1 /*
   2  * Copyright (c) 2000, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "c1/c1_CodeStubs.hpp"
  29 #include "c1/c1_Compilation.hpp"
  30 #include "c1/c1_LIRAssembler.hpp"
  31 #include "c1/c1_MacroAssembler.hpp"
  32 #include "c1/c1_Runtime1.hpp"
  33 #include "c1/c1_ValueStack.hpp"
  34 #include "ci/ciArrayKlass.hpp"
  35 #include "ci/ciInstance.hpp"
  36 #include "compiler/oopMap.hpp"
  37 #include "gc/shared/collectedHeap.hpp"
  38 #include "gc/shared/gc_globals.hpp"
  39 #include "nativeInst_x86.hpp"
  40 #include "oops/objArrayKlass.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/safepointMechanism.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "runtime/stubRoutines.hpp"
  45 #include "utilities/powerOfTwo.hpp"
  46 #include "vmreg_x86.inline.hpp"
  47 
  48 
  49 // These masks are used to provide 128-bit aligned bitmasks to the XMM
  50 // instructions, to allow sign-masking or sign-bit flipping.  They allow
  51 // fast versions of NegF/NegD and AbsF/AbsD.
  52 
  53 // Note: 'double' and 'long long' have 32-bits alignment on x86.
  54 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
  55   // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
  56   // of 128-bits operands for SSE instructions.
  57   jlong *operand = (jlong*)(((intptr_t)adr) & ((intptr_t)(~0xF)));
  58   // Store the value to a 128-bits operand.
  59   operand[0] = lo;
  60   operand[1] = hi;
  61   return operand;
  62 }
  63 
  64 // Buffer for 128-bits masks used by SSE instructions.
  65 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
  66 
  67 // Static initialization during VM startup.
  68 static jlong *float_signmask_pool  = double_quadword(&fp_signmask_pool[1*2],         CONST64(0x7FFFFFFF7FFFFFFF),         CONST64(0x7FFFFFFF7FFFFFFF));
  69 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2],         CONST64(0x7FFFFFFFFFFFFFFF),         CONST64(0x7FFFFFFFFFFFFFFF));
  70 static jlong *float_signflip_pool  = double_quadword(&fp_signmask_pool[3*2], (jlong)UCONST64(0x8000000080000000), (jlong)UCONST64(0x8000000080000000));
  71 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], (jlong)UCONST64(0x8000000000000000), (jlong)UCONST64(0x8000000000000000));
  72 
  73 
  74 NEEDS_CLEANUP // remove this definitions ?
  75 const Register SYNC_header = rax;   // synchronization header
  76 const Register SHIFT_count = rcx;   // where count for shift operations must be
  77 
  78 #define __ _masm->
  79 
  80 
  81 static void select_different_registers(Register preserve,
  82                                        Register extra,
  83                                        Register &tmp1,
  84                                        Register &tmp2) {
  85   if (tmp1 == preserve) {
  86     assert_different_registers(tmp1, tmp2, extra);
  87     tmp1 = extra;
  88   } else if (tmp2 == preserve) {
  89     assert_different_registers(tmp1, tmp2, extra);
  90     tmp2 = extra;
  91   }
  92   assert_different_registers(preserve, tmp1, tmp2);
  93 }
  94 
  95 
  96 
  97 static void select_different_registers(Register preserve,
  98                                        Register extra,
  99                                        Register &tmp1,
 100                                        Register &tmp2,
 101                                        Register &tmp3) {
 102   if (tmp1 == preserve) {
 103     assert_different_registers(tmp1, tmp2, tmp3, extra);
 104     tmp1 = extra;
 105   } else if (tmp2 == preserve) {
 106     assert_different_registers(tmp1, tmp2, tmp3, extra);
 107     tmp2 = extra;
 108   } else if (tmp3 == preserve) {
 109     assert_different_registers(tmp1, tmp2, tmp3, extra);
 110     tmp3 = extra;
 111   }
 112   assert_different_registers(preserve, tmp1, tmp2, tmp3);
 113 }
 114 
 115 
 116 
 117 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
 118   if (opr->is_constant()) {
 119     LIR_Const* constant = opr->as_constant_ptr();
 120     switch (constant->type()) {
 121       case T_INT: {
 122         return true;
 123       }
 124 
 125       default:
 126         return false;
 127     }
 128   }
 129   return false;
