1 /*
   2  * Copyright (c) 1997, 2019, 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_MACROASSEMBLER_X86_HPP
  26 #define CPU_X86_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   static bool uses_implicit_null_check(void* address);
 100 
 101   // Required platform-specific helpers for Label::patch_instructions.
 102   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 103   void pd_patch_instruction(address branch, address target, const char* file, int line) {
 104     unsigned char op = branch[0];
 105     assert(op == 0xE8 /* call */ ||
 106         op == 0xE9 /* jmp */ ||
 107         op == 0xEB /* short jmp */ ||
 108         (op & 0xF0) == 0x70 /* short jcc */ ||
 109         op == 0x0F && (branch[1] & 0xF0) == 0x80 /* jcc */ ||
 110         op == 0xC7 && branch[1] == 0xF8 /* xbegin */,
 111         "Invalid opcode at patch point");
 112 
 113     if (op == 0xEB || (op & 0xF0) == 0x70) {
 114       // short offset operators (jmp and jcc)
 115       char* disp = (char*) &branch[1];
 116       int imm8 = target - (address) &disp[1];
 117       guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset at %s:%d", file, line);
 118       *disp = imm8;
 119     } else {
 120       int* disp = (int*) &branch[(op == 0x0F || op == 0xC7)? 2: 1];
 121       int imm32 = target - (address) &disp[1];
 122       *disp = imm32;
 123     }
 124   }
 125 
 126   // The following 4 methods return the offset of the appropriate move instruction
 127 
 128   // Support for fast byte/short loading with zero extension (depending on particular CPU)
 129   int load_unsigned_byte(Register dst, Address src);
 130   int load_unsigned_short(Register dst, Address src);
 131 
 132   // Support for fast byte/short loading with sign extension (depending on particular CPU)
 133   int load_signed_byte(Register dst, Address src);
 134   int load_signed_short(Register dst, Address src);
 135 
 136   // Support for sign-extension (hi:lo = extend_sign(lo))
 137   void extend_sign(Register hi, Register lo);
 138 
 139   // Load and store values by size and signed-ness
 140   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
 141   void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
 142 
 143   // Support for inc/dec with optimal instruction selection depending on value
 144 
 145   void increment(Register reg, int value = 1) { LP64_ONLY(incrementq(reg, value)) NOT_LP64(incrementl(reg, value)) ; }
 146   void decrement(Register reg, int value = 1) { LP64_ONLY(decrementq(reg, value)) NOT_LP64(decrementl(reg, value)) ; }
 147 
 148   void decrementl(Address dst, int value = 1);
 149   void decrementl(Register reg, int value = 1);
 150 
 151   void decrementq(Register reg, int value = 1);
 152   void decrementq(Address dst, int value = 1);
 153 
 154   void incrementl(Address dst, int value = 1);
 155   void incrementl(Register reg, int value = 1);
 156 
 157   void incrementq(Register reg, int value = 1);
 158   void incrementq(Address dst, int value = 1);
 159 
 160 #ifdef COMPILER2
 161   // special instructions for EVEX
 162   void setvectmask(Register dst, Register src);
 163   void restorevectmask();
 164 #endif
 165 
 166   // Support optimal SSE move instructions.
 167   void movflt(XMMRegister dst, XMMRegister src) {
 168     if (dst-> encoding() == src->encoding()) return;
 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 (dst-> encoding() == src->encoding()) return;
 178     if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
 179     else                       { movsd (dst, src); return; }
 180   }
 181 
 182   void movdbl(XMMRegister dst, AddressLiteral src);
 183 
 184   void movdbl(XMMRegister dst, Address src) {
 185     if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
 186     else                         { movlpd(dst, src); return; }
 187   }
 188   void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
 189 
 190   void incrementl(AddressLiteral dst);
 191   void incrementl(ArrayAddress dst);
 192 
 193   void incrementq(AddressLiteral dst);
 194 
 195   // Alignment
 196   void align(int modulus);
 197   void align(int modulus, int target);
 198 
 199   // A 5 byte nop that is safe for patching (see patch_verified_entry)
 200   void fat_nop();
 201 
 202   // Stack frame creation/removal
 203   void enter();
 204   void leave();
 205 
 206   // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information)
 207   // The pointer will be loaded into the thread register.
 208   void get_thread(Register thread);
 209 
 210 
 211   // Support for VM calls
 212   //
 213   // It is imperative that all calls into the VM are handled via the call_VM macros.
 214   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 215   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 216 
 217 
 218   void call_VM(Register oop_result,
 219                address entry_point,
 220                bool check_exceptions = true);
 221   void call_VM(Register oop_result,
 222                address entry_point,
 223                Register arg_1,
 224                bool check_exceptions = true);
 225   void call_VM(Register oop_result,
 226                address entry_point,
 227                Register arg_1, Register arg_2,
 228                bool check_exceptions = true);
 229   void call_VM(Register oop_result,
 230                address entry_point,
 231                Register arg_1, Register arg_2, Register arg_3,
 232                bool check_exceptions = true);
 233 
 234   // Overloadings with last_Java_sp
 235   void call_VM(Register oop_result,
 236                Register last_java_sp,
 237                address entry_point,
 238                int number_of_arguments = 0,
 239                bool check_exceptions = true);
 240   void call_VM(Register oop_result,
 241                Register last_java_sp,
 242                address entry_point,
 243                Register arg_1, bool
 244                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,
 249                bool check_exceptions = true);
 250   void call_VM(Register oop_result,
 251                Register last_java_sp,
 252                address entry_point,
 253                Register arg_1, Register arg_2, Register arg_3,
 254                bool check_exceptions = true);
 255 
 256   void get_vm_result  (Register oop_result, Register thread);
 257   void get_vm_result_2(Register metadata_result, Register thread);
 258 
 259   // These always tightly bind to MacroAssembler::call_VM_base
 260   // bypassing the virtual implementation
 261   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 262   void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, 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, 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, bool check_exceptions = true);
 265   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);
 266 
 267   void call_VM_leaf0(address entry_point);
 268   void call_VM_leaf(address entry_point,
 269                     int number_of_arguments = 0);
 270   void call_VM_leaf(address entry_point,
 271                     Register arg_1);
 272   void call_VM_leaf(address entry_point,
 273                     Register arg_1, Register arg_2);
 274   void call_VM_leaf(address entry_point,
 275                     Register arg_1, Register arg_2, Register arg_3);
 276 
 277   // These always tightly bind to MacroAssembler::call_VM_leaf_base
 278   // bypassing the virtual implementation
 279   void super_call_VM_leaf(address entry_point);
 280   void super_call_VM_leaf(address entry_point, Register arg_1);
 281   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 282   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 283   void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 284 
 285   // last Java Frame (fills frame anchor)
 286   void set_last_Java_frame(Register thread,
 287                            Register last_java_sp,
 288                            Register last_java_fp,
 289                            address last_java_pc);
 290 
 291   // thread in the default location (r15_thread on 64bit)
 292   void set_last_Java_frame(Register last_java_sp,
 293                            Register last_java_fp,
 294                            address last_java_pc);
 295 
 296   void reset_last_Java_frame(Register thread, bool clear_fp);
 297 
 298   // thread in the default location (r15_thread on 64bit)
 299   void reset_last_Java_frame(bool clear_fp);
 300 
 301   // jobjects
 302   void clear_jweak_tag(Register possibly_jweak);
 303   void resolve_jobject(Register value, Register thread, Register tmp);
 304 
 305   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 306   void c2bool(Register x);
 307 
 308   // C++ bool manipulation
 309 
 310   void movbool(Register dst, Address src);
 311   void movbool(Address dst, bool boolconst);
 312   void movbool(Address dst, Register src);
 313   void testbool(Register dst);
 314 
 315   void resolve_oop_handle(Register result, Register tmp = rscratch2);
 316   void resolve_weak_handle(Register result, Register tmp);
 317   void load_mirror(Register mirror, Register method, Register tmp = rscratch2);
 318   void load_method_holder_cld(Register rresult, Register rmethod);
 319 
 320   void load_method_holder(Register holder, Register method);
 321 
 322   // oop manipulations
 323   void load_klass(Register dst, Register src);
 324   void store_klass(Register dst, Register src);
 325 
 326   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 327                       Register tmp1, Register thread_tmp);
 328   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register src,
 329                        Register tmp1, Register tmp2);
 330 
 331   // Resolves obj access. Result is placed in the same register.
 332   // All other registers are preserved.
 333   void resolve(DecoratorSet decorators, Register obj);
 334 
 335   void load_heap_oop(Register dst, Address src, Register tmp1 = noreg,
 336                      Register thread_tmp = noreg, DecoratorSet decorators = 0);
 337   void load_heap_oop_not_null(Register dst, Address src, Register tmp1 = noreg,
 338                               Register thread_tmp = noreg, DecoratorSet decorators = 0);
 339   void store_heap_oop(Address dst, Register src, Register tmp1 = noreg,
 340                       Register tmp2 = noreg, DecoratorSet decorators = 0);
 341 
 342   // Used for storing NULL. All other oop constants should be
 343   // stored using routines that take a jobject.
