(0xDF); 3082 emit_operand32(rdi, adr); 3083 } 3084 3085 void Assembler::fistp_s(Address adr) { 3086 InstructionMark im(this); 3087 emit_byte(0xDB); 3088 emit_operand32(rbx, adr); 3089 } 3090 3091 void Assembler::fld1() { 3092 emit_byte(0xD9); 3093 emit_byte(0xE8); 3094 } 3095 3096 void Assembler::fld_d(Address adr) { 3097 InstructionMark im(this); 3098 emit_byte(0xDD); 3099 emit_operand32(rax, adr); 3100 } 3101 3102 void Assembler::fld_s(Address adr) { 3103 InstructionMark im(this); 3104 emit_byte(0xD9); 3105 emit_operand32(rax, adr); 3106 } 3107 3108 3109 void Assembler::fld_s(int index) { 3110 emit_farith(0xD9, 0xC0, index); 3111 } 3112 3113 void Assembler::fld_x(Address adr) { 3114 InstructionMark im(this); 3115 emit_byte(0xDB); 3116 emit_operand32(rbp, adr); 3117 } 3118 3119 void Assembler::fldcw(Address src) { 3120 InstructionMark im(this); 3121 emit_byte(0xd9); 3122 emit_operand32(rbp, src); 3123 } 3124 3125 void Assembler::fldenv(Address src) { 3126 InstructionMark im(this); 3127 emit_byte(0xD9); 3128 emit_operand32(rsp, src); 3129 } 3130 3131 void Assembler::fldlg2() { 3132 emit_byte(0xD9); 3133 emit_byte(0xEC); 3134 } 3135 3136 void Assembler::fldln2() { 3137 emit_byte(0xD9); 3138 emit_byte(0xED); 3139 } 3140 3141 void Assembler::fldz() { 3142 emit_byte(0xD9); 3143 emit_byte(0xEE); 3144 } 3145 3146 void Assembler::flog() { 3147 fldln2(); 3148 fxch(); 3149 fyl2x(); 3150 } 3151 3152 void Assembler::flog10() { 3153 fldlg2(); 3154 fxch(); 3155 fyl2x(); 3156 } 3157 3158 void Assembler::fmul(int i) { 3159 emit_farith(0xD8, 0xC8, i); 3160 } 3161 3162 void Assembler::fmul_d(Address src) { 3163 InstructionMark im(this); 3164 emit_byte(0xDC); 3165 emit_operand32(rcx, src); 3166 } 3167 3168 void Assembler::fmul_s(Address src) { 3169 InstructionMark im(this); 3170 emit_byte(0xD8); 3171 emit_operand32(rcx, src); 3172 } 3173 3174 void Assembler::fmula(int i) { 3175 emit_farith(0xDC, 0xC8, i); 3176 } 3177 3178 void Assembler::fmulp(int i) { 3179 emit_farith(0xDE, 0xC8, i); 3180 } 3181 3182 void Assembler::fnsave(Address dst) { 3183 InstructionMark im(this); 3184 emit_byte(0xDD); 3185 emit_operand32(rsi, dst); 3186 } 3187 3188 void Assembler::fnstcw(Address src) { 3189 InstructionMark im(this); 3190 emit_byte(0x9B); 3191 emit_byte(0xD9); 3192 emit_operand32(rdi, src); 3193 } 3194 3195 void Assembler::fnstsw_ax() { 3196 emit_byte(0xdF); 3197 emit_byte(0xE0); 3198 } 3199 3200 void Assembler::fprem() { 3201 emit_byte(0xD9); 3202 emit_byte(0xF8); 3203 } 3204 3205 void Assembler::fprem1() { 3206 emit_byte(0xD9); 3207 emit_byte(0xF5); 3208 } 3209 3210 void Assembler::frstor(Address src) { 3211 InstructionMark im(this); 3212 emit_byte(0xDD); 3213 emit_operand32(rsp, src); 3214 } 3215 3216 void Assembler::fsin() { 3217 emit_byte(0xD9); 3218 emit_byte(0xFE); 3219 } 3220 3221 void Assembler::fsqrt() { 3222 emit_byte(0xD9); 3223 emit_byte(0xFA); 3224 } 3225 3226 void Assembler::fst_d(Address adr) { 3227 InstructionMark im(this); 3228 emit_byte(0xDD); 3229 emit_operand32(rdx, adr); 3230 } 3231 3232 void Assembler::fst_s(Address adr) { 3233 InstructionMark im(this); 3234 emit_byte(0xD9); 3235 emit_operand32(rdx, adr); 3236 } 3237 3238 void Assembler::fstp_d(Address adr) { 3239 InstructionMark im(this); 3240 emit_byte(0xDD); 3241 emit_operand32(rbx, adr); 3242 } 3243 3244 void Assembler::fstp_d(int index) { 3245 emit_farith(0xDD, 0xD8, index); 3246 } 3247 3248 void Assembler::fstp_s(Address adr) { 3249 InstructionMark im(this); 3250 emit_byte(0xD9); 3251 emit_operand32(rbx, adr); 3252 } 3253 3254 void Assembler::fstp_x(Address adr) { 3255 InstructionMark im(this); 3256 emit_byte(0xDB); 3257 emit_operand32(rdi, adr); 3258 } 3259 3260 void Assembler::fsub(int i) { 3261 emit_farith(0xD8, 0xE0, i); 3262 } 3263 3264 void Assembler::fsub_d(Address src) { 3265 InstructionMark im(this); 3266 emit_byte(0xDC); 3267 emit_operand32(rsp, src); 3268 } 3269 3270 void Assembler::fsub_s(Address src) { 3271 InstructionMark im(this); 3272 emit_byte(0xD8); 3273 emit_operand32(rsp, src); 3274 } 3275 3276 void Assembler::fsuba(int i) { 3277 emit_farith(0xDC, 0xE8, i); 3278 } 3279 3280 void Assembler::fsubp(int i) { 3281 emit_farith(0xDE, 0xE8, i); // ST(0) <- ST(0) - ST(1) and pop (Intel manual wrong) 3282 } 3283 3284 void Assembler::fsubr(int i) { 3285 emit_farith(0xD8, 0xE8, i); 3286 } 3287 3288 void Assembler::fsubr_d(Address src) { 3289 InstructionMark im(this); 3290 emit_byte(0xDC); 3291 emit_operand32(rbp, src); 3292 } 3293 3294 void Assembler::fsubr_s(Address src) { 3295 InstructionMark im(this); 3296 emit_byte(0xD8); 3297 emit_operand32(rbp, src); 3298 } 3299 3300 void Assembler::fsubra(int i) { 3301 emit_farith(0xDC, 0xE0, i); 3302 } 3303 3304 void Assembler::fsubrp(int i) { 3305 emit_farith(0xDE, 0xE0, i); // ST(0) <- ST(1) - ST(0) and pop (Intel manual wrong) 3306 } 3307 3308 void Assembler::ftan() { 3309 emit_byte(0xD9); 3310 emit_byte(0xF2); 3311 emit_byte(0xDD); 3312 emit_byte(0xD8); 3313 } 3314 3315 void Assembler::ftst() { 3316 emit_byte(0xD9); 3317 emit_byte(0xE4); 3318 } 3319 3320 void Assembler::fucomi(int i) { 3321 // make sure the instruction is supported (introduced for P6, together with cmov) 3322 guarantee(VM_Version::supports_cmov(), "illegal instruction"); 3323 emit_farith(0xDB, 0xE8, i); 3324 } 3325 3326 void Assembler::fucomip(int i) { 3327 // make sure the instruction is supported (introduced for P6, together with cmov) 3328 guarantee(VM_Version::supports_cmov(), "illegal instruction"); 3329 emit_farith(0xDF, 0xE8, i); 3330 } 3331 3332 void Assembler::fwait() { 3333 emit_byte(0x9B); 3334 } 3335 3336 void Assembler::fxch(int i) { 3337 emit_farith(0xD9, 0xC8, i); 3338 } 3339 3340 void Assembler::fyl2x() { 3341 emit_byte(0xD9); 3342 emit_byte(0xF1); 3343 } 3344 3345 3346 #ifndef _LP64 3347 3348 void Assembler::incl(Register dst) { 3349 // Don't use it directly. Use MacroAssembler::incrementl() instead. 3350 emit_byte(0x40 | dst->encoding()); 3351 } 3352 3353 void Assembler::lea(Register dst, Address src) { 3354 leal(dst, src); 3355 } 3356 3357 void Assembler::mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec) { 3358 InstructionMark im(this); 3359 emit_byte(0xC7); 3360 emit_operand(rax, dst); 3361 emit_data((int)imm32, rspec, 0); 3362 } 3363 3364 void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) { 3365 InstructionMark im(this); 3366 int encode = prefix_and_encode(dst->encoding()); 3367 emit_byte(0xB8 | encode); 3368 emit_data((int)imm32, rspec, 0); 3369 } 3370 3371 void Assembler::popa() { // 32bit 3372 emit_byte(0x61); 3373 } 3374 3375 void Assembler::push_literal32(int32_t imm32, RelocationHolder const& rspec) { 3376 InstructionMark im(this); 3377 emit_byte(0x68); 3378 emit_data(imm32, rspec, 0); 3379 } 3380 3381 void Assembler::pusha() { // 32bit 3382 emit_byte(0x60); 3383 } 3384 3385 void Assembler::set_byte_if_not_zero(Register dst) { 3386 emit_byte(0x0F); 3387 emit_byte(0x95); 3388 emit_byte(0xE0 | dst->encoding()); 3389 } 3390 3391 void Assembler::shldl(Register dst, Register src) { 3392 emit_byte(0x0F); 3393 emit_byte(0xA5); 3394 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); 3395 } 3396 3397 void Assembler::shrdl(Register dst, Register src) { 3398 emit_byte(0x0F); 3399 emit_byte(0xAD); 3400 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); 3401 } 3402 3403 #else // LP64 3404 3405 void Assembler::set_byte_if_not_zero(Register dst) { 3406 int enc = prefix_and_encode(dst->encoding(), true); 3407 emit_byte(0x0F); 3408 emit_byte(0x95); 3409 emit_byte(0xE0 | enc); 3410 } 3411 3412 // 64bit only pieces of the assembler 3413 // This should only be used by 64bit instructions that can use rip-relative 3414 // it cannot be used by instructions that want an immediate value. 3415 3416 bool Assembler::reachable(AddressLiteral adr) { 3417 int64_t disp; 3418 // None will force a 64bit literal to the code stream. Likely a placeholder 3419 // for something that will be patched later and we need to certain it will 3420 // always be reachable. 3421 if (adr.reloc() == relocInfo::none) { 3422 return false; 3423 } 3424 if (adr.reloc() == relocInfo::internal_word_type) { 3425 // This should be rip relative and easily reachable. 3426 return true; 3427 } 3428 if (adr.reloc() == relocInfo::virtual_call_type || 3429 adr.reloc() == relocInfo::opt_virtual_call_type || 3430 adr.reloc() == relocInfo::static_call_type || 3431 adr.reloc() == relocInfo::static_stub_type ) { 3432 // This should be rip relative within the code cache and easily 3433 // reachable until we get huge code caches. (At which point 3434 // ic code is going to have issues). 3435 return true; 3436 } 3437 if (adr.reloc() != relocInfo::external_word_type && 3438 adr.reloc() != relocInfo::poll_return_type && // these are really external_word but need special 3439 adr.reloc() != relocInfo::poll_type && // relocs to identify them 3440 adr.reloc() != relocInfo::runtime_call_type ) { 3441 return false; 3442 } 3443 3444 // Stress the correction code 3445 if (ForceUnreachable) { 3446 // Must be runtimecall reloc, see if it is in the codecache 3447 // Flipping stuff in the codecache to be unreachable causes issues 3448 // with things like inline caches where the additional instructions 3449 // are not handled. 3450 if (CodeCache::find_blob(adr._target) == NULL) { 3451 return false; 3452 } 3453 } 3454 // For external_word_type/runtime_call_type if it is reachable from where we 3455 // are now (possibly a temp buffer) and where we might end up 3456 // anywhere in the codeCache then we are always reachable. 3457 // This would have to change if we ever save/restore shared code 3458 // to be more pessimistic. 3459 3460 disp = (int64_t)adr._target - ((int64_t)CodeCache::low_bound() + sizeof(int)); 3461 if (!is_simm32(disp)) return false; 3462 disp = (int64_t)adr._target - ((int64_t)CodeCache::high_bound() + sizeof(int)); 3463 if (!is_simm32(disp)) return false; 3464 3465 disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int)); 3466 3467 // Because rip relative is a disp + address_of_next_instruction and we 3468 // don't know the value of address_of_next_instruction we apply a fudge factor 3469 // to make sure we will be ok no matter the size of the instruction we get placed into. 3470 // We don't have to fudge the checks above here because they are already worst case. 3471 3472 // 12 == override/rex byte, opcode byte, rm byte, sib byte, a 4-byte disp , 4-byte literal 3473 // + 4 because better safe than sorry. 3474 const int fudge = 12 + 4; 3475 if (disp < 0) { 3476 disp -= fudge; 3477 } else { 3478 disp += fudge; 3479 } 3480 return is_simm32(disp); 3481 } 3482 3483 void Assembler::emit_data64(jlong data, 3484 relocInfo::relocType rtype, 3485 int format) { 3486 if (rtype == relocInfo::none) { 3487 emit_long64(data); 3488 } else { 3489 emit_data64(data, Relocation::spec_simple(rtype), format); 3490 } 3491 } 3492 3493 void Assembler::emit_data64(jlong data, 3494 RelocationHolder const& rspec, 3495 int format) { 3496 assert(imm_operand == 0, "default format must be immediate in this file"); 3497 assert(imm_operand == format, "must be immediate"); 3498 assert(inst_mark() != NULL, "must be inside InstructionMark"); 3499 // Do not use AbstractAssembler::relocate, which is not intended for 3500 // embedded words. Instead, relocate to the enclosing instruction. 3501 code_section()->relocate(inst_mark(), rspec, format); 3502 #ifdef ASSERT 3503 check_relocation(rspec, format); 3504 #endif 3505 emit_long64(data); 3506 } 3507 3508 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) { 3509 if (reg_enc >= 8) { 3510 prefix(REX_B); 3511 reg_enc -= 8; 3512 } else if (byteinst && reg_enc >= 4) { 3513 prefix(REX); 3514 } 3515 return reg_enc; 3516 } 3517 3518 int Assembler::prefixq_and_encode(int reg_enc) { 3519 if (reg_enc < 8) { 3520 prefix(REX_W); 3521 } else { 3522 prefix(REX_WB); 3523 reg_enc -= 8; 3524 } 3525 return reg_enc; 3526 } 3527 3528 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) { 3529 if (dst_enc < 8) { 3530 if (src_enc >= 8) { 3531 prefix(REX_B); 3532 src_enc -= 8; 3533 } else if (byteinst && src_enc >= 4) { 3534 prefix(REX); 3535 } 3536 } else { 3537 if (src_enc < 8) { 3538 prefix(REX_R); 3539 } else { 3540 prefix(REX_RB); 3541 src_enc -= 8; 3542 } 3543 dst_enc -= 8; 3544 } 3545 return dst_enc << 3 | src_enc; 3546 } 3547 3548 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) { 3549 if (dst_enc < 8) { 3550 if (src_enc < 8) { 3551 prefix(REX_W); 3552 } else { 3553 prefix(REX_WB); 3554 src_enc -= 8; 3555 } 3556 } else { 3557 if (src_enc < 8) { 3558 prefix(REX_WR); 3559 } else { 3560 prefix(REX_WRB); 3561 src_enc -= 8; 3562 } 3563 dst_enc -= 8; 3564 } 3565 return dst_enc << 3 | src_enc; 3566 } 3567 3568 void Assembler::prefix(Register reg) { 3569 if (reg->encoding() >= 8) { 3570 prefix(REX_B); 3571 } 3572 } 3573 3574 void Assembler::prefix(Address adr) { 3575 if (adr.base_needs_rex()) { 3576 if (adr.index_needs_rex()) { 3577 prefix(REX_XB); 3578 } else { 3579 prefix(REX_B); 3580 } 3581 } else { 3582 if (adr.index_needs_rex()) { 3583 prefix(REX_X); 3584 } 3585 } 3586 } 3587 3588 void Assembler::prefixq(Address adr) { 3589 if (adr.base_needs_rex()) { 3590 if (adr.index_needs_rex()) { 3591 prefix(REX_WXB); 3592 } else { 3593 prefix(REX_WB); 3594 } 3595 } else { 3596 if (adr.index_needs_rex()) { 3597 prefix(REX_WX); 3598 } else { 3599 prefix(REX_W); 3600 } 3601 } 3602 } 3603 3604 3605 void Assembler::prefix(Address adr, Register reg, bool byteinst) { 3606 if (reg->encoding() < 8) { 3607 if (adr.base_needs_rex()) { 3608 if (adr.index_needs_rex()) { 3609 prefix(REX_XB); 3610 } else { 3611 prefix(REX_B); 3612 } 3613 } else { 3614 if (adr.index_needs_rex()) { 3615 prefix(REX_X); 3616 } else if (reg->encoding() >= 4 ) { 3617 prefix(REX); 3618 } 3619 } 3620 } else { 3621 if (adr.base_needs_rex()) { 3622 if (adr.index_needs_rex()) { 3623 prefix(REX_RXB); 3624 } else { 3625 prefix(REX_RB); 3626 } 3627 } else { 3628 if (adr.index_needs_rex()) { 3629 prefix(REX_RX); 3630 } else { 3631 prefix(REX_R); 3632 } 3633 } 3634 } 3635 } 3636 3637 void Assembler::prefixq(Address adr, Register src) { 3638 if (src->encoding() < 8) { 3639 if (adr.base_needs_rex()) { 3640 if (adr.index_needs_rex()) { 3641 prefix(REX_WXB); 3642 } else { 3643 prefix(REX_WB); 3644 } 3645 } else { 3646 if (adr.index_needs_rex()) { 3647 prefix(REX_WX); 3648 } else { 3649 prefix(REX_W); 3650 } 3651 } 3652 } else { 3653 if (adr.base_needs_rex()) { 3654 if (adr.index_needs_rex()) { 3655 prefix(REX_WRXB); 3656 } else { 3657 prefix(REX_WRB); 3658 } 3659 } else { 3660 if (adr.index_needs_rex()) { 3661 prefix(REX_WRX); 3662 } else { 3663 prefix(REX_WR); 3664 } 3665 } 3666 } 3667 } 3668 3669 void Assembler::prefix(Address adr, XMMRegister reg) { 3670 if (reg->encoding() < 8) { 3671 if (adr.base_needs_rex()) { 3672 if (adr.index_needs_rex()) { 3673 prefix(REX_XB); 3674 } else { 3675 prefix(REX_B); 3676 } 3677 } else { 3678 if (adr.index_needs_rex()) { 3679 prefix(REX_X); 3680 } 3681 } 3682 } else { 3683 if (adr.base_needs_rex()) { 3684 if (adr.index_needs_rex()) { 3685 prefix(REX_RXB); 3686 } else { 3687 prefix(REX_RB); 3688 } 3689 } else { 3690 if (adr.index_needs_rex()) { 3691 prefix(REX_RX); 3692 } else { 3693 prefix(REX_R); 3694 } 3695 } 3696 } 3697 } 3698 3699 void Assembler::adcq(Register dst, int32_t imm32) { 3700 (void) prefixq_and_encode(dst->encoding()); 3701 emit_arith(0x81, 0xD0, dst, imm32); 3702 } 3703 3704 void Assembler::adcq(Register dst, Address src) { 3705 InstructionMark im(this); 3706 prefixq(src, dst); 3707 emit_byte(0x13); 3708 emit_operand(dst, src); 3709 } 3710 3711 void Assembler::adcq(Register dst, Register src) { 3712 (int) prefixq_and_encode(dst->encoding(), src->encoding()); 3713 emit_arith(0x13, 0xC0, dst, src); 3714 } 3715 3716 void Assembler::addq(Address dst, int32_t imm32) { 3717 InstructionMark im(this); 3718 prefixq(dst); 3719 emit_arith_operand(0x81, rax, dst,imm32); 3720 } 3721 3722 void Assembler::addq(Address dst, Register src) { 3723 InstructionMark im(this); 3724 prefixq(dst, src); 3725 emit_byte(0x01); 3726 emit_operand(src, dst); 3727 } 3728 3729 void Assembler::addq(Register dst, int32_t imm32) { 3730 (void) prefixq_and_encode(dst->encoding()); 3731 emit_arith(0x81, 0xC0, dst, imm32); 3732 } 3733 3734 void Assembler::addq(Register dst, Address src) { 3735 InstructionMark im(this); 3736 prefixq(src, dst); 3737 emit_byte(0x03); 3738 emit_operand(dst, src); 3739 } 3740 3741 void Assembler::addq(Register dst, Register src) { 3742 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 3743 emit_arith(0x03, 0xC0, dst, src); 3744 } 3745 3746 void Assembler::andq(Register dst, int32_t imm32) { 3747 (void) prefixq_and_encode(dst->encoding()); 3748 emit_arith(0x81, 0xE0, dst, imm32); 3749 } 3750 3751 void Assembler::andq(Register dst, Address src) { 3752 InstructionMark im(this); 3753 prefixq(src, dst); 3754 emit_byte(0x23); 3755 emit_operand(dst, src); 3756 } 3757 3758 void Assembler::andq(Register dst, Register src) { 3759 (int) prefixq_and_encode(dst->encoding(), src->encoding()); 3760 emit_arith(0x23, 0xC0, dst, src); 3761 } 3762 3763 void Assembler::bsfq(Register dst, Register src) { 3764 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3765 emit_byte(0x0F); 3766 emit_byte(0xBC); 3767 emit_byte(0xC0 | encode); 3768 } 3769 3770 void Assembler::bsrq(Register dst, Register src) { 3771 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT"); 3772 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3773 emit_byte(0x0F); 3774 emit_byte(0xBD); 3775 emit_byte(0xC0 | encode); 3776 } 3777 3778 void Assembler::bswapq(Register reg) { 3779 int encode = prefixq_and_encode(reg->encoding()); 3780 emit_byte(0x0F); 3781 emit_byte(0xC8 | encode); 3782 } 3783 3784 void Assembler::cdqq() { 3785 prefix(REX_W); 3786 emit_byte(0x99); 3787 } 3788 3789 void Assembler::clflush(Address adr) { 3790 prefix(adr); 3791 emit_byte(0x0F); 3792 emit_byte(0xAE); 3793 emit_operand(rdi, adr); 3794 } 3795 3796 void Assembler::cmovq(Condition cc, Register dst, Register src) { 3797 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3798 emit_byte(0x0F); 3799 emit_byte(0x40 | cc); 3800 emit_byte(0xC0 | encode); 3801 } 3802 3803 void Assembler::cmovq(Condition cc, Register dst, Address src) { 3804 InstructionMark im(this); 3805 prefixq(src, dst); 3806 emit_byte(0x0F); 3807 emit_byte(0x40 | cc); 3808 emit_operand(dst, src); 3809 } 3810 3811 void Assembler::cmpq(Address dst, int32_t imm32) { 3812 InstructionMark im(this); 3813 prefixq(dst); 3814 emit_byte(0x81); 3815 emit_operand(rdi, dst, 4); 3816 emit_long(imm32); 3817 } 3818 3819 void Assembler::cmpq(Register dst, int32_t imm32) { 3820 (void) prefixq_and_encode(dst->encoding()); 3821 emit_arith(0x81, 0xF8, dst, imm32); 3822 } 3823 3824 void Assembler::cmpq(Address dst, Register src) { 3825 InstructionMark im(this); 3826 prefixq(dst, src); 3827 emit_byte(0x3B); 3828 emit_operand(src, dst); 3829 } 3830 3831 void Assembler::cmpq(Register dst, Register src) { 3832 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 3833 emit_arith(0x3B, 0xC0, dst, src); 3834 } 3835 3836 void Assembler::cmpq(Register dst, Address src) { 3837 InstructionMark im(this); 3838 prefixq(src, dst); 3839 emit_byte(0x3B); 3840 emit_operand(dst, src); 3841 } 3842 3843 void Assembler::cmpxchgq(Register reg, Address adr) { 3844 InstructionMark im(this); 3845 prefixq(adr, reg); 3846 emit_byte(0x0F); 3847 emit_byte(0xB1); 3848 emit_operand(reg, adr); 3849 } 3850 3851 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) { 3852 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 3853 emit_byte(0xF2); 3854 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3855 emit_byte(0x0F); 3856 emit_byte(0x2A); 3857 emit_byte(0xC0 | encode); 3858 } 3859 3860 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) { 3861 NOT_LP64(assert(VM_Version::supports_sse(), "")); 3862 emit_byte(0xF3); 3863 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3864 emit_byte(0x0F); 3865 emit_byte(0x2A); 3866 emit_byte(0xC0 | encode); 3867 } 3868 3869 void Assembler::cvttsd2siq(Register dst, XMMRegister src) { 3870 NOT_LP64(assert(VM_Version::supports_sse2(), "")); 3871 emit_byte(0xF2); 3872 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3873 emit_byte(0x0F); 3874 emit_byte(0x2C); 3875 emit_byte(0xC0 | encode); 3876 } 3877 3878 void Assembler::cvttss2siq(Register dst, XMMRegister src) { 3879 NOT_LP64(assert(VM_Version::supports_sse(), "")); 3880 emit_byte(0xF3); 3881 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3882 emit_byte(0x0F); 3883 emit_byte(0x2C); 3884 emit_byte(0xC0 | encode); 3885 } 3886 3887 void Assembler::decl(Register dst) { 3888 // Don't use it directly. Use MacroAssembler::decrementl() instead. 3889 // Use two-byte form (one-byte form is a REX prefix in 64-bit mode) 3890 int encode = prefix_and_encode(dst->encoding()); 3891 emit_byte(0xFF); 3892 emit_byte(0xC8 | encode); 3893 } 3894 3895 void Assembler::decq(Register dst) { 3896 // Don't use it directly. Use MacroAssembler::decrementq() instead. 3897 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) 3898 int encode = prefixq_and_encode(dst->encoding()); 3899 emit_byte(0xFF); 3900 emit_byte(0xC8 | encode); 3901 } 3902 3903 void Assembler::decq(Address dst) { 3904 // Don't use it directly. Use MacroAssembler::decrementq() instead. 3905 InstructionMark im(this); 3906 prefixq(dst); 3907 emit_byte(0xFF); 3908 emit_operand(rcx, dst); 3909 } 3910 3911 void Assembler::fxrstor(Address src) { 3912 prefixq(src); 3913 emit_byte(0x0F); 3914 emit_byte(0xAE); 3915 emit_operand(as_Register(1), src); 3916 } 3917 3918 void Assembler::fxsave(Address dst) { 3919 prefixq(dst); 3920 emit_byte(0x0F); 3921 emit_byte(0xAE); 3922 emit_operand(as_Register(0), dst); 3923 } 3924 3925 void Assembler::idivq(Register src) { 3926 int encode = prefixq_and_encode(src->encoding()); 3927 emit_byte(0xF7); 3928 emit_byte(0xF8 | encode); 3929 } 3930 3931 void Assembler::imulq(Register dst, Register src) { 3932 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3933 emit_byte(0x0F); 3934 emit_byte(0xAF); 3935 emit_byte(0xC0 | encode); 3936 } 3937 3938 void Assembler::imulq(Register dst, Register src, int value) { 3939 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 3940 if (is8bit(value)) { 3941 emit_byte(0x6B); 3942 emit_byte(0xC0 | encode); 3943 emit_byte(value & 0xFF); 3944 } else { 3945 emit_byte(0x69); 3946 emit_byte(0xC0 | encode); 3947 emit_long(value); 3948 } 3949 } 3950 3951 void Assembler::incl(Register dst) { 3952 // Don't use it directly. Use MacroAssembler::incrementl() instead. 3953 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) 3954 int encode = prefix_and_encode(dst->encoding()); 3955 emit_byte(0xFF); 3956 emit_byte(0xC0 | encode); 3957 } 3958 3959 void Assembler::incq(Register dst) { 3960 // Don't use it directly. Use MacroAssembler::incrementq() instead. 3961 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) 3962 int encode = prefixq_and_encode(dst->encoding()); 3963 emit_byte(0xFF); 3964 emit_byte(0xC0 | encode); 3965 } 3966 3967 void Assembler::incq(Address dst) { 3968 // Don't use it directly. Use MacroAssembler::incrementq() instead. 