1 /* 2 * Copyright (c) 1997, 2018, 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 public: 42 // Support for VM calls 43 // 44 // This is the base routine called by the different versions of call_VM_leaf. The interpreter 45 // may customize this version by overriding it for its purposes (e.g., to save/restore 46 // additional registers when doing a VM call). 47 48 virtual void call_VM_leaf_base( 49 address entry_point, // the entry point 50 int number_of_arguments // the number of arguments to pop after the call 51 ); 52 53 protected: 54 // This is the base routine called by the different versions of call_VM. The interpreter 55 // may customize this version by overriding it for its purposes (e.g., to save/restore 56 // additional registers when doing a VM call). 57 // 58 // If no java_thread register is specified (noreg) than rdi will be used instead. call_VM_base 59 // returns the register which contains the thread upon return. If a thread register has been 60 // specified, the return value will correspond to that register. If no last_java_sp is specified 61 // (noreg) than rsp will be used instead. 62 virtual void call_VM_base( // returns the register containing the thread upon return 63 Register oop_result, // where an oop-result ends up if any; use noreg otherwise 64 Register java_thread, // the thread if computed before ; use noreg otherwise 65 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise 66 address entry_point, // the entry point 67 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call 68 bool check_exceptions // whether to check for pending exceptions after return 69 ); 70 71 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true); 72 73 // helpers for FPU flag access 74 // tmp is a temporary register, if none is available use noreg 75 void save_rax (Register tmp); 76 void restore_rax(Register tmp); 77 78 public: 79 MacroAssembler(CodeBuffer* code) : Assembler(code) {} 80 81 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code. 82 // The implementation is only non-empty for the InterpreterMacroAssembler, 83 // as only the interpreter handles PopFrame and ForceEarlyReturn requests. 84 virtual void check_and_handle_popframe(Register java_thread); 85 virtual void check_and_handle_earlyret(Register java_thread); 86 87 Address as_Address(AddressLiteral adr); 88 Address as_Address(ArrayAddress adr); 89 90 // Support for NULL-checks 91 // 92 // Generates code that causes a NULL OS exception if the content of reg is NULL. 93 // If the accessed location is M[reg + offset] and the offset is known, provide the 94 // offset. No explicit code generation is needed if the offset is within a certain 95 // range (0 <= offset <= page_size). 96 97 void null_check(Register reg, int offset = -1); 98 static bool needs_explicit_null_check(intptr_t offset); 99 100 // Required platform-specific helpers for Label::patch_instructions. 101 // They _shadow_ the declarations in AbstractAssembler, which are undefined. 102 void pd_patch_instruction(address branch, address target, const char* file, int line) { 103 unsigned char op = branch[0]; 104 assert(op == 0xE8 /* call */ || 105 op == 0xE9 /* jmp */ || 106 op == 0xEB /* short jmp */ || 107 (op & 0xF0) == 0x70 /* short jcc */ || 108 op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ || 109 op == 0xC7 && branch[1] == 0xF8 /* xbegin */, 110 "Invalid opcode at patch point"); 111 112 if (op == 0xEB || (op & 0xF0) == 0x70) { 113 // short offset operators (jmp and jcc) 114 char* disp = (char*) &branch[1]; 115 int imm8 = target - (address) &disp[1]; 116 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset at %s:%d", file, line); 117 *disp = imm8; 118 } else { 119 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1]; 120 int imm32 = target - (address) &disp[1]; 121 *disp = imm32; 122 } 123 } 124 125 // The following 4 methods return the offset of the appropriate move instruction 126 127 // Support for fast byte/short loading with zero extension (depending on particular CPU) 128 int load_unsigned_byte(Register dst, Address src); 129 int load_unsigned_short(Register dst, Address src); 130 131 // Support for fast byte/short loading with sign extension (depending on particular CPU) 132 int load_signed_byte(Register dst, Address src); 133 int load_signed_short(Register dst, Address src); 134 135 // Support for sign-extension (hi:lo = extend_sign(lo)) 136 void extend_sign(Register hi, Register lo); 137 138 // Load and store values by size and signed-ness 139 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg); 140 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg); 141 142 // Support for inc/dec with optimal instruction selection depending on value 143 144 void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; } 145 void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; } 146 147 void decrementl(Address dst, int value = 1); 148 void decrementl(Register reg, int value = 1); 149 150 void decrementq(Register reg, int value = 1); 151 void decrementq(Address dst, int value = 1); 152 153 void incrementl(Address dst, int value = 1); 154 void incrementl(Register reg, int value = 1); 155 156 void incrementq(Register reg, int value = 1); 157 void incrementq(Address dst, int value = 1); 158 159 // special instructions for EVEX 160 void setvectmask(Register dst, Register src); 161 void restorevectmask(); 162 163 // Support optimal SSE move instructions. 164 void movflt(XMMRegister dst, XMMRegister src) { 165 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; } 166 else { movss (dst, src); return; } 167 } 168 void movflt(XMMRegister dst, Address src) { movss(dst, src); } 169 void movflt(XMMRegister dst, AddressLiteral src); 170 void movflt(Address dst, XMMRegister src) { movss(dst, src); } 171 172 void movdbl(XMMRegister dst, XMMRegister src) { 173 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; } 174 else { movsd (dst, src); return; } 175 } 176 177 void movdbl(XMMRegister dst, AddressLiteral src); 178 179 void movdbl(XMMRegister dst, Address src) { 180 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; } 181 else { movlpd(dst, src); return; } 182 } 183 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); } 184 185 void incrementl(AddressLiteral dst); 186 void incrementl(ArrayAddress dst); 187 188 void incrementq(AddressLiteral dst); 189 190 // Alignment 191 void align(int modulus); 192 void align(int modulus, int target); 193 194 // A 5 byte nop that is safe for patching (see patch_verified_entry) 195 void fat_nop(); 196 197 // Stack frame creation/removal 198 void enter(); 199 void leave(); 200 201 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information) 202 // The pointer will be loaded into the thread register. 203 void get_thread(Register thread); 204 205 206 // Support for VM calls 207 // 208 // It is imperative that all calls into the VM are handled via the call_VM macros. 209 // They make sure that the stack linkage is setup correctly. call_VM's correspond 210 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points. 211 212 213 void call_VM(Register oop_result, 214 address entry_point, 215 bool check_exceptions = true); 216 void call_VM(Register oop_result, 217 address entry_point, 218 Register arg_1, 219 bool check_exceptions = true); 220 void call_VM(Register oop_result, 221 address entry_point, 222 Register arg_1, Register arg_2, 223 bool check_exceptions = true); 224 void call_VM(Register oop_result, 225 address entry_point, 226 Register arg_1, Register arg_2, Register arg_3, 227 bool check_exceptions = true); 228 229 // Overloadings with last_Java_sp 230 void call_VM(Register oop_result, 231 Register last_java_sp, 232 address entry_point, 233 int number_of_arguments = 0, 234 bool check_exceptions = true); 235 void call_VM(Register oop_result, 236 Register last_java_sp, 237 address entry_point, 238 Register arg_1, bool 239 check_exceptions = true); 240 void call_VM(Register oop_result, 241 Register last_java_sp, 242 address entry_point, 243 Register arg_1, Register arg_2, 244 bool check_exceptions = true); 245 void call_VM(Register oop_result, 246 Register last_java_sp, 247 address entry_point, 248 Register arg_1, Register arg_2, Register arg_3, 249 bool check_exceptions = true); 250 251 void get_vm_result (Register oop_result, Register thread); 252 void get_vm_result_2(Register metadata_result, Register thread); 253 254 // These always tightly bind to MacroAssembler::call_VM_base 255 // bypassing the virtual implementation 256 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true); 257 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true); 258 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true); 