1 /* 2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef CPU_X86_VM_MACROASSEMBLER_X86_HPP 26 #define CPU_X86_VM_MACROASSEMBLER_X86_HPP 27 28 #include "asm/assembler.hpp" 29 #include "utilities/macros.hpp" 30 #include "runtime/rtmLocking.hpp" 31 32 // MacroAssembler extends Assembler by frequently used macros. 33 // 34 // Instructions for which a 'better' code sequence exists depending 35 // on arguments should also go in here. 36 37 class MacroAssembler: public Assembler { 38 friend class LIR_Assembler; 39 friend class Runtime1; // as_Address() 40 41 protected: 42 43 Address as_Address(AddressLiteral adr); 44 Address as_Address(ArrayAddress adr); 45 46 // Support for VM calls 47 // 48 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 49 // may customize this version by overriding it for its purposes (e.g., to save/restore 50 // additional registers when doing a VM call). 51 #ifdef CC_INTERP 52 // c++ interpreter never wants to use interp_masm version of call_VM 53 #define VIRTUAL 54 #else 55 #define VIRTUAL virtual 56 #endif 57 58 #define COMMA , 59 60 VIRTUAL void call_VM_leaf_base( 61 address entry_point, // the entry point 62 int number_of_arguments // the number of arguments to pop after the call 63 ); 64 65 // This is the base routine called by the different versions of call_VM. The interpreter 66 // may customize this version by overriding it for its purposes (e.g., to save/restore 67 // additional registers when doing a VM call). 68 // 69 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base 70 // returns the register which contains the thread upon return. If a thread register has been 71 // specified, the return value will correspond to that register. If no last_java_sp is specified 72 // (noreg) than rsp will be used instead. 73 VIRTUAL void call_VM_base( // returns the register containing the thread upon return 74 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 75 Register java_thread, // the thread if computed before ; use noreg otherwise 76 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 77 address entry_point, // the entry point 78 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 79 bool check_exceptions // whether to check for pending exceptions after return 80 ); 81 82 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 83 // The implementation is only non-empty for the InterpreterMacroAssembler, 84 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 85 virtual void check_and_handle_popframe(Register java_thread); 86 virtual void check_and_handle_earlyret(Register java_thread); 87 88 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 89 90 // helpers for FPU flag access 91 // tmp is a temporary register, if none is available use noreg 92 void save_rax (Register tmp); 93 void restore_rax(Register tmp); 94 95 public: 96 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 97 98 // Support for NULL-checks 99 // 100 // Generates code that causes a NULL OS exception if the content of reg is NULL. 101 // If the accessed location is M[reg + offset] and the offset is known, provide the 102 // offset. No explicit code generation is needed if the offset is within a certain 103 // range (0 <= offset <= page_size). 104 105 void null_check(Register reg, int offset = -1); 106 static bool needs_explicit_null_check(intptr_t offset); 107 108 // Required platform-specific helpers for Label::patch_instructions. 109 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 110 void pd_patch_instruction(address branch, address target) { 111 unsigned char op = branch[0]; 112 assert(op == 0xE8 /* call */ || 113 op == 0xE9 /* jmp */ || 114 op == 0xEB /* short jmp */ || 115 (op & 0xF0) == 0x70 /* short jcc */ || 116 op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ || 117 op == 0xC7 && branch[1] == 0xF8 /* xbegin */, 118 "Invalid opcode at patch point"); 119 120 if (op == 0xEB || (op & 0xF0) == 0x70) { 121 // short offset operators (jmp and jcc) 122 char* disp = (char*) &branch[1]; 123 int imm8 = target - (address) &disp[1]; 124 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset"); 125 *disp = imm8; 126 } else { 127 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1]; 128 int imm32 = target - (address) &disp[1]; 129 *disp = imm32; 130 } 131 } 132 133 // The following 4 methods return the offset of the appropriate move instruction 134 135 // Support for fast byte/short loading with zero extension (depending on particular CPU) 136 int load_unsigned_byte(Register dst, Address src); 137 int load_unsigned_short(Register dst, Address src); 138 139 // Support for fast byte/short loading with sign extension (depending on particular CPU) 140 int load_signed_byte(Register dst, Address src); 141 int load_signed_short(Register dst, Address src); 142 143 // Support for sign-extension (hi:lo = extend_sign(lo)) 144 void extend_sign(Register hi, Register lo); 145 146 // Load and store values by size and signed-ness 147 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); 148 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); 149 150 // Support for inc/dec with optimal instruction selection depending on value 151 152 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } 153 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } 154 155 void decrementl(Address dst, int value = 1); 156 void decrementl(Register reg, int value = 1); 157 158 void decrementq(Register reg, int value = 1); 159 void decrementq(Address dst, int value = 1); 160 161 void incrementl(Address dst, int value = 1); 162 void incrementl(Register reg, int value = 1); 163 164 void incrementq(Register reg, int value = 1); 165 void incrementq(Address dst, int value = 1); 166 167 // Support optimal SSE move instructions. 168 void movflt(XMMRegister dst, XMMRegister src) { 169 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } 170 else { movss (dst, src); return; } 171 } 172 void movflt(XMMRegister dst, Address src) { movss(dst, src); } 173 void movflt(XMMRegister dst, AddressLiteral src); 174 void movflt(Address dst, XMMRegister src) { movss(dst, src); } 175 176 void movdbl(XMMRegister dst, XMMRegister src) { 177 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } 178 else { movsd (dst, src); return; } 179 } 180 181 void movdbl(XMMRegister dst, AddressLiteral src); 182 183 void movdbl(XMMRegister dst, Address src) { 184 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } 185 else { movlpd(dst, src); return; } 186 } 187 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } 188 189 void incrementl(AddressLiteral dst); 190 void incrementl(ArrayAddress dst); 191 192 void incrementq(AddressLiteral dst); 193 194 // Alignment 195 void align(int modulus); 196 void align(int modulus, int target); 197 198 // A 5 byte nop that is safe for patching (see patch_verified_entry) 199 void fat_nop(); 200 201 // Stack frame creation/removal 202 void enter(); 203 void leave(); 204 205 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 206 // The pointer will be loaded into the thread register. 207 void get_thread(Register thread); 208 209 210 // Support for VM calls 211 // 212 // It is imperative that all calls into the VM are handled via the call_VM macros. 