 130 }
 131 
 132 
 133 LIR_Opr LIR_Assembler::receiverOpr() {
 134   return FrameMap::receiver_opr;
 135 }
 136 
 137 LIR_Opr LIR_Assembler::osrBufferPointer() {
 138   return FrameMap::as_pointer_opr(receiverOpr()->as_register());
 139 }
 140 
 141 //--------------fpu register translations-----------------------
 142 
 143 
 144 address LIR_Assembler::float_constant(float f) {
 145   address const_addr = __ float_constant(f);
 146   if (const_addr == nullptr) {
 147     bailout("const section overflow");
 148     return __ code()->consts()->start();
 149   } else {
 150     return const_addr;
 151   }
 152 }
 153 
 154 
 155 address LIR_Assembler::double_constant(double d) {
 156   address const_addr = __ double_constant(d);
 157   if (const_addr == nullptr) {
 158     bailout("const section overflow");
 159     return __ code()->consts()->start();
 160   } else {
 161     return const_addr;
 162   }
 163 }
 164 
 165 #ifndef _LP64
 166 void LIR_Assembler::fpop() {
 167   __ fpop();
 168 }
 169 
 170 void LIR_Assembler::fxch(int i) {
 171   __ fxch(i);
 172 }
 173 
 174 void LIR_Assembler::fld(int i) {
 175   __ fld_s(i);
 176 }
 177 
 178 void LIR_Assembler::ffree(int i) {
 179   __ ffree(i);
 180 }
 181 #endif // !_LP64
 182 
 183 void LIR_Assembler::breakpoint() {
 184   __ int3();
 185 }
 186 
 187 void LIR_Assembler::push(LIR_Opr opr) {
 188   if (opr->is_single_cpu()) {
 189     __ push_reg(opr->as_register());
 190   } else if (opr->is_double_cpu()) {
 191     NOT_LP64(__ push_reg(opr->as_register_hi()));
 192     __ push_reg(opr->as_register_lo());
 193   } else if (opr->is_stack()) {
 194     __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
 195   } else if (opr->is_constant()) {
 196     LIR_Const* const_opr = opr->as_constant_ptr();
 197     if (const_opr->type() == T_OBJECT) {
 198       __ push_oop(const_opr->as_jobject(), rscratch1);
 199     } else if (const_opr->type() == T_INT) {
 200       __ push_jint(const_opr->as_jint());
 201     } else {
 202       ShouldNotReachHere();
 203     }
 204 
 205   } else {
 206     ShouldNotReachHere();
 207   }
 208 }
 209 
 210 void LIR_Assembler::pop(LIR_Opr opr) {
 211   if (opr->is_single_cpu()) {
 212     __ pop_reg(opr->as_register());
 213   } else {
 214     ShouldNotReachHere();
 215   }
 216 }
 217 
 218 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
 219   return addr->base()->is_illegal() && addr->index()->is_illegal();
 220 }
 221 
 222 //-------------------------------------------
 223 
 224 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 225   return as_Address(addr, rscratch1);
 226 }
 227 
 228 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
 229   if (addr->base()->is_illegal()) {
 230     assert(addr->index()->is_illegal(), "must be illegal too");
 231     AddressLiteral laddr((address)addr->disp(), relocInfo::none);
 232     if (! __ reachable(laddr)) {
 233       __ movptr(tmp, laddr.addr());
 234       Address res(tmp, 0);
 235       return res;
 236     } else {
 237       return __ as_Address(laddr);
 238     }
 239   }
 240 
 241   Register base = addr->base()->as_pointer_register();
 242 
 243   if (addr->index()->is_illegal()) {
 244     return Address( base, addr->disp());
 245   } else if (addr->index()->is_cpu_register()) {
 246     Register index = addr->index()->as_pointer_register();
 247     return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
 248   } else if (addr->index()->is_constant()) {
 249     intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
 250     assert(Assembler::is_simm32(addr_offset), "must be");
 251 
 252     return Address(base, addr_offset);
 253   } else {
 254     Unimplemented();
 255     return Address();
 256   }
 257 }
 258 
 259 
 260 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 261   Address base = as_Address(addr);
 262   return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
 263 }
 264 
 265 
 266 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 267   return as_Address(addr);
 268 }
 269 
 270 
 271 void LIR_Assembler::osr_entry() {