259 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); 260 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); 261 262 void call_VM_leaf0(address entry_point); 263 void call_VM_leaf(address entry_point, 264 int number_of_arguments = 0); 265 void call_VM_leaf(address entry_point, 266 Register arg_1); 267 void call_VM_leaf(address entry_point, 268 Register arg_1, Register arg_2); 269 void call_VM_leaf(address entry_point, 270 Register arg_1, Register arg_2, Register arg_3); 271 272 // These always tightly bind to MacroAssembler::call_VM_leaf_base 273 // bypassing the virtual implementation 274 void super_call_VM_leaf(address entry_point); 275 void super_call_VM_leaf(address entry_point, Register arg_1); 276 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2); 277 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3); 278 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4); 279 280 // last Java Frame (fills frame anchor) 281 void set_last_Java_frame(Register thread, 282 Register last_java_sp, 283 Register last_java_fp, 284 address last_java_pc); 285 286 // thread in the default location (r15_thread on 64bit) 287 void set_last_Java_frame(Register last_java_sp, 288 Register last_java_fp, 289 address last_java_pc); 290 291 void reset_last_Java_frame(Register thread, bool clear_fp); 292 293 // thread in the default location (r15_thread on 64bit) 294 void reset_last_Java_frame(bool clear_fp); 295 296 // jobjects 297 void clear_jweak_tag(Register possibly_jweak); 298 void resolve_jobject(Register value, Register thread, Register tmp); 299 300 // C 'boolean' to Java boolean: x == 0 ? 0 : 1 301 void c2bool(Register x); 302 303 // C++ bool manipulation 304 305 void movbool(Register dst, Address src); 306 void movbool(Address dst, bool boolconst); 307 void movbool(Address dst, Register src); 308 void testbool(Register dst); 309 310 void resolve_oop_handle(Register result, Register tmp = rscratch2); 311 void load_mirror(Register mirror, Register method, Register tmp = rscratch2); 312 313 // oop manipulations 314 void load_klass(Register dst, Register src); 315 void store_klass(Register dst, Register src); 316 317 void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src, 318 Register tmp1, Register thread_tmp); 319 void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register src, 320 Register tmp1, Register tmp2); 321 322 // Resolves obj access. Result is placed in the same register. 323 // All other registers are preserved. 324 void resolve(DecoratorSet decorators, Register obj); 325 326 void load_heap_oop(Register dst, Address src, Register tmp1 = noreg, 327 Register thread_tmp = noreg, DecoratorSet decorators = 0); 328 void load_heap_oop_not_null(Register dst, Address src, Register tmp1 = noreg, 329 Register thread_tmp = noreg, DecoratorSet decorators = 0); 330 void store_heap_oop(Address dst, Register src, Register tmp1 = noreg, 331 Register tmp2 = noreg, DecoratorSet decorators = 0); 332 333 // Used for storing NULL. All other oop constants should be 334 // stored using routines that take a jobject. 335 void store_heap_oop_null(Address dst); 336 337 void load_prototype_header(Register dst, Register src); 338 339 #ifdef _LP64 340 void store_klass_gap(Register dst, Register src); 341 342 // This dummy is to prevent a call to store_heap_oop from 343 // converting a zero (like NULL) into a Register by giving 344 // the compiler two choices it can't resolve 345 346 void store_heap_oop(Address dst, void* dummy); 347 348 void encode_heap_oop(Register r); 349 void decode_heap_oop(Register r); 350 void encode_heap_oop_not_null(Register r); 351 void decode_heap_oop_not_null(Register r); 352 void encode_heap_oop_not_null(Register dst, Register src); 353 void decode_heap_oop_not_null(Register dst, Register src); 354 355 void set_narrow_oop(Register dst, jobject obj); 356 void set_narrow_oop(Address dst, jobject obj); 357 void cmp_narrow_oop(Register dst, jobject obj); 358 void cmp_narrow_oop(Address dst, jobject obj); 359 360 void encode_klass_not_null(Register r); 361 void decode_klass_not_null(Register r); 362 void encode_klass_not_null(Register dst, Register src); 363 void decode_klass_not_null(Register dst, Register src); 364 void set_narrow_klass(Register dst, Klass* k); 365 void set_narrow_klass(Address dst, Klass* k); 366 void cmp_narrow_klass(Register dst, Klass* k); 367 void cmp_narrow_klass(Address dst, Klass* k); 368 369 // Returns the byte size of the instructions generated by decode_klass_not_null() 370 // when compressed klass pointers are being used. 371 static int instr_size_for_decode_klass_not_null(); 372 373 // if heap base register is used - reinit it with the correct value 374 void reinit_heapbase(); 375 376 DEBUG_ONLY(void verify_heapbase(const char* msg);) 377 378 #endif // _LP64 379 380 // Int division/remainder for Java 381 // (as idivl, but checks for special case as described in JVM spec.) 382 // returns idivl instruction offset for implicit exception handling 383 int corrected_idivl(Register reg); 384 385 // Long division/remainder for Java 386 // (as idivq, but checks for special case as described in JVM spec.) 387 // returns idivq instruction offset for implicit exception handling 388 int corrected_idivq(Register reg); 389 390 void int3(); 391 392 // Long operation macros for a 32bit cpu 393 // Long negation for Java 394 void lneg(Register hi, Register lo); 395 396 // Long multiplication for Java 397 // (destroys contents of eax, ebx, ecx and edx) 398 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y 399 400 // Long shifts for Java 401 // (semantics as described in JVM spec.) 402 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f) 403 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f) 404 405 // Long compare for Java 406 // (semantics as described in JVM spec.) 407 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y) 408 409 410 // misc 411 412 // Sign extension 413 void sign_extend_short(Register reg); 414 void sign_extend_byte(Register reg); 415 416 // Division by power of 2, rounding towards 0 417 void division_with_shift(Register reg, int shift_value); 418 419 // Compares the top-most stack entries on the FPU stack and sets the eflags as follows: 420 // 421 // CF (corresponds to C0) if x < y 422 // PF (corresponds to C2) if unordered 423 // ZF (corresponds to C3) if x = y 424 // 425 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 426 // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code) 427 void fcmp(Register tmp); 428 // Variant of the above which allows y to be further down the stack 429 // and which only pops x and y if specified. If pop_right is 430 // specified then pop_left must also be specified. 431 void fcmp(Register tmp, int index, bool pop_left, bool pop_right); 432 433 // Floating-point comparison for Java 434 // Compares the top-most stack entries on the FPU stack and stores the result in dst. 435 // The arguments are in reversed order on the stack (i.e., top of stack is first argument). 436 // (semantics as described in JVM spec.) 437 void fcmp2int(Register dst, bool unordered_is_less); 438 // Variant of the above which allows y to be further down the stack 439 // and which only pops x and y if specified. If pop_right is 440 // specified then pop_left must also be specified. 441 void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right); 442 443 // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards) 444 // tmp is a temporary register, if none is available use noreg 445 void fremr(Register tmp); 446 447 // dst = c = a * b + c 448 void fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c); 449 void fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c); 450 451 void vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len); 452 void vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len); 453 void vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len); 454 void vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len); 455 456 457 // same as fcmp2int, but using SSE2 458 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 459 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less); 460 461 // branch to L if FPU flag C2 is set/not set 462 // tmp is a temporary register, if none is available use noreg 463 void jC2 (Register tmp, Label& L); 464 void jnC2(Register tmp, Label& L); 465 466 // Pop ST (ffree & fincstp combined) 467 void fpop(); 468 469 // Load float value from 'address'. If UseSSE >= 1, the value is loaded into 470 // register xmm0. Otherwise, the value is loaded onto the FPU stack. 