3969 InstructionMark im(this); 3970 prefixq(dst); 3971 emit_byte(0xFF); 3972 emit_operand(rax, dst); 3973 } 3974 3975 void Assembler::lea(Register dst, Address src) { 3976 leaq(dst, src); 3977 } 3978 3979 void Assembler::leaq(Register dst, Address src) { 3980 InstructionMark im(this); 3981 prefixq(src, dst); 3982 emit_byte(0x8D); 3983 emit_operand(dst, src); 3984 } 3985 3986 void Assembler::mov64(Register dst, int64_t imm64) { 3987 InstructionMark im(this); 3988 int encode = prefixq_and_encode(dst->encoding()); 3989 emit_byte(0xB8 | encode); 3990 emit_long64(imm64); 3991 } 3992 3993 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) { 3994 InstructionMark im(this); 3995 int encode = prefixq_and_encode(dst->encoding()); 3996 emit_byte(0xB8 | encode); 3997 emit_data64(imm64, rspec); 3998 } 3999 4000 void Assembler::mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec) { 4001 InstructionMark im(this); 4002 int encode = prefix_and_encode(dst->encoding()); 4003 emit_byte(0xB8 | encode); 4004 emit_data((int)imm32, rspec, narrow_oop_operand); 4005 } 4006 4007 void Assembler::mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec) { 4008 InstructionMark im(this); 4009 prefix(dst); 4010 emit_byte(0xC7); 4011 emit_operand(rax, dst, 4); 4012 emit_data((int)imm32, rspec, narrow_oop_operand); 4013 } 4014 4015 void Assembler::cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec) { 4016 InstructionMark im(this); 4017 int encode = prefix_and_encode(src1->encoding()); 4018 emit_byte(0x81); 4019 emit_byte(0xF8 | encode); 4020 emit_data((int)imm32, rspec, narrow_oop_operand); 4021 } 4022 4023 void Assembler::cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec) { 4024 InstructionMark im(this); 4025 prefix(src1); 4026 emit_byte(0x81); 4027 emit_operand(rax, src1, 4); 4028 emit_data((int)imm32, rspec, narrow_oop_operand); 4029 } 4030 4031 void Assembler::lzcntq(Register dst, Register src) { 4032 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR"); 4033 emit_byte(0xF3); 4034 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4035 emit_byte(0x0F); 4036 emit_byte(0xBD); 4037 emit_byte(0xC0 | encode); 4038 } 4039 4040 void Assembler::movdq(XMMRegister dst, Register src) { 4041 // table D-1 says MMX/SSE2 4042 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), "")); 4043 emit_byte(0x66); 4044 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4045 emit_byte(0x0F); 4046 emit_byte(0x6E); 4047 emit_byte(0xC0 | encode); 4048 } 4049 4050 void Assembler::movdq(Register dst, XMMRegister src) { 4051 // table D-1 says MMX/SSE2 4052 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), "")); 4053 emit_byte(0x66); 4054 // swap src/dst to get correct prefix 4055 int encode = prefixq_and_encode(src->encoding(), dst->encoding()); 4056 emit_byte(0x0F); 4057 emit_byte(0x7E); 4058 emit_byte(0xC0 | encode); 4059 } 4060 4061 void Assembler::movq(Register dst, Register src) { 4062 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4063 emit_byte(0x8B); 4064 emit_byte(0xC0 | encode); 4065 } 4066 4067 void Assembler::movq(Register dst, Address src) { 4068 InstructionMark im(this); 4069 prefixq(src, dst); 4070 emit_byte(0x8B); 4071 emit_operand(dst, src); 4072 } 4073 4074 void Assembler::movq(Address dst, Register src) { 4075 InstructionMark im(this); 4076 prefixq(dst, src); 4077 emit_byte(0x89); 4078 emit_operand(src, dst); 4079 } 4080 4081 void Assembler::movsbq(Register dst, Address src) { 4082 InstructionMark im(this); 4083 prefixq(src, dst); 4084 emit_byte(0x0F); 4085 emit_byte(0xBE); 4086 emit_operand(dst, src); 4087 } 4088 4089 void Assembler::movsbq(Register dst, Register src) { 4090 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4091 emit_byte(0x0F); 4092 emit_byte(0xBE); 4093 emit_byte(0xC0 | encode); 4094 } 4095 4096 void Assembler::movslq(Register dst, int32_t imm32) { 4097 // dbx shows movslq(rcx, 3) as movq $0x0000000049000000,(%rbx) 4098 // and movslq(r8, 3); as movl $0x0000000048000000,(%rbx) 4099 // as a result we shouldn't use until tested at runtime... 4100 ShouldNotReachHere(); 4101 InstructionMark im(this); 4102 int encode = prefixq_and_encode(dst->encoding()); 4103 emit_byte(0xC7 | encode); 4104 emit_long(imm32); 4105 } 4106 4107 void Assembler::movslq(Address dst, int32_t imm32) { 4108 assert(is_simm32(imm32), "lost bits"); 4109 InstructionMark im(this); 4110 prefixq(dst); 4111 emit_byte(0xC7); 4112 emit_operand(rax, dst, 4); 4113 emit_long(imm32); 4114 } 4115 4116 void Assembler::movslq(Register dst, Address src) { 4117 InstructionMark im(this); 4118 prefixq(src, dst); 4119 emit_byte(0x63); 4120 emit_operand(dst, src); 4121 } 4122 4123 void Assembler::movslq(Register dst, Register src) { 4124 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4125 emit_byte(0x63); 4126 emit_byte(0xC0 | encode); 4127 } 4128 4129 void Assembler::movswq(Register dst, Address src) { 4130 InstructionMark im(this); 4131 prefixq(src, dst); 4132 emit_byte(0x0F); 4133 emit_byte(0xBF); 4134 emit_operand(dst, src); 4135 } 4136 4137 void Assembler::movswq(Register dst, Register src) { 4138 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4139 emit_byte(0x0F); 4140 emit_byte(0xBF); 4141 emit_byte(0xC0 | encode); 4142 } 4143 4144 void Assembler::movzbq(Register dst, Address src) { 4145 InstructionMark im(this); 4146 prefixq(src, dst); 4147 emit_byte(0x0F); 4148 emit_byte(0xB6); 4149 emit_operand(dst, src); 4150 } 4151 4152 void Assembler::movzbq(Register dst, Register src) { 4153 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4154 emit_byte(0x0F); 4155 emit_byte(0xB6); 4156 emit_byte(0xC0 | encode); 4157 } 4158 4159 void Assembler::movzwq(Register dst, Address src) { 4160 InstructionMark im(this); 4161 prefixq(src, dst); 4162 emit_byte(0x0F); 4163 emit_byte(0xB7); 4164 emit_operand(dst, src); 4165 } 4166 4167 void Assembler::movzwq(Register dst, Register src) { 4168 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4169 emit_byte(0x0F); 4170 emit_byte(0xB7); 4171 emit_byte(0xC0 | encode); 4172 } 4173 4174 void Assembler::negq(Register dst) { 4175 int encode = prefixq_and_encode(dst->encoding()); 4176 emit_byte(0xF7); 4177 emit_byte(0xD8 | encode); 4178 } 4179 4180 void Assembler::notq(Register dst) { 4181 int encode = prefixq_and_encode(dst->encoding()); 4182 emit_byte(0xF7); 4183 emit_byte(0xD0 | encode); 4184 } 4185 4186 void Assembler::orq(Address dst, int32_t imm32) { 4187 InstructionMark im(this); 4188 prefixq(dst); 4189 emit_byte(0x81); 4190 emit_operand(rcx, dst, 4); 4191 emit_long(imm32); 4192 } 4193 4194 void Assembler::orq(Register dst, int32_t imm32) { 4195 (void) prefixq_and_encode(dst->encoding()); 4196 emit_arith(0x81, 0xC8, dst, imm32); 4197 } 4198 4199 void Assembler::orq(Register dst, Address src) { 4200 InstructionMark im(this); 4201 prefixq(src, dst); 4202 emit_byte(0x0B); 4203 emit_operand(dst, src); 4204 } 4205 4206 void Assembler::orq(Register dst, Register src) { 4207 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 4208 emit_arith(0x0B, 0xC0, dst, src); 4209 } 4210 4211 void Assembler::popa() { // 64bit 4212 movq(r15, Address(rsp, 0)); 4213 movq(r14, Address(rsp, wordSize)); 4214 movq(r13, Address(rsp, 2 * wordSize)); 4215 movq(r12, Address(rsp, 3 * wordSize)); 4216 movq(r11, Address(rsp, 4 * wordSize)); 4217 movq(r10, Address(rsp, 5 * wordSize)); 4218 movq(r9, Address(rsp, 6 * wordSize)); 4219 movq(r8, Address(rsp, 7 * wordSize)); 4220 movq(rdi, Address(rsp, 8 * wordSize)); 4221 movq(rsi, Address(rsp, 9 * wordSize)); 4222 movq(rbp, Address(rsp, 10 * wordSize)); 4223 // skip rsp 4224 movq(rbx, Address(rsp, 12 * wordSize)); 4225 movq(rdx, Address(rsp, 13 * wordSize)); 4226 movq(rcx, Address(rsp, 14 * wordSize)); 4227 movq(rax, Address(rsp, 15 * wordSize)); 4228 4229 addq(rsp, 16 * wordSize); 4230 } 4231 4232 void Assembler::popcntq(Register dst, Address src) { 4233 assert(VM_Version::supports_popcnt(), "must support"); 4234 InstructionMark im(this); 4235 emit_byte(0xF3); 4236 prefixq(src, dst); 4237 emit_byte(0x0F); 4238 emit_byte(0xB8); 4239 emit_operand(dst, src); 4240 } 4241 4242 void Assembler::popcntq(Register dst, Register src) { 4243 assert(VM_Version::supports_popcnt(), "must support"); 4244 emit_byte(0xF3); 4245 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4246 emit_byte(0x0F); 4247 emit_byte(0xB8); 4248 emit_byte(0xC0 | encode); 4249 } 4250 4251 void Assembler::popq(Address dst) { 4252 InstructionMark im(this); 4253 prefixq(dst); 4254 emit_byte(0x8F); 4255 emit_operand(rax, dst); 4256 } 4257 4258 void Assembler::pusha() { // 64bit 4259 // we have to store original rsp. ABI says that 128 bytes 4260 // below rsp are local scratch. 4261 movq(Address(rsp, -5 * wordSize), rsp); 4262 4263 subq(rsp, 16 * wordSize); 4264 4265 movq(Address(rsp, 15 * wordSize), rax); 4266 movq(Address(rsp, 14 * wordSize), rcx); 4267 movq(Address(rsp, 13 * wordSize), rdx); 4268 movq(Address(rsp, 12 * wordSize), rbx); 4269 // skip rsp 4270 movq(Address(rsp, 10 * wordSize), rbp); 4271 movq(Address(rsp, 9 * wordSize), rsi); 4272 movq(Address(rsp, 8 * wordSize), rdi); 4273 movq(Address(rsp, 7 * wordSize), r8); 4274 movq(Address(rsp, 6 * wordSize), r9); 4275 movq(Address(rsp, 5 * wordSize), r10); 4276 movq(Address(rsp, 4 * wordSize), r11); 4277 movq(Address(rsp, 3 * wordSize), r12); 4278 movq(Address(rsp, 2 * wordSize), r13); 4279 movq(Address(rsp, wordSize), r14); 4280 movq(Address(rsp, 0), r15); 4281 } 4282 4283 void Assembler::pushq(Address src) { 4284 InstructionMark im(this); 4285 prefixq(src); 4286 emit_byte(0xFF); 4287 emit_operand(rsi, src); 4288 } 4289 4290 void Assembler::rclq(Register dst, int imm8) { 4291 assert(isShiftCount(imm8 >> 1), "illegal shift count"); 4292 int encode = prefixq_and_encode(dst->encoding()); 4293 if (imm8 == 1) { 4294 emit_byte(0xD1); 4295 emit_byte(0xD0 | encode); 4296 } else { 4297 emit_byte(0xC1); 4298 emit_byte(0xD0 | encode); 4299 emit_byte(imm8); 4300 } 4301 } 4302 void Assembler::sarq(Register dst, int imm8) { 4303 assert(isShiftCount(imm8 >> 1), "illegal shift count"); 4304 int encode = prefixq_and_encode(dst->encoding()); 4305 if (imm8 == 1) { 4306 emit_byte(0xD1); 4307 emit_byte(0xF8 | encode); 4308 } else { 4309 emit_byte(0xC1); 4310 emit_byte(0xF8 | encode); 4311 emit_byte(imm8); 4312 } 4313 } 4314 4315 void Assembler::sarq(Register dst) { 4316 int encode = prefixq_and_encode(dst->encoding()); 4317 emit_byte(0xD3); 4318 emit_byte(0xF8 | encode); 4319 } 4320 void Assembler::sbbq(Address dst, int32_t imm32) { 4321 InstructionMark im(this); 4322 prefixq(dst); 4323 emit_arith_operand(0x81, rbx, dst, imm32); 4324 } 4325 4326 void Assembler::sbbq(Register dst, int32_t imm32) { 4327 (void) prefixq_and_encode(dst->encoding()); 4328 emit_arith(0x81, 0xD8, dst, imm32); 4329 } 4330 4331 void Assembler::sbbq(Register dst, Address src) { 4332 InstructionMark im(this); 4333 prefixq(src, dst); 4334 emit_byte(0x1B); 4335 emit_operand(dst, src); 4336 } 4337 4338 void Assembler::sbbq(Register dst, Register src) { 4339 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 4340 emit_arith(0x1B, 0xC0, dst, src); 4341 } 4342 4343 void Assembler::shlq(Register dst, int imm8) { 4344 assert(isShiftCount(imm8 >> 1), "illegal shift count"); 4345 int encode = prefixq_and_encode(dst->encoding()); 4346 if (imm8 == 1) { 4347 emit_byte(0xD1); 4348 emit_byte(0xE0 | encode); 4349 } else { 4350 emit_byte(0xC1); 4351 emit_byte(0xE0 | encode); 4352 emit_byte(imm8); 4353 } 4354 } 4355 4356 void Assembler::shlq(Register dst) { 4357 int encode = prefixq_and_encode(dst->encoding()); 4358 emit_byte(0xD3); 4359 emit_byte(0xE0 | encode); 4360 } 4361 4362 void Assembler::shrq(Register dst, int imm8) { 4363 assert(isShiftCount(imm8 >> 1), "illegal shift count"); 4364 int encode = prefixq_and_encode(dst->encoding()); 4365 emit_byte(0xC1); 4366 emit_byte(0xE8 | encode); 4367 emit_byte(imm8); 4368 } 4369 4370 void Assembler::shrq(Register dst) { 4371 int encode = prefixq_and_encode(dst->encoding()); 4372 emit_byte(0xD3); 4373 emit_byte(0xE8 | encode); 4374 } 4375 4376 void Assembler::subq(Address dst, int32_t imm32) { 4377 InstructionMark im(this); 4378 prefixq(dst); 4379 if (is8bit(imm32)) { 4380 emit_byte(0x83); 4381 emit_operand(rbp, dst, 1); 4382 emit_byte(imm32 & 0xFF); 4383 } else { 4384 emit_byte(0x81); 4385 emit_operand(rbp, dst, 4); 4386 emit_long(imm32); 4387 } 4388 } 4389 4390 void Assembler::subq(Register dst, int32_t imm32) { 4391 (void) prefixq_and_encode(dst->encoding()); 4392 emit_arith(0x81, 0xE8, dst, imm32); 4393 } 4394 4395 void Assembler::subq(Address dst, Register src) { 4396 InstructionMark im(this); 4397 prefixq(dst, src); 4398 emit_byte(0x29); 4399 emit_operand(src, dst); 4400 } 4401 4402 void Assembler::subq(Register dst, Address src) { 4403 InstructionMark im(this); 4404 prefixq(src, dst); 4405 emit_byte(0x2B); 4406 emit_operand(dst, src); 4407 } 4408 4409 void Assembler::subq(Register dst, Register src) { 4410 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 4411 emit_arith(0x2B, 0xC0, dst, src); 4412 } 4413 4414 void Assembler::testq(Register dst, int32_t imm32) { 4415 // not using emit_arith because test 4416 // doesn't support sign-extension of 4417 // 8bit operands 4418 int encode = dst->encoding(); 4419 if (encode == 0) { 4420 prefix(REX_W); 4421 emit_byte(0xA9); 4422 } else { 4423 encode = prefixq_and_encode(encode); 4424 emit_byte(0xF7); 4425 emit_byte(0xC0 | encode); 4426 } 4427 emit_long(imm32); 4428 } 4429 4430 void Assembler::testq(Register dst, Register src) { 4431 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 4432 emit_arith(0x85, 0xC0, dst, src); 4433 } 4434 4435 void Assembler::xaddq(Address dst, Register src) { 4436 InstructionMark im(this); 4437 prefixq(dst, src); 4438 emit_byte(0x0F); 4439 emit_byte(0xC1); 4440 emit_operand(src, dst); 4441 } 4442 4443 void Assembler::xchgq(Register dst, Address src) { 4444 InstructionMark im(this); 4445 prefixq(src, dst); 4446 emit_byte(0x87); 4447 emit_operand(dst, src); 4448 } 4449 4450 void Assembler::xchgq(Register dst, Register src) { 4451 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 4452 emit_byte(0x87); 4453 emit_byte(0xc0 | encode); 4454 } 4455 4456 void Assembler::xorq(Register dst, Register src) { 4457 (void) prefixq_and_encode(dst->encoding(), src->encoding()); 4458 emit_arith(0x33, 0xC0, dst, src); 4459 } 4460 4461 void Assembler::xorq(Register dst, Address src) { 4462 InstructionMark im(this); 4463 prefixq(src, dst); 4464 emit_byte(0x33); 4465 emit_operand(dst, src); 4466 } 4467 4468 #endif // !LP64 4469 4470 static Assembler::Condition reverse[] = { 4471 Assembler::noOverflow /* overflow = 0x0 */ , 4472 Assembler::overflow /* noOverflow = 0x1 */ , 4473 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ , 4474 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ , 4475 Assembler::notZero /* zero = 0x4, equal = 0x4 */ , 4476 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ , 4477 Assembler::above /* belowEqual = 0x6 */ , 4478 Assembler::belowEqual /* above = 0x7 */ , 4479 Assembler::positive /* negative = 0x8 */ , 4480 Assembler::negative /* positive = 0x9 */ , 4481 Assembler::noParity /* parity = 0xa */ , 4482 Assembler::parity /* noParity = 0xb */ , 4483 Assembler::greaterEqual /* less = 0xc */ , 4484 Assembler::less /* greaterEqual = 0xd */ , 4485 Assembler::greater /* lessEqual = 0xe */ , 4486 Assembler::lessEqual /* greater = 0xf, */ 4487 4488 }; 4489 4490 4491 // Implementation of MacroAssembler 4492 4493 // First all the versions that have distinct versions depending on 32/64 bit 4494 // Unless the difference is trivial (1 line or so). 4495 4496 #ifndef _LP64 4497 4498 // 32bit versions 4499 4500 Address MacroAssembler::as_Address(AddressLiteral adr) { 4501 return Address(adr.target(), adr.rspec()); 4502 } 4503 4504 Address MacroAssembler::as_Address(ArrayAddress adr) { 4505 return Address::make_array(adr); 4506 } 4507 4508 int MacroAssembler::biased_locking_enter(Register lock_reg, 4509 Register obj_reg, 4510 Register swap_reg, 4511 Register tmp_reg, 4512 bool swap_reg_contains_mark, 4513 Label& done, 4514 Label* slow_case, 4515 BiasedLockingCounters* counters) { 4516 assert(UseBiasedLocking, "why call this otherwise?"); 4517 assert(swap_reg == rax, "swap_reg must be rax, for cmpxchg"); 4518 assert_different_registers(lock_reg, obj_reg, swap_reg); 4519 4520 if (PrintBiasedLockingStatistics && counters == NULL) 4521 counters = BiasedLocking::counters(); 4522 4523 bool need_tmp_reg = false; 4524 if (tmp_reg == noreg) { 4525 need_tmp_reg = true; 4526 tmp_reg = lock_reg; 4527 } else { 4528 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); 4529 } 4530 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); 4531 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); 4532 Address klass_addr (obj_reg, oopDesc::klass_offset_in_bytes()); 4533 Address saved_mark_addr(lock_reg, 0); 4534 4535 // Biased locking 4536 // See whether the lock is currently biased toward our thread and 4537 // whether the epoch is still valid 4538 // Note that the runtime guarantees sufficient alignment of JavaThread 4539 // pointers to allow age to be placed into low bits 4540 // First check to see whether biasing is even enabled for this object 4541 Label cas_label; 4542 int null_check_offset = -1; 4543 if (!swap_reg_contains_mark) { 4544 null_check_offset = offset(); 4545 movl(swap_reg, mark_addr); 4546 } 4547 if (need_tmp_reg) { 4548 push(tmp_reg); 4549 } 4550 movl(tmp_reg, swap_reg); 4551 andl(tmp_reg, markOopDesc::biased_lock_mask_in_place); 4552 cmpl(tmp_reg, markOopDesc::biased_lock_pattern); 4553 if (need_tmp_reg) { 4554 pop(tmp_reg); 4555 } 4556 jcc(Assembler::notEqual, cas_label); 4557 // The bias pattern is present in the object's header. Need to check 4558 // whether the bias owner and the epoch are both still current. 4559 // Note that because there is no current thread register on x86 we 4560 // need to store off the mark word we read out of the object to 4561 // avoid reloading it and needing to recheck invariants below. This 4562 // store is unfortunate but it makes the overall code shorter and 4563 // simpler. 4564 movl(saved_mark_addr, swap_reg); 4565 if (need_tmp_reg) { 4566 push(tmp_reg); 4567 } 4568 get_thread(tmp_reg); 4569 xorl(swap_reg, tmp_reg); 4570 if (swap_reg_contains_mark) { 4571 null_check_offset = offset(); 4572 } 4573 movl(tmp_reg, klass_addr); 4574 xorl(swap_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 4575 andl(swap_reg, ~((int) markOopDesc::age_mask_in_place)); 4576 if (need_tmp_reg) { 4577 pop(tmp_reg); 4578 } 4579 if (counters != NULL) { 4580 cond_inc32(Assembler::zero, 4581 ExternalAddress((address)counters->biased_lock_entry_count_addr())); 4582 } 4583 jcc(Assembler::equal, done); 4584 4585 Label try_revoke_bias; 4586 Label try_rebias; 4587 4588 // At this point we know that the header has the bias pattern and 4589 // that we are not the bias owner in the current epoch. We need to 4590 // figure out more details about the state of the header in order to 4591 // know what operations can be legally performed on the object's 4592 // header. 4593 4594 // If the low three bits in the xor result aren't clear, that means 4595 // the prototype header is no longer biased and we have to revoke 4596 // the bias on this object. 4597 testl(swap_reg, markOopDesc::biased_lock_mask_in_place); 4598 jcc(Assembler::notZero, try_revoke_bias); 4599 4600 // Biasing is still enabled for this data type. See whether the 4601 // epoch of the current bias is still valid, meaning that the epoch 4602 // bits of the mark word are equal to the epoch bits of the 4603 // prototype header. (Note that the prototype header's epoch bits 4604 // only change at a safepoint.) If not, attempt to rebias the object 4605 // toward the current thread. Note that we must be absolutely sure 4606 // that the current epoch is invalid in order to do this because 4607 // otherwise the manipulations it performs on the mark word are 4608 // illegal. 4609 testl(swap_reg, markOopDesc::epoch_mask_in_place); 4610 jcc(Assembler::notZero, try_rebias); 4611 4612 // The epoch of the current bias is still valid but we know nothing 4613 // about the owner; it might be set or it might be clear. Try to 4614 // acquire the bias of the object using an atomic operation. If this 4615 // fails we will go in to the runtime to revoke the object's bias. 4616 // Note that we first construct the presumed unbiased header so we 4617 // don't accidentally blow away another thread's valid bias. 4618 movl(swap_reg, saved_mark_addr); 4619 andl(swap_reg, 4620 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); 4621 if (need_tmp_reg) { 4622 push(tmp_reg); 4623 } 4624 get_thread(tmp_reg); 4625 orl(tmp_reg, swap_reg); 4626 if (os::is_MP()) { 4627 lock(); 4628 } 4629 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); 4630 if (need_tmp_reg) { 4631 pop(tmp_reg); 4632 } 4633 // If the biasing toward our thread failed, this means that 4634 // another thread succeeded in biasing it toward itself and we 4635 // need to revoke that bias. The revocation will occur in the 4636 // interpreter runtime in the slow case. 4637 if (counters != NULL) { 4638 cond_inc32(Assembler::zero, 4639 ExternalAddress((address)counters->anonymously_biased_lock_entry_count_addr())); 4640 } 4641 if (slow_case != NULL) { 4642 jcc(Assembler::notZero, *slow_case); 4643 } 4644 jmp(done); 4645 4646 bind(try_rebias); 4647 // At this point we know the epoch has expired, meaning that the 4648 // current "bias owner", if any, is actually invalid. Under these 4649 // circumstances _only_, we are allowed to use the current header's 4650 // value as the comparison value when doing the cas to acquire the 4651 // bias in the current epoch. In other words, we allow transfer of 4652 // the bias from one thread to another directly in this situation. 4653 // 4654 // FIXME: due to a lack of registers we currently blow away the age 4655 // bits in this situation. Should attempt to preserve them. 4656 if (need_tmp_reg) { 4657 push(tmp_reg); 4658 } 4659 get_thread(tmp_reg); 4660 movl(swap_reg, klass_addr); 4661 orl(tmp_reg, Address(swap_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 4662 movl(swap_reg, saved_mark_addr); 4663 if (os::is_MP()) { 4664 lock(); 4665 } 4666 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); 4667 if (need_tmp_reg) { 4668 pop(tmp_reg); 4669 } 4670 // If the biasing toward our thread failed, then another thread 4671 // succeeded in biasing it toward itself and we need to revoke that 4672 // bias. The revocation will occur in the runtime in the slow case. 4673 if (counters != NULL) { 4674 cond_inc32(Assembler::zero, 4675 ExternalAddress((address)counters->rebiased_lock_entry_count_addr())); 4676 } 4677 if (slow_case != NULL) { 4678 jcc(Assembler::notZero, *slow_case); 4679 } 4680 jmp(done); 4681 4682 bind(try_revoke_bias); 4683 // The prototype mark in the klass doesn't have the bias bit set any 4684 // more, indicating that objects of this data type are not supposed 4685 // to be biased any more. We are going to try to reset the mark of 4686 // this object to the prototype value and fall through to the 4687 // CAS-based locking scheme. Note that if our CAS fails, it means 4688 // that another thread raced us for the privilege of revoking the 4689 // bias of this particular object, so it's okay to continue in the 4690 // normal locking code. 4691 // 4692 // FIXME: due to a lack of registers we currently blow away the age 4693 // bits in this situation. Should attempt to preserve them. 4694 movl(swap_reg, saved_mark_addr); 4695 if (need_tmp_reg) { 4696 push(tmp_reg); 4697 } 4698 movl(tmp_reg, klass_addr); 4699 movl(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 4700 if (os::is_MP()) { 4701 lock(); 4702 } 4703 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); 4704 if (need_tmp_reg) { 4705 pop(tmp_reg); 4706 } 4707 // Fall through to the normal CAS-based lock, because no matter what 4708 // the result of the above CAS, some thread must have succeeded in 4709 // removing the bias bit from the object's header. 