 344   void store_heap_oop_null(Address dst);
 345 
 346   void load_prototype_header(Register dst, Register src);
 347 
 348 #ifdef _LP64
 349   void store_klass_gap(Register dst, Register src);
 350 
 351   // This dummy is to prevent a call to store_heap_oop from
 352   // converting a zero (like NULL) into a Register by giving
 353   // the compiler two choices it can't resolve
 354 
 355   void store_heap_oop(Address dst, void* dummy);
 356 
 357   void encode_heap_oop(Register r);
 358   void decode_heap_oop(Register r);
 359   void encode_heap_oop_not_null(Register r);
 360   void decode_heap_oop_not_null(Register r);
 361   void encode_heap_oop_not_null(Register dst, Register src);
 362   void decode_heap_oop_not_null(Register dst, Register src);
 363 
 364   void set_narrow_oop(Register dst, jobject obj);
 365   void set_narrow_oop(Address dst, jobject obj);
 366   void cmp_narrow_oop(Register dst, jobject obj);
 367   void cmp_narrow_oop(Address dst, jobject obj);
 368 
 369   void encode_klass_not_null(Register r);
 370   void decode_klass_not_null(Register r);
 371   void encode_klass_not_null(Register dst, Register src);
 372   void decode_klass_not_null(Register dst, Register src);
 373   void set_narrow_klass(Register dst, Klass* k);
 374   void set_narrow_klass(Address dst, Klass* k);
 375   void cmp_narrow_klass(Register dst, Klass* k);
 376   void cmp_narrow_klass(Address dst, Klass* k);
 377 
 378   // Returns the byte size of the instructions generated by decode_klass_not_null()
 379   // when compressed klass pointers are being used.
 380   static int instr_size_for_decode_klass_not_null();
 381 
 382   // if heap base register is used - reinit it with the correct value
 383   void reinit_heapbase();
 384 
 385   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 386 
 387 #endif // _LP64
 388 
 389   // Int division/remainder for Java
 390   // (as idivl, but checks for special case as described in JVM spec.)
 391   // returns idivl instruction offset for implicit exception handling
 392   int corrected_idivl(Register reg);
 393 
 394   // Long division/remainder for Java
 395   // (as idivq, but checks for special case as described in JVM spec.)
 396   // returns idivq instruction offset for implicit exception handling
 397   int corrected_idivq(Register reg);
 398 
 399   void int3();
 400 
 401   // Long operation macros for a 32bit cpu
 402   // Long negation for Java
 403   void lneg(Register hi, Register lo);
 404 
 405   // Long multiplication for Java
 406   // (destroys contents of eax, ebx, ecx and edx)
 407   void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
 408 
 409   // Long shifts for Java
 410   // (semantics as described in JVM spec.)
 411   void lshl(Register hi, Register lo);                               // hi:lo << (rcx & 0x3f)
 412   void lshr(Register hi, Register lo, bool sign_extension = false);  // hi:lo >> (rcx & 0x3f)
 413 
 414   // Long compare for Java
 415   // (semantics as described in JVM spec.)
 416   void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
 417 
 418 
 419   // misc
 420 
 421   // Sign extension
 422   void sign_extend_short(Register reg);
 423   void sign_extend_byte(Register reg);
 424 
 425   // Division by power of 2, rounding towards 0
 426   void division_with_shift(Register reg, int shift_value);
 427 
 428   // Compares the top-most stack entries on the FPU stack and sets the eflags as follows:
 429   //
 430   // CF (corresponds to C0) if x < y
 431   // PF (corresponds to C2) if unordered
 432   // ZF (corresponds to C3) if x = y
 433   //
 434   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
 435   // tmp is a temporary register, if none is available use noreg (only matters for non-P6 code)
 436   void fcmp(Register tmp);
 437   // Variant of the above which allows y to be further down the stack
 438   // and which only pops x and y if specified. If pop_right is
 439   // specified then pop_left must also be specified.
 440   void fcmp(Register tmp, int index, bool pop_left, bool pop_right);
 441 
 442   // Floating-point comparison for Java
 443   // Compares the top-most stack entries on the FPU stack and stores the result in dst.
 444   // The arguments are in reversed order on the stack (i.e., top of stack is first argument).
 445   // (semantics as described in JVM spec.)
 446   void fcmp2int(Register dst, bool unordered_is_less);
 447   // Variant of the above which allows y to be further down the stack
 448   // and which only pops x and y if specified. If pop_right is
 449   // specified then pop_left must also be specified.
 450   void fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right);
 451 
 452   // Floating-point remainder for Java (ST0 = ST0 fremr ST1, ST1 is empty afterwards)
 453   // tmp is a temporary register, if none is available use noreg
 454   void fremr(Register tmp);
 455 
 456   // dst = c = a * b + c
 457   void fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
 458   void fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
 459 
 460   void vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
 461   void vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
 462   void vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
 463   void vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
 464 
 465 
 466   // same as fcmp2int, but using SSE2
 467   void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
 468   void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
 469 
 470   // branch to L if FPU flag C2 is set/not set
 471   // tmp is a temporary register, if none is available use noreg
 472   void jC2 (Register tmp, Label& L);
 473   void jnC2(Register tmp, Label& L);
 474 
 475   // Pop ST (ffree & fincstp combined)
 476   void fpop();
 477 
 478   // Load float value from 'address'. If UseSSE >= 1, the value is loaded into
 479   // register xmm0. Otherwise, the value is loaded onto the FPU stack.
 480   void load_float(Address src);
 481 
 482   // Store float value to 'address'. If UseSSE >= 1, the value is stored
 483   // from register xmm0. Otherwise, the value is stored from the FPU stack.
 484   void store_float(Address dst);
 485 
 486   // Load double value from 'address'. If UseSSE >= 2, the value is loaded into
 487   // register xmm0. Otherwise, the value is loaded onto the FPU stack.
 488   void load_double(Address src);
 489 
 490   // Store double value to 'address'. If UseSSE >= 2, the value is stored
 491   // from register xmm0. Otherwise, the value is stored from the FPU stack.
 492   void store_double(Address dst);
 493 
 494   // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack
 495   void push_fTOS();
 496 
 497   // pops double TOS element from CPU stack and pushes on FPU stack
 498   void pop_fTOS();
 499 
 500   void empty_FPU_stack();
 501 
 502   void push_IU_state();
 503   void pop_IU_state();
 504 
 505   void push_FPU_state();
 506   void pop_FPU_state();
 507 
 508   void push_CPU_state();
 509   void pop_CPU_state();
 510 
 511   // Round up to a power of two
 512   void round_to(Register reg, int modulus);
 513 
 514   // Callee saved registers handling
 515   void push_callee_saved_registers();
 516   void pop_callee_saved_registers();
 517 
 518   // allocation
 519   void eden_allocate(
 520     Register thread,                   // Current thread
 521     Register obj,                      // result: pointer to object after successful allocation
 522     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 523     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 524     Register t1,                       // temp register
 525     Label&   slow_case                 // continuation point if fast allocation fails
 526   );
 527   void tlab_allocate(
 528     Register thread,                   // Current thread
 529     Register obj,                      // result: pointer to object after successful allocation
 530     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 531     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 532     Register t1,                       // temp register
 533     Register t2,                       // temp register
 534     Label&   slow_case                 // continuation point if fast allocation fails
 535   );
 536   void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp);
 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                                bool return_method = true);
 546 
 547   // virtual method calling
 548   void lookup_virtual_method(Register recv_klass,
 549                              RegisterOrConstant vtable_index,
 550                              Register method_result);
 551 
 552   // Test sub_klass against super_klass, with fast and slow paths.
 553 
 554   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 555   // One of the three labels can be NULL, meaning take the fall-through.
 556   // If super_check_offset is -1, the value is loaded up from super_klass.
 557   // No registers are killed, except temp_reg.
 558   void check_klass_subtype_fast_path(Register sub_klass,
 559                                      Register super_klass,
 560                                      Register temp_reg,
 561                                      Label* L_success,
 562                                      Label* L_failure,
 563                                      Label* L_slow_path,
 564                 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
 565 
 566   // The rest of the type check; must be wired to a corresponding fast path.
 567   // It does not repeat the fast path logic, so don't use it standalone.
 568   // The temp_reg and temp2_reg can be noreg, if no temps are available.
 569   // Updates the sub's secondary super cache as necessary.
 570   // If set_cond_codes, condition codes will be Z on success, NZ on failure.