 272   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 273   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 274   ValueStack* entry_state = osr_entry->state();
 275   int number_of_locks = entry_state->locks_size();
 276 
 277   // we jump here if osr happens with the interpreter
 278   // state set up to continue at the beginning of the
 279   // loop that triggered osr - in particular, we have
 280   // the following registers setup:
 281   //
 282   // rcx: osr buffer
 283   //
 284 
 285   // build frame
 286   ciMethod* m = compilation()->method();
 287   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 288 
 289   // OSR buffer is
 290   //
 291   // locals[nlocals-1..0]
 292   // monitors[0..number_of_locks]
 293   //
 294   // locals is a direct copy of the interpreter frame so in the osr buffer
 295   // so first slot in the local array is the last local from the interpreter
 296   // and last slot is local[0] (receiver) from the interpreter
 297   //
 298   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 299   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 300   // in the interpreter frame (the method lock if a sync method)
 301 
 302   // Initialize monitors in the compiled activation.
 303   //   rcx: pointer to osr buffer
 304   //
 305   // All other registers are dead at this point and the locals will be
 306   // copied into place by code emitted in the IR.
 307 
 308   Register OSR_buf = osrBufferPointer()->as_pointer_register();
 309   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 310     int monitor_offset = BytesPerWord * method()->max_locals() +
 311       (BasicObjectLock::size() * BytesPerWord) * (number_of_locks - 1);
 312     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 313     // the OSR buffer using 2 word entries: first the lock and then
 314     // the oop.
 315     for (int i = 0; i < number_of_locks; i++) {
 316       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 317 #ifdef ASSERT
 318       // verify the interpreter's monitor has a non-null object
 319       {
 320         Label L;
 321         __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), NULL_WORD);
 322         __ jcc(Assembler::notZero, L);
 323         __ stop("locked object is null");
 324         __ bind(L);
 325       }
 326 #endif
 327       __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
 328       __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
 329       __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
 330       __ movptr(frame_map()->address_for_monitor_object(i), rbx);
 331     }
 332   }
 333 }
 334 
 335 
 336 // inline cache check; done before the frame is built.
 337 int LIR_Assembler::check_icache() {
 338   return __ ic_check(CodeEntryAlignment);
 339 }
 340 
 341 void LIR_Assembler::clinit_barrier(ciMethod* method) {
 342   assert(VM_Version::supports_fast_class_init_checks(), "sanity");
 343   assert(!method->holder()->is_not_initialized(), "initialization should have been started");
 344 
 345   Label L_skip_barrier;
 346   Register klass = rscratch1;
 347   Register thread = LP64_ONLY( r15_thread ) NOT_LP64( noreg );
 348   assert(thread != noreg, "x86_32 not implemented");
 349 
 350   __ mov_metadata(klass, method->holder()->constant_encoding());
 351   __ clinit_barrier(klass, thread, &L_skip_barrier /*L_fast_path*/);
 352 
 353   __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
 354 
 355   __ bind(L_skip_barrier);
 356 }
 357 
 358 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
 359   jobject o = nullptr;
 360   PatchingStub* patch = new PatchingStub(_masm, patching_id(info));
 361   __ movoop(reg, o);
 362   patching_epilog(patch, lir_patch_normal, reg, info);
 363 }
 364 
 365 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
 366   Metadata* o = nullptr;
 367   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
 368   __ mov_metadata(reg, o);
 369   patching_epilog(patch, lir_patch_normal, reg, info);
 370 }
 371 
 372 // This specifies the rsp decrement needed to build the frame
 373 int LIR_Assembler::initial_frame_size_in_bytes() const {
 374   // if rounding, must let FrameMap know!