213 // They make sure that the stack linkage is setup correctly. call_VM's correspond 214 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 215 216 217 void call_VM(Register oop_result, 218 address entry_point, 219 bool check_exceptions = true); 220 void call_VM(Register oop_result, 221 address entry_point, 222 Register arg_1, 223 bool check_exceptions = true); 224 void call_VM(Register oop_result, 225 address entry_point, 226 Register arg_1, Register arg_2, 227 bool check_exceptions = true); 228 void call_VM(Register oop_result, 229 address entry_point, 230 Register arg_1, Register arg_2, Register arg_3, 231 bool check_exceptions = true); 232 233 // Overloadings with last_Java_sp 234 void call_VM(Register oop_result, 235 Register last_java_sp, 236 address entry_point, 237 int number_of_arguments = 0, 238 bool check_exceptions = true); 239 void call_VM(Register oop_result, 240 Register last_java_sp, 241 address entry_point, 242 Register arg_1, bool 243 check_exceptions = true); 244 void call_VM(Register oop_result, 245 Register last_java_sp, 246 address entry_point, 247 Register arg_1, Register arg_2, 248 bool check_exceptions = true); 249 void call_VM(Register oop_result, 250 Register last_java_sp, 251 address entry_point, 252 Register arg_1, Register arg_2, Register arg_3, 253 bool check_exceptions = true); 254 255 void get_vm_result (Register oop_result, Register thread); 256 void get_vm_result_2(Register metadata_result, Register thread); 257 258 // These always tightly bind to MacroAssembler::call_VM_base 259 // bypassing the virtual implementation 260 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); 261 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); 262 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); 263 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true); 264 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true); 265 266 void call_VM_leaf(address entry_point, 267 int number_of_arguments = 0); 268 void call_VM_leaf(address entry_point, 269 Register arg_1); 270 void call_VM_leaf(address entry_point, 271 Register arg_1, Register arg_2); 272 void call_VM_leaf(address entry_point, 273 Register arg_1, Register arg_2, Register arg_3); 274 275 // These always tightly bind to MacroAssembler::call_VM_leaf_base 276 // bypassing the virtual implementation 277 void super_call_VM_leaf(address entry_point); 278 void super_call_VM_leaf(address entry_point, Register arg_1); 279 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); 280 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); 281 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); 282 283 // last Java Frame (fills frame anchor) 284 void set_last_Java_frame(Register thread, 285 Register last_java_sp, 286 Register last_java_fp, 287 address last_java_pc); 288 289 // thread in the default location (r15_thread on 64bit) 290 void set_last_Java_frame(Register last_java_sp, 291 Register last_java_fp, 292 address last_java_pc); 293 294 void reset_last_Java_frame(Register thread, bool clear_fp, bool clear_pc); 295 296 // thread in the default location (r15_thread on 64bit) 297 void reset_last_Java_frame(bool clear_fp, bool clear_pc); 298 299 // Stores 300 void store_check(Register obj); // store check for obj - register is destroyed afterwards 301 void store_check(Register obj, Address dst); // same as above, dst is exact store location (reg. is destroyed) 302 303 #if INCLUDE_ALL_GCS 304 305 void g1_write_barrier_pre(Register obj, 306 Register pre_val, 307 Register thread, 308 Register tmp, 309 bool tosca_live, 310 bool expand_call); 311 312 void g1_write_barrier_post(Register store_addr, 313 Register new_val, 314 Register thread, 315 Register tmp, 316 Register tmp2); 317 318 #endif // INCLUDE_ALL_GCS 319 320 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 321 void c2bool(Register x); 322 323 // C++ bool manipulation 324 325 void movbool(Register dst, Address src); 326 void movbool(Address dst, bool boolconst); 327 void movbool(Address dst, Register src); 328 void testbool(Register dst); 329 330 // oop manipulations 331 void load_klass(Register dst, Register src); 332 void store_klass(Register dst, Register src); 333 334 void load_heap_oop(Register dst, Address src); 335 void load_heap_oop_not_null(Register dst, Address src); 336 void store_heap_oop(Address dst, Register src); 337 void cmp_heap_oop(Register src1, Address src2, Register tmp = noreg); 338 339 // Used for storing NULL. All other oop constants should be 340 // stored using routines that take a jobject. 341 void store_heap_oop_null(Address dst); 342 343 void load_prototype_header(Register dst, Register src); 344 345 #ifdef _LP64 346 void store_klass_gap(Register dst, Register src); 347 348 // This dummy is to prevent a call to store_heap_oop from 349 // converting a zero (like NULL) into a Register by giving 350 // the compiler two choices it can't resolve 351 352 void store_heap_oop(Address dst, void* dummy); 353 354 void encode_heap_oop(Register r); 355 void decode_heap_oop(Register r); 356 void encode_heap_oop_not_null(Register r); 357 void decode_heap_oop_not_null(Register r); 358 void encode_heap_oop_not_null(Register dst, Register src); 359 void decode_heap_oop_not_null(Register dst, Register src); 360 361 void set_narrow_oop(Register dst, jobject obj); 362 void set_narrow_oop(Address dst, jobject obj); 363 void cmp_narrow_oop(Register dst, jobject obj); 364 void cmp_narrow_oop(Address dst, jobject obj); 365 366 void encode_klass_not_null(Register r); 367 void decode_klass_not_null(Register r); 368 void encode_klass_not_null(Register dst, Register src); 369 void decode_klass_not_null(Register dst, Register src); 370 void set_narrow_klass(Register dst, Klass* k); 371 void set_narrow_klass(Address dst, Klass* k); 372 void cmp_narrow_klass(Register dst, Klass* k); 373 void cmp_narrow_klass(Address dst, Klass* k); 374 375 // Returns the byte size of the instructions generated by decode_klass_not_null() 376 // when compressed klass pointers are being used. 377 static int instr_size_for_decode_klass_not_null(); 378 379 // if heap base register is used - reinit it with the correct value 380 void reinit_heapbase(); 381 382 DEBUG_ONLY(void verify_heapbase(const char* msg);) 383 384 #endif // _LP64 385 386 // Int division/remainder for Java 387 // (as idivl, but checks for special case as described in JVM spec.) 388 // returns idivl instruction offset for implicit exception handling 389 int corrected_idivl(Register reg); 390 391 // Long division/remainder for Java 392 // (as idivq, but checks for special case as described in JVM spec.) 393 // returns idivq instruction offset for implicit exception handling 394 int corrected_idivq(Register reg); 395 396 void int3(); 397 398 // Long operation macros for a 32bit cpu 399 // Long negation for Java 400 void lneg(Register hi, Register lo); 401 402 // Long multiplication for Java 403 // (destroys contents of eax, ebx, ecx and edx) 404 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y 405 406 // Long shifts for Java 407 // (semantics as described in JVM spec.) 408 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) 409 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) 410 411 // Long compare for Java 412 // (semantics as described in JVM spec.) 413 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) 414 415 416 // misc 417 418 // Sign extension 419 void sign_extend_short(Register reg); 420 void sign_extend_byte(Register reg); 421 422 // Division by power of 2, rounding towards 0 423 void division_with_shift(Register reg, int shift_value); 424 425 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: 426 // 427 // CF (corresponds to C0) if x < y 428 // PF (corresponds to C2) if unordered 429 // ZF (corresponds to C3) if x = y 430 // 431 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 432 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) 433 void fcmp(Register tmp); 434 // Variant of the above which allows y to be further down the stack 435 // and which only pops x and y if specified. If pop_right is 436 // specified then pop_left must also be specified. 