 571   void check_klass_subtype_slow_path(Register sub_klass,
 572                                      Register super_klass,
 573                                      Register temp_reg,
 574                                      Register temp2_reg,
 575                                      Label* L_success,
 576                                      Label* L_failure,
 577                                      bool set_cond_codes = false);
 578 
 579   // Simplified, combined version, good for typical uses.
 580   // Falls through on failure.
 581   void check_klass_subtype(Register sub_klass,
 582                            Register super_klass,
 583                            Register temp_reg,
 584                            Label& L_success);
 585 
 586   void clinit_barrier(Register klass,
 587                       Register thread,
 588                       Label* L_fast_path = NULL,
 589                       Label* L_slow_path = NULL);
 590 
 591   // method handles (JSR 292)
 592   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
 593 
 594   //----
 595   void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0
 596 
 597   // Debugging
 598 
 599   // only if +VerifyOops
 600   // TODO: Make these macros with file and line like sparc version!
 601   void verify_oop(Register reg, const char* s = "broken oop");
 602   void verify_oop_addr(Address addr, const char * s = "broken oop addr");
 603 
 604   // TODO: verify method and klass metadata (compare against vptr?)
 605   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
 606   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
 607 
 608 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
 609 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
 610 
 611   // only if +VerifyFPU
 612   void verify_FPU(int stack_depth, const char* s = "illegal FPU state");
 613 
 614   // Verify or restore cpu control state after JNI call
 615   void restore_cpu_control_state_after_jni();
 616 
 617   // prints msg, dumps registers and stops execution
 618   void stop(const char* msg);
 619 
 620   // prints msg and continues
 621   void warn(const char* msg);
 622 
 623   // dumps registers and other state
 624   void print_state();
 625 
 626   static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
 627   static void debug64(char* msg, int64_t pc, int64_t regs[]);
 628   static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip);
 629   static void print_state64(int64_t pc, int64_t regs[]);
 630 
 631   void os_breakpoint();
 632 
 633   void untested()                                { stop("untested"); }
 634 
 635   void unimplemented(const char* what = "");
 636 
 637   void should_not_reach_here()                   { stop("should not reach here"); }
 638 
 639   void print_CPU_state();
 640 
 641   // Stack overflow checking
 642   void bang_stack_with_offset(int offset) {
 643     // stack grows down, caller passes positive offset
 644     assert(offset > 0, "must bang with negative offset");
 645     movl(Address(rsp, (-offset)), rax);
 646   }
 647 
 648   // Writes to stack successive pages until offset reached to check for
 649   // stack overflow + shadow pages.  Also, clobbers tmp
 650   void bang_stack_size(Register size, Register tmp);
 651 
 652   // Check for reserved stack access in method being exited (for JIT)
 653   void reserved_stack_check();
 654 
 655   virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
 656                                                 Register tmp,
 657                                                 int offset);
 658 
 659   // If thread_reg is != noreg the code assumes the register passed contains
 660   // the thread (required on 64 bit).
 661   void safepoint_poll(Label& slow_path, Register thread_reg, Register temp_reg);
 662 
 663   void verify_tlab();
 664 
 665   // Biased locking support
 666   // lock_reg and obj_reg must be loaded up with the appropriate values.
 667   // swap_reg must be rax, and is killed.
 668   // tmp_reg is optional. If it is supplied (i.e., != noreg) it will
 669   // be killed; if not supplied, push/pop will be used internally to
 670   // allocate a temporary (inefficient, avoid if possible).
 671   // Optional slow case is for implementations (interpreter and C1) which branch to
 672   // slow case directly. Leaves condition codes set for C2's Fast_Lock node.
 673   // Returns offset of first potentially-faulting instruction for null
 674   // check info (currently consumed only by C1). If
 675   // swap_reg_contains_mark is true then returns -1 as it is assumed
 676   // the calling code has already passed any potential faults.
 677   int biased_locking_enter(Register lock_reg, Register obj_reg,
 678                            Register swap_reg, Register tmp_reg,
 679                            bool swap_reg_contains_mark,
 680                            Label& done, Label* slow_case = NULL,
 681                            BiasedLockingCounters* counters = NULL);
 682   void biased_locking_exit (Register obj_reg, Register temp_reg, Label& done);
 683 #ifdef COMPILER2
 684   // Code used by cmpFastLock and cmpFastUnlock mach instructions in .ad file.
 685   // See full desription in macroAssembler_x86.cpp.
 686   void fast_lock(Register obj, Register box, Register tmp,
 687                  Register scr, Register cx1, Register cx2,
 688                  BiasedLockingCounters* counters,
 689                  RTMLockingCounters* rtm_counters,
 690                  RTMLockingCounters* stack_rtm_counters,
 691                  Metadata* method_data,
 692                  bool use_rtm, bool profile_rtm);
 693   void fast_unlock(Register obj, Register box, Register tmp, bool use_rtm);
 694 #if INCLUDE_RTM_OPT
 695   void rtm_counters_update(Register abort_status, Register rtm_counters);
 696   void branch_on_random_using_rdtsc(Register tmp, Register scr, int count, Label& brLabel);
 697   void rtm_abort_ratio_calculation(Register tmp, Register rtm_counters_reg,
 698                                    RTMLockingCounters* rtm_counters,
 699                                    Metadata* method_data);
 700   void rtm_profiling(Register abort_status_Reg, Register rtm_counters_Reg,
 701                      RTMLockingCounters* rtm_counters, Metadata* method_data, bool profile_rtm);
 702   void rtm_retry_lock_on_abort(Register retry_count, Register abort_status, Label& retryLabel);
 703   void rtm_retry_lock_on_busy(Register retry_count, Register box, Register tmp, Register scr, Label& retryLabel);
 704   void rtm_stack_locking(Register obj, Register tmp, Register scr,
 705                          Register retry_on_abort_count,
 706                          RTMLockingCounters* stack_rtm_counters,
 707                          Metadata* method_data, bool profile_rtm,
 708                          Label& DONE_LABEL, Label& IsInflated);
 709   void rtm_inflated_locking(Register obj, Register box, Register tmp,
 710                             Register scr, Register retry_on_busy_count,
 711                             Register retry_on_abort_count,
 712                             RTMLockingCounters* rtm_counters,
 713                             Metadata* method_data, bool profile_rtm,
 714                             Label& DONE_LABEL);
 715 #endif
 716 #endif
 717 
 718   Condition negate_condition(Condition cond);
 719 
 720   // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
 721   // operands. In general the names are modified to avoid hiding the instruction in Assembler
 722   // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
 723   // here in MacroAssembler. The major exception to this rule is call
 724 
 725   // Arithmetics
 726 
 727 
 728   void addptr(Address dst, int32_t src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)) ; }
 729   void addptr(Address dst, Register src);
 730 
 731   void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
 732   void addptr(Register dst, int32_t src);
 733   void addptr(Register dst, Register src);
 734   void addptr(Register dst, RegisterOrConstant src) {
 735     if (src.is_constant()) addptr(dst, (int) src.as_constant());
 736     else                   addptr(dst,       src.as_register());
 737   }
 738 
 739   void andptr(Register dst, int32_t src);
 740   void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
 741 
 742   void cmp8(AddressLiteral src1, int imm);
 743 
 744   // renamed to drag out the casting of address to int32_t/intptr_t
 745   void cmp32(Register src1, int32_t imm);
 746 
 747   void cmp32(AddressLiteral src1, int32_t imm);
 748   // compare reg - mem, or reg - &mem
 749   void cmp32(Register src1, AddressLiteral src2);
 750 
 751   void cmp32(Register src1, Address src2);
 752 
 753 #ifndef _LP64
 754   void cmpklass(Address dst, Metadata* obj);
 755   void cmpklass(Register dst, Metadata* obj);
 756   void cmpoop(Address dst, jobject obj);
 757   void cmpoop_raw(Address dst, jobject obj);
 758 #endif // _LP64
 759 
 760   void cmpoop(Register src1, Register src2);
 761   void cmpoop(Register src1, Address src2);
 762   void cmpoop(Register dst, jobject obj);
 763   void cmpoop_raw(Register dst, jobject obj);
 764 
 765   // NOTE src2 must be the lval. This is NOT an mem-mem compare
 766   void cmpptr(Address src1, AddressLiteral src2);
 767 
 768   void cmpptr(Register src1, AddressLiteral src2);
 769 
 770   void cmpptr(Register src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
 771   void cmpptr(Register src1, Address src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
 772   // void cmpptr(Address src1, Register src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
 773 
 774   void cmpptr(Register src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
 775   void cmpptr(Address src1, int32_t src2) { LP64_ONLY(cmpq(src1, src2)) NOT_LP64(cmpl(src1, src2)) ; }
 776 
 777   // cmp64 to avoild hiding cmpq
 778   void cmp64(Register src1, AddressLiteral src);
 779 
 780   void cmpxchgptr(Register reg, Address adr);
 781 
 782   void locked_cmpxchgptr(Register reg, AddressLiteral adr);
 783 
 784 
 785   void imulptr(Register dst, Register src) { LP64_ONLY(imulq(dst, src)) NOT_LP64(imull(dst, src)); }
 786   void imulptr(Register dst, Register src, int imm32) { LP64_ONLY(imulq(dst, src, imm32)) NOT_LP64(imull(dst, src, imm32)); }
 787 
 788 
 789   void negptr(Register dst) { LP64_ONLY(negq(dst)) NOT_LP64(negl(dst)); }
 790 
 791   void notptr(Register dst) { LP64_ONLY(notq(dst)) NOT_LP64(notl(dst)); }
 792 
 793   void shlptr(Register dst, int32_t shift);
 794   void shlptr(Register dst) { LP64_ONLY(shlq(dst)) NOT_LP64(shll(dst)); }
 795 
 796   void shrptr(Register dst, int32_t shift);
 797   void shrptr(Register dst) { LP64_ONLY(shrq(dst)) NOT_LP64(shrl(dst)); }
 798 
 799   void sarptr(Register dst) { LP64_ONLY(sarq(dst)) NOT_LP64(sarl(dst)); }
 800   void sarptr(Register dst, int32_t src) { LP64_ONLY(sarq(dst, src)) NOT_LP64(sarl(dst, src)); }
 801 
 802   void subptr(Address dst, int32_t src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
 803 
 804   void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
 805   void subptr(Register dst, int32_t src);
 806   // Force generation of a 4 byte immediate value even if it fits into 8bit
 807   void subptr_imm32(Register dst, int32_t src);
 808   void subptr(Register dst, Register src);
 809   void subptr(Register dst, RegisterOrConstant src) {
 810     if (src.is_constant()) subptr(dst, (int) src.as_constant());
 811     else                   subptr(dst,       src.as_register());
 812   }
 813 
 814   void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
 815   void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
 816 
 817   void xchgptr(Register src1, Register src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
 818   void xchgptr(Register src1, Address src2) { LP64_ONLY(xchgq(src1, src2)) NOT_LP64(xchgl(src1, src2)) ; }
 819 
 820   void xaddptr(Address src1, Register src2) { LP64_ONLY(xaddq(src1, src2)) NOT_LP64(xaddl(src1, src2)) ; }
 821 
 822 
 823 
 824   // Helper functions for statistics gathering.