 375 
 376   // The frame_map records size in slots (32bit word)
 377 
 378   // subtract two words to account for return address and link
 379   return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word))  * VMRegImpl::stack_slot_size;
 380 }
 381 
 382 
 383 int LIR_Assembler::emit_exception_handler() {
 384   // generate code for exception handler
 385   address handler_base = __ start_a_stub(exception_handler_size());
 386   if (handler_base == nullptr) {
 387     // not enough space left for the handler
 388     bailout("exception handler overflow");
 389     return -1;
 390   }
 391 
 392   int offset = code_offset();
 393 
 394   // the exception oop and pc are in rax, and rdx
 395   // no other registers need to be preserved, so invalidate them
 396   __ invalidate_registers(false, true, true, false, true, true);
 397 
 398   // check that there is really an exception
 399   __ verify_not_null_oop(rax);
 400 
 401   // search an exception handler (rax: exception oop, rdx: throwing pc)
 402   __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
 403   __ should_not_reach_here();
 404   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 405   __ end_a_stub();
 406 
 407   return offset;
 408 }
 409 
 410 
 411 // Emit the code to remove the frame from the stack in the exception
 412 // unwind path.
 413 int LIR_Assembler::emit_unwind_handler() {
 414 #ifndef PRODUCT
 415   if (CommentedAssembly) {
 416     _masm->block_comment("Unwind handler");
 417   }
 418 #endif
 419 
 420   int offset = code_offset();
 421 
 422   // Fetch the exception from TLS and clear out exception related thread state
 423   Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
 424   NOT_LP64(__ get_thread(thread));
 425   __ movptr(rax, Address(thread, JavaThread::exception_oop_offset()));
 426   __ movptr(Address(thread, JavaThread::exception_oop_offset()), NULL_WORD);
 427   __ movptr(Address(thread, JavaThread::exception_pc_offset()), NULL_WORD);
 428 
 429   __ bind(_unwind_handler_entry);
 430   __ verify_not_null_oop(rax);
 431   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 432     __ mov(rbx, rax);  // Preserve the exception (rbx is always callee-saved)
 433   }
 434 
 435   // Perform needed unlocking
 436   MonitorExitStub* stub = nullptr;
 437   if (method()->is_synchronized()) {
 438     monitor_address(0, FrameMap::rax_opr);
 439     stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
 440     if (LockingMode == LM_MONITOR) {
 441       __ jmp(*stub->entry());
 442     } else {
 443       __ unlock_object(rdi, rsi, rax, *stub->entry());
 444     }
 445     __ bind(*stub->continuation());
 446   }
 447 
 448   if (compilation()->env()->dtrace_method_probes()) {
 449 #ifdef _LP64
 450     __ mov(rdi, r15_thread);
 451     __ mov_metadata(rsi, method()->constant_encoding());
 452 #else
 453     __ get_thread(rax);
 454     __ movptr(Address(rsp, 0), rax);
 455     __ mov_metadata(Address(rsp, sizeof(void*)), method()->constant_encoding(), noreg);
 456 #endif
 457     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
 458   }
 459 
 460   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 461     __ mov(rax, rbx);  // Restore the exception
 462   }
 463 
 464   // remove the activation and dispatch to the unwind handler
 465   __ remove_frame(initial_frame_size_in_bytes());
 466   __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
 467 
 468   // Emit the slow path assembly
 469   if (stub != nullptr) {
 470     stub->emit_code(this);
 471   }
 472 
 473   return offset;
 474 }
 475 
 476 
 477 int LIR_Assembler::emit_deopt_handler() {