471 void load_float(Address src); 472 473 // Store float value to 'address'. If UseSSE >= 1, the value is stored 474 // from register xmm0. Otherwise, the value is stored from the FPU stack. 475 void store_float(Address dst); 476 477 // Load double value from 'address'. If UseSSE >= 2, the value is loaded into 478 // register xmm0. Otherwise, the value is loaded onto the FPU stack. 479 void load_double(Address src); 480 481 // Store double value to 'address'. If UseSSE >= 2, the value is stored 482 // from register xmm0. Otherwise, the value is stored from the FPU stack. 483 void store_double(Address dst); 484 485 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack 486 void push_fTOS(); 487 488 // pops double TOS element from CPU stack and pushes on FPU stack 489 void pop_fTOS(); 490 491 void empty_FPU_stack(); 492 493 void push_IU_state(); 494 void pop_IU_state(); 495 496 void push_FPU_state(); 497 void pop_FPU_state(); 498 499 void push_CPU_state(); 500 void pop_CPU_state(); 501 502 // Round up to a power of two 503 void round_to(Register reg, int modulus); 504 505 // Callee saved registers handling 506 void push_callee_saved_registers(); 507 void pop_callee_saved_registers(); 508 509 // allocation 510 void eden_allocate( 511 Register thread, // Current thread 512 Register obj, // result: pointer to object after successful allocation 513 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 514 int con_size_in_bytes, // object size in bytes if known at compile time 515 Register t1, // temp register 516 Label& slow_case // continuation point if fast allocation fails 517 ); 518 void tlab_allocate( 519 Register thread, // Current thread 520 Register obj, // result: pointer to object after successful allocation 521 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise 522 int con_size_in_bytes, // object size in bytes if known at compile time 523 Register t1, // temp register 524 Register t2, // temp register 525 Label& slow_case // continuation point if fast allocation fails 526 ); 527 void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp); 528 529 // interface method calling 530 void lookup_interface_method(Register recv_klass, 531 Register intf_klass, 532 RegisterOrConstant itable_index, 533 Register method_result, 534 Register scan_temp, 535 Label& no_such_interface, 536 bool return_method = true); 537 538 // virtual method calling 539 void lookup_virtual_method(Register recv_klass, 540 RegisterOrConstant vtable_index, 541 Register method_result); 542 543 // Test sub_klass against super_klass, with fast and slow paths. 544 545 // The fast path produces a tri-state answer: yes / no / maybe-slow. 546 // One of the three labels can be NULL, meaning take the fall-through. 547 // If super_check_offset is -1, the value is loaded up from super_klass. 548 // No registers are killed, except temp_reg. 549 void check_klass_subtype_fast_path(Register sub_klass, 550 Register super_klass, 551 Register temp_reg, 552 Label* L_success, 553 Label* L_failure, 554 Label* L_slow_path, 555 RegisterOrConstant super_check_offset = RegisterOrConstant(-1)); 556 557 // The rest of the type check; must be wired to a corresponding fast path. 558 // It does not repeat the fast path logic, so don't use it standalone. 559 // The temp_reg and temp2_reg can be noreg, if no temps are available. 560 // Updates the sub's secondary super cache as necessary. 561 // If set_cond_codes, condition codes will be Z on success, NZ on failure. 562 void check_klass_subtype_slow_path(Register sub_klass, 563 Register super_klass, 564 Register temp_reg, 565 Register temp2_reg, 566 Label* L_success, 567 Label* L_failure, 568 bool set_cond_codes = false); 569 570 // Simplified, combined version, good for typical uses. 571 // Falls through on failure. 572 void check_klass_subtype(Register sub_klass, 573 Register super_klass, 574 Register temp_reg, 575 Label& L_success); 576 577 // method handles (JSR 292) 578 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0); 579 580 //---- 581 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 582 583 // Debugging 584 585 // only if +VerifyOops 586 // TODO: Make these macros with file and line like sparc version! 587 void verify_oop(Register reg, const char* s = "broken oop"); 588 void verify_oop_addr(Address addr, const char * s = "broken oop addr"); 589 590 // TODO: verify method and klass metadata (compare against vptr?) 591 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {} 592 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){} 593 594 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__) 595 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__) 596 597 // only if +VerifyFPU 598 void verify_FPU(int stack_depth, const char* s = "illegal FPU state"); 599 600 // Verify or restore cpu control state after JNI call 601 void restore_cpu_control_state_after_jni(); 602 603 // prints msg, dumps registers and stops execution 604 void stop(const char* msg); 605 606 // prints msg and continues 607 void warn(const char* msg); 608 609 // dumps registers and other state 610 void print_state(); 611 612 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg); 613 static void debug64(char* msg, int64_t pc, int64_t regs[]); 614 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip); 615 static void print_state64(int64_t pc, int64_t regs[]); 616 617 void os_breakpoint(); 618 619 void untested() { stop("untested"); } 620 621 void unimplemented(const char* what = ""); 622 623 void should_not_reach_here() { stop("should not reach here"); } 624 625 void print_CPU_state(); 626 627 // Stack overflow checking 628 void bang_stack_with_offset(int offset) { 629 // stack grows down, caller passes positive offset 630 assert(offset > 0, "must bang with negative offset"); 631 movl(Address(rsp, (-offset)), rax); 632 } 633 634 // Writes to stack successive pages until offset reached to check for 635 // stack overflow + shadow pages. Also, clobbers tmp 636 void bang_stack_size(Register size, Register tmp); 637 638 // Check for reserved stack access in method being exited (for JIT) 639 void reserved_stack_check(); 640 641 virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, 642 Register tmp, 643 int offset); 644 645 // Support for serializing memory accesses between threads 646 void serialize_memory(Register thread, Register tmp); 647 648 // If thread_reg is != noreg the code assumes the register passed contains 649 // the thread (required on 64 bit). 650 void safepoint_poll(Label& slow_path, Register thread_reg, Register temp_reg); 651 652 void verify_tlab(); 653 654 // Biased locking support 655 // lock_reg and obj_reg must be loaded up with the appropriate values. 656 // swap_reg must be rax, and is killed. 657 // tmp_reg is optional. If it is supplied (i.e., != noreg) it will 658 // be killed; if not supplied, push/pop will be used internally to 659 // allocate a temporary (inefficient, avoid if possible). 660 // Optional slow case is for implementations (interpreter and C1) which branch to 661 // slow case directly. Leaves condition codes set for C2's Fast_Lock node. 662 // Returns offset of first potentially-faulting instruction for null 663 // check info (currently consumed only by C1). If 664 // swap_reg_contains_mark is true then returns -1 as it is assumed 665 // the calling code has already passed any potential faults. 666 int biased_locking_enter(Register lock_reg, Register obj_reg, 667 Register swap_reg, Register tmp_reg, 668 bool swap_reg_contains_mark, 669 Label& done, Label* slow_case = NULL, 670 BiasedLockingCounters* counters = NULL); 671 void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done); 672 #ifdef COMPILER2 673 // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file. 674 // See full desription in macroAssembler_x86.cpp. 675 void fast_lock(Register obj, Register box, Register tmp, 676 Register scr, Register cx1, Register cx2, 677 BiasedLockingCounters* counters, 678 RTMLockingCounters* rtm_counters, 679 RTMLockingCounters* stack_rtm_counters, 680 Metadata* method_data, 681 bool use_rtm, bool profile_rtm); 682 void fast_unlock(Register obj, Register box, Register tmp, bool use_rtm); 683 #if INCLUDE_RTM_OPT 684 void rtm_counters_update(Register abort_status, Register rtm_counters); 685 void branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel); 686 void rtm_abort_ratio_calculation(Register tmp, Register rtm_counters_reg, 687 RTMLockingCounters* rtm_counters, 688 Metadata* method_data); 689 void rtm_profiling(Register abort_status_Reg, Register rtm_counters_Reg, 690 RTMLockingCounters* rtm_counters, Metadata* method_data, bool profile_rtm); 691 void rtm_retry_lock_on_abort(Register retry_count, Register abort_status, Label& retryLabel); 