 825   // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
 826   void cond_inc32(Condition cond, AddressLiteral counter_addr);
 827   // Unconditional atomic increment.
 828   void atomic_incl(Address counter_addr);
 829   void atomic_incl(AddressLiteral counter_addr, Register scr = rscratch1);
 830 #ifdef _LP64
 831   void atomic_incq(Address counter_addr);
 832   void atomic_incq(AddressLiteral counter_addr, Register scr = rscratch1);
 833 #endif
 834   void atomic_incptr(AddressLiteral counter_addr, Register scr = rscratch1) { LP64_ONLY(atomic_incq(counter_addr, scr)) NOT_LP64(atomic_incl(counter_addr, scr)) ; }
 835   void atomic_incptr(Address counter_addr) { LP64_ONLY(atomic_incq(counter_addr)) NOT_LP64(atomic_incl(counter_addr)) ; }
 836 
 837   void lea(Register dst, AddressLiteral adr);
 838   void lea(Address dst, AddressLiteral adr);
 839   void lea(Register dst, Address adr) { Assembler::lea(dst, adr); }
 840 
 841   void leal32(Register dst, Address src) { leal(dst, src); }
 842 
 843   // Import other testl() methods from the parent class or else
 844   // they will be hidden by the following overriding declaration.
 845   using Assembler::testl;
 846   void testl(Register dst, AddressLiteral src);
 847 
 848   void orptr(Register dst, Address src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
 849   void orptr(Register dst, Register src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
 850   void orptr(Register dst, int32_t src) { LP64_ONLY(orq(dst, src)) NOT_LP64(orl(dst, src)); }
 851   void orptr(Address dst, int32_t imm32) { LP64_ONLY(orq(dst, imm32)) NOT_LP64(orl(dst, imm32)); }
 852 
 853   void testptr(Register src, int32_t imm32) {  LP64_ONLY(testq(src, imm32)) NOT_LP64(testl(src, imm32)); }
 854   void testptr(Register src1, Address src2) { LP64_ONLY(testq(src1, src2)) NOT_LP64(testl(src1, src2)); }
 855   void testptr(Register src1, Register src2);
 856 
 857   void xorptr(Register dst, Register src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
 858   void xorptr(Register dst, Address src) { LP64_ONLY(xorq(dst, src)) NOT_LP64(xorl(dst, src)); }
 859 
 860   // Calls
 861 
 862   void call(Label& L, relocInfo::relocType rtype);
 863   void call(Register entry);
 864 
 865   // NOTE: this call transfers to the effective address of entry NOT
 866   // the address contained by entry. This is because this is more natural
 867   // for jumps/calls.
 868   void call(AddressLiteral entry);
 869 
 870   // Emit the CompiledIC call idiom
 871   void ic_call(address entry, jint method_index = 0);
 872 
 873   // Jumps
 874 
 875   // NOTE: these jumps tranfer to the effective address of dst NOT
 876   // the address contained by dst. This is because this is more natural
 877   // for jumps/calls.
 878   void jump(AddressLiteral dst);
 879   void jump_cc(Condition cc, AddressLiteral dst);
 880 
 881   // 32bit can do a case table jump in one instruction but we no longer allow the base
 882   // to be installed in the Address class. This jump will tranfers to the address
 883   // contained in the location described by entry (not the address of entry)
 884   void jump(ArrayAddress entry);
 885 
 886   // Floating
 887 
 888   void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
 889   void andpd(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
 890   void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); }
 891 
 892   void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
 893   void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
 894   void andps(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
 895 
 896   void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
 897   void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
 898   void comiss(XMMRegister dst, AddressLiteral src);
 899 
 900   void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
 901   void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
 902   void comisd(XMMRegister dst, AddressLiteral src);
 903 
 904   void fadd_s(Address src)        { Assembler::fadd_s(src); }
 905   void fadd_s(AddressLiteral src) { Assembler::fadd_s(as_Address(src)); }
 906 
 907   void fldcw(Address src) { Assembler::fldcw(src); }
 908   void fldcw(AddressLiteral src);
 909 
 910   void fld_s(int index)   { Assembler::fld_s(index); }
 911   void fld_s(Address src) { Assembler::fld_s(src); }
 912   void fld_s(AddressLiteral src);
 913 
 914   void fld_d(Address src) { Assembler::fld_d(src); }
 915   void fld_d(AddressLiteral src);
 916 
 917   void fld_x(Address src) { Assembler::fld_x(src); }
 918   void fld_x(AddressLiteral src);
 919 
 920   void fmul_s(Address src)        { Assembler::fmul_s(src); }
 921   void fmul_s(AddressLiteral src) { Assembler::fmul_s(as_Address(src)); }
 922 
 923   void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
 924   void ldmxcsr(AddressLiteral src);
 925 
 926 #ifdef _LP64
 927  private:
 928   void sha256_AVX2_one_round_compute(
 929     Register  reg_old_h,
 930     Register  reg_a,
 931     Register  reg_b,
 932     Register  reg_c,
 933     Register  reg_d,
 934     Register  reg_e,
 935     Register  reg_f,
 936     Register  reg_g,
 937     Register  reg_h,
 938     int iter);
 939   void sha256_AVX2_four_rounds_compute_first(int start);
 940   void sha256_AVX2_four_rounds_compute_last(int start);
 941   void sha256_AVX2_one_round_and_sched(
 942         XMMRegister xmm_0,     /* == ymm4 on 0, 1, 2, 3 iterations, then rotate 4 registers left on 4, 8, 12 iterations */
 943         XMMRegister xmm_1,     /* ymm5 */  /* full cycle is 16 iterations */
 944         XMMRegister xmm_2,     /* ymm6 */
 945         XMMRegister xmm_3,     /* ymm7 */
 946         Register    reg_a,      /* == eax on 0 iteration, then rotate 8 register right on each next iteration */
 947         Register    reg_b,      /* ebx */    /* full cycle is 8 iterations */
 948         Register    reg_c,      /* edi */
 949         Register    reg_d,      /* esi */
 950         Register    reg_e,      /* r8d */
 951         Register    reg_f,      /* r9d */
 952         Register    reg_g,      /* r10d */
 953         Register    reg_h,      /* r11d */
 954         int iter);
 955 
 956   void addm(int disp, Register r1, Register r2);
 957   void gfmul(XMMRegister tmp0, XMMRegister t);
 958   void schoolbookAAD(int i, Register subkeyH, XMMRegister data, XMMRegister tmp0,
 959                      XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3);
 960   void generateHtbl_one_block(Register htbl);
 961   void generateHtbl_eight_blocks(Register htbl);
 962  public:
 963   void sha256_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
 964                    XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
 965                    Register buf, Register state, Register ofs, Register limit, Register rsp,
 966                    bool multi_block, XMMRegister shuf_mask);
 967   void avx_ghash(Register state, Register htbl, Register data, Register blocks);
 968 #endif
 969 
 970 #ifdef _LP64
 971  private:
 972   void sha512_AVX2_one_round_compute(Register old_h, Register a, Register b, Register c, Register d,
 973                                      Register e, Register f, Register g, Register h, int iteration);
 974 
 975   void sha512_AVX2_one_round_and_schedule(XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
 976                                           Register a, Register b, Register c, Register d, Register e, Register f,
 977                                           Register g, Register h, int iteration);
 978 
 979   void addmq(int disp, Register r1, Register r2);
 980  public:
 981   void sha512_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
 982                    XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
 983                    Register buf, Register state, Register ofs, Register limit, Register rsp, bool multi_block,
 984                    XMMRegister shuf_mask);
 985 #endif
 986 
 987   void fast_sha1(XMMRegister abcd, XMMRegister e0, XMMRegister e1, XMMRegister msg0,
 988                  XMMRegister msg1, XMMRegister msg2, XMMRegister msg3, XMMRegister shuf_mask,
 989                  Register buf, Register state, Register ofs, Register limit, Register rsp,
 990                  bool multi_block);
 991 
 992 #ifdef _LP64
 993   void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
 994                    XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
 995                    Register buf, Register state, Register ofs, Register limit, Register rsp,
 996                    bool multi_block, XMMRegister shuf_mask);
 997 #else
 998   void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
 999                    XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
1000                    Register buf, Register state, Register ofs, Register limit, Register rsp,
1001                    bool multi_block);
1002 #endif
1003 