437 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); 438 439 // Floating-point comparison for Java 440 // Compares the top-most stack entries on the FPU stack and stores the result in dst. 441 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 442 // (semantics as described in JVM spec.) 443 void fcmp2int(Register dst, bool unordered_is_less); 444 // Variant of the above which allows y to be further down the stack 445 // and which only pops x and y if specified. If pop_right is 446 // specified then pop_left must also be specified. 447 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); 448 449 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) 450 // tmp is a temporary register, if none is available use noreg 451 void fremr(Register tmp); 452 453 454 // same as fcmp2int, but using SSE2 455 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 456 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 457 458 // Inlined sin/cos generator for Java; must not use CPU instruction 459 // directly on Intel as it does not have high enough precision 460 // outside of the range [-pi/4, pi/4]. Extra argument indicate the 461 // number of FPU stack slots in use; all but the topmost will 462 // require saving if a slow case is necessary. Assumes argument is 463 // on FP TOS; result is on FP TOS. No cpu registers are changed by 464 // this code. 465 void trigfunc(char trig, int num_fpu_regs_in_use = 1); 466 467 // branch to L if FPU flag C2 is set/not set 468 // tmp is a temporary register, if none is available use noreg 469 void jC2 (Register tmp, Label& L); 470 void jnC2(Register tmp, Label& L); 471 472 // Pop ST (ffree & fincstp combined) 473 void fpop(); 474 475 // Load float value from 'address'. If UseSSE >= 1, the value is loaded into 476 // register xmm0. Otherwise, the value is loaded onto the FPU stack. 477 void load_float(Address src); 478 479 // Store float value to 'address'. If UseSSE >= 1, the value is stored 480 // from register xmm0. Otherwise, the value is stored from the FPU stack. 481 void store_float(Address dst); 482 483 // Load double value from 'address'. If UseSSE >= 2, the value is loaded into 484 // register xmm0. Otherwise, the value is loaded onto the FPU stack. 485 void load_double(Address src); 486 487 // Store double value to 'address'. If UseSSE >= 2, the value is stored 488 // from register xmm0. Otherwise, the value is stored from the FPU stack. 489 void store_double(Address dst); 490 491 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack 492 void push_fTOS(); 493 494 // pops double TOS element from CPU stack and pushes on FPU stack 495 void pop_fTOS(); 496 497 void empty_FPU_stack(); 498 499 void push_IU_state(); 500 void pop_IU_state(); 501 502 void push_FPU_state(); 503 void pop_FPU_state(); 504 505 void push_CPU_state(); 506 void pop_CPU_state(); 507 508 // Round up to a power of two 509 void round_to(Register reg, int modulus); 510 511 // Callee saved registers handling 512 void push_callee_saved_registers(); 513 void pop_callee_saved_registers(); 514 515 // allocation 516 void eden_allocate( 517 Register obj, // result: pointer to object after successful allocation 518 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 519 int con_size_in_bytes, // object size in bytes if known at compile time 520 Register t1, // temp register 521 Label& slow_case // continuation point if fast allocation fails 522 ); 523 void tlab_allocate( 524 Register obj, // result: pointer to object after successful allocation 525 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 526 int con_size_in_bytes, // object size in bytes if known at compile time 527 Register t1, // temp register 528 Register t2, // temp register 529 Label& slow_case // continuation point if fast allocation fails 530 ); 531 Register tlab_refill(Label& retry_tlab, Label& try_eden, Label& slow_case); // returns TLS address 532 void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp); 533 534 void incr_allocated_bytes(Register thread, 535 Register var_size_in_bytes, int con_size_in_bytes, 536 Register t1 = noreg); 537 538 // interface method calling 539 void lookup_interface_method(Register recv_klass, 540 Register intf_klass, 541 RegisterOrConstant itable_index, 542 Register method_result, 543 Register scan_temp, 544 Label& no_such_interface); 545 546 // virtual method calling 547 void lookup_virtual_method(Register recv_klass, 548 RegisterOrConstant vtable_index, 549 Register method_result); 550 551 // Test sub_klass against super_klass, with fast and slow paths. 552 553 // The fast path produces a tri-state answer: yes / no / maybe-slow. 554 // One of the three labels can be NULL, meaning take the fall-through. 555 // If super_check_offset is -1, the value is loaded up from super_klass. 556 // No registers are killed, except temp_reg. 557 void check_klass_subtype_fast_path(Register sub_klass, 558 Register super_klass, 559 Register temp_reg, 560 Label* L_success, 561 Label* L_failure, 562 Label* L_slow_path, 563 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 564 565 // The rest of the type check; must be wired to a corresponding fast path. 566 // It does not repeat the fast path logic, so don't use it standalone. 567 // The temp_reg and temp2_reg can be noreg, if no temps are available. 568 // Updates the sub's secondary super cache as necessary. 569 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 570 void check_klass_subtype_slow_path(Register sub_klass, 571 Register super_klass, 572 Register temp_reg, 573 Register temp2_reg, 574 Label* L_success, 575 Label* L_failure, 576 bool set_cond_codes = false); 577 578 // Simplified, combined version, good for typical uses. 579 // Falls through on failure. 580 void check_klass_subtype(Register sub_klass, 581 Register super_klass, 582 Register temp_reg, 583 Label& L_success); 584 585 // method handles (JSR 292) 586 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 587 588 //---- 589 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 590 591 // Debugging 592 593 // only if +VerifyOops 594 // TODO: Make these macros with file and line like sparc version! 