692 void rtm_retry_lock_on_busy(Register retry_count, Register box, Register tmp, Register scr, Label& retryLabel); 693 void rtm_stack_locking(Register obj, Register tmp, Register scr, 694 Register retry_on_abort_count, 695 RTMLockingCounters* stack_rtm_counters, 696 Metadata* method_data, bool profile_rtm, 697 Label& DONE_LABEL, Label& IsInflated); 698 void rtm_inflated_locking(Register obj, Register box, Register tmp, 699 Register scr, Register retry_on_busy_count, 700 Register retry_on_abort_count, 701 RTMLockingCounters* rtm_counters, 702 Metadata* method_data, bool profile_rtm, 703 Label& DONE_LABEL); 704 #endif 705 #endif 706 707 Condition negate_condition(Condition cond); 708 709 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit 710 // operands. In general the names are modified to avoid hiding the instruction in Assembler 711 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers 712 // here in MacroAssembler. The major exception to this rule is call 713 714 // Arithmetics 715 716 717 void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; } 718 void addptr(Address dst, Register src); 719 720 void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); } 721 void addptr(Register dst, int32_t src); 722 void addptr(Register dst, Register src); 723 void addptr(Register dst, RegisterOrConstant src) { 724 if (src.is_constant()) addptr(dst, (int) src.as_constant()); 725 else addptr(dst, src.as_register()); 726 } 727 728 void andptr(Register dst, int32_t src); 729 void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; } 730 731 void cmp8(AddressLiteral src1, int imm); 732 733 // renamed to drag out the casting of address to int32_t/intptr_t 734 void cmp32(Register src1, int32_t imm); 735 736 void cmp32(AddressLiteral src1, int32_t imm); 737 // compare reg - mem, or reg - &mem 738 void cmp32(Register src1, AddressLiteral src2); 739 740 void cmp32(Register src1, Address src2); 741 742 #ifndef _LP64 743 void cmpklass(Address dst, Metadata* obj); 744 void cmpklass(Register dst, Metadata* obj); 745 void cmpoop(Address dst, jobject obj); 746 void cmpoop_raw(Address dst, jobject obj); 747 #endif // _LP64 748 749 void cmpoop(Register src1, Register src2); 750 void cmpoop(Register src1, Address src2); 751 void cmpoop(Register dst, jobject obj); 752 void cmpoop_raw(Register dst, jobject obj); 753 754 // NOTE src2 must be the lval. This is NOT an mem-mem compare 755 void cmpptr(Address src1, AddressLiteral src2); 756 757 void cmpptr(Register src1, AddressLiteral src2); 758 759 void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 760 void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 761 // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 762 763 void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 764 void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; } 765 766 // cmp64 to avoild hiding cmpq 767 void cmp64(Register src1, AddressLiteral src); 768 769 void cmpxchgptr(Register reg, Address adr); 770 771 void locked_cmpxchgptr(Register reg, AddressLiteral adr); 772 773 774 void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); } 775 void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); } 776 777 778 void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); } 779 780 void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); } 781 782 void shlptr(Register dst, int32_t shift); 783 void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); } 784 785 void shrptr(Register dst, int32_t shift); 786 void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); } 787 788 void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); } 789 void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); } 790 791 void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 792 793 void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); } 794 void subptr(Register dst, int32_t src); 795 // Force generation of a 4 byte immediate value even if it fits into 8bit 796 void subptr_imm32(Register dst, int32_t src); 797 void subptr(Register dst, Register src); 798 void subptr(Register dst, RegisterOrConstant src) { 799 if (src.is_constant()) subptr(dst, (int) src.as_constant()); 800 else subptr(dst, src.as_register()); 801 } 802 803 void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 804 void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); } 805 806 void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 807 void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; } 808 809 void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; } 810 811 812 813 // Helper functions for statistics gathering. 814 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes. 815 void cond_inc32(Condition cond, AddressLiteral counter_addr); 816 // Unconditional atomic increment. 817 void atomic_incl(Address counter_addr); 818 void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1); 819 #ifdef _LP64 820 void atomic_incq(Address counter_addr); 821 void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1); 822 #endif 823 void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; } 824 void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; } 825 826 void lea(Register dst, AddressLiteral adr); 827 void lea(Address dst, AddressLiteral adr); 828 void lea(Register dst, Address adr) { Assembler::lea(dst, adr); } 829 830 void leal32(Register dst, Address src) { leal(dst, src); } 831 832 // Import other testl() methods from the parent class or else 833 // they will be hidden by the following overriding declaration. 834 using Assembler::testl; 835 void testl(Register dst, AddressLiteral src); 836 837 void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 838 void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 839 void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); } 840 void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); } 841 842 void testptr(Register src, int32_t imm32) { LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); } 843 void testptr(Register src1, Address src2) { LP64_ONLY(testq(src1, src2)) NOT_LP64(testl(src1, src2)); } 844 void testptr(Register src1, Register src2); 845 846 void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 847 void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); } 848 849 // Calls 850 851 void call(Label& L, relocInfo::relocType rtype); 852 void call(Register entry); 853 854 // NOTE: this call transfers to the effective address of entry NOT 855 // the address contained by entry. This is because this is more natural 856 // for jumps/calls. 857 void call(AddressLiteral entry); 858 859 // Emit the CompiledIC call idiom 860 void ic_call(address entry, jint method_index = 0); 861 862 // Jumps 863 864 // NOTE: these jumps tranfer to the effective address of dst NOT 865 // the address contained by dst. This is because this is more natural 866 // for jumps/calls. 867 void jump(AddressLiteral dst); 868 void jump_cc(Condition cc, AddressLiteral dst); 869 870 // 32bit can do a case table jump in one instruction but we no longer allow the base 871 // to be installed in the Address class. This jump will tranfers to the address 872 // contained in the location described by entry (not the address of entry) 873 void jump(ArrayAddress entry); 874 875 // Floating 876 877 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); } 878 void andpd(XMMRegister dst, AddressLiteral src); 879 void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); } 880 881 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); } 882 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); } 883 void andps(XMMRegister dst, AddressLiteral src); 884 885 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); } 886 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); } 887 void comiss(XMMRegister dst, AddressLiteral src); 888 889 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); } 890 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); } 891 void comisd(XMMRegister dst, AddressLiteral src); 892 893 void fadd_s(Address src) { Assembler::fadd_s(src); } 894 void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); } 895 896 void fldcw(Address src) { Assembler::fldcw(src); } 897 void fldcw(AddressLiteral src); 898 899 void fld_s(int index) { Assembler::fld_s(index); } 900 void fld_s(Address src) { Assembler::fld_s(src); } 901 void fld_s(AddressLiteral src); 902 903 void fld_d(Address src) { Assembler::fld_d(src); } 904 void fld_d(AddressLiteral src); 905 906 void fld_x(Address src) { Assembler::fld_x(src); } 907 void fld_x(AddressLiteral src); 908 909 void fmul_s(Address src) { Assembler::fmul_s(src); } 910 void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); } 