1004   void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1005                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1006                 Register rax, Register rcx, Register rdx, Register tmp);
1007 
1008 #ifdef _LP64
1009   void fast_log(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1010                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1011                 Register rax, Register rcx, Register rdx, Register tmp1, Register tmp2);
1012 
1013   void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1014                   XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1015                   Register rax, Register rcx, Register rdx, Register r11);
1016 
1017   void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
1018                 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
1019                 Register rdx, Register tmp1, Register tmp2, Register tmp3, Register tmp4);
1020 
1021   void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1022                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1023                 Register rax, Register rbx, Register rcx, Register rdx, Register tmp1, Register tmp2,
1024                 Register tmp3, Register tmp4);
1025 
1026   void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1027                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1028                 Register rax, Register rcx, Register rdx, Register tmp1,
1029                 Register tmp2, Register tmp3, Register tmp4);
1030   void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1031                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1032                 Register rax, Register rcx, Register rdx, Register tmp1,
1033                 Register tmp2, Register tmp3, Register tmp4);
1034 #else
1035   void fast_log(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 tmp1);
1038 
1039   void fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1040                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1041                 Register rax, Register rcx, Register rdx, Register tmp);
1042 
1043   void fast_pow(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4,
1044                 XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register rax, Register rcx,
1045                 Register rdx, Register tmp);
1046 
1047   void fast_sin(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1048                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1049                 Register rax, Register rbx, Register rdx);
1050 
1051   void fast_cos(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1052                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1053                 Register rax, Register rcx, Register rdx, Register tmp);
1054 
1055   void libm_sincos_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1056                         Register edx, Register ebx, Register esi, Register edi,
1057                         Register ebp, Register esp);
1058 
1059   void libm_reduce_pi04l(Register eax, Register ecx, Register edx, Register ebx,
1060                          Register esi, Register edi, Register ebp, Register esp);
1061 
1062   void libm_tancot_huge(XMMRegister xmm0, XMMRegister xmm1, Register eax, Register ecx,
1063                         Register edx, Register ebx, Register esi, Register edi,
1064                         Register ebp, Register esp);
1065 
1066   void fast_tan(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1067                 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1068                 Register rax, Register rcx, Register rdx, Register tmp);
1069 #endif
1070 
1071   void increase_precision();
1072   void restore_precision();
1073 
1074 private:
1075 
1076   // these are private because users should be doing movflt/movdbl
1077 
1078   void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
1079   void movss(Address dst, XMMRegister src)     { Assembler::movss(dst, src); }
1080   void movss(XMMRegister dst, Address src)     { Assembler::movss(dst, src); }
1081   void movss(XMMRegister dst, AddressLiteral src);
1082 
1083   void movlpd(XMMRegister dst, Address src)    {Assembler::movlpd(dst, src); }
1084   void movlpd(XMMRegister dst, AddressLiteral src);
1085 
1086 public:
1087 
1088   void addsd(XMMRegister dst, XMMRegister src)    { Assembler::addsd(dst, src); }
1089   void addsd(XMMRegister dst, Address src)        { Assembler::addsd(dst, src); }
1090   void addsd(XMMRegister dst, AddressLiteral src);
1091 
1092   void addss(XMMRegister dst, XMMRegister src)    { Assembler::addss(dst, src); }
1093   void addss(XMMRegister dst, Address src)        { Assembler::addss(dst, src); }
1094   void addss(XMMRegister dst, AddressLiteral src);
1095 
1096   void addpd(XMMRegister dst, XMMRegister src)    { Assembler::addpd(dst, src); }
1097   void addpd(XMMRegister dst, Address src)        { Assembler::addpd(dst, src); }
1098   void addpd(XMMRegister dst, AddressLiteral src);
1099 
1100   void divsd(XMMRegister dst, XMMRegister src)    { Assembler::divsd(dst, src); }
1101   void divsd(XMMRegister dst, Address src)        { Assembler::divsd(dst, src); }
1102   void divsd(XMMRegister dst, AddressLiteral src);
1103 
1104   void divss(XMMRegister dst, XMMRegister src)    { Assembler::divss(dst, src); }
1105   void divss(XMMRegister dst, Address src)        { Assembler::divss(dst, src); }
1106   void divss(XMMRegister dst, AddressLiteral src);
1107 
1108   // Move Unaligned Double Quadword
1109   void movdqu(Address     dst, XMMRegister src);
1110   void movdqu(XMMRegister dst, Address src);
1111   void movdqu(XMMRegister dst, XMMRegister src);
1112   void movdqu(XMMRegister dst, AddressLiteral src, Register scratchReg = rscratch1);
1113   // AVX Unaligned forms
1114   void vmovdqu(Address     dst, XMMRegister src);
1115   void vmovdqu(XMMRegister dst, Address src);
1116   void vmovdqu(XMMRegister dst, XMMRegister src);
1117   void vmovdqu(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1118   void evmovdquq(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1119   void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1120   void evmovdquq(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1121   void evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch);
1122 
1123   // Move Aligned Double Quadword
1124   void movdqa(XMMRegister dst, Address src)       { Assembler::movdqa(dst, src); }
1125   void movdqa(XMMRegister dst, XMMRegister src)   { Assembler::movdqa(dst, src); }
1126   void movdqa(XMMRegister dst, AddressLiteral src);
1127 
1128   void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
1129   void movsd(Address dst, XMMRegister src)     { Assembler::movsd(dst, src); }
1130   void movsd(XMMRegister dst, Address src)     { Assembler::movsd(dst, src); }
1131   void movsd(XMMRegister dst, AddressLiteral src);
1132 
1133   void mulpd(XMMRegister dst, XMMRegister src)    { Assembler::mulpd(dst, src); }
1134   void mulpd(XMMRegister dst, Address src)        { Assembler::mulpd(dst, src); }
1135   void mulpd(XMMRegister dst, AddressLiteral src);
1136 
1137   void mulsd(XMMRegister dst, XMMRegister src)    { Assembler::mulsd(dst, src); }
1138   void mulsd(XMMRegister dst, Address src)        { Assembler::mulsd(dst, src); }
1139   void mulsd(XMMRegister dst, AddressLiteral src);
1140 
1141   void mulss(XMMRegister dst, XMMRegister src)    { Assembler::mulss(dst, src); }
1142   void mulss(XMMRegister dst, Address src)        { Assembler::mulss(dst, src); }
1143   void mulss(XMMRegister dst, AddressLiteral src);
1144 
1145   // Carry-Less Multiplication Quadword
1146   void pclmulldq(XMMRegister dst, XMMRegister src) {
1147     // 0x00 - multiply lower 64 bits [0:63]
1148     Assembler::pclmulqdq(dst, src, 0x00);
1149   }
1150   void pclmulhdq(XMMRegister dst, XMMRegister src) {
1151     // 0x11 - multiply upper 64 bits [64:127]
1152     Assembler::pclmulqdq(dst, src, 0x11);
1153   }
1154 
1155   void pcmpeqb(XMMRegister dst, XMMRegister src);
1156   void pcmpeqw(XMMRegister dst, XMMRegister src);
1157 
1158   void pcmpestri(XMMRegister dst, Address src, int imm8);
1159   void pcmpestri(XMMRegister dst, XMMRegister src, int imm8);
1160 
1161   void pmovzxbw(XMMRegister dst, XMMRegister src);
1162   void pmovzxbw(XMMRegister dst, Address src);
1163 
1164   void pmovmskb(Register dst, XMMRegister src);
1165 
1166   void ptest(XMMRegister dst, XMMRegister src);
1167 
1168   void sqrtsd(XMMRegister dst, XMMRegister src)    { Assembler::sqrtsd(dst, src); }
1169   void sqrtsd(XMMRegister dst, Address src)        { Assembler::sqrtsd(dst, src); }
1170   void sqrtsd(XMMRegister dst, AddressLiteral src);
1171 
1172   void sqrtss(XMMRegister dst, XMMRegister src)    { Assembler::sqrtss(dst, src); }
1173   void sqrtss(XMMRegister dst, Address src)        { Assembler::sqrtss(dst, src); }
1174   void sqrtss(XMMRegister dst, AddressLiteral src);