4710 if (counters != NULL) { 4711 cond_inc32(Assembler::zero, 4712 ExternalAddress((address)counters->revoked_lock_entry_count_addr())); 4713 } 4714 4715 bind(cas_label); 4716 4717 return null_check_offset; 4718 } 4719 void MacroAssembler::call_VM_leaf_base(address entry_point, 4720 int number_of_arguments) { 4721 call(RuntimeAddress(entry_point)); 4722 increment(rsp, number_of_arguments * wordSize); 4723 } 4724 4725 void MacroAssembler::cmpoop(Address src1, jobject obj) { 4726 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); 4727 } 4728 4729 void MacroAssembler::cmpoop(Register src1, jobject obj) { 4730 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); 4731 } 4732 4733 void MacroAssembler::extend_sign(Register hi, Register lo) { 4734 // According to Intel Doc. AP-526, "Integer Divide", p.18. 4735 if (VM_Version::is_P6() && hi == rdx && lo == rax) { 4736 cdql(); 4737 } else { 4738 movl(hi, lo); 4739 sarl(hi, 31); 4740 } 4741 } 4742 4743 void MacroAssembler::fat_nop() { 4744 // A 5 byte nop that is safe for patching (see patch_verified_entry) 4745 emit_byte(0x26); // es: 4746 emit_byte(0x2e); // cs: 4747 emit_byte(0x64); // fs: 4748 emit_byte(0x65); // gs: 4749 emit_byte(0x90); 4750 } 4751 4752 void MacroAssembler::jC2(Register tmp, Label& L) { 4753 // set parity bit if FPU flag C2 is set (via rax) 4754 save_rax(tmp); 4755 fwait(); fnstsw_ax(); 4756 sahf(); 4757 restore_rax(tmp); 4758 // branch 4759 jcc(Assembler::parity, L); 4760 } 4761 4762 void MacroAssembler::jnC2(Register tmp, Label& L) { 4763 // set parity bit if FPU flag C2 is set (via rax) 4764 save_rax(tmp); 4765 fwait(); fnstsw_ax(); 4766 sahf(); 4767 restore_rax(tmp); 4768 // branch 4769 jcc(Assembler::noParity, L); 4770 } 4771 4772 // 32bit can do a case table jump in one instruction but we no longer allow the base 4773 // to be installed in the Address class 4774 void MacroAssembler::jump(ArrayAddress entry) { 4775 jmp(as_Address(entry)); 4776 } 4777 4778 // Note: y_lo will be destroyed 4779 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) { 4780 // Long compare for Java (semantics as described in JVM spec.) 4781 Label high, low, done; 4782 4783 cmpl(x_hi, y_hi); 4784 jcc(Assembler::less, low); 4785 jcc(Assembler::greater, high); 4786 // x_hi is the return register 4787 xorl(x_hi, x_hi); 4788 cmpl(x_lo, y_lo); 4789 jcc(Assembler::below, low); 4790 jcc(Assembler::equal, done); 4791 4792 bind(high); 4793 xorl(x_hi, x_hi); 4794 increment(x_hi); 4795 jmp(done); 4796 4797 bind(low); 4798 xorl(x_hi, x_hi); 4799 decrementl(x_hi); 4800 4801 bind(done); 4802 } 4803 4804 void MacroAssembler::lea(Register dst, AddressLiteral src) { 4805 mov_literal32(dst, (int32_t)src.target(), src.rspec()); 4806 } 4807 4808 void MacroAssembler::lea(Address dst, AddressLiteral adr) { 4809 // leal(dst, as_Address(adr)); 4810 // see note in movl as to why we must use a move 4811 mov_literal32(dst, (int32_t) adr.target(), adr.rspec()); 4812 } 4813 4814 void MacroAssembler::leave() { 4815 mov(rsp, rbp); 4816 pop(rbp); 4817 } 4818 4819 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) { 4820 // Multiplication of two Java long values stored on the stack 4821 // as illustrated below. Result is in rdx:rax. 4822 // 4823 // rsp ---> [ ?? ] \ \ 4824 // .... | y_rsp_offset | 4825 // [ y_lo ] / (in bytes) | x_rsp_offset 4826 // [ y_hi ] | (in bytes) 4827 // .... | 4828 // [ x_lo ] / 4829 // [ x_hi ] 4830 // .... 4831 // 4832 // Basic idea: lo(result) = lo(x_lo * y_lo) 4833 // hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi) 4834 Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset); 4835 Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset); 4836 Label quick; 4837 // load x_hi, y_hi and check if quick 4838 // multiplication is possible 4839 movl(rbx, x_hi); 4840 movl(rcx, y_hi); 4841 movl(rax, rbx); 4842 orl(rbx, rcx); // rbx, = 0 <=> x_hi = 0 and y_hi = 0 4843 jcc(Assembler::zero, quick); // if rbx, = 0 do quick multiply 4844 // do full multiplication 4845 // 1st step 4846 mull(y_lo); // x_hi * y_lo 4847 movl(rbx, rax); // save lo(x_hi * y_lo) in rbx, 4848 // 2nd step 4849 movl(rax, x_lo); 4850 mull(rcx); // x_lo * y_hi 4851 addl(rbx, rax); // add lo(x_lo * y_hi) to rbx, 4852 // 3rd step 4853 bind(quick); // note: rbx, = 0 if quick multiply! 4854 movl(rax, x_lo); 4855 mull(y_lo); // x_lo * y_lo 4856 addl(rdx, rbx); // correct hi(x_lo * y_lo) 4857 } 4858 4859 void MacroAssembler::lneg(Register hi, Register lo) { 4860 negl(lo); 4861 adcl(hi, 0); 4862 negl(hi); 4863 } 4864 4865 void MacroAssembler::lshl(Register hi, Register lo) { 4866 // Java shift left long support (semantics as described in JVM spec., p.305) 4867 // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n)) 4868 // shift value is in rcx ! 4869 assert(hi != rcx, "must not use rcx"); 4870 assert(lo != rcx, "must not use rcx"); 4871 const Register s = rcx; // shift count 4872 const int n = BitsPerWord; 4873 Label L; 4874 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) 4875 cmpl(s, n); // if (s < n) 4876 jcc(Assembler::less, L); // else (s >= n) 4877 movl(hi, lo); // x := x << n 4878 xorl(lo, lo); 4879 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! 4880 bind(L); // s (mod n) < n 4881 shldl(hi, lo); // x := x << s 4882 shll(lo); 4883 } 4884 4885 4886 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) { 4887 // Java shift right long support (semantics as described in JVM spec., p.306 & p.310) 4888 // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n)) 4889 assert(hi != rcx, "must not use rcx"); 4890 assert(lo != rcx, "must not use rcx"); 4891 const Register s = rcx; // shift count 4892 const int n = BitsPerWord; 4893 Label L; 4894 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) 4895 cmpl(s, n); // if (s < n) 4896 jcc(Assembler::less, L); // else (s >= n) 4897 movl(lo, hi); // x := x >> n 4898 if (sign_extension) sarl(hi, 31); 4899 else xorl(hi, hi); 4900 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! 4901 bind(L); // s (mod n) < n 4902 shrdl(lo, hi); // x := x >> s 4903 if (sign_extension) sarl(hi); 4904 else shrl(hi); 4905 } 4906 4907 void MacroAssembler::movoop(Register dst, jobject obj) { 4908 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); 4909 } 4910 4911 void MacroAssembler::movoop(Address dst, jobject obj) { 4912 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); 4913 } 4914 4915 void MacroAssembler::movptr(Register dst, AddressLiteral src) { 4916 if (src.is_lval()) { 4917 mov_literal32(dst, (intptr_t)src.target(), src.rspec()); 4918 } else { 4919 movl(dst, as_Address(src)); 4920 } 4921 } 4922 4923 void MacroAssembler::movptr(ArrayAddress dst, Register src) { 4924 movl(as_Address(dst), src); 4925 } 4926 4927 void MacroAssembler::movptr(Register dst, ArrayAddress src) { 4928 movl(dst, as_Address(src)); 4929 } 4930 4931 // src should NEVER be a real pointer. Use AddressLiteral for true pointers 4932 void MacroAssembler::movptr(Address dst, intptr_t src) { 4933 movl(dst, src); 4934 } 4935 4936 4937 void MacroAssembler::pop_callee_saved_registers() { 4938 pop(rcx); 4939 pop(rdx); 4940 pop(rdi); 4941 pop(rsi); 4942 } 4943 4944 void MacroAssembler::pop_fTOS() { 4945 fld_d(Address(rsp, 0)); 4946 addl(rsp, 2 * wordSize); 4947 } 4948 4949 void MacroAssembler::push_callee_saved_registers() { 4950 push(rsi); 4951 push(rdi); 4952 push(rdx); 4953 push(rcx); 4954 } 4955 4956 void MacroAssembler::push_fTOS() { 4957 subl(rsp, 2 * wordSize); 4958 fstp_d(Address(rsp, 0)); 4959 } 4960 4961 4962 void MacroAssembler::pushoop(jobject obj) { 4963 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate()); 4964 } 4965 4966 4967 void MacroAssembler::pushptr(AddressLiteral src) { 4968 if (src.is_lval()) { 4969 push_literal32((int32_t)src.target(), src.rspec()); 4970 } else { 4971 pushl(as_Address(src)); 4972 } 4973 } 4974 4975 void MacroAssembler::set_word_if_not_zero(Register dst) { 4976 xorl(dst, dst); 4977 set_byte_if_not_zero(dst); 4978 } 4979 4980 static void pass_arg0(MacroAssembler* masm, Register arg) { 4981 masm->push(arg); 4982 } 4983 4984 static void pass_arg1(MacroAssembler* masm, Register arg) { 4985 masm->push(arg); 4986 } 4987 4988 static void pass_arg2(MacroAssembler* masm, Register arg) { 4989 masm->push(arg); 4990 } 4991 4992 static void pass_arg3(MacroAssembler* masm, Register arg) { 4993 masm->push(arg); 4994 } 4995 4996 #ifndef PRODUCT 4997 extern "C" void findpc(intptr_t x); 4998 #endif 4999 5000 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) { 5001 // In order to get locks to work, we need to fake a in_VM state 5002 JavaThread* thread = JavaThread::current(); 5003 JavaThreadState saved_state = thread->thread_state(); 5004 thread->set_thread_state(_thread_in_vm); 5005 if (ShowMessageBoxOnError) { 5006 JavaThread* thread = JavaThread::current(); 5007 JavaThreadState saved_state = thread->thread_state(); 5008 thread->set_thread_state(_thread_in_vm); 5009 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { 5010 ttyLocker ttyl; 5011 BytecodeCounter::print(); 5012 } 5013 // To see where a verify_oop failed, get $ebx+40/X for this frame. 5014 // This is the value of eip which points to where verify_oop will return. 5015 if (os::message_box(msg, "Execution stopped, print registers?")) { 5016 ttyLocker ttyl; 5017 tty->print_cr("eip = 0x%08x", eip); 5018 #ifndef PRODUCT 5019 if ((WizardMode || Verbose) && PrintMiscellaneous) { 5020 tty->cr(); 5021 findpc(eip); 5022 tty->cr(); 5023 } 5024 #endif 5025 tty->print_cr("rax = 0x%08x", rax); 5026 tty->print_cr("rbx = 0x%08x", rbx); 5027 tty->print_cr("rcx = 0x%08x", rcx); 5028 tty->print_cr("rdx = 0x%08x", rdx); 5029 tty->print_cr("rdi = 0x%08x", rdi); 5030 tty->print_cr("rsi = 0x%08x", rsi); 5031 tty->print_cr("rbp = 0x%08x", rbp); 5032 tty->print_cr("rsp = 0x%08x", rsp); 5033 BREAKPOINT; 5034 assert(false, "start up GDB"); 5035 } 5036 } else { 5037 ttyLocker ttyl; 5038 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg); 5039 assert(false, "DEBUG MESSAGE"); 5040 } 5041 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); 5042 } 5043 5044 void MacroAssembler::stop(const char* msg) { 5045 ExternalAddress message((address)msg); 5046 // push address of message 5047 pushptr(message.addr()); 5048 { Label L; call(L, relocInfo::none); bind(L); } // push eip 5049 pusha(); // push registers 5050 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); 5051 hlt(); 5052 } 5053 5054 void MacroAssembler::warn(const char* msg) { 5055 push_CPU_state(); 5056 5057 ExternalAddress message((address) msg); 5058 // push address of message 5059 pushptr(message.addr()); 5060 5061 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning))); 5062 addl(rsp, wordSize); // discard argument 5063 pop_CPU_state(); 5064 } 5065 5066 #else // _LP64 5067 5068 // 64 bit versions 5069 5070 Address MacroAssembler::as_Address(AddressLiteral adr) { 5071 // amd64 always does this as a pc-rel 5072 // we can be absolute or disp based on the instruction type 5073 // jmp/call are displacements others are absolute 5074 assert(!adr.is_lval(), "must be rval"); 5075 assert(reachable(adr), "must be"); 5076 return Address((int32_t)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc()); 5077 5078 } 5079 5080 Address MacroAssembler::as_Address(ArrayAddress adr) { 5081 AddressLiteral base = adr.base(); 5082 lea(rscratch1, base); 5083 Address index = adr.index(); 5084 assert(index._disp == 0, "must not have disp"); // maybe it can? 5085 Address array(rscratch1, index._index, index._scale, index._disp); 5086 return array; 5087 } 5088 5089 int MacroAssembler::biased_locking_enter(Register lock_reg, 5090 Register obj_reg, 5091 Register swap_reg, 5092 Register tmp_reg, 5093 bool swap_reg_contains_mark, 5094 Label& done, 5095 Label* slow_case, 5096 BiasedLockingCounters* counters) { 5097 assert(UseBiasedLocking, "why call this otherwise?"); 5098 assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq"); 5099 assert(tmp_reg != noreg, "tmp_reg must be supplied"); 5100 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); 5101 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); 5102 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); 5103 Address saved_mark_addr(lock_reg, 0); 5104 5105 if (PrintBiasedLockingStatistics && counters == NULL) 5106 counters = BiasedLocking::counters(); 5107 5108 // Biased locking 5109 // See whether the lock is currently biased toward our thread and 5110 // whether the epoch is still valid 5111 // Note that the runtime guarantees sufficient alignment of JavaThread 5112 // pointers to allow age to be placed into low bits 5113 // First check to see whether biasing is even enabled for this object 5114 Label cas_label; 5115 int null_check_offset = -1; 5116 if (!swap_reg_contains_mark) { 5117 null_check_offset = offset(); 5118 movq(swap_reg, mark_addr); 5119 } 5120 movq(tmp_reg, swap_reg); 5121 andq(tmp_reg, markOopDesc::biased_lock_mask_in_place); 5122 cmpq(tmp_reg, markOopDesc::biased_lock_pattern); 5123 jcc(Assembler::notEqual, cas_label); 5124 // The bias pattern is present in the object's header. Need to check 5125 // whether the bias owner and the epoch are both still current. 5126 load_prototype_header(tmp_reg, obj_reg); 5127 orq(tmp_reg, r15_thread); 5128 xorq(tmp_reg, swap_reg); 5129 andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place)); 5130 if (counters != NULL) { 5131 cond_inc32(Assembler::zero, 5132 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); 5133 } 5134 jcc(Assembler::equal, done); 5135 5136 Label try_revoke_bias; 5137 Label try_rebias; 5138 5139 // At this point we know that the header has the bias pattern and 5140 // that we are not the bias owner in the current epoch. We need to 5141 // figure out more details about the state of the header in order to 5142 // know what operations can be legally performed on the object's 5143 // header. 5144 5145 // If the low three bits in the xor result aren't clear, that means 5146 // the prototype header is no longer biased and we have to revoke 5147 // the bias on this object. 5148 testq(tmp_reg, markOopDesc::biased_lock_mask_in_place); 5149 jcc(Assembler::notZero, try_revoke_bias); 5150 5151 // Biasing is still enabled for this data type. See whether the 5152 // epoch of the current bias is still valid, meaning that the epoch 5153 // bits of the mark word are equal to the epoch bits of the 5154 // prototype header. (Note that the prototype header's epoch bits 5155 // only change at a safepoint.) If not, attempt to rebias the object 5156 // toward the current thread. Note that we must be absolutely sure 5157 // that the current epoch is invalid in order to do this because 5158 // otherwise the manipulations it performs on the mark word are 5159 // illegal. 5160 testq(tmp_reg, markOopDesc::epoch_mask_in_place); 5161 jcc(Assembler::notZero, try_rebias); 5162 5163 // The epoch of the current bias is still valid but we know nothing 5164 // about the owner; it might be set or it might be clear. Try to 5165 // acquire the bias of the object using an atomic operation. If this 5166 // fails we will go in to the runtime to revoke the object's bias. 5167 // Note that we first construct the presumed unbiased header so we 5168 // don't accidentally blow away another thread's valid bias. 5169 andq(swap_reg, 5170 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); 5171 movq(tmp_reg, swap_reg); 5172 orq(tmp_reg, r15_thread); 5173 if (os::is_MP()) { 5174 lock(); 5175 } 5176 cmpxchgq(tmp_reg, Address(obj_reg, 0)); 5177 // If the biasing toward our thread failed, this means that 5178 // another thread succeeded in biasing it toward itself and we 5179 // need to revoke that bias. The revocation will occur in the 5180 // interpreter runtime in the slow case. 5181 if (counters != NULL) { 5182 cond_inc32(Assembler::zero, 5183 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); 5184 } 5185 if (slow_case != NULL) { 5186 jcc(Assembler::notZero, *slow_case); 5187 } 5188 jmp(done); 5189 5190 bind(try_rebias); 5191 // At this point we know the epoch has expired, meaning that the 5192 // current "bias owner", if any, is actually invalid. Under these 5193 // circumstances _only_, we are allowed to use the current header's 5194 // value as the comparison value when doing the cas to acquire the 5195 // bias in the current epoch. In other words, we allow transfer of 5196 // the bias from one thread to another directly in this situation. 5197 // 5198 // FIXME: due to a lack of registers we currently blow away the age 5199 // bits in this situation. Should attempt to preserve them. 5200 load_prototype_header(tmp_reg, obj_reg); 5201 orq(tmp_reg, r15_thread); 5202 if (os::is_MP()) { 5203 lock(); 5204 } 5205 cmpxchgq(tmp_reg, Address(obj_reg, 0)); 5206 // If the biasing toward our thread failed, then another thread 5207 // succeeded in biasing it toward itself and we need to revoke that 5208 // bias. The revocation will occur in the runtime in the slow case. 5209 if (counters != NULL) { 5210 cond_inc32(Assembler::zero, 5211 ExternalAddress((address) counters->rebiased_lock_entry_count_addr())); 5212 } 5213 if (slow_case != NULL) { 5214 jcc(Assembler::notZero, *slow_case); 5215 } 5216 jmp(done); 5217 5218 bind(try_revoke_bias); 5219 // The prototype mark in the klass doesn't have the bias bit set any 5220 // more, indicating that objects of this data type are not supposed 5221 // to be biased any more. We are going to try to reset the mark of 5222 // this object to the prototype value and fall through to the 5223 // CAS-based locking scheme. Note that if our CAS fails, it means 5224 // that another thread raced us for the privilege of revoking the 5225 // bias of this particular object, so it's okay to continue in the 5226 // normal locking code. 5227 // 5228 // FIXME: due to a lack of registers we currently blow away the age 5229 // bits in this situation. Should attempt to preserve them. 5230 load_prototype_header(tmp_reg, obj_reg); 5231 if (os::is_MP()) { 5232 lock(); 5233 } 5234 cmpxchgq(tmp_reg, Address(obj_reg, 0)); 5235 // Fall through to the normal CAS-based lock, because no matter what 5236 // the result of the above CAS, some thread must have succeeded in 5237 // removing the bias bit from the object's header. 5238 if (counters != NULL) { 5239 cond_inc32(Assembler::zero, 5240 ExternalAddress((address) counters->revoked_lock_entry_count_addr())); 5241 } 5242 5243 bind(cas_label); 5244 5245 return null_check_offset; 5246 } 5247 5248 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) { 5249 Label L, E; 5250 5251 #ifdef _WIN64 5252 // Windows always allocates space for it's register args 5253 assert(num_args <= 4, "only register arguments supported"); 5254 subq(rsp, frame::arg_reg_save_area_bytes); 5255 #endif 5256 5257 // Align stack if necessary 5258 testl(rsp, 15); 5259 jcc(Assembler::zero, L); 5260 5261 subq(rsp, 8); 5262 { 5263 call(RuntimeAddress(entry_point)); 5264 } 5265 addq(rsp, 8); 5266 jmp(E); 5267 5268 bind(L); 5269 { 5270 call(RuntimeAddress(entry_point)); 5271 } 5272 5273 bind(E); 5274 5275 #ifdef _WIN64 5276 // restore stack pointer 5277 addq(rsp, frame::arg_reg_save_area_bytes); 5278 #endif 5279 5280 } 5281 5282 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) { 5283 assert(!src2.is_lval(), "should use cmpptr"); 5284 5285 if (reachable(src2)) { 5286 cmpq(src1, as_Address(src2)); 5287 } else { 5288 lea(rscratch1, src2); 5289 Assembler::cmpq(src1, Address(rscratch1, 0)); 5290 } 5291 } 5292 5293 int MacroAssembler::corrected_idivq(Register reg) { 5294 // Full implementation of Java ldiv and lrem; checks for special 5295 // case as described in JVM spec., p.243 & p.271. The function 5296 // returns the (pc) offset of the idivl instruction - may be needed 5297 // for implicit exceptions. 