911 912 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); } 913 void ldmxcsr(AddressLiteral src); 914 915 #ifdef _LP64 916 private: 917 void sha256_AVX2_one_round_compute( 918 Register reg_old_h, 919 Register reg_a, 920 Register reg_b, 921 Register reg_c, 922 Register reg_d, 923 Register reg_e, 924 Register reg_f, 925 Register reg_g, 926 Register reg_h, 927 int iter); 928 void sha256_AVX2_four_rounds_compute_first(int start); 929 void sha256_AVX2_four_rounds_compute_last(int start); 930 void sha256_AVX2_one_round_and_sched( 931 XMMRegister xmm_0, /* == ymm4 on 0, 1, 2, 3 iterations, then rotate 4 registers left on 4, 8, 12 iterations */ 932 XMMRegister xmm_1, /* ymm5 */ /* full cycle is 16 iterations */ 933 XMMRegister xmm_2, /* ymm6 */ 934 XMMRegister xmm_3, /* ymm7 */ 935 Register reg_a, /* == eax on 0 iteration, then rotate 8 register right on each next iteration */ 936 Register reg_b, /* ebx */ /* full cycle is 8 iterations */ 937 Register reg_c, /* edi */ 938 Register reg_d, /* esi */ 939 Register reg_e, /* r8d */ 940 Register reg_f, /* r9d */ 941 Register reg_g, /* r10d */ 942 Register reg_h, /* r11d */ 943 int iter); 944 945 void addm(int disp, Register r1, Register r2); 946 947 public: 948 void sha256_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0, 949 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4, 950 Register buf, Register state, Register ofs, Register limit, Register rsp, 951 bool multi_block, XMMRegister shuf_mask); 952 #endif 953 954 #ifdef _LP64 955 private: 956 void sha512_AVX2_one_round_compute(Register old_h, Register a, Register b, Register c, Register d, 957 Register e, Register f, Register g, Register h, int iteration); 958 959 void sha512_AVX2_one_round_and_schedule(XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 960 Register a, Register b, Register c, Register d, Register e, Register f, 961 Register g, Register h, int iteration); 962 963 void addmq(int disp, Register r1, Register r2); 964 public: 965 void sha512_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0, 966 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4, 967 Register buf, Register state, Register ofs, Register limit, Register rsp, bool multi_block, 968 XMMRegister shuf_mask); 969 #endif 970 971 void fast_sha1(XMMRegister abcd, XMMRegister e0, XMMRegister e1, XMMRegister msg0, 972 XMMRegister msg1, XMMRegister msg2, XMMRegister msg3, XMMRegister shuf_mask, 973 Register buf, Register state, Register ofs, Register limit, Register rsp, 974 bool multi_block); 975 976 #ifdef _LP64 977 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0, 978 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4, 979 Register buf, Register state, Register ofs, Register limit, Register rsp, 980 bool multi_block, XMMRegister shuf_mask); 981 #else 982 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0, 983 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4, 984 Register buf, Register state, Register ofs, Register limit, Register rsp, 985 bool multi_block); 986 #endif 987 988 void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 989 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 990 Register rax, Register rcx, Register rdx, Register tmp); 991 992 #ifdef _LP64 993 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 994 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 995 Register rax, Register rcx, Register rdx, Register tmp1, Register tmp2); 996 997 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 998 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 999 Register rax, Register rcx, Register rdx, Register r11); 1000 1001 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4, 1002 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx, 1003 Register rdx, Register tmp1, Register tmp2, Register tmp3, Register tmp4); 1004 1005 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1006 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1007 Register rax, Register rbx, Register rcx, Register rdx, Register tmp1, Register tmp2, 1008 Register tmp3, Register tmp4); 1009 1010 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1011 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1012 Register rax, Register rcx, Register rdx, Register tmp1, 1013 Register tmp2, Register tmp3, Register tmp4); 1014 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1015 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1016 Register rax, Register rcx, Register rdx, Register tmp1, 1017 Register tmp2, Register tmp3, Register tmp4); 1018 #else 1019 void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1020 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1021 Register rax, Register rcx, Register rdx, Register tmp1); 1022 1023 void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1024 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1025 Register rax, Register rcx, Register rdx, Register tmp); 1026 1027 void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4, 1028 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx, 1029 Register rdx, Register tmp); 1030 1031 void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1032 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1033 Register rax, Register rbx, Register rdx); 1034 1035 void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1036 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1037 Register rax, Register rcx, Register rdx, Register tmp); 1038 1039 void libm_sincos_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx, 1040 Register edx, Register ebx, Register esi, Register edi, 1041 Register ebp, Register esp); 1042 1043 void libm_reduce_pi04l(Register eax, Register ecx, Register edx, Register ebx, 1044 Register esi, Register edi, Register ebp, Register esp); 1045 1046 void libm_tancot_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx, 1047 Register edx, Register ebx, Register esi, Register edi, 1048 Register ebp, Register esp); 1049 1050 void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, 1051 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, 1052 Register rax, Register rcx, Register rdx, Register tmp); 1053 #endif 1054 1055 void increase_precision(); 1056 void restore_precision(); 1057 1058 private: 1059 1060 // these are private because users should be doing movflt/movdbl 1061 1062 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); } 1063 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); } 1064 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); } 1065 void movss(XMMRegister dst, AddressLiteral src); 1066 1067 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); } 1068 void movlpd(XMMRegister dst, AddressLiteral src); 1069 1070 public: 1071 1072 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); } 1073 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); } 1074 void addsd(XMMRegister dst, AddressLiteral src); 1075 1076 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); } 1077 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); } 1078 void addss(XMMRegister dst, AddressLiteral src); 1079 1080 void addpd(XMMRegister dst, XMMRegister src) { Assembler::addpd(dst, src); } 1081 void addpd(XMMRegister dst, Address src) { Assembler::addpd(dst, src); } 1082 void addpd(XMMRegister dst, AddressLiteral src); 1083 1084 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); } 1085 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); } 1086 void divsd(XMMRegister dst, AddressLiteral src); 1087 1088 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); } 1089 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); } 1090 void divss(XMMRegister dst, AddressLiteral src); 1091 1092 // Move Unaligned Double Quadword 1093 void movdqu(Address dst, XMMRegister src); 1094 void movdqu(XMMRegister dst, Address src); 1095 void movdqu(XMMRegister dst, XMMRegister src); 1096 void movdqu(XMMRegister dst, AddressLiteral src, Register scratchReg = rscratch1); 1097 // AVX Unaligned forms 1098 void vmovdqu(Address dst, XMMRegister src); 1099 void vmovdqu(XMMRegister dst, Address src); 1100 void vmovdqu(XMMRegister dst, XMMRegister src); 1101 void vmovdqu(XMMRegister dst, AddressLiteral src); 1102 void evmovdquq(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); } 1103 void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); } 1104 void evmovdquq(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); } 1105 void evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch); 1106 1107 // Move Aligned Double Quadword 1108 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); } 1109 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); } 