1175 
1176   void subsd(XMMRegister dst, XMMRegister src)    { Assembler::subsd(dst, src); }
1177   void subsd(XMMRegister dst, Address src)        { Assembler::subsd(dst, src); }
1178   void subsd(XMMRegister dst, AddressLiteral src);
1179 
1180   void subss(XMMRegister dst, XMMRegister src)    { Assembler::subss(dst, src); }
1181   void subss(XMMRegister dst, Address src)        { Assembler::subss(dst, src); }
1182   void subss(XMMRegister dst, AddressLiteral src);
1183 
1184   void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
1185   void ucomiss(XMMRegister dst, Address src)     { Assembler::ucomiss(dst, src); }
1186   void ucomiss(XMMRegister dst, AddressLiteral src);
1187 
1188   void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1189   void ucomisd(XMMRegister dst, Address src)     { Assembler::ucomisd(dst, src); }
1190   void ucomisd(XMMRegister dst, AddressLiteral src);
1191 
1192   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1193   void xorpd(XMMRegister dst, XMMRegister src);
1194   void xorpd(XMMRegister dst, Address src)     { Assembler::xorpd(dst, src); }
1195   void xorpd(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1196 
1197   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1198   void xorps(XMMRegister dst, XMMRegister src);
1199   void xorps(XMMRegister dst, Address src)     { Assembler::xorps(dst, src); }
1200   void xorps(XMMRegister dst, AddressLiteral src, Register scratch_reg = rscratch1);
1201 
1202   // Shuffle Bytes
1203   void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); }
1204   void pshufb(XMMRegister dst, Address src)     { Assembler::pshufb(dst, src); }
1205   void pshufb(XMMRegister dst, AddressLiteral src);
1206   // AVX 3-operands instructions
1207 
1208   void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); }
1209   void vaddsd(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vaddsd(dst, nds, src); }
1210   void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1211 
1212   void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); }
1213   void vaddss(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vaddss(dst, nds, src); }
1214   void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1215 
1216   void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1217   void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len);
1218 
1219   void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1220   void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1221 
1222   void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1223   void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1224 
1225   void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1226   void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1227   void vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1228 
1229   void vpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len);
1230   void vpbroadcastw(XMMRegister dst, Address src, int vector_len) { Assembler::vpbroadcastw(dst, src, vector_len); }
1231 
1232   void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1233 
1234   void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1235 
1236   void vpmovzxbw(XMMRegister dst, Address src, int vector_len);
1237   void vpmovzxbw(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vpmovzxbw(dst, src, vector_len); }
1238 
1239   void vpmovmskb(Register dst, XMMRegister src);
1240 
1241   void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1242   void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1243 
1244   void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1245   void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1246 
1247   void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1248   void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1249 
1250   void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1251   void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1252 
1253   void evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1254   void evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1255 
1256   void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1257   void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1258 
1259   void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1260   void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1261 
1262   void vptest(XMMRegister dst, XMMRegister src);
1263 
1264   void punpcklbw(XMMRegister dst, XMMRegister src);
1265   void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); }
1266 
1267   void pshufd(XMMRegister dst, Address src, int mode);
1268   void pshufd(XMMRegister dst, XMMRegister src, int mode) { Assembler::pshufd(dst, src, mode); }
1269 
1270   void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1271   void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); }
1272 
1273   void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1274   void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len)     { Assembler::vandpd(dst, nds, src, vector_len); }
1275   void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1276 
1277   void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1278   void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len)     { Assembler::vandps(dst, nds, src, vector_len); }
1279   void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1280 
1281   void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); }
1282   void vdivsd(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vdivsd(dst, nds, src); }
1283   void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1284 
1285   void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); }
1286   void vdivss(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vdivss(dst, nds, src); }
1287   void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1288 
1289   void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); }
1290   void vmulsd(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vmulsd(dst, nds, src); }
1291   void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1292 
1293   void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); }
1294   void vmulss(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vmulss(dst, nds, src); }
1295   void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1296 
1297   void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); }
1298   void vsubsd(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vsubsd(dst, nds, src); }
1299   void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1300 
1301   void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); }
1302   void vsubss(XMMRegister dst, XMMRegister nds, Address src)     { Assembler::vsubss(dst, nds, src); }
1303   void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1304 
1305   void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1306   void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src);
1307 
1308   // AVX Vector instructions
1309 
1310   void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1311   void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1312   void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1313 
1314   void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1315   void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1316   void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1317 
1318   void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1319     if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1320       Assembler::vpxor(dst, nds, src, vector_len);
1321     else
1322       Assembler::vxorpd(dst, nds, src, vector_len);
1323   }
1324   void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
1325     if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1326       Assembler::vpxor(dst, nds, src, vector_len);
1327     else
1328       Assembler::vxorpd(dst, nds, src, vector_len);
1329   }
1330   void vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register scratch_reg = rscratch1);
1331 
1332   // Simple version for AVX2 256bit vectors
1333   void vpxor(XMMRegister dst, XMMRegister src) { Assembler::vpxor(dst, dst, src, true); }
1334   void vpxor(XMMRegister dst, Address src) { Assembler::vpxor(dst, dst, src, true); }
1335 
1336   void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8) {
1337     if (UseAVX > 2) {
1338       Assembler::vinserti32x4(dst, dst, src, imm8);
1339     } else if (UseAVX > 1) {
1340       // vinserti128 is available only in AVX2
1341       Assembler::vinserti128(dst, nds, src, imm8);
1342     } else {
1343       Assembler::vinsertf128(dst, nds, src, imm8);
1344     }
1345   }
1346 
1347   void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8) {
1348     if (UseAVX > 2) {
1349       Assembler::vinserti32x4(dst, dst, src, imm8);
1350     } else if (UseAVX > 1) {
1351       // vinserti128 is available only in AVX2
1352       Assembler::vinserti128(dst, nds, src, imm8);
1353     } else {