595 void verify_oop(Register reg, const char* s = "broken oop"); 596 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 597 598 // TODO: verify method and klass metadata (compare against vptr?) 599 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {} 600 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){} 601 602 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__) 603 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__) 604 605 // only if +VerifyFPU 606 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 607 608 // Verify or restore cpu control state after JNI call 609 void restore_cpu_control_state_after_jni(); 610 611 // prints msg, dumps registers and stops execution 612 void stop(const char* msg); 613 614 // prints msg and continues 615 void warn(const char* msg); 616 617 // dumps registers and other state 618 void print_state(); 619 620 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); 621 static void debug64(char* msg, int64_t pc, int64_t regs[]); 622 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip); 623 static void print_state64(int64_t pc, int64_t regs[]); 624 625 void os_breakpoint(); 626 627 void untested() { stop("untested"); } 628 629 void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } 630 631 void should_not_reach_here() { stop("should not reach here"); } 632 633 void print_CPU_state(); 634 635 // Stack overflow checking 636 void bang_stack_with_offset(int offset) { 637 // stack grows down, caller passes positive offset 638 assert(offset > 0, "must bang with negative offset"); 639 movl(Address(rsp, (-offset)), rax); 640 } 641 642 // Writes to stack successive pages until offset reached to check for 643 // stack overflow + shadow pages. Also, clobbers tmp 644 void bang_stack_size(Register size, Register tmp); 645 646 // Check for reserved stack access in method being exited (for JIT) 647 void reserved_stack_check(); 648 649 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 650 Register tmp, 651 int offset); 652 653 // Support for serializing memory accesses between threads 654 void serialize_memory(Register thread, Register tmp); 655 656 void verify_tlab(); 657 658 // Biased locking support 659 // lock_reg and obj_reg must be loaded up with the appropriate values. 660 // swap_reg must be rax, and is killed. 661 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will 662 // be killed; if not supplied, push/pop will be used internally to 663 // allocate a temporary (inefficient, avoid if possible). 664 // Optional slow case is for implementations (interpreter and C1) which branch to 665 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 666 // Returns offset of first potentially-faulting instruction for null 667 // check info (currently consumed only by C1). If 668 // swap_reg_contains_mark is true then returns -1 as it is assumed 669 // the calling code has already passed any potential faults. 670 int biased_locking_enter(Register lock_reg, Register obj_reg, 671 Register swap_reg, Register tmp_reg, 672 bool swap_reg_contains_mark, 673 Label& done, Label* slow_case = NULL, 674 BiasedLockingCounters* counters = NULL); 675 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 676 #ifdef COMPILER2 677 // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file. 678 // See full desription in macroAssembler_x86.cpp. 679 void fast_lock(Register obj, Register box, Register tmp, 680 Register scr, Register cx1, Register cx2, 681 BiasedLockingCounters* counters, 682 RTMLockingCounters* rtm_counters, 683 RTMLockingCounters* stack_rtm_counters, 684 Metadata* method_data, 685 bool use_rtm, bool profile_rtm); 686 void fast_unlock(Register obj, Register box, Register tmp, bool use_rtm); 687 #if INCLUDE_RTM_OPT 688 void rtm_counters_update(Register abort_status, Register rtm_counters); 689 void branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel); 690 void rtm_abort_ratio_calculation(Register tmp, Register rtm_counters_reg, 691 RTMLockingCounters* rtm_counters, 692 Metadata* method_data); 693 void rtm_profiling(Register abort_status_Reg, Register rtm_counters_Reg, 694 RTMLockingCounters* rtm_counters, Metadata* method_data, bool profile_rtm); 695 void rtm_retry_lock_on_abort(Register retry_count, Register abort_status, Label& retryLabel); 696 void rtm_retry_lock_on_busy(Register retry_count, Register box, Register tmp, Register scr, Label& retryLabel); 697 void rtm_stack_locking(Register obj, Register tmp, Register scr, 698 Register retry_on_abort_count, 699 RTMLockingCounters* stack_rtm_counters, 700 Metadata* method_data, bool profile_rtm, 701 Label& DONE_LABEL, Label& IsInflated); 702 void rtm_inflated_locking(Register obj, Register box, Register tmp, 703 Register scr, Register retry_on_busy_count, 704 Register retry_on_abort_count, 705 RTMLockingCounters* rtm_counters, 706 Metadata* method_data, bool profile_rtm, 707 Label& DONE_LABEL); 708 #endif 709 #endif 710 711 Condition negate_condition(Condition cond); 712 713 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit 714 // operands. In general the names are modified to avoid hiding the instruction in Assembler 715 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers 716 // here in MacroAssembler. The major exception to this rule is call 717 718 // Arithmetics 719 720 721 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } 722 void addptr(Address dst, Register src); 723 724 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } 725 void addptr(Register dst, int32_t src); 726 void addptr(Register dst, Register src); 727 void addptr(Register dst, RegisterOrConstant src) { 728 if (src.is_constant()) addptr(dst, (int) src.as_constant()); 729 else addptr(dst, src.as_register()); 730 } 731 732 void andptr(Register dst, int32_t src); 733 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } 734 735 void cmp8(AddressLiteral src1, int imm); 736 737 // renamed to drag out the casting of address to int32_t/intptr_t 738 void cmp32(Register src1, int32_t imm); 739 740 void cmp32(AddressLiteral src1, int32_t imm); 741 // compare reg - mem, or reg - &mem 742 void cmp32(Register src1, AddressLiteral src2); 743 744 void cmp32(Register src1, Address src2); 745 746 #ifndef _LP64 747 void cmpklass(Address dst, Metadata* obj); 748 void cmpklass(Register dst, Metadata* obj); 749 void cmpoop(Address dst, jobject obj); 750 void cmpoop(Register dst, jobject obj); 751 #endif // _LP64 752 753 // NOTE src2 must be the lval. This is NOT an mem-mem compare 754 void cmpptr(Address src1, AddressLiteral src2); 755 756 void cmpptr(Register src1, AddressLiteral src2); 757 758 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 759 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 760 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 761 762 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 763 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 764 765 // cmp64 to avoild hiding cmpq 766 void cmp64(Register src1, AddressLiteral src); 767 768 void cmpxchgptr(Register reg, Address adr); 769 770 void locked_cmpxchgptr(Register reg, AddressLiteral adr); 771 772 773 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } 774 