1110 void movdqa(XMMRegister dst, AddressLiteral src); 1111 1112 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); } 1113 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); } 1114 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); } 1115 void movsd(XMMRegister dst, AddressLiteral src); 1116 1117 void mulpd(XMMRegister dst, XMMRegister src) { Assembler::mulpd(dst, src); } 1118 void mulpd(XMMRegister dst, Address src) { Assembler::mulpd(dst, src); } 1119 void mulpd(XMMRegister dst, AddressLiteral src); 1120 1121 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); } 1122 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); } 1123 void mulsd(XMMRegister dst, AddressLiteral src); 1124 1125 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); } 1126 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); } 1127 void mulss(XMMRegister dst, AddressLiteral src); 1128 1129 // Carry-Less Multiplication Quadword 1130 void pclmulldq(XMMRegister dst, XMMRegister src) { 1131 // 0x00 - multiply lower 64 bits [0:63] 1132 Assembler::pclmulqdq(dst, src, 0x00); 1133 } 1134 void pclmulhdq(XMMRegister dst, XMMRegister src) { 1135 // 0x11 - multiply upper 64 bits [64:127] 1136 Assembler::pclmulqdq(dst, src, 0x11); 1137 } 1138 1139 void pcmpeqb(XMMRegister dst, XMMRegister src); 1140 void pcmpeqw(XMMRegister dst, XMMRegister src); 1141 1142 void pcmpestri(XMMRegister dst, Address src, int imm8); 1143 void pcmpestri(XMMRegister dst, XMMRegister src, int imm8); 1144 1145 void pmovzxbw(XMMRegister dst, XMMRegister src); 1146 void pmovzxbw(XMMRegister dst, Address src); 1147 1148 void pmovmskb(Register dst, XMMRegister src); 1149 1150 void ptest(XMMRegister dst, XMMRegister src); 1151 1152 void sqrtsd(XMMRegister dst, XMMRegister src) { Assembler::sqrtsd(dst, src); } 1153 void sqrtsd(XMMRegister dst, Address src) { Assembler::sqrtsd(dst, src); } 1154 void sqrtsd(XMMRegister dst, AddressLiteral src); 1155 1156 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); } 1157 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); } 1158 void sqrtss(XMMRegister dst, AddressLiteral src); 1159 1160 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); } 1161 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); } 1162 void subsd(XMMRegister dst, AddressLiteral src); 1163 1164 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); } 1165 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); } 1166 void subss(XMMRegister dst, AddressLiteral src); 1167 1168 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); } 1169 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); } 1170 void ucomiss(XMMRegister dst, AddressLiteral src); 1171 1172 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); } 1173 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); } 1174 void ucomisd(XMMRegister dst, AddressLiteral src); 1175 1176 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values 1177 void xorpd(XMMRegister dst, XMMRegister src); 1178 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); } 1179 void xorpd(XMMRegister dst, AddressLiteral src); 1180 1181 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values 1182 void xorps(XMMRegister dst, XMMRegister src); 1183 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); } 1184 void xorps(XMMRegister dst, AddressLiteral src); 1185 1186 // Shuffle Bytes 1187 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); } 1188 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); } 1189 void pshufb(XMMRegister dst, AddressLiteral src); 1190 // AVX 3-operands instructions 1191 1192 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); } 1193 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); } 1194 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1195 1196 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); } 1197 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); } 1198 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1199 1200 void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len); 1201 void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len); 1202 1203 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1204 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1205 1206 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1207 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1208 1209 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); } 1210 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); } 1211 void vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1212 1213 void vpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len); 1214 void vpbroadcastw(XMMRegister dst, Address src, int vector_len) { Assembler::vpbroadcastw(dst, src, vector_len); } 1215 1216 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1217 1218 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1219 1220 void vpmovzxbw(XMMRegister dst, Address src, int vector_len); 1221 void vpmovzxbw(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vpmovzxbw(dst, src, vector_len); } 1222 1223 void vpmovmskb(Register dst, XMMRegister src); 1224 1225 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1226 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1227 1228 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1229 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1230 1231 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len); 1232 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len); 1233 1234 void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1235 void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1236 1237 void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1238 void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1239 1240 void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len); 1241 void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len); 1242 1243 void vptest(XMMRegister dst, XMMRegister src); 1244 1245 void punpcklbw(XMMRegister dst, XMMRegister src); 1246 void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); } 1247 1248 void pshufd(XMMRegister dst, Address src, int mode); 1249 void pshufd(XMMRegister dst, XMMRegister src, int mode) { Assembler::pshufd(dst, src, mode); } 1250 1251 void pshuflw(XMMRegister dst, XMMRegister src, int mode); 1252 void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); } 1253 1254 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); } 1255 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); } 1256 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1257 1258 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); } 1259 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); } 1260 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1261 1262 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); } 1263 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); } 1264 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1265 1266 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); } 1267 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); } 1268 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1269 1270 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); } 1271 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); } 1272 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1273 1274 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); } 1275 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); } 1276 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1277 1278 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); } 1279 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); } 1280 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1281 1282 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); } 1283 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); } 1284 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1285 1286 void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1287 void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src); 1288 1289 // AVX Vector instructions 1290 1291 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); } 1292 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); } 1293 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1294 1295 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); } 1296 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); } 1297 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len); 1298 1299 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { 1300 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2 1301 Assembler::vpxor(dst, nds, src, vector_len); 1302 else 1303 Assembler::vxorpd(dst, nds, src, vector_len); 1304 } 1305 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { 1306 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2 1307 Assembler::vpxor(dst, nds, src, vector_len); 1308 else 1309 Assembler::vxorpd(dst, nds, src, vector_len); 1310 } 1311 1312 // Simple version for AVX2 256bit vectors 1313 void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); } 1314 void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); } 1315 1316 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8) { 1317 if (UseAVX > 2) { 1318 Assembler::vinserti32x4(dst, dst, src, imm8); 1319 } else if (UseAVX > 1) { 1320 // vinserti128 is available only in AVX2 1321 Assembler::vinserti128(dst, nds, src, imm8); 1322 } else { 1323 Assembler::vinsertf128(dst, nds, src, imm8); 1324 } 1325 } 1326 1327 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8) { 1328 if (UseAVX > 2) { 1329 Assembler::vinserti32x4(dst, dst, src, imm8); 1330 } else if (UseAVX > 1) { 1331 // vinserti128 is available only in AVX2 1332 Assembler::vinserti128(dst, nds, src, imm8); 1333 } else { 1334 Assembler::vinsertf128(dst, nds, src, imm8); 1335 } 1336 } 1337 1338 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8) { 1339 if (UseAVX > 2) { 1340 Assembler::vextracti32x4(dst, src, imm8); 1341 } else if (UseAVX > 1) { 1342 // vextracti128 is available only in AVX2 1343 Assembler::vextracti128(dst, src, imm8); 1344 } else { 1345 Assembler::vextractf128(dst, src, imm8); 1346 } 1347 } 1348 1349 void vextracti128(Address dst, XMMRegister src, uint8_t imm8) { 1350 if (UseAVX > 2) { 1351 Assembler::vextracti32x4(dst, src, imm8); 1352 } else if (UseAVX > 1) { 1353 // vextracti128 is available only in AVX2 1354 Assembler::vextracti128(dst, src, imm8); 1355 } else { 1356 Assembler::vextractf128(dst, src, imm8); 1357 } 1358 } 1359 1360 // 128bit copy to/from high 128 bits of 256bit (YMM) vector registers 1361 void vinserti128_high(XMMRegister dst, XMMRegister src) { 1362 vinserti128(dst, dst, src, 1); 1363 } 1364 void vinserti128_high(XMMRegister dst, Address src) { 1365 vinserti128(dst, dst, src, 1); 1366 } 1367 void vextracti128_high(XMMRegister dst, XMMRegister src) { 1368 vextracti128(dst, src, 1); 1369 } 1370 void vextracti128_high(Address dst, XMMRegister src) { 1371 vextracti128(dst, src, 1); 1372 } 1373 1374 void vinsertf128_high(XMMRegister dst, XMMRegister src) { 1375 if (UseAVX > 2) { 1376 Assembler::vinsertf32x4(dst, dst, src, 1); 1377 } else { 1378 Assembler::vinsertf128(dst, dst, src, 1); 1379 } 1380 } 1381 1382 void vinsertf128_high(XMMRegister dst, Address src) { 1383 if (UseAVX > 2) { 1384 Assembler::vinsertf32x4(dst, dst, src, 1); 1385 } else { 1386 Assembler::vinsertf128(dst, dst, src, 1); 1387 } 1388 } 1389 1390 void vextractf128_high(XMMRegister dst, XMMRegister src) { 1391 if (UseAVX > 2) { 1392 Assembler::vextractf32x4(dst, src, 1); 1393 } else { 1394 Assembler::vextractf128(dst, src, 1); 1395 } 1396 } 1397 1398 void vextractf128_high(Address dst, XMMRegister src) { 1399 if (UseAVX > 2) { 1400 Assembler::vextractf32x4(dst, src, 1); 1401 } else { 1402 Assembler::vextractf128(dst, src, 1); 1403 } 1404 } 1405 1406 // 256bit copy to/from high 256 bits of 512bit (ZMM) vector registers 1407 void vinserti64x4_high(XMMRegister dst, XMMRegister src) { 1408 Assembler::vinserti64x4(dst, dst, src, 1); 1409 } 1410 void vinsertf64x4_high(XMMRegister dst, XMMRegister src) { 1411 Assembler::vinsertf64x4(dst, dst, src, 1); 1412 } 1413 void vextracti64x4_high(XMMRegister dst, XMMRegister src) { 1414 Assembler::vextracti64x4(dst, src, 1); 1415 } 1416 void vextractf64x4_high(XMMRegister dst, XMMRegister src) { 1417 Assembler::vextractf64x4(dst, src, 1); 1418 } 1419 void vextractf64x4_high(Address dst, XMMRegister src) { 1420 Assembler::vextractf64x4(dst, src, 1); 1421 } 1422 void vinsertf64x4_high(XMMRegister dst, Address src) { 1423 Assembler::vinsertf64x4(dst, dst, src, 1); 1424 } 1425 1426 // 128bit copy to/from low 128 bits of 256bit (YMM) vector registers 1427 void vinserti128_low(XMMRegister dst, XMMRegister src) { 1428 vinserti128(dst, dst, src, 0); 1429 } 1430 void vinserti128_low(XMMRegister dst, Address src) { 1431 vinserti128(dst, dst, src, 0); 1432 } 1433 void vextracti128_low(XMMRegister dst, XMMRegister src) { 1434 vextracti128(dst, src, 0); 1435 } 1436 void vextracti128_low(Address dst, XMMRegister src) { 1437 vextracti128(dst, src, 0); 1438 } 1439 1440 void vinsertf128_low(XMMRegister dst, XMMRegister src) { 1441 if (UseAVX > 2) { 1442 Assembler::vinsertf32x4(dst, dst, src, 0); 1443 } else { 1444 Assembler::vinsertf128(dst, dst, src, 0); 1445 } 1446 } 1447 1448 void vinsertf128_low(XMMRegister dst, Address src) { 1449 if (UseAVX > 2) { 1450 Assembler::vinsertf32x4(dst, dst, src, 0); 1451 } else { 1452 Assembler::vinsertf128(dst, dst, src, 0); 1453 } 1454 } 1455 1456 void vextractf128_low(XMMRegister dst, XMMRegister src) { 1457 if (UseAVX > 2) { 1458 Assembler::vextractf32x4(dst, src, 0); 1459 } else { 1460 Assembler::vextractf128(dst, src, 0); 1461 } 1462 } 1463 1464 void vextractf128_low(Address dst, XMMRegister src) { 1465 if (UseAVX > 2) { 1466 Assembler::vextractf32x4(dst, src, 0); 1467 } else { 1468 Assembler::vextractf128(dst, src, 0); 1469 } 1470 } 1471 1472 // 256bit copy to/from low 256 bits of 512bit (ZMM) vector registers 1473 void vinserti64x4_low(XMMRegister dst, XMMRegister src) { 1474 Assembler::vinserti64x4(dst, dst, src, 0); 1475 } 1476 void vinsertf64x4_low(XMMRegister dst, XMMRegister src) { 1477 Assembler::vinsertf64x4(dst, dst, src, 0); 1478 } 1479 void vextracti64x4_low(XMMRegister dst, XMMRegister src) { 1480 Assembler::vextracti64x4(dst, src, 0); 1481 } 1482 void vextractf64x4_low(XMMRegister dst, XMMRegister src) { 1483 Assembler::vextractf64x4(dst, src, 0); 1484 } 1485 void vextractf64x4_low(Address dst, XMMRegister src) { 1486 Assembler::vextractf64x4(dst, src, 0); 1487 } 1488 void vinsertf64x4_low(XMMRegister dst, Address src) { 1489 Assembler::vinsertf64x4(dst, dst, src, 0); 1490 } 1491 1492 // Carry-Less Multiplication Quadword 1493 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) { 1494 // 0x00 - multiply lower 64 bits [0:63] 1495 Assembler::vpclmulqdq(dst, nds, src, 0x00); 1496 } 1497 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) { 1498 // 0x11 - multiply upper 64 bits [64:127] 1499 Assembler::vpclmulqdq(dst, nds, src, 0x11); 1500 } 1501 void evpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { 1502 // 0x00 - multiply lower 64 bits [0:63] 1503 Assembler::evpclmulqdq(dst, nds, src, 0x00, vector_len); 1504 } 1505 void evpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { 1506 // 0x11 - multiply upper 64 bits [64:127] 1507 Assembler::evpclmulqdq(dst, nds, src, 0x11, vector_len); 1508 } 1509 1510 // Data 1511 1512 void cmov32( Condition cc, Register dst, Address src); 1513 void cmov32( Condition cc, Register dst, Register src); 1514 1515 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); } 1516 1517 void cmovptr(Condition cc, Register dst, Address src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 1518 void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); } 1519 1520 void movoop(Register dst, jobject obj); 1521 void movoop(Address dst, jobject obj); 1522 1523 void mov_metadata(Register dst, Metadata* obj); 1524 void mov_metadata(Address dst, Metadata* obj); 1525 1526 void movptr(ArrayAddress dst, Register src); 1527 // can this do an lea? 1528 void movptr(Register dst, ArrayAddress src); 1529 1530 void movptr(Register dst, Address src); 1531 1532 #ifdef _LP64 1533 void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1); 1534 #else 1535 void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit 1536 #endif 1537 1538 void movptr(Register dst, intptr_t src); 1539 void movptr(Register dst, Register src); 1540 void movptr(Address dst, intptr_t src); 1541 1542 void movptr(Address dst, Register src); 1543 1544 void movptr(Register dst, RegisterOrConstant src) { 1545 if (src.is_constant()) movptr(dst, src.as_constant()); 1546 else movptr(dst, src.as_register()); 1547 } 1548 1549 #ifdef _LP64 1550 // Generally the next two are only used for moving NULL 1551 // Although there are situations in initializing the mark word where 1552 // they could be used. They are dangerous. 1553 1554 // They only exist on LP64 so that int32_t and intptr_t are not the same 1555 // and we have ambiguous declarations. 1556 1557 void movptr(Address dst, int32_t imm32); 1558 void movptr(Register dst, int32_t imm32); 1559 #endif // _LP64 1560 1561 // to avoid hiding movl 1562 void mov32(AddressLiteral dst, Register src); 1563 void mov32(Register dst, AddressLiteral src); 1564 1565 // to avoid hiding movb 1566 void movbyte(ArrayAddress dst, int src); 1567 1568 // Import other mov() methods from the parent class or else 1569 // they will be hidden by the following overriding declaration. 