1354       Assembler::vinsertf128(dst, nds, src, imm8);
1355     }
1356   }
1357 
1358   void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8) {
1359     if (UseAVX > 2) {
1360       Assembler::vextracti32x4(dst, src, imm8);
1361     } else if (UseAVX > 1) {
1362       // vextracti128 is available only in AVX2
1363       Assembler::vextracti128(dst, src, imm8);
1364     } else {
1365       Assembler::vextractf128(dst, src, imm8);
1366     }
1367   }
1368 
1369   void vextracti128(Address dst, XMMRegister src, uint8_t imm8) {
1370     if (UseAVX > 2) {
1371       Assembler::vextracti32x4(dst, src, imm8);
1372     } else if (UseAVX > 1) {
1373       // vextracti128 is available only in AVX2
1374       Assembler::vextracti128(dst, src, imm8);
1375     } else {
1376       Assembler::vextractf128(dst, src, imm8);
1377     }
1378   }
1379 
1380   // 128bit copy to/from high 128 bits of 256bit (YMM) vector registers
1381   void vinserti128_high(XMMRegister dst, XMMRegister src) {
1382     vinserti128(dst, dst, src, 1);
1383   }
1384   void vinserti128_high(XMMRegister dst, Address src) {
1385     vinserti128(dst, dst, src, 1);
1386   }
1387   void vextracti128_high(XMMRegister dst, XMMRegister src) {
1388     vextracti128(dst, src, 1);
1389   }
1390   void vextracti128_high(Address dst, XMMRegister src) {
1391     vextracti128(dst, src, 1);
1392   }
1393 
1394   void vinsertf128_high(XMMRegister dst, XMMRegister src) {
1395     if (UseAVX > 2) {
1396       Assembler::vinsertf32x4(dst, dst, src, 1);
1397     } else {
1398       Assembler::vinsertf128(dst, dst, src, 1);
1399     }
1400   }
1401 
1402   void vinsertf128_high(XMMRegister dst, Address src) {
1403     if (UseAVX > 2) {
1404       Assembler::vinsertf32x4(dst, dst, src, 1);
1405     } else {
1406       Assembler::vinsertf128(dst, dst, src, 1);
1407     }
1408   }
1409 
1410   void vextractf128_high(XMMRegister dst, XMMRegister src) {
1411     if (UseAVX > 2) {
1412       Assembler::vextractf32x4(dst, src, 1);
1413     } else {
1414       Assembler::vextractf128(dst, src, 1);
1415     }
1416   }
1417 
1418   void vextractf128_high(Address dst, XMMRegister src) {
1419     if (UseAVX > 2) {
1420       Assembler::vextractf32x4(dst, src, 1);
1421     } else {
1422       Assembler::vextractf128(dst, src, 1);
1423     }
1424   }
1425 
1426   // 256bit copy to/from high 256 bits of 512bit (ZMM) vector registers
1427   void vinserti64x4_high(XMMRegister dst, XMMRegister src) {
1428     Assembler::vinserti64x4(dst, dst, src, 1);
1429   }
1430   void vinsertf64x4_high(XMMRegister dst, XMMRegister src) {
1431     Assembler::vinsertf64x4(dst, dst, src, 1);
1432   }
1433   void vextracti64x4_high(XMMRegister dst, XMMRegister src) {
1434     Assembler::vextracti64x4(dst, src, 1);
1435   }
1436   void vextractf64x4_high(XMMRegister dst, XMMRegister src) {
1437     Assembler::vextractf64x4(dst, src, 1);
1438   }
1439   void vextractf64x4_high(Address dst, XMMRegister src) {
1440     Assembler::vextractf64x4(dst, src, 1);
1441   }
1442   void vinsertf64x4_high(XMMRegister dst, Address src) {
1443     Assembler::vinsertf64x4(dst, dst, src, 1);
1444   }
1445 
1446   // 128bit copy to/from low 128 bits of 256bit (YMM) vector registers
1447   void vinserti128_low(XMMRegister dst, XMMRegister src) {
1448     vinserti128(dst, dst, src, 0);
1449   }
1450   void vinserti128_low(XMMRegister dst, Address src) {
1451     vinserti128(dst, dst, src, 0);
1452   }
1453   void vextracti128_low(XMMRegister dst, XMMRegister src) {
1454     vextracti128(dst, src, 0);
1455   }
1456   void vextracti128_low(Address dst, XMMRegister src) {
1457     vextracti128(dst, src, 0);
1458   }
1459 
1460   void vinsertf128_low(XMMRegister dst, XMMRegister src) {
1461     if (UseAVX > 2) {
1462       Assembler::vinsertf32x4(dst, dst, src, 0);
1463     } else {
1464       Assembler::vinsertf128(dst, dst, src, 0);
1465     }
1466   }
1467 
1468   void vinsertf128_low(XMMRegister dst, Address src) {
1469     if (UseAVX > 2) {
1470       Assembler::vinsertf32x4(dst, dst, src, 0);
1471     } else {
1472       Assembler::vinsertf128(dst, dst, src, 0);
1473     }
1474   }
1475 
1476   void vextractf128_low(XMMRegister dst, XMMRegister src) {
1477     if (UseAVX > 2) {
1478       Assembler::vextractf32x4(dst, src, 0);
1479     } else {
1480       Assembler::vextractf128(dst, src, 0);
1481     }
1482   }
1483 
1484   void vextractf128_low(Address dst, XMMRegister src) {
1485     if (UseAVX > 2) {
1486       Assembler::vextractf32x4(dst, src, 0);
1487     } else {
1488       Assembler::vextractf128(dst, src, 0);
1489     }
1490   }
1491 
1492   // 256bit copy to/from low 256 bits of 512bit (ZMM) vector registers
1493   void vinserti64x4_low(XMMRegister dst, XMMRegister src) {
1494     Assembler::vinserti64x4(dst, dst, src, 0);
1495   }
1496   void vinsertf64x4_low(XMMRegister dst, XMMRegister src) {
1497     Assembler::vinsertf64x4(dst, dst, src, 0);
1498   }
1499   void vextracti64x4_low(XMMRegister dst, XMMRegister src) {
1500     Assembler::vextracti64x4(dst, src, 0);
1501   }
1502   void vextractf64x4_low(XMMRegister dst, XMMRegister src) {
1503     Assembler::vextractf64x4(dst, src, 0);
1504   }
1505   void vextractf64x4_low(Address dst, XMMRegister src) {
1506     Assembler::vextractf64x4(dst, src, 0);
1507   }
1508   void vinsertf64x4_low(XMMRegister dst, Address src) {
1509     Assembler::vinsertf64x4(dst, dst, src, 0);
1510   }
1511 
1512   // Carry-Less Multiplication Quadword
1513   void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1514     // 0x00 - multiply lower 64 bits [0:63]
1515     Assembler::vpclmulqdq(dst, nds, src, 0x00);
1516   }
1517   void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1518     // 0x11 - multiply upper 64 bits [64:127]
1519     Assembler::vpclmulqdq(dst, nds, src, 0x11);
1520   }
1521   void vpclmullqhqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1522     // 0x10 - multiply nds[0:63] and src[64:127]
1523     Assembler::vpclmulqdq(dst, nds, src, 0x10);
1524   }
1525   void vpclmulhqlqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1526     //0x01 - multiply nds[64:127] and src[0:63]
1527     Assembler::vpclmulqdq(dst, nds, src, 0x01);
1528   }
1529 
1530   void evpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1531     // 0x00 - multiply lower 64 bits [0:63]
1532     Assembler::evpclmulqdq(dst, nds, src, 0x00, vector_len);
1533   }
1534   void evpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1535     // 0x11 - multiply upper 64 bits [64:127]
1536     Assembler::evpclmulqdq(dst, nds, src, 0x11, vector_len);
1537   }
1538 
1539   // Data
1540 
1541   void cmov32( Condition cc, Register dst, Address  src);
1542   void cmov32( Condition cc, Register dst, Register src);
1543 
1544   void cmov(   Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); }
1545 
1546   void cmovptr(Condition cc, Register dst, Address  src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1547   void cmovptr(Condition cc, Register dst, Register src) { LP64_ONLY(cmovq(cc, dst, src)) NOT_LP64(cmov32(cc, dst, src)); }
1548 
1549   void movoop(Register dst, jobject obj);
1550   void movoop(Address dst, jobject obj);
1551 
1552   void mov_metadata(Register dst, Metadata* obj);
1553   void mov_metadata(Address dst, Metadata* obj);
1554 
1555   void movptr(ArrayAddress dst, Register src);
1556   // can this do an lea?
1557   void movptr(Register dst, ArrayAddress src);
1558 
1559   void movptr(Register dst, Address src);
1560 
1561 #ifdef _LP64
1562   void movptr(Register dst, AddressLiteral src, Register scratch=rscratch1);
1563 #else
1564   void movptr(Register dst, AddressLiteral src, Register scratch=noreg); // Scratch reg is ignored in 32-bit
1565 #endif
1566 
1567   void movptr(Register dst, intptr_t src);
1568   void movptr(Register dst, Register src);
1569   void movptr(Address dst, intptr_t src);
1570 
1571   void movptr(Address dst, Register src);
1572 
1573   void movptr(Register dst, RegisterOrConstant src) {
1574     if (src.is_constant()) movptr(dst, src.as_constant());
1575     else                   movptr(dst, src.as_register());
1576   }
1577 
1578 #ifdef _LP64
1579   // Generally the next two are only used for moving NULL
1580   // Although there are situations in initializing the mark word where
1581   // they could be used. They are dangerous.
1582 
1583   // They only exist on LP64 so that int32_t and intptr_t are not the same
1584   // and we have ambiguous declarations.
1585 
1586   void movptr(Address dst, int32_t imm32);
1587   void movptr(Register dst, int32_t imm32);
1588 #endif // _LP64
1589 
1590   // to avoid hiding movl
1591   void mov32(AddressLiteral dst, Register src);
1592   void mov32(Register dst, AddressLiteral src);
1593 
1594   // to avoid hiding movb
1595   void movbyte(ArrayAddress dst, int src);
1596 
1597   // Import other mov() methods from the parent class or else
1598   // they will be hidden by the following overriding declaration.