void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); } 775 776 777 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } 778 779 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } 780 781 void shlptr(Register dst, int32_t shift); 782 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } 783 784 void shrptr(Register dst, int32_t shift); 785 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } 786 787 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } 788 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } 789 790 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 791 792 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 793 void subptr(Register dst, int32_t src); 794 // Force generation of a 4 byte immediate value even if it fits into 8bit 795 void subptr_imm32(Register dst, int32_t src); 796 void subptr(Register dst, Register src); 797 void subptr(Register dst, RegisterOrConstant src) { 798 if (src.is_constant()) subptr(dst, (int) src.as_constant()); 799 else subptr(dst, src.as_register()); 800 } 801 802 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 803 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 804 805 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 806 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 807 808 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } 809 810 811 812 // Helper functions for statistics gathering. 813 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. 814 void cond_inc32(Condition cond, AddressLiteral counter_addr); 815 // Unconditional atomic increment. 816 void atomic_incl(Address counter_addr); 817 void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1); 818 #ifdef _LP64 819 void atomic_incq(Address counter_addr); 820 void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1); 821 #endif 822 void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; } 823 void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; } 824 825 void lea(Register dst, AddressLiteral adr); 826 void lea(Address dst, AddressLiteral adr); 827 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } 828 829 void leal32(Register dst, Address src) { leal(dst, src); } 830 831 // Import other testl() methods from the parent class or else 832 // they will be hidden by the following overriding declaration. 833 using Assembler::testl; 834 void testl(Register dst, AddressLiteral src); 835 836 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 837 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 838 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 839 void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); } 840 841 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } 842 void testptr(Register src1, Register src2); 843 844 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 845 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 846 847 // Calls 848 849 void call(Label& L, relocInfo::relocType rtype); 850 void call(Register entry); 851 852 // NOTE: this call tranfers to the effective address of entry NOT 853 // the address contained by entry. This is because this is more natural 854 // for jumps/calls. 855 void call(AddressLiteral entry); 856 857 // Emit the CompiledIC call idiom 858 void ic_call(address entry, jint method_index = 0); 859 860 // Jumps 861 862 // NOTE: these jumps tranfer to the effective address of dst NOT 863 // the address contained by dst. This is because this is more natural 864 // for jumps/calls. 865 void jump(AddressLiteral dst); 866 void jump_cc(Condition cc, AddressLiteral dst); 867 868 // 32bit can do a case table jump in one instruction but we no longer allow the base 869 // to be installed in the Address class. This jump will tranfers to the address 870 // contained in the location described by entry (not the address of entry) 871 void jump(ArrayAddress entry); 872 873 // Floating 874 875 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } 876 void andpd(XMMRegister dst, AddressLiteral src); 877 void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); } 878 879 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); } 880 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); } 881 void andps(XMMRegister dst, AddressLiteral src); 882 883 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); } 884 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } 885 void comiss(XMMRegister dst, AddressLiteral src); 886 887 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); } 888 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } 889 void comisd(XMMRegister dst, AddressLiteral src); 890 891 void fadd_s(Address src) { Assembler::fadd_s(src); } 892 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); } 893 894 void fldcw(Address src) { Assembler::fldcw(src); } 895 void fldcw(AddressLiteral src); 896 897 void fld_s(int index) { Assembler::fld_s(index); } 898 void fld_s(Address src) { Assembler::fld_s(src); } 899 void fld_s(AddressLiteral src); 900 901 void fld_d(Address src) { Assembler::fld_d(src); } 902 void fld_d(AddressLiteral src); 903 904 void fld_x(Address src) { Assembler::fld_x(src); } 905 void fld_x(AddressLiteral src); 906 907 void fmul_s(Address src) { Assembler::fmul_s(src); } 908 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); } 909 910 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } 911 void ldmxcsr(AddressLiteral src); 912 913 void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 914 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 915 Register rax, Register rcx, Register rdx, Register tmp); 916 917 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 918 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 919 Register rax, Register rcx, Register rdx, Register tmp1 LP64_ONLY(COMMA Register tmp2)); 920 921 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4, 922 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx, 923 Register rdx NOT_LP64(COMMA Register tmp) LP64_ONLY(COMMA Register tmp1) 924 LP64_ONLY(COMMA Register tmp2) LP64_ONLY(COMMA Register tmp3) LP64_ONLY(COMMA Register tmp4)); 925 926 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 927 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 928 Register rax, Register rbx LP64_ONLY(COMMA Register rcx), Register rdx 929 LP64_ONLY(COMMA Register tmp1) LP64_ONLY(COMMA Register tmp2) 930 LP64_ONLY(COMMA Register tmp3) LP64_ONLY(COMMA Register tmp4)); 931 932 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 933 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 934 Register rax, Register rcx, Register rdx NOT_LP64(COMMA Register tmp) 935 LP64_ONLY(COMMA Register r8) LP64_ONLY(COMMA Register r9) 936 LP64_ONLY(COMMA Register r10) LP64_ONLY(COMMA Register r11)); 937 938 #ifndef _LP64 939 void libm_sincos_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx, 940 Register edx, Register ebx, Register esi, Register edi, 941 Register ebp, Register esp); 942 void libm_reduce_pi04l(Register eax, Register ecx, Register edx, Register ebx, 943 Register esi, Register edi, Register ebp, Register esp); 944 #endif 945 946 void increase_precision(); 947 void restore_precision(); 948 949 private: 950 951 // call