5298 // 5299 // normal case special case 5300 // 5301 // input : rax: dividend min_long 5302 // reg: divisor (may not be eax/edx) -1 5303 // 5304 // output: rax: quotient (= rax idiv reg) min_long 5305 // rdx: remainder (= rax irem reg) 0 5306 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register"); 5307 static const int64_t min_long = 0x8000000000000000; 5308 Label normal_case, special_case; 5309 5310 // check for special case 5311 cmp64(rax, ExternalAddress((address) &min_long)); 5312 jcc(Assembler::notEqual, normal_case); 5313 xorl(rdx, rdx); // prepare rdx for possible special case (where 5314 // remainder = 0) 5315 cmpq(reg, -1); 5316 jcc(Assembler::equal, special_case); 5317 5318 // handle normal case 5319 bind(normal_case); 5320 cdqq(); 5321 int idivq_offset = offset(); 5322 idivq(reg); 5323 5324 // normal and special case exit 5325 bind(special_case); 5326 5327 return idivq_offset; 5328 } 5329 5330 void MacroAssembler::decrementq(Register reg, int value) { 5331 if (value == min_jint) { subq(reg, value); return; } 5332 if (value < 0) { incrementq(reg, -value); return; } 5333 if (value == 0) { ; return; } 5334 if (value == 1 && UseIncDec) { decq(reg) ; return; } 5335 /* else */ { subq(reg, value) ; return; } 5336 } 5337 5338 void MacroAssembler::decrementq(Address dst, int value) { 5339 if (value == min_jint) { subq(dst, value); return; } 5340 if (value < 0) { incrementq(dst, -value); return; } 5341 if (value == 0) { ; return; } 5342 if (value == 1 && UseIncDec) { decq(dst) ; return; } 5343 /* else */ { subq(dst, value) ; return; } 5344 } 5345 5346 void MacroAssembler::fat_nop() { 5347 // A 5 byte nop that is safe for patching (see patch_verified_entry) 5348 // Recommened sequence from 'Software Optimization Guide for the AMD 5349 // Hammer Processor' 5350 emit_byte(0x66); 5351 emit_byte(0x66); 5352 emit_byte(0x90); 5353 emit_byte(0x66); 5354 emit_byte(0x90); 5355 } 5356 5357 void MacroAssembler::incrementq(Register reg, int value) { 5358 if (value == min_jint) { addq(reg, value); return; } 5359 if (value < 0) { decrementq(reg, -value); return; } 5360 if (value == 0) { ; return; } 5361 if (value == 1 && UseIncDec) { incq(reg) ; return; } 5362 /* else */ { addq(reg, value) ; return; } 5363 } 5364 5365 void MacroAssembler::incrementq(Address dst, int value) { 5366 if (value == min_jint) { addq(dst, value); return; } 5367 if (value < 0) { decrementq(dst, -value); return; } 5368 if (value == 0) { ; return; } 5369 if (value == 1 && UseIncDec) { incq(dst) ; return; } 5370 /* else */ { addq(dst, value) ; return; } 5371 } 5372 5373 // 32bit can do a case table jump in one instruction but we no longer allow the base 5374 // to be installed in the Address class 5375 void MacroAssembler::jump(ArrayAddress entry) { 5376 lea(rscratch1, entry.base()); 5377 Address dispatch = entry.index(); 5378 assert(dispatch._base == noreg, "must be"); 5379 dispatch._base = rscratch1; 5380 jmp(dispatch); 5381 } 5382 5383 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) { 5384 ShouldNotReachHere(); // 64bit doesn't use two regs 5385 cmpq(x_lo, y_lo); 5386 } 5387 5388 void MacroAssembler::lea(Register dst, AddressLiteral src) { 5389 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); 5390 } 5391 5392 void MacroAssembler::lea(Address dst, AddressLiteral adr) { 5393 mov_literal64(rscratch1, (intptr_t)adr.target(), adr.rspec()); 5394 movptr(dst, rscratch1); 5395 } 5396 5397 void MacroAssembler::leave() { 5398 // %%% is this really better? Why not on 32bit too? 5399 emit_byte(0xC9); // LEAVE 5400 } 5401 5402 void MacroAssembler::lneg(Register hi, Register lo) { 5403 ShouldNotReachHere(); // 64bit doesn't use two regs 5404 negq(lo); 5405 } 5406 5407 void MacroAssembler::movoop(Register dst, jobject obj) { 5408 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate()); 5409 } 5410 5411 void MacroAssembler::movoop(Address dst, jobject obj) { 5412 mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate()); 5413 movq(dst, rscratch1); 5414 } 5415 5416 void MacroAssembler::movptr(Register dst, AddressLiteral src) { 5417 if (src.is_lval()) { 5418 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); 5419 } else { 5420 if (reachable(src)) { 5421 movq(dst, as_Address(src)); 5422 } else { 5423 lea(rscratch1, src); 5424 movq(dst, Address(rscratch1,0)); 5425 } 5426 } 5427 } 5428 5429 void MacroAssembler::movptr(ArrayAddress dst, Register src) { 5430 movq(as_Address(dst), src); 5431 } 5432 5433 void MacroAssembler::movptr(Register dst, ArrayAddress src) { 5434 movq(dst, as_Address(src)); 5435 } 5436 5437 // src should NEVER be a real pointer. Use AddressLiteral for true pointers 5438 void MacroAssembler::movptr(Address dst, intptr_t src) { 5439 mov64(rscratch1, src); 5440 movq(dst, rscratch1); 5441 } 5442 5443 // These are mostly for initializing NULL 5444 void MacroAssembler::movptr(Address dst, int32_t src) { 5445 movslq(dst, src); 5446 } 5447 5448 void MacroAssembler::movptr(Register dst, int32_t src) { 5449 mov64(dst, (intptr_t)src); 5450 } 5451 5452 void MacroAssembler::pushoop(jobject obj) { 5453 movoop(rscratch1, obj); 5454 push(rscratch1); 5455 } 5456 5457 void MacroAssembler::pushptr(AddressLiteral src) { 5458 lea(rscratch1, src); 5459 if (src.is_lval()) { 5460 push(rscratch1); 5461 } else { 5462 pushq(Address(rscratch1, 0)); 5463 } 5464 } 5465 5466 void MacroAssembler::reset_last_Java_frame(bool clear_fp, 5467 bool clear_pc) { 5468 // we must set sp to zero to clear frame 5469 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); 5470 // must clear fp, so that compiled frames are not confused; it is 5471 // possible that we need it only for debugging 5472 if (clear_fp) { 5473 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); 5474 } 5475 5476 if (clear_pc) { 5477 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); 5478 } 5479 } 5480 5481 void MacroAssembler::set_last_Java_frame(Register last_java_sp, 5482 Register last_java_fp, 5483 address last_java_pc) { 5484 // determine last_java_sp register 5485 if (!last_java_sp->is_valid()) { 5486 last_java_sp = rsp; 5487 } 5488 5489 // last_java_fp is optional 5490 if (last_java_fp->is_valid()) { 5491 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), 5492 last_java_fp); 5493 } 5494 5495 // last_java_pc is optional 5496 if (last_java_pc != NULL) { 5497 Address java_pc(r15_thread, 5498 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()); 5499 lea(rscratch1, InternalAddress(last_java_pc)); 5500 movptr(java_pc, rscratch1); 5501 } 5502 5503 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp); 5504 } 5505 5506 static void pass_arg0(MacroAssembler* masm, Register arg) { 5507 if (c_rarg0 != arg ) { 5508 masm->mov(c_rarg0, arg); 5509 } 5510 } 5511 5512 static void pass_arg1(MacroAssembler* masm, Register arg) { 5513 if (c_rarg1 != arg ) { 5514 masm->mov(c_rarg1, arg); 5515 } 5516 } 5517 5518 static void pass_arg2(MacroAssembler* masm, Register arg) { 5519 if (c_rarg2 != arg ) { 5520 masm->mov(c_rarg2, arg); 5521 } 5522 } 5523 5524 static void pass_arg3(MacroAssembler* masm, Register arg) { 5525 if (c_rarg3 != arg ) { 5526 masm->mov(c_rarg3, arg); 5527 } 5528 } 5529 5530 void MacroAssembler::stop(const char* msg) { 5531 address rip = pc(); 5532 pusha(); // get regs on stack 5533 lea(c_rarg0, ExternalAddress((address) msg)); 5534 lea(c_rarg1, InternalAddress(rip)); 5535 movq(c_rarg2, rsp); // pass pointer to regs array 5536 andq(rsp, -16); // align stack as required by ABI 5537 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64))); 5538 hlt(); 5539 } 5540 5541 void MacroAssembler::warn(const char* msg) { 5542 push(r12); 5543 movq(r12, rsp); 5544 andq(rsp, -16); // align stack as required by push_CPU_state and call 5545 5546 push_CPU_state(); // keeps alignment at 16 bytes 5547 lea(c_rarg0, ExternalAddress((address) msg)); 5548 call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0); 5549 pop_CPU_state(); 5550 5551 movq(rsp, r12); 5552 pop(r12); 5553 } 5554 5555 #ifndef PRODUCT 5556 extern "C" void findpc(intptr_t x); 5557 #endif 5558 5559 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) { 5560 // In order to get locks to work, we need to fake a in_VM state 5561 if (ShowMessageBoxOnError ) { 5562 JavaThread* thread = JavaThread::current(); 5563 JavaThreadState saved_state = thread->thread_state(); 5564 thread->set_thread_state(_thread_in_vm); 5565 #ifndef PRODUCT 5566 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { 5567 ttyLocker ttyl; 5568 BytecodeCounter::print(); 5569 } 5570 #endif 5571 // To see where a verify_oop failed, get $ebx+40/X for this frame. 5572 // XXX correct this offset for amd64 5573 // This is the value of eip which points to where verify_oop will return. 5574 if (os::message_box(msg, "Execution stopped, print registers?")) { 5575 ttyLocker ttyl; 5576 tty->print_cr("rip = 0x%016lx", pc); 5577 #ifndef PRODUCT 5578 tty->cr(); 5579 findpc(pc); 5580 tty->cr(); 5581 #endif 5582 tty->print_cr("rax = 0x%016lx", regs[15]); 5583 tty->print_cr("rbx = 0x%016lx", regs[12]); 5584 tty->print_cr("rcx = 0x%016lx", regs[14]); 5585 tty->print_cr("rdx = 0x%016lx", regs[13]); 5586 tty->print_cr("rdi = 0x%016lx", regs[8]); 5587 tty->print_cr("rsi = 0x%016lx", regs[9]); 5588 tty->print_cr("rbp = 0x%016lx", regs[10]); 5589 tty->print_cr("rsp = 0x%016lx", regs[11]); 5590 tty->print_cr("r8 = 0x%016lx", regs[7]); 5591 tty->print_cr("r9 = 0x%016lx", regs[6]); 5592 tty->print_cr("r10 = 0x%016lx", regs[5]); 5593 tty->print_cr("r11 = 0x%016lx", regs[4]); 5594 tty->print_cr("r12 = 0x%016lx", regs[3]); 5595 tty->print_cr("r13 = 0x%016lx", regs[2]); 5596 tty->print_cr("r14 = 0x%016lx", regs[1]); 5597 tty->print_cr("r15 = 0x%016lx", regs[0]); 5598 BREAKPOINT; 5599 } 5600 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); 5601 } else { 5602 ttyLocker ttyl; 5603 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", 5604 msg); 5605 } 5606 } 5607 5608 #endif // _LP64 5609 5610 // Now versions that are common to 32/64 bit 5611 5612 void MacroAssembler::addptr(Register dst, int32_t imm32) { 5613 LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32)); 5614 } 5615 5616 void MacroAssembler::addptr(Register dst, Register src) { 5617 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); 5618 } 5619 5620 void MacroAssembler::addptr(Address dst, Register src) { 5621 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); 5622 } 5623 5624 void MacroAssembler::align(int modulus) { 5625 if (offset() % modulus != 0) { 5626 nop(modulus - (offset() % modulus)); 5627 } 5628 } 5629 5630 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) { 5631 if (reachable(src)) { 5632 andpd(dst, as_Address(src)); 5633 } else { 5634 lea(rscratch1, src); 5635 andpd(dst, Address(rscratch1, 0)); 5636 } 5637 } 5638 5639 void MacroAssembler::andptr(Register dst, int32_t imm32) { 5640 LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32)); 5641 } 5642 5643 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) { 5644 pushf(); 5645 if (os::is_MP()) 5646 lock(); 5647 incrementl(counter_addr); 5648 popf(); 5649 } 5650 5651 // Writes to stack successive pages until offset reached to check for 5652 // stack overflow + shadow pages. This clobbers tmp. 5653 void MacroAssembler::bang_stack_size(Register size, Register tmp) { 5654 movptr(tmp, rsp); 5655 // Bang stack for total size given plus shadow page size. 5656 // Bang one page at a time because large size can bang beyond yellow and 5657 // red zones. 5658 Label loop; 5659 bind(loop); 5660 movl(Address(tmp, (-os::vm_page_size())), size ); 5661 subptr(tmp, os::vm_page_size()); 5662 subl(size, os::vm_page_size()); 5663 jcc(Assembler::greater, loop); 5664 5665 // Bang down shadow pages too. 5666 // The -1 because we already subtracted 1 page. 5667 for (int i = 0; i< StackShadowPages-1; i++) { 5668 // this could be any sized move but this is can be a debugging crumb 5669 // so the bigger the better. 5670 movptr(Address(tmp, (-i*os::vm_page_size())), size ); 5671 } 5672 } 5673 5674 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) { 5675 assert(UseBiasedLocking, "why call this otherwise?"); 5676 5677 // Check for biased locking unlock case, which is a no-op 5678 // Note: we do not have to check the thread ID for two reasons. 5679 // First, the interpreter checks for IllegalMonitorStateException at 5680 // a higher level. Second, if the bias was revoked while we held the 5681 // lock, the object could not be rebiased toward another thread, so 5682 // the bias bit would be clear. 5683 movptr(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 5684 andptr(temp_reg, markOopDesc::biased_lock_mask_in_place); 5685 cmpptr(temp_reg, markOopDesc::biased_lock_pattern); 5686 jcc(Assembler::equal, done); 5687 } 5688 5689 void MacroAssembler::c2bool(Register x) { 5690 // implements x == 0 ? 0 : 1 5691 // note: must only look at least-significant byte of x 5692 // since C-style booleans are stored in one byte 5693 // only! (was bug) 5694 andl(x, 0xFF); 5695 setb(Assembler::notZero, x); 5696 } 5697 5698 // Wouldn't need if AddressLiteral version had new name 5699 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) { 5700 Assembler::call(L, rtype); 5701 } 5702 5703 void MacroAssembler::call(Register entry) { 5704 Assembler::call(entry); 5705 } 5706 5707 void MacroAssembler::call(AddressLiteral entry) { 5708 if (reachable(entry)) { 5709 Assembler::call_literal(entry.target(), entry.rspec()); 5710 } else { 5711 lea(rscratch1, entry); 5712 Assembler::call(rscratch1); 5713 } 5714 } 5715 5716 // Implementation of call_VM versions 5717 5718 void MacroAssembler::call_VM(Register oop_result, 5719 address entry_point, 5720 bool check_exceptions) { 5721 Label C, E; 5722 call(C, relocInfo::none); 5723 jmp(E); 5724 5725 bind(C); 5726 call_VM_helper(oop_result, entry_point, 0, check_exceptions); 5727 ret(0); 5728 5729 bind(E); 5730 } 5731 5732 void MacroAssembler::call_VM(Register oop_result, 5733 address entry_point, 5734 Register arg_1, 5735 bool check_exceptions) { 5736 Label C, E; 5737 call(C, relocInfo::none); 5738 jmp(E); 5739 5740 bind(C); 5741 pass_arg1(this, arg_1); 5742 call_VM_helper(oop_result, entry_point, 1, check_exceptions); 5743 ret(0); 5744 5745 bind(E); 5746 } 5747 5748 void MacroAssembler::call_VM(Register oop_result, 5749 address entry_point, 5750 Register arg_1, 5751 Register arg_2, 5752 bool check_exceptions) { 5753 Label C, E; 5754 call(C, relocInfo::none); 5755 jmp(E); 5756 5757 bind(C); 5758 5759 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); 5760 5761 pass_arg2(this, arg_2); 5762 pass_arg1(this, arg_1); 5763 call_VM_helper(oop_result, entry_point, 2, check_exceptions); 5764 ret(0); 5765 5766 bind(E); 5767 } 5768 5769 void MacroAssembler::call_VM(Register oop_result, 5770 address entry_point, 5771 Register arg_1, 5772 Register arg_2, 5773 Register arg_3, 5774 bool check_exceptions) { 5775 Label C, E; 5776 call(C, relocInfo::none); 5777 jmp(E); 5778 5779 bind(C); 5780 5781 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg")); 5782 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg")); 5783 pass_arg3(this, arg_3); 5784 5785 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); 5786 pass_arg2(this, arg_2); 5787 5788 pass_arg1(this, arg_1); 5789 call_VM_helper(oop_result, entry_point, 3, check_exceptions); 5790 ret(0); 5791 5792 bind(E); 5793 } 5794 5795 void MacroAssembler::call_VM(Register oop_result, 5796 Register last_java_sp, 5797 address entry_point, 5798 int number_of_arguments, 5799 bool check_exceptions) { 5800 Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg); 5801 call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions); 5802 } 5803 5804 void MacroAssembler::call_VM(Register oop_result, 5805 Register last_java_sp, 5806 address entry_point, 5807 Register arg_1, 5808 bool check_exceptions) { 5809 pass_arg1(this, arg_1); 5810 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions); 5811 } 5812 5813 void MacroAssembler::call_VM(Register oop_result, 5814 Register last_java_sp, 5815 address entry_point, 5816 Register arg_1, 5817 Register arg_2, 5818 bool check_exceptions) { 5819 5820 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); 5821 pass_arg2(this, arg_2); 5822 pass_arg1(this, arg_1); 5823 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions); 5824 } 5825 5826 void MacroAssembler::call_VM(Register oop_result, 5827 Register last_java_sp, 5828 address entry_point, 5829 Register arg_1, 5830 Register arg_2, 5831 Register arg_3, 5832 bool check_exceptions) { 5833 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg")); 5834 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg")); 5835 pass_arg3(this, arg_3); 5836 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); 5837 pass_arg2(this, arg_2); 5838 pass_arg1(this, arg_1); 5839 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions); 5840 } 5841 5842 void MacroAssembler::call_VM_base(Register oop_result, 5843 Register java_thread, 5844 Register last_java_sp, 5845 address entry_point, 5846 int number_of_arguments, 5847 bool check_exceptions) { 5848 // determine java_thread register 5849 if (!java_thread->is_valid()) { 5850 #ifdef _LP64 5851 java_thread = r15_thread; 5852 #else 5853 java_thread = rdi; 5854 get_thread(java_thread); 5855 #endif // LP64 5856 } 5857 // determine last_java_sp register 5858 if (!last_java_sp->is_valid()) { 5859 last_java_sp = rsp; 5860 } 5861 // debugging support 5862 assert(number_of_arguments >= 0 , "cannot have negative number of arguments"); 5863 LP64_ONLY(assert(java_thread == r15_thread, "unexpected register")); 5864 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result"); 5865 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp"); 5866 5867 // push java thread (becomes first argument of C function) 5868 5869 NOT_LP64(push(java_thread); number_of_arguments++); 5870 LP64_ONLY(mov(c_rarg0, r15_thread)); 5871 5872 // set last Java frame before call 5873 assert(last_java_sp != rbp, "can't use ebp/rbp"); 5874 5875 // Only interpreter should have to set fp 5876 set_last_Java_frame(java_thread, last_java_sp, rbp, NULL); 5877 5878 // do the call, remove parameters 5879 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 5880 5881 // restore the thread (cannot use the pushed argument since arguments 5882 // may be overwritten by C code generated by an optimizing compiler); 5883 // however can use the register value directly if it is callee saved. 5884 if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) { 5885 // rdi & rsi (also r15) are callee saved -> nothing to do 5886 #ifdef ASSERT 5887 guarantee(java_thread != rax, "change this code"); 5888 push(rax); 5889 { Label L; 5890 get_thread(rax); 5891 cmpptr(java_thread, rax); 5892 jcc(Assembler::equal, L); 5893 stop("MacroAssembler::call_VM_base: rdi not callee saved?"); 5894 bind(L); 5895 } 5896 pop(rax); 5897 #endif 5898 } else { 5899 get_thread(java_thread); 5900 } 5901 // reset last Java frame 5902 // Only interpreter should have to clear fp 5903 reset_last_Java_frame(java_thread, true, false); 5904 5905 #ifndef CC_INTERP 5906 // C++ interp handles this in the interpreter 5907 check_and_handle_popframe(java_thread); 5908 check_and_handle_earlyret(java_thread); 5909 #endif /* CC_INTERP */ 5910 5911 if (check_exceptions) { 5912 // check for pending exceptions (java_thread is set upon return) 5913 cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); 5914 #ifndef _LP64 5915 jump_cc(Assembler::notEqual, 5916 RuntimeAddress(StubRoutines::forward_exception_entry())); 5917 #else 5918 // This used to conditionally jump to forward_exception however it is 5919 // possible if we relocate that the branch will not reach. So we must jump 5920 // around so we can always reach 5921 5922 Label ok; 5923 jcc(Assembler::equal, ok); 5924 jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 5925 bind(ok); 5926 #endif // LP64 5927 } 5928 5929 // get oop result if there is one and reset the value in the thread 5930 if (oop_result->is_valid()) { 5931 movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset())); 5932 movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD); 5933 verify_oop(oop_result, "broken oop in call_VM_base"); 5934 } 5935 } 5936 5937 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) { 5938 5939 // Calculate the value for last_Java_sp 5940 // somewhat subtle. call_VM does an intermediate call 5941 // which places a return address on the stack just under the 5942 // stack pointer as the user finsihed with it. This allows 5943 // use to retrieve last_Java_pc from last_Java_sp[-1]. 5944 // On 32bit we then have to push additional args on the stack to accomplish 5945 // the actual requested call. On 64bit call_VM only can use register args 5946 // so the only extra space is the return address that call_VM created. 5947 // This hopefully explains the calculations here. 5948 5949 #ifdef _LP64 5950 // We've pushed one address, correct last_Java_sp 5951 lea(rax, Address(rsp, wordSize)); 5952 #else 5953 lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize)); 5954 #endif // LP64 5955 5956 call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions); 5957 5958 } 5959 5960 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) { 5961 call_VM_leaf_base(entry_point, number_of_arguments); 5962 } 5963 5964 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) { 5965 pass_arg0(this, arg_0); 5966 call_VM_leaf(entry_point, 1); 5967 } 5968 5969 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) { 5970 5971 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg")); 5972 pass_arg1(this, arg_1); 5973 pass_arg0(this, arg_0); 5974 call_VM_leaf(entry_point, 2); 5975 } 5976 5977 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) { 5978 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg")); 5979 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); 5980 pass_arg2(this, arg_2); 5981 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg")); 5982 pass_arg1(this, arg_1); 5983 pass_arg0(this, arg_0); 5984 call_VM_leaf(entry_point, 3); 5985 } 5986 5987 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { 5988 } 5989 5990 void MacroAssembler::check_and_handle_popframe(Register java_thread) { 5991 } 5992 5993 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) { 5994 if (reachable(src1)) { 5995 cmpl(as_Address(src1), imm); 5996 } else { 5997 lea(rscratch1, src1); 5998 cmpl(Address(rscratch1, 0), imm); 5999 } 6000 } 6001 6002 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) { 6003 assert(!src2.is_lval(), "use cmpptr"); 6004 if (reachable(src2)) { 6005 cmpl(src1, as_Address(src2)); 6006 } else { 6007 lea(rscratch1, src2); 6008 cmpl(src1, Address(rscratch1, 0)); 6009 } 6010 } 6011 6012 void MacroAssembler::cmp32(Register src1, int32_t imm) { 6013 Assembler::cmpl(src1, imm); 6014 } 6015 6016 void MacroAssembler::cmp32(Register src1, Address src2) { 6017 Assembler::cmpl(src1, src2); 6018 } 6019 6020 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { 6021 ucomisd(opr1, opr2); 6022 6023 Label L; 6024 if (unordered_is_less) { 6025 movl(dst, -1); 6026 jcc(Assembler::parity, L); 6027 jcc(Assembler::below , L); 6028 movl(dst, 0); 6029 jcc(Assembler::equal , L); 6030 increment(dst); 6031 } else { // unordered is greater 6032 movl(dst, 1); 6033 jcc(Assembler::parity, L); 6034 jcc(Assembler::above , L); 6035 movl(dst, 0); 6036 jcc(Assembler::equal , L); 6037 decrementl(dst); 6038 } 6039 bind(L); 6040 } 6041 6042 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { 6043 ucomiss(opr1, opr2); 6044 6045 Label L; 6046 if (unordered_is_less) { 6047 movl(dst, -1); 6048 jcc(Assembler::parity, L); 6049 jcc(Assembler::below , L); 6050 movl(dst, 0); 6051 jcc(Assembler::equal , L); 6052 increment(dst); 6053 } else { // unordered is greater 6054 movl(dst, 1); 6055 jcc(Assembler::parity, L); 6056 jcc(Assembler::above , L); 6057 movl(dst, 0); 6058 jcc(Assembler::equal , L); 6059 decrementl(dst); 6060 } 6061 bind(L); 6062 } 6063 6064 6065 void MacroAssembler::cmp8(AddressLiteral src1, int imm) { 6066 if (reachable(src1)) { 6067 cmpb(as_Address(src1), imm); 6068 } else { 6069 lea(rscratch1, src1); 6070 cmpb(Address(rscratch1, 0), imm); 6071 } 6072 } 6073 6074 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) { 6075 #ifdef _LP64 6076 if (src2.is_lval()) { 6077 movptr(rscratch1, src2); 6078 Assembler::cmpq(src1, rscratch1); 6079 } else if (reachable(src2)) { 6080 cmpq(src1, as_Address(src2)); 6081 } else { 6082 lea(rscratch1, src2); 6083 Assembler::cmpq(src1, Address(rscratch1, 0)); 6084 } 6085 #else 6086 if (src2.is_lval()) { 6087 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); 6088 } else { 6089 cmpl(src1, as_Address(src2)); 6090 } 6091 #endif // _LP64 6092 } 6093 6094 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) { 6095 assert(src2.is_lval(), "not a mem-mem compare"); 6096 #ifdef _LP64 6097 // moves src2's literal address 6098 movptr(rscratch1, src2); 6099 Assembler::cmpq(src1, rscratch1); 6100 #else 6101 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); 6102 #endif // _LP64 6103 } 6104 6105 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr) { 6106 if (reachable(adr)) { 6107 if (os::is_MP()) 6108 lock(); 6109 cmpxchgptr(reg, as_Address(adr)); 6110 } else { 6111 lea(rscratch1, adr); 6112 if (os::is_MP()) 6113 lock(); 6114 cmpxchgptr(reg, Address(rscratch1, 0)); 6115 } 6116 } 6117 6118 void MacroAssembler::cmpxchgptr(Register reg, Address adr) { 6119 LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr)); 6120 } 6121 6122 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) { 6123 if (reachable(src)) { 6124 comisd(dst, as_Address(src)); 6125 } else { 6126 lea(rscratch1, src); 6127 comisd(dst, Address(rscratch1, 0)); 6128 } 6129 } 6130 6131 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) { 6132 if (reachable(src)) { 6133 comiss(dst, as_Address(src)); 6134 } else { 6135 lea(rscratch1, src); 6136 comiss(dst, Address(rscratch1, 0)); 6137 } 6138 } 6139 6140 6141 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) { 6142 Condition negated_cond = negate_condition(cond); 6143 Label L; 6144 jcc(negated_cond, L); 6145 atomic_incl(counter_addr); 6146 bind(L); 6147 } 6148 6149 int MacroAssembler::corrected_idivl(Register reg) { 6150 // Full implementation of Java idiv and irem; checks for 6151 // special case as described in JVM spec., p.243 & p.271. 