1570 using Assembler::movdl; 1571 using Assembler::movq; 1572 void movdl(XMMRegister dst, AddressLiteral src); 1573 void movq(XMMRegister dst, AddressLiteral src); 1574 1575 // Can push value or effective address 1576 void pushptr(AddressLiteral src); 1577 1578 void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); } 1579 void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); } 1580 1581 void pushoop(jobject obj); 1582 void pushklass(Metadata* obj); 1583 1584 // sign extend as need a l to ptr sized element 1585 void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); } 1586 void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); } 1587 1588 // C2 compiled method's prolog code. 1589 void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b); 1590 1591 // clear memory of size 'cnt' qwords, starting at 'base'; 1592 // if 'is_large' is set, do not try to produce short loop 1593 void clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, bool is_large); 1594 1595 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM registers 1596 void xmm_clear_mem(Register base, Register cnt, XMMRegister xtmp); 1597 1598 #ifdef COMPILER2 1599 void string_indexof_char(Register str1, Register cnt1, Register ch, Register result, 1600 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp); 1601 1602 // IndexOf strings. 1603 // Small strings are loaded through stack if they cross page boundary. 1604 void string_indexof(Register str1, Register str2, 1605 Register cnt1, Register cnt2, 1606 int int_cnt2, Register result, 1607 XMMRegister vec, Register tmp, 1608 int ae); 1609 1610 // IndexOf for constant substrings with size >= 8 elements 1611 // which don't need to be loaded through stack. 1612 void string_indexofC8(Register str1, Register str2, 1613 Register cnt1, Register cnt2, 1614 int int_cnt2, Register result, 1615 XMMRegister vec, Register tmp, 1616 int ae); 1617 1618 // Smallest code: we don't need to load through stack, 1619 // check string tail. 1620 1621 // helper function for string_compare 1622 void load_next_elements(Register elem1, Register elem2, Register str1, Register str2, 1623 Address::ScaleFactor scale, Address::ScaleFactor scale1, 1624 Address::ScaleFactor scale2, Register index, int ae); 1625 // Compare strings. 1626 void string_compare(Register str1, Register str2, 1627 Register cnt1, Register cnt2, Register result, 1628 XMMRegister vec1, int ae); 1629 1630 // Search for Non-ASCII character (Negative byte value) in a byte array, 1631 // return true if it has any and false otherwise. 1632 void has_negatives(Register ary1, Register len, 1633 Register result, Register tmp1, 1634 XMMRegister vec1, XMMRegister vec2); 1635 1636 // Compare char[] or byte[] arrays. 1637 void arrays_equals(bool is_array_equ, Register ary1, Register ary2, 1638 Register limit, Register result, Register chr, 1639 XMMRegister vec1, XMMRegister vec2, bool is_char); 1640 1641 #endif 1642 1643 // Fill primitive arrays 1644 void generate_fill(BasicType t, bool aligned, 1645 Register to, Register value, Register count, 1646 Register rtmp, XMMRegister xtmp); 1647 1648 void encode_iso_array(Register src, Register dst, Register len, 1649 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3, 1650 XMMRegister tmp4, Register tmp5, Register result); 1651 1652 #ifdef _LP64 1653 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2); 1654 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart, 1655 Register y, Register y_idx, Register z, 1656 Register carry, Register product, 1657 Register idx, Register kdx); 1658 void multiply_add_128_x_128(Register x_xstart, Register y, Register z, 1659 Register yz_idx, Register idx, 1660 Register carry, Register product, int offset); 1661 void multiply_128_x_128_bmi2_loop(Register y, Register z, 1662 Register carry, Register carry2, 1663 Register idx, Register jdx, 1664 Register yz_idx1, Register yz_idx2, 1665 Register tmp, Register tmp3, Register tmp4); 1666 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z, 1667 Register yz_idx, Register idx, Register jdx, 1668 Register carry, Register product, 1669 Register carry2); 1670 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen, 1671 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5); 1672 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3, 1673 Register tmp4, Register tmp5, Register rdxReg, Register raxReg); 1674 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry, 1675 Register tmp2); 1676 void multiply_add_64(Register sum, Register op1, Register op2, Register carry, 1677 Register rdxReg, Register raxReg); 1678 void add_one_64(Register z, Register zlen, Register carry, Register tmp1); 1679 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, 1680 Register tmp3, Register tmp4); 1681 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2, 1682 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg); 1683 1684 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1, 1685 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, 1686 Register raxReg); 1687 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1, 1688 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg, 1689 Register raxReg); 1690 void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale, 1691 Register result, Register tmp1, Register tmp2, 1692 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3); 1693 #endif 1694 1695 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic. 1696 void update_byte_crc32(Register crc, Register val, Register table); 1697 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp); 1698 // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic 1699 // Note on a naming convention: 1700 // Prefix w = register only used on a Westmere+ architecture 1701 // Prefix n = register only used on a Nehalem architecture 1702 #ifdef _LP64 1703 void crc32c_ipl_alg4(Register in_out, uint32_t n, 1704 Register tmp1, Register tmp2, Register tmp3); 1705 #else 1706 void crc32c_ipl_alg4(Register in_out, uint32_t n, 1707 Register tmp1, Register tmp2, Register tmp3, 1708 XMMRegister xtmp1, XMMRegister xtmp2); 1709 #endif 1710 void crc32c_pclmulqdq(XMMRegister w_xtmp1, 1711 Register in_out, 1712 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported, 1713 XMMRegister w_xtmp2, 1714 Register tmp1, 1715 Register n_tmp2, Register n_tmp3); 1716 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, 1717 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1718 Register tmp1, Register tmp2, 1719 Register n_tmp3); 1720 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, 1721 Register in_out1, Register in_out2, Register in_out3, 1722 Register tmp1, Register tmp2, Register tmp3, 1723 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1724 Register tmp4, Register tmp5, 1725 Register n_tmp6); 1726 void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2, 1727 Register tmp1, Register tmp2, Register tmp3, 1728 Register tmp4, Register tmp5, Register tmp6, 1729 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3, 1730 bool is_pclmulqdq_supported); 1731 // Fold 128-bit data chunk 1732 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset); 1733 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf); 1734 // Fold 8-bit data 1735 void fold_8bit_crc32(Register crc, Register table, Register tmp); 1736 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp); 1737 void fold_128bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset); 1738 1739 // Compress char[] array to byte[]. 1740 void char_array_compress(Register src, Register dst, Register len, 1741 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3, 1742 XMMRegister tmp4, Register tmp5, Register result); 1743 1744 // Inflate byte[] array to char[]. 1745 void byte_array_inflate(Register src, Register dst, Register len, 1746 XMMRegister tmp1, Register tmp2); 1747 1748 }; 1749 1750 /** 1751 * class SkipIfEqual: 1752 * 1753 * Instantiating this class will result in assembly code being output that will 1754 * jump around any code emitted between the creation of the instance and it's 1755 * automatic destruction at the end of a scope block, depending on the value of 1756 * the flag passed to the constructor, which will be checked at run-time. 1757 */ 1758 class SkipIfEqual { 1759 private: 1760 MacroAssembler* _masm; 1761 Label _label; 1762 1763 public: 1764 SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value); 1765 ~SkipIfEqual(); 1766 }; 1767 1768 #endif // CPU_X86_VM_MACROASSEMBLER_X86_HPP