1599   using Assembler::movdl;
1600   using Assembler::movq;
1601   void movdl(XMMRegister dst, AddressLiteral src);
1602   void movq(XMMRegister dst, AddressLiteral src);
1603 
1604   // Can push value or effective address
1605   void pushptr(AddressLiteral src);
1606 
1607   void pushptr(Address src) { LP64_ONLY(pushq(src)) NOT_LP64(pushl(src)); }
1608   void popptr(Address src) { LP64_ONLY(popq(src)) NOT_LP64(popl(src)); }
1609 
1610   void pushoop(jobject obj);
1611   void pushklass(Metadata* obj);
1612 
1613   // sign extend as need a l to ptr sized element
1614   void movl2ptr(Register dst, Address src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(movl(dst, src)); }
1615   void movl2ptr(Register dst, Register src) { LP64_ONLY(movslq(dst, src)) NOT_LP64(if (dst != src) movl(dst, src)); }
1616 
1617 #ifdef COMPILER2
1618   // Generic instructions support for use in .ad files C2 code generation
1619   void vabsnegd(int opcode, XMMRegister dst, Register scr);
1620   void vabsnegd(int opcode, XMMRegister dst, XMMRegister src, int vector_len, Register scr);
1621   void vabsnegf(int opcode, XMMRegister dst, Register scr);
1622   void vabsnegf(int opcode, XMMRegister dst, XMMRegister src, int vector_len, Register scr);
1623   void vextendbw(bool sign, XMMRegister dst, XMMRegister src, int vector_len);
1624   void vextendbw(bool sign, XMMRegister dst, XMMRegister src);
1625   void vshiftd(int opcode, XMMRegister dst, XMMRegister src);
1626   void vshiftd(int opcode, XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1627   void vshiftw(int opcode, XMMRegister dst, XMMRegister src);
1628   void vshiftw(int opcode, XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1629   void vshiftq(int opcode, XMMRegister dst, XMMRegister src);
1630   void vshiftq(int opcode, XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1631 #endif
1632 
1633   // C2 compiled method's prolog code.
1634   void verified_entry(int framesize, int stack_bang_size, bool fp_mode_24b, bool is_stub);
1635 
1636   // clear memory of size 'cnt' qwords, starting at 'base';
1637   // if 'is_large' is set, do not try to produce short loop
1638   void clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, bool is_large);
1639 
1640   // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM registers
1641   void xmm_clear_mem(Register base, Register cnt, XMMRegister xtmp);
1642 
1643 #ifdef COMPILER2
1644   void string_indexof_char(Register str1, Register cnt1, Register ch, Register result,
1645                            XMMRegister vec1, XMMRegister vec2, XMMRegister vec3, Register tmp);
1646 
1647   // IndexOf strings.
1648   // Small strings are loaded through stack if they cross page boundary.
1649   void string_indexof(Register str1, Register str2,
1650                       Register cnt1, Register cnt2,
1651                       int int_cnt2,  Register result,
1652                       XMMRegister vec, Register tmp,
1653                       int ae);
1654 
1655   // IndexOf for constant substrings with size >= 8 elements
1656   // which don't need to be loaded through stack.
1657   void string_indexofC8(Register str1, Register str2,
1658                       Register cnt1, Register cnt2,
1659                       int int_cnt2,  Register result,
1660                       XMMRegister vec, Register tmp,
1661                       int ae);
1662 
1663     // Smallest code: we don't need to load through stack,
1664     // check string tail.
1665 
1666   // helper function for string_compare
1667   void load_next_elements(Register elem1, Register elem2, Register str1, Register str2,
1668                           Address::ScaleFactor scale, Address::ScaleFactor scale1,
1669                           Address::ScaleFactor scale2, Register index, int ae);
1670   // Compare strings.
1671   void string_compare(Register str1, Register str2,
1672                       Register cnt1, Register cnt2, Register result,
1673                       XMMRegister vec1, int ae);
1674 
1675   // Search for Non-ASCII character (Negative byte value) in a byte array,
1676   // return true if it has any and false otherwise.
1677   void has_negatives(Register ary1, Register len,
1678                      Register result, Register tmp1,
1679                      XMMRegister vec1, XMMRegister vec2);
1680 
1681   // Compare char[] or byte[] arrays.
1682   void arrays_equals(bool is_array_equ, Register ary1, Register ary2,
1683                      Register limit, Register result, Register chr,
1684                      XMMRegister vec1, XMMRegister vec2, bool is_char);
1685 
1686 #endif
1687 
1688   // Fill primitive arrays
1689   void generate_fill(BasicType t, bool aligned,
1690                      Register to, Register value, Register count,
1691                      Register rtmp, XMMRegister xtmp);
1692 
1693   void encode_iso_array(Register src, Register dst, Register len,
1694                         XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1695                         XMMRegister tmp4, Register tmp5, Register result);
1696 
1697 #ifdef _LP64
1698   void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2);
1699   void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1700                              Register y, Register y_idx, Register z,
1701                              Register carry, Register product,
1702                              Register idx, Register kdx);
1703   void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1704                               Register yz_idx, Register idx,
1705                               Register carry, Register product, int offset);
1706   void multiply_128_x_128_bmi2_loop(Register y, Register z,
1707                                     Register carry, Register carry2,
1708                                     Register idx, Register jdx,
1709                                     Register yz_idx1, Register yz_idx2,
1710                                     Register tmp, Register tmp3, Register tmp4);
1711   void multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
1712                                Register yz_idx, Register idx, Register jdx,
1713                                Register carry, Register product,
1714                                Register carry2);
1715   void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register zlen,
1716                        Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
1717   void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
1718                      Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1719   void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry,
1720                             Register tmp2);
1721   void multiply_add_64(Register sum, Register op1, Register op2, Register carry,
1722                        Register rdxReg, Register raxReg);
1723   void add_one_64(Register z, Register zlen, Register carry, Register tmp1);
1724   void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1725                        Register tmp3, Register tmp4);
1726   void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
1727                      Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
1728 
1729   void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1,
1730                Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1731                Register raxReg);
1732   void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1,
1733                Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
1734                Register raxReg);
1735   void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
1736                            Register result, Register tmp1, Register tmp2,
1737                            XMMRegister vec1, XMMRegister vec2, XMMRegister vec3);
1738 #endif
1739 
1740   // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
1741   void update_byte_crc32(Register crc, Register val, Register table);
1742   void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp);
1743   // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic
1744   // Note on a naming convention:
1745   // Prefix w = register only used on a Westmere+ architecture
1746   // Prefix n = register only used on a Nehalem architecture
1747 #ifdef _LP64
1748   void crc32c_ipl_alg4(Register in_out, uint32_t n,
1749                        Register tmp1, Register tmp2, Register tmp3);
1750 #else
1751   void crc32c_ipl_alg4(Register in_out, uint32_t n,
1752                        Register tmp1, Register tmp2, Register tmp3,
1753                        XMMRegister xtmp1, XMMRegister xtmp2);
1754 #endif
1755   void crc32c_pclmulqdq(XMMRegister w_xtmp1,
1756                         Register in_out,
1757                         uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
1758                         XMMRegister w_xtmp2,
1759                         Register tmp1,
1760                         Register n_tmp2, Register n_tmp3);
1761   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,
1762                        XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1763                        Register tmp1, Register tmp2,
1764                        Register n_tmp3);
1765   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,
1766                          Register in_out1, Register in_out2, Register in_out3,
1767                          Register tmp1, Register tmp2, Register tmp3,
1768                          XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1769                          Register tmp4, Register tmp5,
1770                          Register n_tmp6);
1771   void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
1772                             Register tmp1, Register tmp2, Register tmp3,
1773                             Register tmp4, Register tmp5, Register tmp6,
1774                             XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
1775                             bool is_pclmulqdq_supported);
1776   // Fold 128-bit data chunk
1777   void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1778   void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf);
1779   // Fold 8-bit data
1780   void fold_8bit_crc32(Register crc, Register table, Register tmp);
1781   void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp);
1782   void fold_128bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
1783 
1784   // Compress char[] array to byte[].
1785   void char_array_compress(Register src, Register dst, Register len,
1786                            XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1787                            XMMRegister tmp4, Register tmp5, Register result);
1788 
1789   // Inflate byte[] array to char[].
1790   void byte_array_inflate(Register src, Register dst, Register len,
1791                           XMMRegister tmp1, Register tmp2);
1792 
1793 };
1794 
1795 /**
1796  * class SkipIfEqual:
1797  *
1798  * Instantiating this class will result in assembly code being output that will
1799  * jump around any code emitted between the creation of the instance and it's
1800  * automatic destruction at the end of a scope block, depending on the value of
1801  * the flag passed to the constructor, which will be checked at run-time.
1802  */
1803 class SkipIfEqual {
1804  private:
1805   MacroAssembler* _masm;
1806   Label _label;
1807 
1808  public:
1809    SkipIfEqual(MacroAssembler*, const bool* flag_addr, bool value);
1810    ~SkipIfEqual();
1811 };
1812 
1813 #endif // CPU_X86_MACROASSEMBLER_X86_HPP