6152 // The function returns the (pc) offset of the idivl 6153 // instruction - may be needed for implicit exceptions. 6154 // 6155 // normal case special case 6156 // 6157 // input : rax,: dividend min_int 6158 // reg: divisor (may not be rax,/rdx) -1 6159 // 6160 // output: rax,: quotient (= rax, idiv reg) min_int 6161 // rdx: remainder (= rax, irem reg) 0 6162 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register"); 6163 const int min_int = 0x80000000; 6164 Label normal_case, special_case; 6165 6166 // check for special case 6167 cmpl(rax, min_int); 6168 jcc(Assembler::notEqual, normal_case); 6169 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0) 6170 cmpl(reg, -1); 6171 jcc(Assembler::equal, special_case); 6172 6173 // handle normal case 6174 bind(normal_case); 6175 cdql(); 6176 int idivl_offset = offset(); 6177 idivl(reg); 6178 6179 // normal and special case exit 6180 bind(special_case); 6181 6182 return idivl_offset; 6183 } 6184 6185 6186 6187 void MacroAssembler::decrementl(Register reg, int value) { 6188 if (value == min_jint) {subl(reg, value) ; return; } 6189 if (value < 0) { incrementl(reg, -value); return; } 6190 if (value == 0) { ; return; } 6191 if (value == 1 && UseIncDec) { decl(reg) ; return; } 6192 /* else */ { subl(reg, value) ; return; } 6193 } 6194 6195 void MacroAssembler::decrementl(Address dst, int value) { 6196 if (value == min_jint) {subl(dst, value) ; return; } 6197 if (value < 0) { incrementl(dst, -value); return; } 6198 if (value == 0) { ; return; } 6199 if (value == 1 && UseIncDec) { decl(dst) ; return; } 6200 /* else */ { subl(dst, value) ; return; } 6201 } 6202 6203 void MacroAssembler::division_with_shift (Register reg, int shift_value) { 6204 assert (shift_value > 0, "illegal shift value"); 6205 Label _is_positive; 6206 testl (reg, reg); 6207 jcc (Assembler::positive, _is_positive); 6208 int offset = (1 << shift_value) - 1 ; 6209 6210 if (offset == 1) { 6211 incrementl(reg); 6212 } else { 6213 addl(reg, offset); 6214 } 6215 6216 bind (_is_positive); 6217 sarl(reg, shift_value); 6218 } 6219 6220 // !defined(COMPILER2) is because of stupid core builds 6221 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) 6222 void MacroAssembler::empty_FPU_stack() { 6223 if (VM_Version::supports_mmx()) { 6224 emms(); 6225 } else { 6226 for (int i = 8; i-- > 0; ) ffree(i); 6227 } 6228 } 6229 #endif // !LP64 || C1 || !C2 6230 6231 6232 // Defines obj, preserves var_size_in_bytes 6233 void MacroAssembler::eden_allocate(Register obj, 6234 Register var_size_in_bytes, 6235 int con_size_in_bytes, 6236 Register t1, 6237 Label& slow_case) { 6238 assert(obj == rax, "obj must be in rax, for cmpxchg"); 6239 assert_different_registers(obj, var_size_in_bytes, t1); 6240 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { 6241 jmp(slow_case); 6242 } else { 6243 Register end = t1; 6244 Label retry; 6245 bind(retry); 6246 ExternalAddress heap_top((address) Universe::heap()->top_addr()); 6247 movptr(obj, heap_top); 6248 if (var_size_in_bytes == noreg) { 6249 lea(end, Address(obj, con_size_in_bytes)); 6250 } else { 6251 lea(end, Address(obj, var_size_in_bytes, Address::times_1)); 6252 } 6253 // if end < obj then we wrapped around => object too long => slow case 6254 cmpptr(end, obj); 6255 jcc(Assembler::below, slow_case); 6256 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr())); 6257 jcc(Assembler::above, slow_case); 6258 // Compare obj with the top addr, and if still equal, store the new top addr in 6259 // end at the address of the top addr pointer. Sets ZF if was equal, and clears 6260 // it otherwise. Use lock prefix for atomicity on MPs. 6261 locked_cmpxchgptr(end, heap_top); 6262 jcc(Assembler::notEqual, retry); 6263 } 6264 } 6265 6266 void MacroAssembler::enter() { 6267 push(rbp); 6268 mov(rbp, rsp); 6269 } 6270 6271 void MacroAssembler::fcmp(Register tmp) { 6272 fcmp(tmp, 1, true, true); 6273 } 6274 6275 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) { 6276 assert(!pop_right || pop_left, "usage error"); 6277 if (VM_Version::supports_cmov()) { 6278 assert(tmp == noreg, "unneeded temp"); 6279 if (pop_left) { 6280 fucomip(index); 6281 } else { 6282 fucomi(index); 6283 } 6284 if (pop_right) { 6285 fpop(); 6286 } 6287 } else { 6288 assert(tmp != noreg, "need temp"); 6289 if (pop_left) { 6290 if (pop_right) { 6291 fcompp(); 6292 } else { 6293 fcomp(index); 6294 } 6295 } else { 6296 fcom(index); 6297 } 6298 // convert FPU condition into eflags condition via rax, 6299 save_rax(tmp); 6300 fwait(); fnstsw_ax(); 6301 sahf(); 6302 restore_rax(tmp); 6303 } 6304 // condition codes set as follows: 6305 // 6306 // CF (corresponds to C0) if x < y 6307 // PF (corresponds to C2) if unordered 6308 // ZF (corresponds to C3) if x = y 6309 } 6310 6311 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) { 6312 fcmp2int(dst, unordered_is_less, 1, true, true); 6313 } 6314 6315 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) { 6316 fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right); 6317 Label L; 6318 if (unordered_is_less) { 6319 movl(dst, -1); 6320 jcc(Assembler::parity, L); 6321 jcc(Assembler::below , L); 6322 movl(dst, 0); 6323 jcc(Assembler::equal , L); 6324 increment(dst); 6325 } else { // unordered is greater 6326 movl(dst, 1); 6327 jcc(Assembler::parity, L); 6328 jcc(Assembler::above , L); 6329 movl(dst, 0); 6330 jcc(Assembler::equal , L); 6331 decrementl(dst); 6332 } 6333 bind(L); 6334 } 6335 6336 void MacroAssembler::fld_d(AddressLiteral src) { 6337 fld_d(as_Address(src)); 6338 } 6339 6340 void MacroAssembler::fld_s(AddressLiteral src) { 6341 fld_s(as_Address(src)); 6342 } 6343 6344 void MacroAssembler::fld_x(AddressLiteral src) { 6345 Assembler::fld_x(as_Address(src)); 6346 } 6347 6348 void MacroAssembler::fldcw(AddressLiteral src) { 6349 Assembler::fldcw(as_Address(src)); 6350 } 6351 6352 void MacroAssembler::fpop() { 6353 ffree(); 6354 fincstp(); 6355 } 6356 6357 void MacroAssembler::fremr(Register tmp) { 6358 save_rax(tmp); 6359 { Label L; 6360 bind(L); 6361 fprem(); 6362 fwait(); fnstsw_ax(); 6363 #ifdef _LP64 6364 testl(rax, 0x400); 6365 jcc(Assembler::notEqual, L); 6366 #else 6367 sahf(); 6368 jcc(Assembler::parity, L); 6369 #endif // _LP64 6370 } 6371 restore_rax(tmp); 6372 // Result is in ST0. 6373 // Note: fxch & fpop to get rid of ST1 6374 // (otherwise FPU stack could overflow eventually) 6375 fxch(1); 6376 fpop(); 6377 } 6378 6379 6380 void MacroAssembler::incrementl(AddressLiteral dst) { 6381 if (reachable(dst)) { 6382 incrementl(as_Address(dst)); 6383 } else { 6384 lea(rscratch1, dst); 6385 incrementl(Address(rscratch1, 0)); 6386 } 6387 } 6388 6389 void MacroAssembler::incrementl(ArrayAddress dst) { 6390 incrementl(as_Address(dst)); 6391 } 6392 6393 void MacroAssembler::incrementl(Register reg, int value) { 6394 if (value == min_jint) {addl(reg, value) ; return; } 6395 if (value < 0) { decrementl(reg, -value); return; } 6396 if (value == 0) { ; return; } 6397 if (value == 1 && UseIncDec) { incl(reg) ; return; } 6398 /* else */ { addl(reg, value) ; return; } 6399 } 6400 6401 void MacroAssembler::incrementl(Address dst, int value) { 6402 if (value == min_jint) {addl(dst, value) ; return; } 6403 if (value < 0) { decrementl(dst, -value); return; } 6404 if (value == 0) { ; return; } 6405 if (value == 1 && UseIncDec) { incl(dst) ; return; } 6406 /* else */ { addl(dst, value) ; return; } 6407 } 6408 6409 void MacroAssembler::jump(AddressLiteral dst) { 6410 if (reachable(dst)) { 6411 jmp_literal(dst.target(), dst.rspec()); 6412 } else { 6413 lea(rscratch1, dst); 6414 jmp(rscratch1); 6415 } 6416 } 6417 6418 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) { 6419 if (reachable(dst)) { 6420 InstructionMark im(this); 6421 relocate(dst.reloc()); 6422 const int short_size = 2; 6423 const int long_size = 6; 6424 int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos); 6425 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) { 6426 // 0111 tttn #8-bit disp 6427 emit_byte(0x70 | cc); 6428 emit_byte((offs - short_size) & 0xFF); 6429 } else { 6430 // 0000 1111 1000 tttn #32-bit disp 6431 emit_byte(0x0F); 6432 emit_byte(0x80 | cc); 6433 emit_long(offs - long_size); 6434 } 6435 } else { 6436 #ifdef ASSERT 6437 warning("reversing conditional branch"); 6438 #endif /* ASSERT */ 6439 Label skip; 6440 jccb(reverse[cc], skip); 6441 lea(rscratch1, dst); 6442 Assembler::jmp(rscratch1); 6443 bind(skip); 6444 } 6445 } 6446 6447 void MacroAssembler::ldmxcsr(AddressLiteral src) { 6448 if (reachable(src)) { 6449 Assembler::ldmxcsr(as_Address(src)); 6450 } else { 6451 lea(rscratch1, src); 6452 Assembler::ldmxcsr(Address(rscratch1, 0)); 6453 } 6454 } 6455 6456 int MacroAssembler::load_signed_byte(Register dst, Address src) { 6457 int off; 6458 if (LP64_ONLY(true ||) VM_Version::is_P6()) { 6459 off = offset(); 6460 movsbl(dst, src); // movsxb 6461 } else { 6462 off = load_unsigned_byte(dst, src); 6463 shll(dst, 24); 6464 sarl(dst, 24); 6465 } 6466 return off; 6467 } 6468 6469 // Note: load_signed_short used to be called load_signed_word. 6470 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler 6471 // manual, which means 16 bits, that usage is found nowhere in HotSpot code. 6472 // The term "word" in HotSpot means a 32- or 64-bit machine word. 6473 int MacroAssembler::load_signed_short(Register dst, Address src) { 6474 int off; 6475 if (LP64_ONLY(true ||) VM_Version::is_P6()) { 6476 // This is dubious to me since it seems safe to do a signed 16 => 64 bit 6477 // version but this is what 64bit has always done. This seems to imply 6478 // that users are only using 32bits worth. 6479 off = offset(); 6480 movswl(dst, src); // movsxw 6481 } else { 6482 off = load_unsigned_short(dst, src); 6483 shll(dst, 16); 6484 sarl(dst, 16); 6485 } 6486 return off; 6487 } 6488 6489 int MacroAssembler::load_unsigned_byte(Register dst, Address src) { 6490 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, 6491 // and "3.9 Partial Register Penalties", p. 22). 6492 int off; 6493 if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) { 6494 off = offset(); 6495 movzbl(dst, src); // movzxb 6496 } else { 6497 xorl(dst, dst); 6498 off = offset(); 6499 movb(dst, src); 6500 } 6501 return off; 6502 } 6503 6504 // Note: load_unsigned_short used to be called load_unsigned_word. 6505 int MacroAssembler::load_unsigned_short(Register dst, Address src) { 6506 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, 6507 // and "3.9 Partial Register Penalties", p. 22). 6508 int off; 6509 if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) { 6510 off = offset(); 6511 movzwl(dst, src); // movzxw 6512 } else { 6513 xorl(dst, dst); 6514 off = offset(); 6515 movw(dst, src); 6516 } 6517 return off; 6518 } 6519 6520 void MacroAssembler::load_sized_value(Register dst, Address src, 6521 size_t size_in_bytes, bool is_signed) { 6522 switch (size_in_bytes) { 6523 #ifndef _LP64 6524 // For case 8, caller is responsible for manually loading 6525 // the second word into another register. 6526 case 8: movl(dst, src); break; 6527 #else 6528 case 8: movq(dst, src); break; 6529 #endif 6530 case 4: movl(dst, src); break; 6531 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break; 6532 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break; 6533 default: ShouldNotReachHere(); 6534 } 6535 } 6536 6537 void MacroAssembler::mov32(AddressLiteral dst, Register src) { 6538 if (reachable(dst)) { 6539 movl(as_Address(dst), src); 6540 } else { 6541 lea(rscratch1, dst); 6542 movl(Address(rscratch1, 0), src); 6543 } 6544 } 6545 6546 void MacroAssembler::mov32(Register dst, AddressLiteral src) { 6547 if (reachable(src)) { 6548 movl(dst, as_Address(src)); 6549 } else { 6550 lea(rscratch1, src); 6551 movl(dst, Address(rscratch1, 0)); 6552 } 6553 } 6554 6555 // C++ bool manipulation 6556 6557 void MacroAssembler::movbool(Register dst, Address src) { 6558 if(sizeof(bool) == 1) 6559 movb(dst, src); 6560 else if(sizeof(bool) == 2) 6561 movw(dst, src); 6562 else if(sizeof(bool) == 4) 6563 movl(dst, src); 6564 else 6565 // unsupported 6566 ShouldNotReachHere(); 6567 } 6568 6569 void MacroAssembler::movbool(Address dst, bool boolconst) { 6570 if(sizeof(bool) == 1) 6571 movb(dst, (int) boolconst); 6572 else if(sizeof(bool) == 2) 6573 movw(dst, (int) boolconst); 6574 else if(sizeof(bool) == 4) 6575 movl(dst, (int) boolconst); 6576 else 6577 // unsupported 6578 ShouldNotReachHere(); 6579 } 6580 6581 void MacroAssembler::movbool(Address dst, Register src) { 6582 if(sizeof(bool) == 1) 6583 movb(dst, src); 6584 else if(sizeof(bool) == 2) 6585 movw(dst, src); 6586 else if(sizeof(bool) == 4) 6587 movl(dst, src); 6588 else 6589 // unsupported 6590 ShouldNotReachHere(); 6591 } 6592 6593 void MacroAssembler::movbyte(ArrayAddress dst, int src) { 6594 movb(as_Address(dst), src); 6595 } 6596 6597 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) { 6598 if (reachable(src)) { 6599 if (UseXmmLoadAndClearUpper) { 6600 movsd (dst, as_Address(src)); 6601 } else { 6602 movlpd(dst, as_Address(src)); 6603 } 6604 } else { 6605 lea(rscratch1, src); 6606 if (UseXmmLoadAndClearUpper) { 6607 movsd (dst, Address(rscratch1, 0)); 6608 } else { 6609 movlpd(dst, Address(rscratch1, 0)); 6610 } 6611 } 6612 } 6613 6614 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) { 6615 if (reachable(src)) { 6616 movss(dst, as_Address(src)); 6617 } else { 6618 lea(rscratch1, src); 6619 movss(dst, Address(rscratch1, 0)); 6620 } 6621 } 6622 6623 void MacroAssembler::movptr(Register dst, Register src) { 6624 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); 6625 } 6626 6627 void MacroAssembler::movptr(Register dst, Address src) { 6628 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); 6629 } 6630 6631 // src should NEVER be a real pointer. Use AddressLiteral for true pointers 6632 void MacroAssembler::movptr(Register dst, intptr_t src) { 6633 LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src)); 6634 } 6635 6636 void MacroAssembler::movptr(Address dst, Register src) { 6637 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); 6638 } 6639 6640 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) { 6641 if (reachable(src)) { 6642 movss(dst, as_Address(src)); 6643 } else { 6644 lea(rscratch1, src); 6645 movss(dst, Address(rscratch1, 0)); 6646 } 6647 } 6648 6649 void MacroAssembler::null_check(Register reg, int offset) { 6650 if (needs_explicit_null_check(offset)) { 6651 // provoke OS NULL exception if reg = NULL by 6652 // accessing M[reg] w/o changing any (non-CC) registers 6653 // NOTE: cmpl is plenty here to provoke a segv 6654 cmpptr(rax, Address(reg, 0)); 6655 // Note: should probably use testl(rax, Address(reg, 0)); 6656 // may be shorter code (however, this version of 6657 // testl needs to be implemented first) 6658 } else { 6659 // nothing to do, (later) access of M[reg + offset] 6660 // will provoke OS NULL exception if reg = NULL 6661 } 6662 } 6663 6664 void MacroAssembler::os_breakpoint() { 6665 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability 6666 // (e.g., MSVC can't call ps() otherwise) 6667 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint))); 6668 } 6669 6670 void MacroAssembler::pop_CPU_state() { 6671 pop_FPU_state(); 6672 pop_IU_state(); 6673 } 6674 6675 void MacroAssembler::pop_FPU_state() { 6676 NOT_LP64(frstor(Address(rsp, 0));) 6677 LP64_ONLY(fxrstor(Address(rsp, 0));) 6678 addptr(rsp, FPUStateSizeInWords * wordSize); 6679 } 6680 6681 void MacroAssembler::pop_IU_state() { 6682 popa(); 6683 LP64_ONLY(addq(rsp, 8)); 6684 popf(); 6685 } 6686 6687 // Save Integer and Float state 6688 // Warning: Stack must be 16 byte aligned (64bit) 6689 void MacroAssembler::push_CPU_state() { 6690 push_IU_state(); 6691 push_FPU_state(); 6692 } 6693 6694 void MacroAssembler::push_FPU_state() { 6695 subptr(rsp, FPUStateSizeInWords * wordSize); 6696 #ifndef _LP64 6697 fnsave(Address(rsp, 0)); 6698 fwait(); 6699 #else 6700 fxsave(Address(rsp, 0)); 6701 #endif // LP64 6702 } 6703 6704 void MacroAssembler::push_IU_state() { 6705 // Push flags first because pusha kills them 6706 pushf(); 6707 // Make sure rsp stays 16-byte aligned 6708 LP64_ONLY(subq(rsp, 8)); 6709 pusha(); 6710 } 6711 6712 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) { 6713 // determine java_thread register 6714 if (!java_thread->is_valid()) { 6715 java_thread = rdi; 6716 get_thread(java_thread); 6717 } 6718 // we must set sp to zero to clear frame 6719 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); 6720 if (clear_fp) { 6721 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); 6722 } 6723 6724 if (clear_pc) 6725 movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); 6726 6727 } 6728 6729 void MacroAssembler::restore_rax(Register tmp) { 6730 if (tmp == noreg) pop(rax); 6731 else if (tmp != rax) mov(rax, tmp); 6732 } 6733 6734 void MacroAssembler::round_to(Register reg, int modulus) { 6735 addptr(reg, modulus - 1); 6736 andptr(reg, -modulus); 6737 } 6738 6739 void MacroAssembler::save_rax(Register tmp) { 6740 if (tmp == noreg) push(rax); 6741 else if (tmp != rax) mov(tmp, rax); 6742 } 6743 6744 // Write serialization page so VM thread can do a pseudo remote membar. 6745 // We use the current thread pointer to calculate a thread specific 6746 // offset to write to within the page. This minimizes bus traffic 6747 // due to cache line collision. 6748 void MacroAssembler::serialize_memory(Register thread, Register tmp) { 6749 movl(tmp, thread); 6750 shrl(tmp, os::get_serialize_page_shift_count()); 6751 andl(tmp, (os::vm_page_size() - sizeof(int))); 6752 6753 Address index(noreg, tmp, Address::times_1); 6754 ExternalAddress page(os::get_memory_serialize_page()); 6755 6756 // Size of store must match masking code above 6757 movl(as_Address(ArrayAddress(page, index)), tmp); 6758 } 6759 6760 // Calls to C land 6761 // 6762 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded 6763 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp 6764 // has to be reset to 0. This is required to allow proper stack traversal. 6765 void MacroAssembler::set_last_Java_frame(Register java_thread, 6766 Register last_java_sp, 6767 Register last_java_fp, 6768 address last_java_pc) { 6769 // determine java_thread register 6770 if (!java_thread->is_valid()) { 6771 java_thread = rdi; 6772 get_thread(java_thread); 6773 } 6774 // determine last_java_sp register 6775 if (!last_java_sp->is_valid()) { 6776 last_java_sp = rsp; 6777 } 6778 6779 // last_java_fp is optional 6780 6781 if (last_java_fp->is_valid()) { 6782 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp); 6783 } 6784 6785 // last_java_pc is optional 6786 6787 if (last_java_pc != NULL) { 6788 lea(Address(java_thread, 6789 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()), 6790 InternalAddress(last_java_pc)); 6791 6792 } 6793 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp); 6794 } 6795 6796 void MacroAssembler::shlptr(Register dst, int imm8) { 6797 LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8)); 6798 } 6799 6800 void MacroAssembler::shrptr(Register dst, int imm8) { 6801 LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8)); 6802 } 6803 6804 void MacroAssembler::sign_extend_byte(Register reg) { 6805 if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) { 6806 movsbl(reg, reg); // movsxb 6807 } else { 6808 shll(reg, 24); 6809 sarl(reg, 24); 6810 } 6811 } 6812 6813 void MacroAssembler::sign_extend_short(Register reg) { 6814 if (LP64_ONLY(true ||) VM_Version::is_P6()) { 6815 movswl(reg, reg); // movsxw 6816 } else { 6817 shll(reg, 16); 6818 sarl(reg, 16); 6819 } 6820 } 6821 6822 ////////////////////////////////////////////////////////////////////////////////// 6823 #ifndef SERIALGC 6824 6825 void MacroAssembler::g1_write_barrier_pre(Register obj, 6826 #ifndef _LP64 6827 Register thread, 6828 #endif 6829 Register tmp, 6830 Register tmp2, 6831 bool tosca_live) { 6832 LP64_ONLY(Register thread = r15_thread;) 6833 Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 6834 PtrQueue::byte_offset_of_active())); 6835 6836 Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 6837 PtrQueue::byte_offset_of_index())); 6838 Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() + 6839 PtrQueue::byte_offset_of_buf())); 6840 6841 6842 Label done; 6843 Label runtime; 6844 6845 // if (!marking_in_progress) goto done; 6846 if (in_bytes(PtrQueue::byte_width_of_active()) == 4) { 6847 cmpl(in_progress, 0); 6848 } else { 6849 assert(in_bytes(PtrQueue::byte_width_of_active()) == 1, "Assumption"); 6850 cmpb(in_progress, 0); 6851 } 6852 jcc(Assembler::equal, done); 6853 6854 // if (x.f == NULL) goto done; 6855 #ifdef _LP64 6856 load_heap_oop(tmp2, Address(obj, 0)); 6857 #else 6858 movptr(tmp2, Address(obj, 0)); 6859 #endif 6860 cmpptr(tmp2, (int32_t) NULL_WORD); 6861 jcc(Assembler::equal, done); 6862 6863 // Can we store original value in the thread's buffer? 6864 6865 #ifdef _LP64 6866 movslq(tmp, index); 6867 cmpq(tmp, 0); 6868 #else 6869 cmpl(index, 0); 6870 #endif 6871 jcc(Assembler::equal, runtime); 6872 #ifdef _LP64 6873 subq(tmp, wordSize); 6874 movl(index, tmp); 6875 addq(tmp, buffer); 6876 #else 6877 subl(index, wordSize); 6878 movl(tmp, buffer); 6879 addl(tmp, index); 6880 #endif 6881 movptr(Address(tmp, 0), tmp2); 6882 jmp(done); 6883 bind(runtime); 6884 // save the live input values 6885 if(tosca_live) push(rax); 6886 push(obj); 6887 #ifdef _LP64 6888 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, r15_thread); 6889 #else 6890 push(thread); 6891 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, thread); 6892 pop(thread); 6893 #endif 6894 pop(obj); 6895 if(tosca_live) pop(rax); 6896 bind(done); 6897 6898 } 6899 6900 void MacroAssembler::g1_write_barrier_post(Register store_addr, 6901 Register new_val, 6902 #ifndef _LP64 6903 Register thread, 6904 #endif 6905 Register tmp, 6906 Register tmp2) { 6907 6908 LP64_ONLY(Register thread = r15_thread;) 6909 Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 6910 PtrQueue::byte_offset_of_index())); 6911 Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() + 6912 PtrQueue::byte_offset_of_buf())); 6913 BarrierSet* bs = Universe::heap()->barrier_set(); 6914 CardTableModRefBS* ct = (CardTableModRefBS*)bs; 6915 Label done; 6916 Label runtime; 6917 6918 // Does store cross heap regions? 6919 6920 movptr(tmp, store_addr); 6921 xorptr(tmp, new_val); 6922 shrptr(tmp, HeapRegion::LogOfHRGrainBytes); 6923 jcc(Assembler::equal, done); 6924 6925 // crosses regions, storing NULL? 6926 6927 cmpptr(new_val, (int32_t) NULL_WORD); 6928 jcc(Assembler::equal, done); 6929 6930 // storing region crossing non-NULL, is card already dirty? 