runtime as a fallback for trig functions and pow/exp. 952 void fp_runtime_fallback(address runtime_entry, int nb_args, int num_fpu_regs_in_use); 953 954 // these are private because users should be doing movflt/movdbl 955 956 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } 957 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } 958 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } 959 void movss(XMMRegister dst, AddressLiteral src); 960 961 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } 962 void movlpd(XMMRegister dst, AddressLiteral src); 963 964 public: 965 966 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); } 967 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); } 968 void addsd(XMMRegister dst, AddressLiteral src); 969 970 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); } 971 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); } 972 void addss(XMMRegister dst, AddressLiteral src); 973 974 void addpd(XMMRegister dst, XMMRegister src) { Assembler::addpd(dst, src); } 975 void addpd(XMMRegister dst, Address src) { Assembler::addpd(dst, src); } 976 void addpd(XMMRegister dst, AddressLiteral src); 977 978 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); } 979 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); } 980 void divsd(XMMRegister dst, AddressLiteral src); 981 982 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); } 983 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); } 984 void divss(XMMRegister dst, AddressLiteral src); 985 986 // Move Unaligned Double Quadword 987 void movdqu(Address dst, XMMRegister src); 988 void movdqu(XMMRegister dst, Address src); 989 void movdqu(XMMRegister dst, XMMRegister src); 990 void movdqu(XMMRegister dst, AddressLiteral src); 991 // AVX Unaligned forms 992 void vmovdqu(Address dst, XMMRegister src); 993 void vmovdqu(XMMRegister dst, Address src); 994 void vmovdqu(XMMRegister dst, XMMRegister src); 995 void vmovdqu(XMMRegister dst, AddressLiteral src); 996 997 // Move Aligned Double Quadword 998 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); } 999 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); } 1000 void movdqa(XMMRegister dst, AddressLiteral src); 1001 1002 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } 1003 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } 1004 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } 1005 void movsd(XMMRegister dst, AddressLiteral src); 1006 1007 void mulpd(XMMRegister dst, XMMRegister src) { Assembler::mulpd(dst, src); } 1008 void mulpd(XMMRegister dst, Address src) { Assembler::mulpd(dst, src); } 1009 void mulpd(XMMRegister dst, AddressLiteral src); 1010 1011 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); } 1012 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); } 1013 void mulsd(XMMRegister dst, AddressLiteral src); 1014 1015 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); } 1016 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); } 1017 void mulss(XMMRegister dst, AddressLiteral src); 1018 1019 // Carry-Less Multiplication Quadword 1020 void pclmulldq(XMMRegister dst, XMMRegister src) { 1021 // 0x00 - multiply lower 64 bits [0:63] 1022 Assembler::pclmulqdq(dst, src, 0x00); 1023 } 1024 void pclmulhdq(XMMRegister dst, XMMRegister src) { 1025 // 0x11 - multiply upper 64 bits [64:127] 1026 Assembler::pclmulqdq(dst, src, 0x11); 1027 } 1028 1029 void pcmpeqb(XMMRegister dst, XMMRegister src); 1030 void pcmpeqw(XMMRegister dst, XMMRegister src); 1031 1032 void pcmpestri(XMMRegister dst, Address src, int imm8); 1033 void pcmpestri(XMMRegister dst, XMMRegister src, int imm8); 1034 1035 void pmovzxbw(XMMRegister dst, XMMRegister src); 1036 void pmovzxbw(XMMRegister dst, Address src); 1037 1038 void pmovmskb(Register dst, XMMRegister src); 1039 1040 void ptest(XMMRegister dst, XMMRegister src); 1041 1042 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); } 1043 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); } 1044 void sqrtsd(XMMRegister dst, AddressLiteral src); 1045 1046 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); } 1047 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); } 1048 void sqrtss(XMMRegister dst, AddressLiteral src); 1049 1050 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); } 1051 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); } 1052 void subsd(XMMRegister dst, AddressLiteral src); 1053 1054 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); } 1055 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); } 1056 void subss(XMMRegister dst, AddressLiteral src); 1057 1058 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } 1059 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } 1060 void ucomiss(XMMRegister dst, AddressLiteral src); 1061 1062 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } 1063 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } 1064 void ucomisd(XMMRegister dst, AddressLiteral src); 1065 1066 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 1067 void xorpd(XMMRegister dst, XMMRegister src); 1068 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } 1069 void xorpd(XMMRegister dst, AddressLiteral src); 1070 1071 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 1072 void xorps(XMMRegister dst, XMMRegister src); 1073 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } 1074 void xorps(XMMRegister dst, AddressLiteral src); 1075 1076 // Shuffle Bytes 1077 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); } 1078 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); } 1079 void pshufb(XMMRegister dst, AddressLiteral src); 1080 // AVX 3-operands instructions 1081 1082 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); } 1083 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); } 1084 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1085 1086 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); } 1087 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); } 1088 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1089 1090 void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len); 1091 void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len); 1092 1093 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1094 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1095 1096 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1097 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1098 1099 void vpbroadcastw(XMMRegister dst, XMMRegister src); 1100 1101 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1102 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1103 1104 void vpmovzxbw(XMMRegister dst, Address src, int vector_len); 1105 void vpmovmskb(Register dst, XMMRegister src); 1106 1107 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1108 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1109 1110 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1111 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1112 1113 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1114 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1115 1116 void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1117 