6931 6932 ExternalAddress cardtable((address) ct->byte_map_base); 6933 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 6934 #ifdef _LP64 6935 const Register card_addr = tmp; 6936 6937 movq(card_addr, store_addr); 6938 shrq(card_addr, CardTableModRefBS::card_shift); 6939 6940 lea(tmp2, cardtable); 6941 6942 // get the address of the card 6943 addq(card_addr, tmp2); 6944 #else 6945 const Register card_index = tmp; 6946 6947 movl(card_index, store_addr); 6948 shrl(card_index, CardTableModRefBS::card_shift); 6949 6950 Address index(noreg, card_index, Address::times_1); 6951 const Register card_addr = tmp; 6952 lea(card_addr, as_Address(ArrayAddress(cardtable, index))); 6953 #endif 6954 cmpb(Address(card_addr, 0), 0); 6955 jcc(Assembler::equal, done); 6956 6957 // storing a region crossing, non-NULL oop, card is clean. 6958 // dirty card and log. 6959 6960 movb(Address(card_addr, 0), 0); 6961 6962 cmpl(queue_index, 0); 6963 jcc(Assembler::equal, runtime); 6964 subl(queue_index, wordSize); 6965 movptr(tmp2, buffer); 6966 #ifdef _LP64 6967 movslq(rscratch1, queue_index); 6968 addq(tmp2, rscratch1); 6969 movq(Address(tmp2, 0), card_addr); 6970 #else 6971 addl(tmp2, queue_index); 6972 movl(Address(tmp2, 0), card_index); 6973 #endif 6974 jmp(done); 6975 6976 bind(runtime); 6977 // save the live input values 6978 push(store_addr); 6979 push(new_val); 6980 #ifdef _LP64 6981 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, r15_thread); 6982 #else 6983 push(thread); 6984 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread); 6985 pop(thread); 6986 #endif 6987 pop(new_val); 6988 pop(store_addr); 6989 6990 bind(done); 6991 6992 } 6993 6994 #endif // SERIALGC 6995 ////////////////////////////////////////////////////////////////////////////////// 6996 6997 6998 void MacroAssembler::store_check(Register obj) { 6999 // Does a store check for the oop in register obj. The content of 7000 // register obj is destroyed afterwards. 7001 store_check_part_1(obj); 7002 store_check_part_2(obj); 7003 } 7004 7005 void MacroAssembler::store_check(Register obj, Address dst) { 7006 store_check(obj); 7007 } 7008 7009 7010 // split the store check operation so that other instructions can be scheduled inbetween 7011 void MacroAssembler::store_check_part_1(Register obj) { 7012 BarrierSet* bs = Universe::heap()->barrier_set(); 7013 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); 7014 shrptr(obj, CardTableModRefBS::card_shift); 7015 } 7016 7017 void MacroAssembler::store_check_part_2(Register obj) { 7018 BarrierSet* bs = Universe::heap()->barrier_set(); 7019 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); 7020 CardTableModRefBS* ct = (CardTableModRefBS*)bs; 7021 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 7022 7023 // The calculation for byte_map_base is as follows: 7024 // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift); 7025 // So this essentially converts an address to a displacement and 7026 // it will never need to be relocated. On 64bit however the value may be too 7027 // large for a 32bit displacement 7028 7029 intptr_t disp = (intptr_t) ct->byte_map_base; 7030 if (is_simm32(disp)) { 7031 Address cardtable(noreg, obj, Address::times_1, disp); 7032 movb(cardtable, 0); 7033 } else { 7034 // By doing it as an ExternalAddress disp could be converted to a rip-relative 7035 // displacement and done in a single instruction given favorable mapping and 7036 // a smarter version of as_Address. Worst case it is two instructions which 7037 // is no worse off then loading disp into a register and doing as a simple 7038 // Address() as above. 7039 // We can't do as ExternalAddress as the only style since if disp == 0 we'll 7040 // assert since NULL isn't acceptable in a reloci (see 6644928). In any case 7041 // in some cases we'll get a single instruction version. 7042 7043 ExternalAddress cardtable((address)disp); 7044 Address index(noreg, obj, Address::times_1); 7045 movb(as_Address(ArrayAddress(cardtable, index)), 0); 7046 } 7047 } 7048 7049 void MacroAssembler::subptr(Register dst, int32_t imm32) { 7050 LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32)); 7051 } 7052 7053 void MacroAssembler::subptr(Register dst, Register src) { 7054 LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); 7055 } 7056 7057 void MacroAssembler::test32(Register src1, AddressLiteral src2) { 7058 // src2 must be rval 7059 7060 if (reachable(src2)) { 7061 testl(src1, as_Address(src2)); 7062 } else { 7063 lea(rscratch1, src2); 7064 testl(src1, Address(rscratch1, 0)); 7065 } 7066 } 7067 7068 // C++ bool manipulation 7069 void MacroAssembler::testbool(Register dst) { 7070 if(sizeof(bool) == 1) 7071 testb(dst, 0xff); 7072 else if(sizeof(bool) == 2) { 7073 // testw implementation needed for two byte bools 7074 ShouldNotReachHere(); 7075 } else if(sizeof(bool) == 4) 7076 testl(dst, dst); 7077 else 7078 // unsupported 7079 ShouldNotReachHere(); 7080 } 7081 7082 void MacroAssembler::testptr(Register dst, Register src) { 7083 LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src)); 7084 } 7085 7086 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes. 7087 void MacroAssembler::tlab_allocate(Register obj, 7088 Register var_size_in_bytes, 7089 int con_size_in_bytes, 7090 Register t1, 7091 Register t2, 7092 Label& slow_case) { 7093 assert_different_registers(obj, t1, t2); 7094 assert_different_registers(obj, var_size_in_bytes, t1); 7095 Register end = t2; 7096 Register thread = NOT_LP64(t1) LP64_ONLY(r15_thread); 7097 7098 verify_tlab(); 7099 7100 NOT_LP64(get_thread(thread)); 7101 7102 movptr(obj, Address(thread, JavaThread::tlab_top_offset())); 7103 if (var_size_in_bytes == noreg) { 7104 lea(end, Address(obj, con_size_in_bytes)); 7105 } else { 7106 lea(end, Address(obj, var_size_in_bytes, Address::times_1)); 7107 } 7108 cmpptr(end, Address(thread, JavaThread::tlab_end_offset())); 7109 jcc(Assembler::above, slow_case); 7110 7111 // update the tlab top pointer 7112 movptr(Address(thread, JavaThread::tlab_top_offset()), end); 7113 7114 // recover var_size_in_bytes if necessary 7115 if (var_size_in_bytes == end) { 7116 subptr(var_size_in_bytes, obj); 7117 } 7118 verify_tlab(); 7119 } 7120 7121 // Preserves rbx, and rdx. 7122 void MacroAssembler::tlab_refill(Label& retry, 7123 Label& try_eden, 7124 Label& slow_case) { 7125 Register top = rax; 7126 Register t1 = rcx; 7127 Register t2 = rsi; 7128 Register thread_reg = NOT_LP64(rdi) LP64_ONLY(r15_thread); 7129 assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx); 7130 Label do_refill, discard_tlab; 7131 7132 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { 7133 // No allocation in the shared eden. 7134 jmp(slow_case); 7135 } 7136 7137 NOT_LP64(get_thread(thread_reg)); 7138 7139 movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); 7140 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); 7141 7142 // calculate amount of free space 7143 subptr(t1, top); 7144 shrptr(t1, LogHeapWordSize); 7145 7146 // Retain tlab and allocate object in shared space if 7147 // the amount free in the tlab is too large to discard. 7148 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset()))); 7149 jcc(Assembler::lessEqual, discard_tlab); 7150 7151 // Retain 7152 // %%% yuck as movptr... 7153 movptr(t2, (int32_t) ThreadLocalAllocBuffer::refill_waste_limit_increment()); 7154 addptr(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2); 7155 if (TLABStats) { 7156 // increment number of slow_allocations 7157 addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1); 7158 } 7159 jmp(try_eden); 7160 7161 bind(discard_tlab); 7162 if (TLABStats) { 7163 // increment number of refills 7164 addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1); 7165 // accumulate wastage -- t1 is amount free in tlab 7166 addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1); 7167 } 7168 7169 // if tlab is currently allocated (top or end != null) then 7170 // fill [top, end + alignment_reserve) with array object 7171 testptr (top, top); 7172 jcc(Assembler::zero, do_refill); 7173 7174 // set up the mark word 7175 movptr(Address(top, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::prototype()->copy_set_hash(0x2)); 7176 // set the length to the remaining space 7177 subptr(t1, typeArrayOopDesc::header_size(T_INT)); 7178 addptr(t1, (int32_t)ThreadLocalAllocBuffer::alignment_reserve()); 7179 shlptr(t1, log2_intptr(HeapWordSize/sizeof(jint))); 7180 movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1); 7181 // set klass to intArrayKlass 7182 // dubious reloc why not an oop reloc? 7183 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr())); 7184 // store klass last. concurrent gcs assumes klass length is valid if 7185 // klass field is not null. 7186 store_klass(top, t1); 7187 7188 // refill the tlab with an eden allocation 7189 bind(do_refill); 7190 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); 7191 shlptr(t1, LogHeapWordSize); 7192 // add object_size ?? 7193 eden_allocate(top, t1, 0, t2, slow_case); 7194 7195 // Check that t1 was preserved in eden_allocate. 7196 #ifdef ASSERT 7197 if (UseTLAB) { 7198 Label ok; 7199 Register tsize = rsi; 7200 assert_different_registers(tsize, thread_reg, t1); 7201 push(tsize); 7202 movptr(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); 7203 shlptr(tsize, LogHeapWordSize); 7204 cmpptr(t1, tsize); 7205 jcc(Assembler::equal, ok); 7206 stop("assert(t1 != tlab size)"); 7207 should_not_reach_here(); 7208 7209 bind(ok); 7210 pop(tsize); 7211 } 7212 #endif 7213 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top); 7214 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top); 7215 addptr(top, t1); 7216 subptr(top, (int32_t)ThreadLocalAllocBuffer::alignment_reserve_in_bytes()); 7217 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top); 7218 verify_tlab(); 7219 jmp(retry); 7220 } 7221 7222 static const double pi_4 = 0.7853981633974483; 7223 7224 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) { 7225 // A hand-coded argument reduction for values in fabs(pi/4, pi/2) 7226 // was attempted in this code; unfortunately it appears that the 7227 // switch to 80-bit precision and back causes this to be 7228 // unprofitable compared with simply performing a runtime call if 7229 // the argument is out of the (-pi/4, pi/4) range. 7230 7231 Register tmp = noreg; 7232 if (!VM_Version::supports_cmov()) { 7233 // fcmp needs a temporary so preserve rbx, 7234 tmp = rbx; 7235 push(tmp); 7236 } 7237 7238 Label slow_case, done; 7239 7240 ExternalAddress pi4_adr = (address)&pi_4; 7241 if (reachable(pi4_adr)) { 7242 // x ?<= pi/4 7243 fld_d(pi4_adr); 7244 fld_s(1); // Stack: X PI/4 X 7245 fabs(); // Stack: |X| PI/4 X 7246 fcmp(tmp); 7247 jcc(Assembler::above, slow_case); 7248 7249 // fastest case: -pi/4 <= x <= pi/4 7250 switch(trig) { 7251 case 's': 7252 fsin(); 7253 break; 7254 case 'c': 7255 fcos(); 7256 break; 7257 case 't': 7258 ftan(); 7259 break; 7260 default: 7261 assert(false, "bad intrinsic"); 7262 break; 7263 } 7264 jmp(done); 7265 } 7266 7267 // slow case: runtime call 7268 bind(slow_case); 7269 // Preserve registers across runtime call 7270 pusha(); 7271 int incoming_argument_and_return_value_offset = -1; 7272 if (num_fpu_regs_in_use > 1) { 7273 // Must preserve all other FPU regs (could alternatively convert 7274 // SharedRuntime::dsin and dcos into assembly routines known not to trash 7275 // FPU state, but can not trust C compiler) 7276 NEEDS_CLEANUP; 7277 // NOTE that in this case we also push the incoming argument to 7278 // the stack and restore it later; we also use this stack slot to 7279 // hold the return value from dsin or dcos. 7280 for (int i = 0; i < num_fpu_regs_in_use; i++) { 7281 subptr(rsp, sizeof(jdouble)); 7282 fstp_d(Address(rsp, 0)); 7283 } 7284 incoming_argument_and_return_value_offset = sizeof(jdouble)*(num_fpu_regs_in_use-1); 7285 fld_d(Address(rsp, incoming_argument_and_return_value_offset)); 7286 } 7287 subptr(rsp, sizeof(jdouble)); 7288 fstp_d(Address(rsp, 0)); 7289 #ifdef _LP64 7290 movdbl(xmm0, Address(rsp, 0)); 7291 #endif // _LP64 7292 7293 // NOTE: we must not use call_VM_leaf here because that requires a 7294 // complete interpreter frame in debug mode -- same bug as 4387334 7295 // MacroAssembler::call_VM_leaf_base is perfectly safe and will 7296 // do proper 64bit abi 7297 7298 NEEDS_CLEANUP; 7299 // Need to add stack banging before this runtime call if it needs to 7300 // be taken; however, there is no generic stack banging routine at 7301 // the MacroAssembler level 7302 switch(trig) { 7303 case 's': 7304 { 7305 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), 0); 7306 } 7307 break; 7308 case 'c': 7309 { 7310 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), 0); 7311 } 7312 break; 7313 case 't': 7314 { 7315 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 0); 7316 } 7317 break; 7318 default: 7319 assert(false, "bad intrinsic"); 7320 break; 7321 } 7322 #ifdef _LP64 7323 movsd(Address(rsp, 0), xmm0); 7324 fld_d(Address(rsp, 0)); 7325 #endif // _LP64 7326 addptr(rsp, sizeof(jdouble)); 7327 if (num_fpu_regs_in_use > 1) { 7328 // Must save return value to stack and then restore entire FPU stack 7329 fstp_d(Address(rsp, incoming_argument_and_return_value_offset)); 7330 for (int i = 0; i < num_fpu_regs_in_use; i++) { 7331 fld_d(Address(rsp, 0)); 7332 addptr(rsp, sizeof(jdouble)); 7333 } 7334 } 7335 popa(); 7336 7337 // Come here with result in F-TOS 7338 bind(done); 7339 7340 if (tmp != noreg) { 7341 pop(tmp); 7342 } 7343 } 7344 7345 7346 // Look up the method for a megamorphic invokeinterface call. 7347 // The target method is determined by <intf_klass, itable_index>. 7348 // The receiver klass is in recv_klass. 7349 // On success, the result will be in method_result, and execution falls through. 7350 // On failure, execution transfers to the given label. 7351 void MacroAssembler::lookup_interface_method(Register recv_klass, 7352 Register intf_klass, 7353 RegisterOrConstant itable_index, 7354 Register method_result, 7355 Register scan_temp, 7356 Label& L_no_such_interface) { 7357 assert_different_registers(recv_klass, intf_klass, method_result, scan_temp); 7358 assert(itable_index.is_constant() || itable_index.as_register() == method_result, 7359 "caller must use same register for non-constant itable index as for method"); 7360 7361 // Compute start of first itableOffsetEntry (which is at the end of the vtable) 7362 int vtable_base = instanceKlass::vtable_start_offset() * wordSize; 7363 int itentry_off = itableMethodEntry::method_offset_in_bytes(); 7364 int scan_step = itableOffsetEntry::size() * wordSize; 7365 int vte_size = vtableEntry::size() * wordSize; 7366 Address::ScaleFactor times_vte_scale = Address::times_ptr; 7367 assert(vte_size == wordSize, "else adjust times_vte_scale"); 7368 7369 movl(scan_temp, Address(recv_klass, instanceKlass::vtable_length_offset() * wordSize)); 7370 7371 // %%% Could store the aligned, prescaled offset in the klassoop. 7372 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base)); 7373 if (HeapWordsPerLong > 1) { 7374 // Round up to align_object_offset boundary 7375 // see code for instanceKlass::start_of_itable! 7376 round_to(scan_temp, BytesPerLong); 7377 } 7378 7379 // Adjust recv_klass by scaled itable_index, so we can free itable_index. 7380 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below"); 7381 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off)); 7382 7383 // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) { 7384 // if (scan->interface() == intf) { 7385 // result = (klass + scan->offset() + itable_index); 7386 // } 7387 // } 7388 Label search, found_method; 7389 7390 for (int peel = 1; peel >= 0; peel--) { 7391 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset_in_bytes())); 7392 cmpptr(intf_klass, method_result); 7393 7394 if (peel) { 7395 jccb(Assembler::equal, found_method); 7396 } else { 7397 jccb(Assembler::notEqual, search); 7398 // (invert the test to fall through to found_method...) 7399 } 7400 7401 if (!peel) break; 7402 7403 bind(search); 7404 7405 // Check that the previous entry is non-null. A null entry means that 7406 // the receiver class doesn't implement the interface, and wasn't the 7407 // same as when the caller was compiled. 7408 testptr(method_result, method_result); 7409 jcc(Assembler::zero, L_no_such_interface); 7410 addptr(scan_temp, scan_step); 7411 } 7412 7413 bind(found_method); 7414 7415 // Got a hit. 7416 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset_in_bytes())); 7417 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1)); 7418 } 7419 7420 7421 void MacroAssembler::check_klass_subtype(Register sub_klass, 7422 Register super_klass, 7423 Register temp_reg, 7424 Label& L_success) { 7425 Label L_failure; 7426 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, NULL); 7427 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL); 7428 bind(L_failure); 7429 } 7430 7431 7432 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass, 7433 Register super_klass, 7434 Register temp_reg, 7435 Label* L_success, 7436 Label* L_failure, 7437 Label* L_slow_path, 7438 RegisterOrConstant super_check_offset) { 7439 assert_different_registers(sub_klass, super_klass, temp_reg); 7440 bool must_load_sco = (super_check_offset.constant_or_zero() == -1); 7441 if (super_check_offset.is_register()) { 7442 assert_different_registers(sub_klass, super_klass, 7443 super_check_offset.as_register()); 7444 } else if (must_load_sco) { 7445 assert(temp_reg != noreg, "supply either a temp or a register offset"); 7446 } 7447 7448 Label L_fallthrough; 7449 int label_nulls = 0; 7450 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 7451 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 7452 if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; } 7453 assert(label_nulls <= 1, "at most one NULL in the batch"); 7454 7455 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 7456 Klass::secondary_super_cache_offset_in_bytes()); 7457 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 7458 Klass::super_check_offset_offset_in_bytes()); 7459 Address super_check_offset_addr(super_klass, sco_offset); 7460 7461 // Hacked jcc, which "knows" that L_fallthrough, at least, is in 7462 // range of a jccb. If this routine grows larger, reconsider at 7463 // least some of these. 7464 #define local_jcc(assembler_cond, label) \ 7465 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \ 7466 else jcc( assembler_cond, label) /*omit semi*/ 7467 7468 // Hacked jmp, which may only be used just before L_fallthrough. 7469 #define final_jmp(label) \ 7470 if (&(label) == &L_fallthrough) { /*do nothing*/ } \ 7471 else jmp(label) /*omit semi*/ 7472 7473 // If the pointers are equal, we are done (e.g., String[] elements). 7474 // This self-check enables sharing of secondary supertype arrays among 7475 // non-primary types such as array-of-interface. Otherwise, each such 7476 // type would need its own customized SSA. 7477 // We move this check to the front of the fast path because many 7478 // type checks are in fact trivially successful in this manner, 7479 // so we get a nicely predicted branch right at the start of the check. 7480 cmpptr(sub_klass, super_klass); 7481 local_jcc(Assembler::equal, *L_success); 7482 7483 // Check the supertype display: 7484 if (must_load_sco) { 7485 // Positive movl does right thing on LP64. 7486 movl(temp_reg, super_check_offset_addr); 7487 super_check_offset = RegisterOrConstant(temp_reg); 7488 } 7489 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); 7490 cmpptr(super_klass, super_check_addr); // load displayed supertype 7491 7492 // This check has worked decisively for primary supers. 7493 // Secondary supers are sought in the super_cache ('super_cache_addr'). 7494 // (Secondary supers are interfaces and very deeply nested subtypes.) 7495 // This works in the same check above because of a tricky aliasing 7496 // between the super_cache and the primary super display elements. 7497 // (The 'super_check_addr' can address either, as the case requires.) 7498 // Note that the cache is updated below if it does not help us find 7499 // what we need immediately. 7500 // So if it was a primary super, we can just fail immediately. 7501 // Otherwise, it's the slow path for us (no success at this point). 7502 7503 if (super_check_offset.is_register()) { 7504 local_jcc(Assembler::equal, *L_success); 7505 cmpl(super_check_offset.as_register(), sc_offset); 7506 if (L_failure == &L_fallthrough) { 7507 local_jcc(Assembler::equal, *L_slow_path); 7508 } else { 7509 local_jcc(Assembler::notEqual, *L_failure); 7510 final_jmp(*L_slow_path); 7511 } 7512 } else if (super_check_offset.as_constant() == sc_offset) { 7513 // Need a slow path; fast failure is impossible. 7514 if (L_slow_path == &L_fallthrough) { 7515 local_jcc(Assembler::equal, *L_success); 7516 } else { 7517 local_jcc(Assembler::notEqual, *L_slow_path); 7518 final_jmp(*L_success); 7519 } 7520 } else { 7521 // No slow path; it's a fast decision. 7522 if (L_failure == &L_fallthrough) { 7523 local_jcc(Assembler::equal, *L_success); 7524 } else { 7525 local_jcc(Assembler::notEqual, *L_failure); 7526 final_jmp(*L_success); 7527 } 7528 } 7529 7530 bind(L_fallthrough); 7531 7532 #undef local_jcc 7533 #undef final_jmp 7534 } 7535 7536 7537 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass, 7538 Register super_klass, 7539 Register temp_reg, 7540 Register temp2_reg, 7541 Label* L_success, 7542 Label* L_failure, 7543 bool set_cond_codes) { 7544 assert_different_registers(sub_klass, super_klass, temp_reg); 7545 if (temp2_reg != noreg) 7546 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg); 7547 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg) 7548 7549 Label L_fallthrough; 7550 int label_nulls = 0; 7551 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 7552 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 7553 assert(label_nulls <= 1, "at most one NULL in the batch"); 7554 7555 // a couple of useful fields in sub_klass: 7556 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 7557 Klass::secondary_supers_offset_in_bytes()); 7558 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 7559 Klass::secondary_super_cache_offset_in_bytes()); 7560 Address secondary_supers_addr(sub_klass, ss_offset); 7561 Address super_cache_addr( sub_klass, sc_offset); 7562 7563 // Do a linear scan of the secondary super-klass chain. 7564 // This code is rarely used, so simplicity is a virtue here. 7565 // The repne_scan instruction uses fixed registers, which we must spill. 7566 // Don't worry too much about pre-existing connections with the input regs. 7567 7568 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super) 7569 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter) 7570 7571 // Get super_klass value into rax (even if it was in rdi or rcx). 