void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1118 1119 void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1120 void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1121 1122 void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1123 void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1124 1125 void vptest(XMMRegister dst, XMMRegister src); 1126 1127 void punpcklbw(XMMRegister dst, XMMRegister src); 1128 void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); } 1129 1130 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1131 void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); } 1132 1133 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); } 1134 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); } 1135 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1136 1137 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); } 1138 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); } 1139 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1140 1141 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); } 1142 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); } 1143 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1144 1145 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); } 1146 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); } 1147 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1148 1149 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); } 1150 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); } 1151 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1152 1153 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); } 1154 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); } 1155 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1156 1157 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); } 1158 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); } 1159 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1160 1161 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); } 1162 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); } 1163 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1164 1165 void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1166 void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1167 1168 // AVX Vector instructions 1169 1170 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); } 1171 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); } 1172 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1173 1174 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); } 1175 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); } 1176 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1177 1178 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { 1179 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2 1180 Assembler::vpxor(dst, nds, src, vector_len); 1181 else 1182 Assembler::vxorpd(dst, nds, src, vector_len); 1183 } 1184 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { 1185 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2 1186 Assembler::vpxor(dst, nds, src, vector_len); 1187 else 1188 Assembler::vxorpd(dst, nds, src, vector_len); 1189 } 1190 1191 // Simple version for AVX2 256bit vectors 1192 void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); } 1193 void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); } 1194 1195 // Move packed integer values from low 128 bit to hign 128 bit in 256 bit vector. 1196 void vinserti128h(XMMRegister dst, XMMRegister nds, XMMRegister src) { 1197 if (UseAVX > 1) // vinserti128h is available only in AVX2 1198 Assembler::vinserti128h(dst, nds, src); 1199 else 1200 Assembler::vinsertf128h(dst, nds, src); 1201 } 1202 1203 // Carry-Less Multiplication Quadword 1204 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) { 1205 // 0x00 - multiply lower 64 bits [0:63] 1206 Assembler::vpclmulqdq(dst, nds, src, 0x00); 1207 } 1208 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) { 1209 // 0x11 - multiply upper 64 bits [64:127] 1210 Assembler::vpclmulqdq(dst, nds, src, 0x11); 1211 } 1212 1213 // Data 1214 1215 void cmov32( Condition cc, Register dst, Address src); 1216 void cmov32( Condition cc, Register dst, Register src); 1217 1218 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); } 1219 1220 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 1221 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 1222 1223 void movoop(Register dst, jobject obj); 1224 void movoop(Address dst, jobject obj); 1225 1226 void mov_metadata(Register dst, Metadata* obj); 1227 void mov_metadata(Address dst, Metadata* obj); 1228 1229 void movptr(ArrayAddress dst, Register src); 1230 // can this do an lea? 1231 void movptr(Register dst, ArrayAddress src); 1232 1233 void movptr(Register dst, Address src); 1234 1235 #ifdef _LP64 1236 void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1); 1237 #else 1238 void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit 1239 #endif 1240 1241 void movptr(Register dst, intptr_t src); 1242 void movptr(Register dst, Register src); 1243 void movptr(Address dst, intptr_t src); 1244 1245 void movptr(Address dst, Register src); 1246 1247 void movptr(Register dst, RegisterOrConstant src) { 1248 if (src.is_constant()) movptr(dst, src.as_constant()); 1249 else movptr(dst, src.as_register()); 1250 } 1251 1252 #ifdef _LP64 1253 // Generally the next two are only used for moving NULL 1254 // Although there are situations in initializing the mark word where 1255 // they could be used. They are dangerous. 1256 1257 // They only exist on LP64 so that int32_t and intptr_t are not the same 1258 // and we have ambiguous declarations. 1259 1260 void movptr(Address dst, int32_t imm32); 1261 void movptr(Register dst, int32_t imm32); 1262 #endif // _LP64 1263 1264 // to avoid hiding movl 1265 void mov32(AddressLiteral dst, Register src); 1266 void mov32(Register dst, AddressLiteral src); 1267 1268 // to avoid hiding movb 1269 void movbyte(ArrayAddress dst, int src); 1270 1271 // Import other mov() methods from the parent class or else 1272 // they will be hidden by the following overriding declaration. 1273 using Assembler::movdl; 1274 using Assembler::movq; 1275 void movdl(XMMRegister dst, AddressLiteral src); 1276 void movq(XMMRegister dst, AddressLiteral src); 1277 1278 // Can push value or effective address 1279 void pushptr(AddressLiteral src); 1280 1281 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } 1282 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } 1283 1284 void pushoop(jobject obj); 1285 void pushklass(Metadata* obj); 1286 1287 // sign extend as need a l to ptr sized element 1288 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } 1289 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } 1290 1291 // C2 compiled method's prolog code. 1292 void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b); 1293 1294 // clear memory of size 'cnt' qwords, starting at 'base'. 1295 void clear_mem(Register base, Register cnt, Register rtmp); 1296 1297 #ifdef COMPILER2 1298 void string_indexof_char(Register str1, Register cnt1, Register ch, Register result, 1299 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp); 1300 1301 // IndexOf strings. 1302 // Small strings are loaded through stack if they cross page boundary. 1303 void string_indexof(Register str1, Register str2, 1304 Register cnt1, Register cnt2, 1305 int int_cnt2, Register result, 1306 XMMRegister vec, Register tmp, 1307 int ae); 1308 1309 // IndexOf for constant substrings with size >= 8 elements 1310 // which don't need to be loaded through stack. 1311 void string_indexofC8(Register str1, Register str2, 1312 Register cnt1, Register cnt2, 1313 int int_cnt2, Register result, 1314 XMMRegister vec, Register tmp, 1315 int ae); 1316 1317 // Smallest code: we don't need to load through stack, 1318 // check string tail. 1319 1320 // helper function for string_compare 1321 void load_next_elements(Register elem1, Register elem2, Register str1, Register str2, 1322 Address::ScaleFactor scale, Address::ScaleFactor scale1, 1323 Address::ScaleFactor scale2, Register index, int ae); 1324 // Compare strings. 1325 void string_compare(Register str1, Register str2, 1326 Register cnt1, Register cnt2, Register result, 1327 XMMRegister vec1, int ae); 1328 1329 // Search for Non-ASCII character (Negative byte value) in a byte array, 1330 // return true if it has any and false otherwise. 1331 void has_negatives(Register ary1, Register len, 1332 Register result, Register tmp1, 1333 XMMRegister vec1, XMMRegister vec2); 1334 1335 // Compare char[] or byte[] arrays. 1336 void arrays_equals(bool is_array_equ, Register ary1, Register ary2, 1337 Register limit, Register result, Register chr, 1338 XMMRegister vec1, XMMRegister vec2, bool is_char); 1339 1340 #endif 1341 1342 // Fill primitive arrays 1343 void generate_fill(BasicType t, bool aligned, 1344 Register to, Register value, Register count, 1345 Register rtmp, XMMRegister xtmp); 1346 1347 void encode_iso_array(Register src, Register dst, Register len, 1348 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3, 1349 XMMRegister tmp4, Register tmp5, Register result); 1350 1351 #ifdef _LP64 1352 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2); 1353 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart, 1354 Register y, Register y_idx, Register z, 1355 Register carry, Register product, 1356 Register idx, Register kdx); 1357 void multiply_add_128_x_128(Register x_xstart, Register y, Register z, 1358 Register yz_idx, Register idx, 1359 Register carry, Register product, int offset); 1360 void multiply_128_x_128_bmi2_loop(Register y, Register z, 1361 Register carry, Register carry2, 1362 Register idx, Register jdx, 1363 Register yz_idx1, Register yz_idx2, 1364 Register tmp, Register tmp3, Register tmp4); 1365 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z, 1366 Register yz_idx, Register idx, Register jdx, 1367 Register carry, Register product, 1368 Register carry2); 1369 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen, 1370 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5); 1371 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3, 1372 Register tmp4, Register tmp5, Register rdxReg, Register raxReg); 1373 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, 1374 Register tmp2); 1375 void multiply_add_64(Register sum, Register op1, Register op2, Register carry, 1376 Register rdxReg, Register raxReg); 1377 void add_one_64(Register z, Register zlen, Register carry, Register tmp1); 1378 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, 1379 Register tmp3, Register tmp4); 1380 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, 1381 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg); 1382 1383 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1, 1384 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, 1385 Register raxReg); 1386 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1, 1387 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, 1388 Register raxReg); 1389 void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale, 1390 Register result, Register tmp1, Register tmp2, 1391 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3); 1392 #endif 1393 1394 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic. 1395 void update_byte_crc32(Register crc, Register val, Register table); 1396 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp); 1397 // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic 1398 // Note on a naming convention: 1399 // Prefix w = register only used on a Westmere+ architecture 1400 // Prefix n = register only used on a Nehalem architecture 1401 #ifdef _LP64 1402 void crc32c_ipl_alg4(Register in_out, uint32_t n, 1403 Register tmp1, Register tmp2, Register tmp3); 1404 #else 1405 void crc32c_ipl_alg4(Register in_out, uint32_t n, 1406 Register tmp1, Register tmp2, Register tmp3, 1407 XMMRegister xtmp1, XMMRegister xtmp2); 1408 #endif 1409 void crc32c_pclmulqdq(XMMRegister w_xtmp1, 1410 Register in_out, 1411 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported, 1412 XMMRegister w_xtmp2, 1413 Register tmp1, 1414 Register n_tmp2, Register n_tmp3); 1415 void crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2, 1416 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1417 Register tmp1, Register tmp2, 1418 Register n_tmp3); 1419 void crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, 1420 Register in_out1, Register in_out2, Register in_out3, 1421 Register tmp1, Register tmp2, Register tmp3, 1422 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1423 Register tmp4, Register tmp5, 1424 Register n_tmp6); 1425 void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2, 1426 Register tmp1, Register tmp2, Register tmp3, 1427 Register tmp4, Register tmp5, Register tmp6, 1428 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1429 bool is_pclmulqdq_supported); 1430 // Fold 128-bit data chunk 1431 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset); 1432 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf); 1433 // Fold 8-bit data 1434 void fold_8bit_crc32(Register crc, Register table, Register tmp); 1435 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp); 1436 1437 // Compress char[] array to byte[]. 1438 void char_array_compress(Register src, Register dst, Register len, 1439 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3, 1440 XMMRegister tmp4, Register tmp5, Register result); 1441 1442 // Inflate byte[] array to char[]. 1443 void byte_array_inflate(Register src, Register dst, Register len, 1444 XMMRegister tmp1, Register tmp2); 1445 1446 #undef VIRTUAL 1447 1448 }; 1449 1450 /** 1451 * class SkipIfEqual: 1452 * 1453 * Instantiating this class will result in assembly code being output that will 1454 * jump around any code emitted between the creation of the instance and it's 1455 * automatic destruction at the end of a scope block, depending on the value of 1456 * the flag passed to the constructor, which will be checked at run-time. 1457 */ 1458 class SkipIfEqual { 1459 private: 1460 MacroAssembler* _masm; 1461 Label _label; 1462 1463 public: 1464 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 1465 ~SkipIfEqual(); 1466 }; 1467 1468 #endif // CPU_X86_VM_MACROASSEMBLER_X86_HPP