7572 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false; 7573 if (super_klass != rax || UseCompressedOops) { 7574 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; } 7575 mov(rax, super_klass); 7576 } 7577 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; } 7578 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; } 7579 7580 #ifndef PRODUCT 7581 int* pst_counter = &SharedRuntime::_partial_subtype_ctr; 7582 ExternalAddress pst_counter_addr((address) pst_counter); 7583 NOT_LP64( incrementl(pst_counter_addr) ); 7584 LP64_ONLY( lea(rcx, pst_counter_addr) ); 7585 LP64_ONLY( incrementl(Address(rcx, 0)) ); 7586 #endif //PRODUCT 7587 7588 // We will consult the secondary-super array. 7589 movptr(rdi, secondary_supers_addr); 7590 // Load the array length. (Positive movl does right thing on LP64.) 7591 movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 7592 // Skip to start of data. 7593 addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 7594 7595 // Scan RCX words at [RDI] for an occurrence of RAX. 7596 // Set NZ/Z based on last compare. 7597 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does 7598 // not change flags (only scas instruction which is repeated sets flags). 7599 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found. 7600 #ifdef _LP64 7601 // This part is tricky, as values in supers array could be 32 or 64 bit wide 7602 // and we store values in objArrays always encoded, thus we need to encode 7603 // the value of rax before repne. Note that rax is dead after the repne. 7604 if (UseCompressedOops) { 7605 encode_heap_oop_not_null(rax); // Changes flags. 7606 // The superclass is never null; it would be a basic system error if a null 7607 // pointer were to sneak in here. Note that we have already loaded the 7608 // Klass::super_check_offset from the super_klass in the fast path, 7609 // so if there is a null in that register, we are already in the afterlife. 7610 testl(rax,rax); // Set Z = 0 7611 repne_scanl(); 7612 } else 7613 #endif // _LP64 7614 { 7615 testptr(rax,rax); // Set Z = 0 7616 repne_scan(); 7617 } 7618 // Unspill the temp. registers: 7619 if (pushed_rdi) pop(rdi); 7620 if (pushed_rcx) pop(rcx); 7621 if (pushed_rax) pop(rax); 7622 7623 if (set_cond_codes) { 7624 // Special hack for the AD files: rdi is guaranteed non-zero. 7625 assert(!pushed_rdi, "rdi must be left non-NULL"); 7626 // Also, the condition codes are properly set Z/NZ on succeed/failure. 7627 } 7628 7629 if (L_failure == &L_fallthrough) 7630 jccb(Assembler::notEqual, *L_failure); 7631 else jcc(Assembler::notEqual, *L_failure); 7632 7633 // Success. Cache the super we found and proceed in triumph. 7634 movptr(super_cache_addr, super_klass); 7635 7636 if (L_success != &L_fallthrough) { 7637 jmp(*L_success); 7638 } 7639 7640 #undef IS_A_TEMP 7641 7642 bind(L_fallthrough); 7643 } 7644 7645 7646 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) { 7647 ucomisd(dst, as_Address(src)); 7648 } 7649 7650 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) { 7651 ucomiss(dst, as_Address(src)); 7652 } 7653 7654 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) { 7655 if (reachable(src)) { 7656 xorpd(dst, as_Address(src)); 7657 } else { 7658 lea(rscratch1, src); 7659 xorpd(dst, Address(rscratch1, 0)); 7660 } 7661 } 7662 7663 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) { 7664 if (reachable(src)) { 7665 xorps(dst, as_Address(src)); 7666 } else { 7667 lea(rscratch1, src); 7668 xorps(dst, Address(rscratch1, 0)); 7669 } 7670 } 7671 7672 void MacroAssembler::verify_oop(Register reg, const char* s) { 7673 if (!VerifyOops) return; 7674 7675 // Pass register number to verify_oop_subroutine 7676 char* b = new char[strlen(s) + 50]; 7677 sprintf(b, "verify_oop: %s: %s", reg->name(), s); 7678 #ifdef _LP64 7679 push(rscratch1); // save r10, trashed by movptr() 7680 #endif 7681 push(rax); // save rax, 7682 push(reg); // pass register argument 7683 ExternalAddress buffer((address) b); 7684 // avoid using pushptr, as it modifies scratch registers 7685 // and our contract is not to modify anything 7686 movptr(rax, buffer.addr()); 7687 push(rax); 7688 // call indirectly to solve generation ordering problem 7689 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); 7690 call(rax); 7691 // Caller pops the arguments (oop, message) and restores rax, r10 7692 } 7693 7694 7695 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr, 7696 Register tmp, 7697 int offset) { 7698 intptr_t value = *delayed_value_addr; 7699 if (value != 0) 7700 return RegisterOrConstant(value + offset); 7701 7702 // load indirectly to solve generation ordering problem 7703 movptr(tmp, ExternalAddress((address) delayed_value_addr)); 7704 7705 #ifdef ASSERT 7706 { Label L; 7707 testptr(tmp, tmp); 7708 if (WizardMode) { 7709 jcc(Assembler::notZero, L); 7710 char* buf = new char[40]; 7711 sprintf(buf, "DelayedValue="INTPTR_FORMAT, delayed_value_addr[1]); 7712 stop(buf); 7713 } else { 7714 jccb(Assembler::notZero, L); 7715 hlt(); 7716 } 7717 bind(L); 7718 } 7719 #endif 7720 7721 if (offset != 0) 7722 addptr(tmp, offset); 7723 7724 return RegisterOrConstant(tmp); 7725 } 7726 7727 7728 // registers on entry: 7729 // - rax ('check' register): required MethodType 7730 // - rcx: method handle 7731 // - rdx, rsi, or ?: killable temp 7732 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg, 7733 Register temp_reg, 7734 Label& wrong_method_type) { 7735 Address type_addr(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)); 7736 // compare method type against that of the receiver 7737 if (UseCompressedOops) { 7738 load_heap_oop(temp_reg, type_addr); 7739 cmpptr(mtype_reg, temp_reg); 7740 } else { 7741 cmpptr(mtype_reg, type_addr); 7742 } 7743 jcc(Assembler::notEqual, wrong_method_type); 7744 } 7745 7746 7747 // A method handle has a "vmslots" field which gives the size of its 7748 // argument list in JVM stack slots. This field is either located directly 7749 // in every method handle, or else is indirectly accessed through the 7750 // method handle's MethodType. This macro hides the distinction. 7751 void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, 7752 Register temp_reg) { 7753 assert_different_registers(vmslots_reg, mh_reg, temp_reg); 7754 // load mh.type.form.vmslots 7755 if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) { 7756 // hoist vmslots into every mh to avoid dependent load chain 7757 movl(vmslots_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg))); 7758 } else { 7759 Register temp2_reg = vmslots_reg; 7760 load_heap_oop(temp2_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg))); 7761 load_heap_oop(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg))); 7762 movl(vmslots_reg, Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg))); 7763 } 7764 } 7765 7766 7767 // registers on entry: 7768 // - rcx: method handle 7769 // - rdx: killable temp (interpreted only) 7770 // - rax: killable temp (compiled only) 7771 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) { 7772 assert(mh_reg == rcx, "caller must put MH object in rcx"); 7773 assert_different_registers(mh_reg, temp_reg); 7774 7775 // pick out the interpreted side of the handler 7776 // NOTE: vmentry is not an oop! 7777 movptr(temp_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg))); 7778 7779 // off we go... 7780 jmp(Address(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes())); 7781 7782 // for the various stubs which take control at this point, 7783 // see MethodHandles::generate_method_handle_stub 7784 } 7785 7786 7787 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot, 7788 int extra_slot_offset) { 7789 // cf. TemplateTable::prepare_invoke(), if (load_receiver). 7790 int stackElementSize = Interpreter::stackElementSize; 7791 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0); 7792 #ifdef ASSERT 7793 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1); 7794 assert(offset1 - offset == stackElementSize, "correct arithmetic"); 7795 #endif 7796 Register scale_reg = noreg; 7797 Address::ScaleFactor scale_factor = Address::no_scale; 7798 if (arg_slot.is_constant()) { 7799 offset += arg_slot.as_constant() * stackElementSize; 7800 } else { 7801 scale_reg = arg_slot.as_register(); 7802 scale_factor = Address::times(stackElementSize); 7803 } 7804 offset += wordSize; // return PC is on stack 7805 return Address(rsp, scale_reg, scale_factor, offset); 7806 } 7807 7808 7809 void MacroAssembler::verify_oop_addr(Address addr, const char* s) { 7810 if (!VerifyOops) return; 7811 7812 // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord); 7813 // Pass register number to verify_oop_subroutine 7814 char* b = new char[strlen(s) + 50]; 7815 sprintf(b, "verify_oop_addr: %s", s); 7816 7817 #ifdef _LP64 7818 push(rscratch1); // save r10, trashed by movptr() 7819 #endif 7820 push(rax); // save rax, 7821 // addr may contain rsp so we will have to adjust it based on the push 7822 // we just did 7823 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which 7824 // stores rax into addr which is backwards of what was intended. 7825 if (addr.uses(rsp)) { 7826 lea(rax, addr); 7827 pushptr(Address(rax, BytesPerWord)); 7828 } else { 7829 pushptr(addr); 7830 } 7831 7832 ExternalAddress buffer((address) b); 7833 // pass msg argument 7834 // avoid using pushptr, as it modifies scratch registers 7835 // and our contract is not to modify anything 7836 movptr(rax, buffer.addr()); 7837 push(rax); 7838 7839 // call indirectly to solve generation ordering problem 7840 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); 7841 call(rax); 7842 // Caller pops the arguments (addr, message) and restores rax, r10. 7843 } 7844 7845 void MacroAssembler::verify_tlab() { 7846 #ifdef ASSERT 7847 if (UseTLAB && VerifyOops) { 7848 Label next, ok; 7849 Register t1 = rsi; 7850 Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread); 7851 7852 push(t1); 7853 NOT_LP64(push(thread_reg)); 7854 NOT_LP64(get_thread(thread_reg)); 7855 7856 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); 7857 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset()))); 7858 jcc(Assembler::aboveEqual, next); 7859 stop("assert(top >= start)"); 7860 should_not_reach_here(); 7861 7862 bind(next); 7863 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); 7864 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); 7865 jcc(Assembler::aboveEqual, ok); 7866 stop("assert(top <= end)"); 7867 should_not_reach_here(); 7868 7869 bind(ok); 7870 NOT_LP64(pop(thread_reg)); 7871 pop(t1); 7872 } 7873 #endif 7874 } 7875 7876 class ControlWord { 7877 public: 7878 int32_t _value; 7879 7880 int rounding_control() const { return (_value >> 10) & 3 ; } 7881 int precision_control() const { return (_value >> 8) & 3 ; } 7882 bool precision() const { return ((_value >> 5) & 1) != 0; } 7883 bool underflow() const { return ((_value >> 4) & 1) != 0; } 7884 bool overflow() const { return ((_value >> 3) & 1) != 0; } 7885 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } 7886 bool denormalized() const { return ((_value >> 1) & 1) != 0; } 7887 bool invalid() const { return ((_value >> 0) & 1) != 0; } 7888 7889 void print() const { 7890 // rounding control 7891 const char* rc; 7892 switch (rounding_control()) { 7893 case 0: rc = "round near"; break; 7894 case 1: rc = "round down"; break; 7895 case 2: rc = "round up "; break; 7896 case 3: rc = "chop "; break; 7897 }; 7898 // precision control 7899 const char* pc; 7900 switch (precision_control()) { 7901 case 0: pc = "24 bits "; break; 7902 case 1: pc = "reserved"; break; 7903 case 2: pc = "53 bits "; break; 7904 case 3: pc = "64 bits "; break; 7905 }; 7906 // flags 7907 char f[9]; 7908 f[0] = ' '; 7909 f[1] = ' '; 7910 f[2] = (precision ()) ? 'P' : 'p'; 7911 f[3] = (underflow ()) ? 'U' : 'u'; 7912 f[4] = (overflow ()) ? 'O' : 'o'; 7913 f[5] = (zero_divide ()) ? 'Z' : 'z'; 7914 f[6] = (denormalized()) ? 'D' : 'd'; 7915 f[7] = (invalid ()) ? 'I' : 'i'; 7916 f[8] = '\x0'; 7917 // output 7918 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc); 7919 } 7920 7921 }; 7922 7923 class StatusWord { 7924 public: 7925 int32_t _value; 7926 7927 bool busy() const { return ((_value >> 15) & 1) != 0; } 7928 bool C3() const { return ((_value >> 14) & 1) != 0; } 7929 bool C2() const { return ((_value >> 10) & 1) != 0; } 7930 bool C1() const { return ((_value >> 9) & 1) != 0; } 7931 bool C0() const { return ((_value >> 8) & 1) != 0; } 7932 int top() const { return (_value >> 11) & 7 ; } 7933 bool error_status() const { return ((_value >> 7) & 1) != 0; } 7934 bool stack_fault() const { return ((_value >> 6) & 1) != 0; } 7935 bool precision() const { return ((_value >> 5) & 1) != 0; } 7936 bool underflow() const { return ((_value >> 4) & 1) != 0; } 7937 bool overflow() const { return ((_value >> 3) & 1) != 0; } 7938 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } 7939 bool denormalized() const { return ((_value >> 1) & 1) != 0; } 7940 bool invalid() const { return ((_value >> 0) & 1) != 0; } 7941 7942 void print() const { 7943 // condition codes 7944 char c[5]; 7945 c[0] = (C3()) ? '3' : '-'; 7946 c[1] = (C2()) ? '2' : '-'; 7947 c[2] = (C1()) ? '1' : '-'; 7948 c[3] = (C0()) ? '0' : '-'; 7949 c[4] = '\x0'; 7950 // flags 7951 char f[9]; 7952 f[0] = (error_status()) ? 'E' : '-'; 7953 f[1] = (stack_fault ()) ? 'S' : '-'; 7954 f[2] = (precision ()) ? 'P' : '-'; 7955 f[3] = (underflow ()) ? 'U' : '-'; 7956 f[4] = (overflow ()) ? 'O' : '-'; 7957 f[5] = (zero_divide ()) ? 'Z' : '-'; 7958 f[6] = (denormalized()) ? 'D' : '-'; 7959 f[7] = (invalid ()) ? 'I' : '-'; 7960 f[8] = '\x0'; 7961 // output 7962 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top()); 7963 } 7964 7965 }; 7966 7967 class TagWord { 7968 public: 7969 int32_t _value; 7970 7971 int tag_at(int i) const { return (_value >> (i*2)) & 3; } 7972 7973 void print() const { 7974 printf("%04x", _value & 0xFFFF); 7975 } 7976 7977 }; 7978 7979 class FPU_Register { 7980 public: 7981 int32_t _m0; 7982 int32_t _m1; 7983 int16_t _ex; 7984 7985 bool is_indefinite() const { 7986 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0; 7987 } 7988 7989 void print() const { 7990 char sign = (_ex < 0) ? '-' : '+'; 7991 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " "; 7992 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind); 7993 }; 7994 7995 }; 7996 7997 class FPU_State { 7998 public: 7999 enum { 8000 register_size = 10, 8001 number_of_registers = 8, 8002 register_mask = 7 8003 }; 8004 8005 ControlWord _control_word; 8006 StatusWord _status_word; 8007 TagWord _tag_word; 8008 int32_t _error_offset; 8009 int32_t _error_selector; 8010 int32_t _data_offset; 8011 int32_t _data_selector; 8012 int8_t _register[register_size * number_of_registers]; 8013 8014 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); } 8015 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; } 8016 8017 const char* tag_as_string(int tag) const { 8018 switch (tag) { 8019 case 0: return "valid"; 8020 case 1: return "zero"; 8021 case 2: return "special"; 8022 case 3: return "empty"; 8023 } 8024 ShouldNotReachHere(); 8025 return NULL; 8026 } 8027 8028 void print() const { 8029 // print computation registers 8030 { int t = _status_word.top(); 8031 for (int i = 0; i < number_of_registers; i++) { 8032 int j = (i - t) & register_mask; 8033 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j); 8034 st(j)->print(); 8035 printf(" %s\n", tag_as_string(_tag_word.tag_at(i))); 8036 } 8037 } 8038 printf("\n"); 8039 // print control registers 8040 printf("ctrl = "); _control_word.print(); printf("\n"); 8041 printf("stat = "); _status_word .print(); printf("\n"); 8042 printf("tags = "); _tag_word .print(); printf("\n"); 8043 } 8044 8045 }; 8046 8047 class Flag_Register { 8048 public: 8049 int32_t _value; 8050 8051 bool overflow() const { return ((_value >> 11) & 1) != 0; } 8052 bool direction() const { return ((_value >> 10) & 1) != 0; } 8053 bool sign() const { return ((_value >> 7) & 1) != 0; } 8054 bool zero() const { return ((_value >> 6) & 1) != 0; } 8055 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; } 8056 bool parity() const { return ((_value >> 2) & 1) != 0; } 8057 bool carry() const { return ((_value >> 0) & 1) != 0; } 8058 8059 void print() const { 8060 // flags 8061 char f[8]; 8062 f[0] = (overflow ()) ? 'O' : '-'; 8063 f[1] = (direction ()) ? 'D' : '-'; 8064 f[2] = (sign ()) ? 'S' : '-'; 8065 f[3] = (zero ()) ? 'Z' : '-'; 8066 f[4] = (auxiliary_carry()) ? 'A' : '-'; 8067 f[5] = (parity ()) ? 'P' : '-'; 8068 f[6] = (carry ()) ? 'C' : '-'; 8069 f[7] = '\x0'; 8070 // output 8071 printf("%08x flags = %s", _value, f); 8072 } 8073 8074 }; 8075 8076 class IU_Register { 8077 public: 8078 int32_t _value; 8079 8080 void print() const { 8081 printf("%08x %11d", _value, _value); 8082 } 8083 8084 }; 8085 8086 class IU_State { 8087 public: 8088 Flag_Register _eflags; 8089 IU_Register _rdi; 8090 IU_Register _rsi; 8091 IU_Register _rbp; 8092 IU_Register _rsp; 8093 IU_Register _rbx; 8094 IU_Register _rdx; 8095 IU_Register _rcx; 8096 IU_Register _rax; 8097 8098 void print() const { 8099 // computation registers 8100 printf("rax, = "); _rax.print(); printf("\n"); 8101 printf("rbx, = "); _rbx.print(); printf("\n"); 8102 printf("rcx = "); _rcx.print(); printf("\n"); 8103 printf("rdx = "); _rdx.print(); printf("\n"); 8104 printf("rdi = "); _rdi.print(); printf("\n"); 8105 printf("rsi = "); _rsi.print(); printf("\n"); 8106 printf("rbp, = "); _rbp.print(); printf("\n"); 8107 printf("rsp = "); _rsp.print(); printf("\n"); 8108 printf("\n"); 8109 // control registers 8110 printf("flgs = "); _eflags.print(); printf("\n"); 8111 } 8112 }; 8113 8114 8115 class CPU_State { 8116 public: 8117 FPU_State _fpu_state; 8118 IU_State _iu_state; 8119 8120 void print() const { 8121 printf("--------------------------------------------------\n"); 8122 _iu_state .print(); 8123 printf("\n"); 8124 _fpu_state.print(); 8125 printf("--------------------------------------------------\n"); 8126 } 8127 8128 }; 8129 8130 8131 static void _print_CPU_state(CPU_State* state) { 8132 state->print(); 8133 }; 8134 8135 8136 void MacroAssembler::print_CPU_state() { 8137 push_CPU_state(); 8138 push(rsp); // pass CPU state 8139 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state))); 8140 addptr(rsp, wordSize); // discard argument 8141 pop_CPU_state(); 8142 } 8143 8144 8145 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) { 8146 static int counter = 0; 8147 FPU_State* fs = &state->_fpu_state; 8148 counter++; 8149 // For leaf calls, only verify that the top few elements remain empty. 8150 // We only need 1 empty at the top for C2 code. 8151 if( stack_depth < 0 ) { 8152 if( fs->tag_for_st(7) != 3 ) { 8153 printf("FPR7 not empty\n"); 8154 state->print(); 8155 assert(false, "error"); 8156 return false; 8157 } 8158 return true; // All other stack states do not matter 8159 } 8160 8161 assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std, 8162 "bad FPU control word"); 8163 8164 // compute stack depth 8165 int i = 0; 8166 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) < 3) i++; 8167 int d = i; 8168 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++; 8169 // verify findings 8170 if (i != FPU_State::number_of_registers) { 8171 // stack not contiguous 8172 printf("%s: stack not contiguous at ST%d\n", s, i); 8173 state->print(); 8174 assert(false, "error"); 8175 return false; 8176 } 8177 // check if computed stack depth corresponds to expected stack depth 8178 if (stack_depth < 0) { 8179 // expected stack depth is -stack_depth or less 8180 if (d > -stack_depth) { 8181 // too many elements on the stack 8182 printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d); 8183 state->print(); 8184 assert(false, "error"); 8185 return false; 8186 } 8187 } else { 8188 // expected stack depth is stack_depth 8189 if (d != stack_depth) { 8190 // wrong stack depth 8191 printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d); 8192 state->print(); 8193 assert(false, "error"); 8194 return false; 8195 } 8196 } 8197 // everything is cool 8198 return true; 8199 } 8200 8201 8202 void MacroAssembler::verify_FPU(int stack_depth, const char* s) { 8203 if (!VerifyFPU) return; 8204 push_CPU_state(); 8205 push(rsp); // pass CPU state 8206 ExternalAddress msg((address) s); 8207 // pass message string s 8208 pushptr(msg.addr()); 8209 push(stack_depth); // pass stack depth 8210 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU))); 8211 addptr(rsp, 3 * wordSize); // discard arguments 8212 // check for error 8213 { Label L; 8214 testl(rax, rax); 8215 jcc(Assembler::notZero, L); 8216 int3(); // break if error condition 8217 bind(L); 8218 } 8219 pop_CPU_state(); 8220 } 8221 8222 void MacroAssembler::load_klass(Register dst, Register src) { 8223 #ifdef _LP64 8224 if (UseCompressedOops) { 8225 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); 8226 decode_heap_oop_not_null(dst); 8227 } else 8228 #endif 8229 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); 8230 } 8231 8232 void MacroAssembler::load_prototype_header(Register dst, Register src) { 8233 #ifdef _LP64 8234 if (UseCompressedOops) { 8235 assert (Universe::heap() != NULL, "java heap should be initialized"); 8236 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); 8237 if (Universe::narrow_oop_shift() != 0) { 8238 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8239 if (LogMinObjAlignmentInBytes == Address::times_8) { 8240 movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 8241 } else { 8242 // OK to use shift since we don't need to preserve flags. 8243 shlq(dst, LogMinObjAlignmentInBytes); 8244 movq(dst, Address(r12_heapbase, dst, Address::times_1, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 8245 } 8246 } else { 8247 movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 8248 } 8249 } else 8250 #endif 8251 { 8252 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); 8253 movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 8254 } 8255 } 8256 8257 void MacroAssembler::store_klass(Register dst, Register src) { 8258 #ifdef _LP64 8259 if (UseCompressedOops) { 8260 encode_heap_oop_not_null(src); 8261 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src); 8262 } else 8263 #endif 8264 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src); 8265 } 8266 8267 void MacroAssembler::load_heap_oop(Register dst, Address src) { 8268 #ifdef _LP64 8269 if (UseCompressedOops) { 8270 movl(dst, src); 8271 decode_heap_oop(dst); 8272 } else 8273 #endif 8274 movptr(dst, src); 8275 } 8276 8277 void MacroAssembler::store_heap_oop(Address dst, Register src) { 8278 #ifdef _LP64 8279 if (UseCompressedOops) { 8280 assert(!dst.uses(src), "not enough registers"); 8281 encode_heap_oop(src); 8282 movl(dst, src); 8283 } else 8284 #endif 8285 movptr(dst, src); 8286 } 8287 8288 // Used for storing NULLs. 8289 void MacroAssembler::store_heap_oop_null(Address dst) { 8290 #ifdef _LP64 8291 if (UseCompressedOops) { 8292 movl(dst, (int32_t)NULL_WORD); 8293 } else { 8294 movslq(dst, (int32_t)NULL_WORD); 8295 } 8296 #else 8297 movl(dst, (int32_t)NULL_WORD); 8298 #endif 8299 } 8300 8301 #ifdef _LP64 8302 void MacroAssembler::store_klass_gap(Register dst, Register src) { 8303 if (UseCompressedOops) { 8304 // Store to klass gap in destination 8305 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src); 8306 } 8307 } 8308 8309 #ifdef ASSERT 8310 void MacroAssembler::verify_heapbase(const char* msg) { 8311 assert (UseCompressedOops, "should be compressed"); 8312 assert (Universe::heap() != NULL, "java heap should be initialized"); 8313 if (CheckCompressedOops) { 8314 Label ok; 8315 push(rscratch1); // cmpptr trashes rscratch1 8316 cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); 8317 jcc(Assembler::equal, ok); 8318 stop(msg); 8319 bind(ok); 8320 pop(rscratch1); 8321 } 8322 } 8323 #endif 8324 8325 // Algorithm must match oop.inline.hpp encode_heap_oop. 8326 void MacroAssembler::encode_heap_oop(Register r) { 8327 #ifdef ASSERT 8328 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?"); 8329 #endif 8330 verify_oop(r, "broken oop in encode_heap_oop"); 8331 if (Universe::narrow_oop_base() == NULL) { 8332 if (Universe::narrow_oop_shift() != 0) { 8333 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8334 shrq(r, LogMinObjAlignmentInBytes); 8335 } 8336 return; 8337 } 8338 testq(r, r); 8339 cmovq(Assembler::equal, r, r12_heapbase); 8340 subq(r, r12_heapbase); 8341 shrq(r, LogMinObjAlignmentInBytes); 8342 } 8343 8344 void MacroAssembler::encode_heap_oop_not_null(Register r) { 8345 #ifdef ASSERT 8346 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?"); 8347 if (CheckCompressedOops) { 8348 Label ok; 8349 testq(r, r); 8350 jcc(Assembler::notEqual, ok); 8351 stop("null oop passed to encode_heap_oop_not_null"); 8352 bind(ok); 8353 } 8354 #endif 8355 verify_oop(r, "broken oop in encode_heap_oop_not_null"); 8356 if (Universe::narrow_oop_base() != NULL) { 8357 subq(r, r12_heapbase); 8358 } 8359 if (Universe::narrow_oop_shift() != 0) { 8360 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8361 shrq(r, LogMinObjAlignmentInBytes); 8362 } 8363 } 8364 8365 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) { 8366 #ifdef ASSERT 8367 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?"); 8368 if (CheckCompressedOops) { 8369 Label ok; 8370 testq(src, src); 8371 jcc(Assembler::notEqual, ok); 8372 stop("null oop passed to encode_heap_oop_not_null2"); 8373 bind(ok); 8374 } 8375 #endif 8376 verify_oop(src, "broken oop in encode_heap_oop_not_null2"); 8377 if (dst != src) { 8378 movq(dst, src); 8379 } 8380 if (Universe::narrow_oop_base() != NULL) { 8381 subq(dst, r12_heapbase); 8382 } 8383 if (Universe::narrow_oop_shift() != 0) { 8384 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8385 shrq(dst, LogMinObjAlignmentInBytes); 8386 } 8387 } 8388 8389 void MacroAssembler::decode_heap_oop(Register r) { 8390 #ifdef ASSERT 8391 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?"); 8392 #endif 8393 if (Universe::narrow_oop_base() == NULL) { 8394 if (Universe::narrow_oop_shift() != 0) { 8395 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8396 shlq(r, LogMinObjAlignmentInBytes); 8397 } 8398 } else { 8399 Label done; 8400 shlq(r, LogMinObjAlignmentInBytes); 8401 jccb(Assembler::equal, done); 8402 addq(r, r12_heapbase); 8403 bind(done); 8404 } 8405 verify_oop(r, "broken oop in decode_heap_oop"); 8406 } 8407 8408 void MacroAssembler::decode_heap_oop_not_null(Register r) { 8409 // Note: it will change flags 8410 assert (UseCompressedOops, "should only be used for compressed headers"); 8411 assert (Universe::heap() != NULL, "java heap should be initialized"); 8412 // Cannot assert, unverified entry point counts instructions (see .ad file) 8413 // vtableStubs also counts instructions in pd_code_size_limit. 8414 // Also do not verify_oop as this is called by verify_oop. 8415 if (Universe::narrow_oop_shift() != 0) { 8416 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8417 shlq(r, LogMinObjAlignmentInBytes); 8418 if (Universe::narrow_oop_base() != NULL) { 8419 addq(r, r12_heapbase); 8420 } 8421 } else { 8422 assert (Universe::narrow_oop_base() == NULL, "sanity"); 8423 } 8424 } 8425 8426 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) { 8427 // Note: it will change flags 8428 assert (UseCompressedOops, "should only be used for compressed headers"); 8429 assert (Universe::heap() != NULL, "java heap should be initialized"); 8430 // Cannot assert, unverified entry point counts instructions (see .ad file) 8431 // vtableStubs also counts instructions in pd_code_size_limit. 8432 // Also do not verify_oop as this is called by verify_oop. 8433 if (Universe::narrow_oop_shift() != 0) { 8434 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 8435 if (LogMinObjAlignmentInBytes == Address::times_8) { 8436 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0)); 8437 } else { 8438 if (dst != src) { 8439 movq(dst, src); 8440 } 8441 shlq(dst, LogMinObjAlignmentInBytes); 8442 if (Universe::narrow_oop_base() != NULL) { 8443 addq(dst, r12_heapbase); 8444 } 8445 } 8446 } else { 8447 assert (Universe::narrow_oop_base() == NULL, "sanity"); 8448 if (dst != src) { 8449 movq(dst, src); 8450 } 8451 } 8452 } 8453 8454 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) { 8455 assert (UseCompressedOops, "should only be used for compressed headers"); 8456 assert (Universe::heap() != NULL, "java heap should be initialized"); 8457 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 8458 int oop_index = oop_recorder()->find_index(obj); 8459 RelocationHolder rspec = oop_Relocation::spec(oop_index); 8460 mov_narrow_oop(dst, oop_index, rspec); 8461 } 8462 8463 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) { 8464 assert (UseCompressedOops, "should only be used for compressed headers"); 8465 assert (Universe::heap() != NULL, "java heap should be initialized"); 8466 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 8467 int oop_index = oop_recorder()->find_index(obj); 8468 RelocationHolder rspec = oop_Relocation::spec(oop_index); 8469 mov_narrow_oop(dst, oop_index, rspec); 8470 } 8471 8472 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) { 8473 assert (UseCompressedOops, "should only be used for compressed headers"); 8474 assert (Universe::heap() != NULL, "java heap should be initialized"); 8475 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 8476 int oop_index = oop_recorder()->find_index(obj); 8477 RelocationHolder rspec = oop_Relocation::spec(oop_index); 8478 Assembler::cmp_narrow_oop(dst, oop_index, rspec); 8479 } 8480 8481 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) { 8482 assert (UseCompressedOops, "should only be used for compressed headers"); 8483 assert (Universe::heap() != NULL, "java heap should be initialized"); 8484 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 8485 int oop_index = oop_recorder()->find_index(obj); 8486 RelocationHolder rspec = oop_Relocation::spec(oop_index); 8487 Assembler::cmp_narrow_oop(dst, oop_index, rspec); 8488 } 8489 8490 void MacroAssembler::reinit_heapbase() { 8491 if (UseCompressedOops) { 8492 movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); 8493 } 8494 } 8495 #endif // _LP64 8496 8497 // IndexOf substring. 8498 void MacroAssembler::string_indexof(Register str1, Register str2, 8499 Register cnt1, Register cnt2, Register result, 8500 XMMRegister vec, Register tmp) { 8501 assert(UseSSE42Intrinsics, "SSE4.2 is required"); 8502 8503 Label RELOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR, 8504 SCAN_SUBSTR, RET_NOT_FOUND, CLEANUP; 8505 8506 push(str1); // string addr 8507 push(str2); // substr addr 8508 push(cnt2); // substr count 8509 jmpb(PREP_FOR_SCAN); 8510 8511 // Substr count saved at sp 8512 // Substr saved at sp+1*wordSize 8513 // String saved at sp+2*wordSize 8514 8515 // Reload substr for rescan 8516 bind(RELOAD_SUBSTR); 8517 movl(cnt2, Address(rsp, 0)); 8518 movptr(str2, Address(rsp, wordSize)); 8519 // We came here after the beginninig of the substring was 8520 // matched but the rest of it was not so we need to search 8521 // again. Start from the next element after the previous match. 8522 subptr(str1, result); // Restore counter 8523 shrl(str1, 1); 8524 addl(cnt1, str1); 8525 decrementl(cnt1); 8526 lea(str1, Address(result, 2)); // Reload string 8527 8528 // Load substr 8529 bind(PREP_FOR_SCAN); 8530 movdqu(vec, Address(str2, 0)); 8531 addl(cnt1, 8); // prime the loop 8532 subptr(str1, 16); 8533 8534 // Scan string for substr in 16-byte vectors 8535 bind(SCAN_TO_SUBSTR); 8536 subl(cnt1, 8); 8537 addptr(str1, 16); 8538 8539 // pcmpestri 8540 // inputs: 8541 // xmm - substring 8542 // rax - substring length (elements count) 8543 // mem - scaned string 8544 // rdx - string length (elements count) 8545 // 0xd - mode: 1100 (substring search) + 01 (unsigned shorts) 8546 // outputs: 8547 // rcx - matched index in string 8548 assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri"); 8549 8550 pcmpestri(vec, Address(str1, 0), 0x0d); 8551 jcc(Assembler::above, SCAN_TO_SUBSTR); // CF == 0 && ZF == 0 8552 jccb(Assembler::aboveEqual, RET_NOT_FOUND); // CF == 0 8553 8554 // Fallthrough: found a potential substr 8555 8556 // Make sure string is still long enough 8557 subl(cnt1, tmp); 8558 cmpl(cnt1, cnt2); 8559 jccb(Assembler::negative, RET_NOT_FOUND); 8560 // Compute start addr of substr 8561 lea(str1, Address(str1, tmp, Address::times_2)); 8562 movptr(result, str1); // save 8563 8564 // Compare potential substr 8565 addl(cnt1, 8); // prime the loop 8566 addl(cnt2, 8); 8567 subptr(str1, 16); 8568 subptr(str2, 16); 8569 8570 // Scan 16-byte vectors of string and substr 8571 bind(SCAN_SUBSTR); 8572 subl(cnt1, 8); 8573 subl(cnt2, 8); 8574 addptr(str1, 16); 8575 addptr(str2, 16); 8576 movdqu(vec, Address(str2, 0)); 8577 pcmpestri(vec, Address(str1, 0), 0x0d); 8578 jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0 8579 jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0 8580 8581 // Compute substr offset 8582 subptr(result, Address(rsp, 2*wordSize)); 8583 shrl(result, 1); // index 8584 jmpb(CLEANUP); 8585 8586 bind(RET_NOT_FOUND); 8587 movl(result, -1); 8588 8589 bind(CLEANUP); 8590 addptr(rsp, 3*wordSize); 8591 } 8592 8593 // Compare strings. 8594 void MacroAssembler::string_compare(Register str1, Register str2, 8595 Register cnt1, Register cnt2, Register result, 8596 XMMRegister vec1, XMMRegister vec2) { 8597 Label LENGTH_DIFF_LABEL, POP_LABEL, DONE_LABEL, WHILE_HEAD_LABEL; 8598 8599 // Compute the minimum of the string lengths and the 8600 // difference of the string lengths (stack). 8601 // Do the conditional move stuff 8602 movl(result, cnt1); 8603 subl(cnt1, cnt2); 8604 push(cnt1); 8605 if (VM_Version::supports_cmov()) { 8606 cmovl(Assembler::lessEqual, cnt2, result); 8607 } else { 8608 Label GT_LABEL; 8609 jccb(Assembler::greater, GT_LABEL); 8610 movl(cnt2, result); 8611 bind(GT_LABEL); 8612 } 8613 8614 // Is the minimum length zero? 8615 testl(cnt2, cnt2); 8616 jcc(Assembler::zero, LENGTH_DIFF_LABEL); 8617 8618 // Load first characters 8619 load_unsigned_short(result, Address(str1, 0)); 8620 load_unsigned_short(cnt1, Address(str2, 0)); 8621 8622 // Compare first characters 8623 subl(result, cnt1); 8624 jcc(Assembler::notZero, POP_LABEL); 8625 decrementl(cnt2); 8626 jcc(Assembler::zero, LENGTH_DIFF_LABEL); 8627 8628 { 8629 // Check after comparing first character to see if strings are equivalent 8630 Label LSkip2; 8631 // Check if the strings start at same location 8632 cmpptr(str1, str2); 8633 jccb(Assembler::notEqual, LSkip2); 8634 8635 // Check if the length difference is zero (from stack) 8636 cmpl(Address(rsp, 0), 0x0); 8637 jcc(Assembler::equal, LENGTH_DIFF_LABEL); 8638 8639 // Strings might not be equivalent 8640 bind(LSkip2); 8641 } 8642 8643 // Advance to next character 8644 addptr(str1, 2); 8645 addptr(str2, 2); 8646 8647 if (UseSSE42Intrinsics) { 8648 // With SSE4.2, use double quad vector compare 8649 Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL; 8650 // Setup to compare 16-byte vectors 8651 movl(cnt1, cnt2); 8652 andl(cnt2, 0xfffffff8); // cnt2 holds the vector count 8653 andl(cnt1, 0x00000007); // cnt1 holds the tail count 8654 testl(cnt2, cnt2); 8655 jccb(Assembler::zero, COMPARE_TAIL); 8656 8657 lea(str2, Address(str2, cnt2, Address::times_2)); 8658 lea(str1, Address(str1, cnt2, Address::times_2)); 8659 negptr(cnt2); 8660 8661 bind(COMPARE_VECTORS); 8662 movdqu(vec1, Address(str1, cnt2, Address::times_2)); 8663 movdqu(vec2, Address(str2, cnt2, Address::times_2)); 8664 pxor(vec1, vec2); 8665 ptest(vec1, vec1); 8666 jccb(Assembler::notZero, VECTOR_NOT_EQUAL); 8667 addptr(cnt2, 8); 8668 jcc(Assembler::notZero, COMPARE_VECTORS); 8669 jmpb(COMPARE_TAIL); 8670 8671 // Mismatched characters in the vectors 8672 bind(VECTOR_NOT_EQUAL); 8673 lea(str1, Address(str1, cnt2, Address::times_2)); 8674 lea(str2, Address(str2, cnt2, Address::times_2)); 8675 movl(cnt1, 8); 8676 8677 // Compare tail (< 8 chars), or rescan last vectors to 8678 // find 1st mismatched characters 8679 bind(COMPARE_TAIL); 8680 testl(cnt1, cnt1); 8681 jccb(Assembler::zero, LENGTH_DIFF_LABEL); 8682 movl(cnt2, cnt1); 8683 // Fallthru to tail compare 8684 } 8685 8686 // Shift str2 and str1 to the end of the arrays, negate min 8687 lea(str1, Address(str1, cnt2, Address::times_2, 0)); 8688 lea(str2, Address(str2, cnt2, Address::times_2, 0)); 8689 negptr(cnt2); 8690 8691 // Compare the rest of the characters 8692 bind(WHILE_HEAD_LABEL); 8693 load_unsigned_short(result, Address(str1, cnt2, Address::times_2, 0)); 8694 load_unsigned_short(cnt1, Address(str2, cnt2, Address::times_2, 0)); 8695 subl(result, cnt1); 8696 jccb(Assembler::notZero, POP_LABEL); 8697 increment(cnt2); 8698 jcc(Assembler::notZero, WHILE_HEAD_LABEL); 8699 8700 // Strings are equal up to min length. Return the length difference. 8701 bind(LENGTH_DIFF_LABEL); 8702 pop(result); 8703 jmpb(DONE_LABEL); 8704 8705 // Discard the stored length difference 8706 bind(POP_LABEL); 8707 addptr(rsp, wordSize); 8708 8709 // That's it 8710 bind(DONE_LABEL); 8711 } 8712 8713 // Compare char[] arrays aligned to 4 bytes or substrings. 8714 void MacroAssembler::char_arrays_equals(bool is_array_equ, Register ary1, Register ary2, 8715 Register limit, Register result, Register chr, 8716 XMMRegister vec1, XMMRegister vec2) { 8717 Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR; 8718 8719 int length_offset = arrayOopDesc::length_offset_in_bytes(); 8720 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR); 8721 8722 // Check the input args 8723 cmpptr(ary1, ary2); 8724 jcc(Assembler::equal, TRUE_LABEL); 8725 8726 if (is_array_equ) { 8727 // Need additional checks for arrays_equals. 8728 testptr(ary1, ary1); 8729 jcc(Assembler::zero, FALSE_LABEL); 8730 testptr(ary2, ary2); 8731 jcc(Assembler::zero, FALSE_LABEL); 8732 8733 // Check the lengths 8734 movl(limit, Address(ary1, length_offset)); 8735 cmpl(limit, Address(ary2, length_offset)); 8736 jcc(Assembler::notEqual, FALSE_LABEL); 8737 } 8738 8739 // count == 0 8740 testl(limit, limit); 8741 jcc(Assembler::zero, TRUE_LABEL); 8742 8743 if (is_array_equ) { 8744 // Load array address 8745 lea(ary1, Address(ary1, base_offset)); 8746 lea(ary2, Address(ary2, base_offset)); 8747 } 8748 8749 shll(limit, 1); // byte count != 0 8750 movl(result, limit); // copy 8751 8752 if (UseSSE42Intrinsics) { 8753 // With SSE4.2, use double quad vector compare 8754 Label COMPARE_WIDE_VECTORS, COMPARE_TAIL; 8755 // Compare 16-byte vectors 8756 andl(result, 0x0000000e); // tail count (in bytes) 8757 andl(limit, 0xfffffff0); // vector count (in bytes) 8758 jccb(Assembler::zero, COMPARE_TAIL); 8759 8760 lea(ary1, Address(ary1, limit, Address::times_1)); 8761 lea(ary2, Address(ary2, limit, Address::times_1)); 8762 negptr(limit); 8763 8764 bind(COMPARE_WIDE_VECTORS); 8765 movdqu(vec1, Address(ary1, limit, Address::times_1)); 8766 movdqu(vec2, Address(ary2, limit, Address::times_1)); 8767 pxor(vec1, vec2); 8768 ptest(vec1, vec1); 8769 jccb(Assembler::notZero, FALSE_LABEL); 8770 addptr(limit, 16); 8771 jcc(Assembler::notZero, COMPARE_WIDE_VECTORS); 8772 8773 bind(COMPARE_TAIL); // limit is zero 8774 movl(limit, result); 8775 // Fallthru to tail compare 8776 } 8777 8778 // Compare 4-byte vectors 8779 andl(limit, 0xfffffffc); // vector count (in bytes) 8780 jccb(Assembler::zero, COMPARE_CHAR); 8781 8782 lea(ary1, Address(ary1, limit, Address::times_1)); 8783 lea(ary2, Address(ary2, limit, Address::times_1)); 8784 negptr(limit); 8785 8786 bind(COMPARE_VECTORS); 8787 movl(chr, Address(ary1, limit, Address::times_1)); 8788 cmpl(chr, Address(ary2, limit, Address::times_1)); 8789 jccb(Assembler::notEqual, FALSE_LABEL); 8790 addptr(limit, 4); 8791 jcc(Assembler::notZero, COMPARE_VECTORS); 8792 8793 // Compare trailing char (final 2 bytes), if any 8794 bind(COMPARE_CHAR); 8795 testl(result, 0x2); // tail char 8796 jccb(Assembler::zero, TRUE_LABEL); 8797 load_unsigned_short(chr, Address(ary1, 0)); 8798 load_unsigned_short(limit, Address(ary2, 0)); 8799 cmpl(chr, limit); 8800 jccb(Assembler::notEqual, FALSE_LABEL); 8801 8802 bind(TRUE_LABEL); 8803 movl(result, 1); // return true 8804 jmpb(DONE); 8805 8806 bind(FALSE_LABEL); 8807 xorl(result, result); // return false 8808 8809 // That's it 8810 bind(DONE); 8811 } 8812 8813 #ifdef PRODUCT 8814 #define BLOCK_COMMENT(str) /* nothing */ 8815 #else 8816 #define BLOCK_COMMENT(str) block_comment(str) 8817 #endif 8818 8819 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 8820 void MacroAssembler::generate_fill(BasicType t, bool aligned, 8821 Register to, Register value, Register count, 8822 Register rtmp, XMMRegister xtmp) { 8823 assert_different_registers(to, value, count, rtmp); 8824 Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte; 8825 Label L_fill_2_bytes, L_fill_4_bytes; 8826 8827 int shift = -1; 8828 switch (t) { 8829 case T_BYTE: 8830 shift = 2; 8831 break; 8832 case T_SHORT: 8833 shift = 1; 8834 break; 8835 case T_INT: 8836 shift = 0; 8837 break; 8838 default: ShouldNotReachHere(); 8839 } 8840 8841 if (t == T_BYTE) { 8842 andl(value, 0xff); 8843 movl(rtmp, value); 8844 shll(rtmp, 8); 8845 orl(value, rtmp); 8846 } 8847 if (t == T_SHORT) { 8848 andl(value, 0xffff); 8849 } 8850 if (t == T_BYTE || t == T_SHORT) { 8851 movl(rtmp, value); 8852 shll(rtmp, 16); 8853 orl(value, rtmp); 8854 } 8855 8856 cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element 8857 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp 8858 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { 8859 // align source address at 4 bytes address boundary 8860 if (t == T_BYTE) { 8861 // One byte misalignment happens only for byte arrays 8862 testptr(to, 1); 8863 jccb(Assembler::zero, L_skip_align1); 8864 movb(Address(to, 0), value); 8865 increment(to); 8866 decrement(count); 8867 BIND(L_skip_align1); 8868 } 8869 // Two bytes misalignment happens only for byte and short (char) arrays 8870 testptr(to, 2); 8871 jccb(Assembler::zero, L_skip_align2); 8872 movw(Address(to, 0), value); 8873 addptr(to, 2); 8874 subl(count, 1<<(shift-1)); 8875 BIND(L_skip_align2); 8876 } 8877 if (UseSSE < 2) { 8878 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes; 8879 // Fill 32-byte chunks 8880 subl(count, 8 << shift); 8881 jcc(Assembler::less, L_check_fill_8_bytes); 8882 align(16); 8883 8884 BIND(L_fill_32_bytes_loop); 8885 8886 for (int i = 0; i < 32; i += 4) { 8887 movl(Address(to, i), value); 8888 } 8889 8890 addptr(to, 32); 8891 subl(count, 8 << shift); 8892 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop); 8893 BIND(L_check_fill_8_bytes); 8894 addl(count, 8 << shift); 8895 jccb(Assembler::zero, L_exit); 8896 jmpb(L_fill_8_bytes); 8897 8898 // 8899 // length is too short, just fill qwords 8900 // 8901 BIND(L_fill_8_bytes_loop); 8902 movl(Address(to, 0), value); 8903 movl(Address(to, 4), value); 8904 addptr(to, 8); 8905 BIND(L_fill_8_bytes); 8906 subl(count, 1 << (shift + 1)); 8907 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop); 8908 // fall through to fill 4 bytes 8909 } else { 8910 Label L_fill_32_bytes; 8911 if (!UseUnalignedLoadStores) { 8912 // align to 8 bytes, we know we are 4 byte aligned to start 8913 testptr(to, 4); 8914 jccb(Assembler::zero, L_fill_32_bytes); 8915 movl(Address(to, 0), value); 8916 addptr(to, 4); 8917 subl(count, 1<<shift); 8918 } 8919 BIND(L_fill_32_bytes); 8920 { 8921 assert( UseSSE >= 2, "supported cpu only" ); 8922 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes; 8923 // Fill 32-byte chunks 8924 movdl(xtmp, value); 8925 pshufd(xtmp, xtmp, 0); 8926 8927 subl(count, 8 << shift); 8928 jcc(Assembler::less, L_check_fill_8_bytes); 8929 align(16); 8930 8931 BIND(L_fill_32_bytes_loop); 8932 8933 if (UseUnalignedLoadStores) { 8934 movdqu(Address(to, 0), xtmp); 8935 movdqu(Address(to, 16), xtmp); 8936 } else { 8937 movq(Address(to, 0), xtmp); 8938 movq(Address(to, 8), xtmp); 8939 movq(Address(to, 16), xtmp); 8940 movq(Address(to, 24), xtmp); 8941 } 8942 8943 addptr(to, 32); 8944 subl(count, 8 << shift); 8945 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop); 8946 BIND(L_check_fill_8_bytes); 8947 addl(count, 8 << shift); 8948 jccb(Assembler::zero, L_exit); 8949 jmpb(L_fill_8_bytes); 8950 8951 // 8952 // length is too short, just fill qwords 8953 // 8954 BIND(L_fill_8_bytes_loop); 8955 movq(Address(to, 0), xtmp); 8956 addptr(to, 8); 8957 BIND(L_fill_8_bytes); 8958 subl(count, 1 << (shift + 1)); 8959 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop); 8960 } 8961 } 8962 // fill trailing 4 bytes 8963 BIND(L_fill_4_bytes); 8964 testl(count, 1<<shift); 8965 jccb(Assembler::zero, L_fill_2_bytes); 8966 movl(Address(to, 0), value); 8967 if (t == T_BYTE || t == T_SHORT) { 8968 addptr(to, 4); 8969 BIND(L_fill_2_bytes); 8970 // fill trailing 2 bytes 8971 testl(count, 1<<(shift-1)); 8972 jccb(Assembler::zero, L_fill_byte); 8973 movw(Address(to, 0), value); 8974 if (t == T_BYTE) { 8975 addptr(to, 2); 8976 BIND(L_fill_byte); 8977 // fill trailing byte 8978 testl(count, 1); 8979 jccb(Assembler::zero, L_exit); 8980 movb(Address(to, 0), value); 8981 } else { 8982 BIND(L_fill_byte); 8983 } 8984 } else { 8985 BIND(L_fill_2_bytes); 8986 } 8987 BIND(L_exit); 8988 } 8989 #undef BIND 8990 #undef BLOCK_COMMENT 8991 8992 8993 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) { 8994 switch (cond) { 8995 // Note some conditions are synonyms for others 8996 case Assembler::zero: return Assembler::notZero; 8997 case Assembler::notZero: return Assembler::zero; 8998 case Assembler::less: return Assembler::greaterEqual; 8999 case Assembler::lessEqual: return Assembler::greater; 9000 case Assembler::greater: return Assembler::lessEqual; 9001 case Assembler::greaterEqual: return Assembler::less; 9002 case Assembler::below: return Assembler::aboveEqual; 9003 case Assembler::belowEqual: return Assembler::above; 9004 case Assembler::above: return Assembler::belowEqual; 9005 case Assembler::aboveEqual: return Assembler::below; 9006 case Assembler::overflow: return Assembler::noOverflow; 9007 case Assembler::noOverflow: return Assembler::overflow; 9008 case Assembler::negative: return Assembler::positive; 9009 case Assembler::positive: return Assembler::negative; 9010 case Assembler::parity: return Assembler::noParity; 9011 case Assembler::noParity: return Assembler::parity; 9012 } 9013 ShouldNotReachHere(); return Assembler::overflow; 9014 } 9015 9016 SkipIfEqual::SkipIfEqual( 9017 MacroAssembler* masm, const bool* flag_addr, bool value) { 9018 _masm = masm; 9019 _masm->cmp8(ExternalAddress((address)flag_addr), value); 9020 _masm->jcc(Assembler::equal, _label); 9021 } 9022 9023 SkipIfEqual::~SkipIfEqual() { 9024 _masm->bind(_label); 9025 }