 478   // generate code for exception handler
 479   address handler_base = __ start_a_stub(deopt_handler_size());
 480   if (handler_base == nullptr) {
 481     // not enough space left for the handler
 482     bailout("deopt handler overflow");
 483     return -1;
 484   }
 485 
 486   int offset = code_offset();
 487   InternalAddress here(__ pc());
 488 
 489   __ pushptr(here.addr(), rscratch1);
 490   __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
 491   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 492   __ end_a_stub();
 493 
 494   return offset;
 495 }
 496 
 497 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
 498   assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
 499   if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
 500     assert(result->fpu() == 0, "result must already be on TOS");
 501   }
 502 
 503   // Pop the stack before the safepoint code
 504   __ remove_frame(initial_frame_size_in_bytes());
 505 
 506   if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
 507     __ reserved_stack_check();
 508   }
 509 
 510   // Note: we do not need to round double result; float result has the right precision
 511   // the poll sets the condition code, but no data registers
 512 
 513 #ifdef _LP64
 514   const Register thread = r15_thread;
 515 #else
 516   const Register thread = rbx;
 517   __ get_thread(thread);
 518 #endif
 519   code_stub->set_safepoint_offset(__ offset());
 520   __ relocate(relocInfo::poll_return_type);
 521   __ safepoint_poll(*code_stub->entry(), thread, true /* at_return */, true /* in_nmethod */);
 522   __ ret(0);
 523 }
 524 
 525 
 526 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
 527   guarantee(info != nullptr, "Shouldn't be null");
 528   int offset = __ offset();
 529 #ifdef _LP64
 530   const Register poll_addr = rscratch1;
 531   __ movptr(poll_addr, Address(r15_thread, JavaThread::polling_page_offset()));
 532 #else
 533   assert(tmp->is_cpu_register(), "needed");
 534   const Register poll_addr = tmp->as_register();
 535   __ get_thread(poll_addr);
 536   __ movptr(poll_addr, Address(poll_addr, in_bytes(JavaThread::polling_page_offset())));
 537 #endif
 538   add_debug_info_for_branch(info);
 539   __ relocate(relocInfo::poll_type);
 540   address pre_pc = __ pc();
 541   __ testl(rax, Address(poll_addr, 0));
 542   address post_pc = __ pc();
 543   guarantee(pointer_delta(post_pc, pre_pc, 1) == 2 LP64_ONLY(+1), "must be exact length");
 544   return offset;
 545 }
 546 
 547 
 548 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 549   if (from_reg != to_reg) __ mov(to_reg, from_reg);
 550 }
 551 
 552 void LIR_Assembler::swap_reg(Register a, Register b) {
 553   __ xchgptr(a, b);
 554 }
 555 
 556 
 557 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 558   assert(src->is_constant(), "should not call otherwise");
 559   assert(dest->is_register(), "should not call otherwise");
 560   LIR_Const* c = src->as_constant_ptr();
 561 
 562   switch (c->type()) {
 563     case T_INT: {
 564       assert(patch_code == lir_patch_none, "no patching handled here");
 565       __ movl(dest->as_register(), c->as_jint());
 566       break;
 567     }
 568 
 569     case T_ADDRESS: {
 570       assert(patch_code == lir_patch_none, "no patching handled here");
 571       __ movptr(dest->as_register(), c->as_jint());
 572       break;
 573     }
 574 
 575     case T_LONG: {
 576       assert(patch_code == lir_patch_none, "no patching handled here");
 577 #ifdef _LP64
 578       __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
 579 #else
 580       __ movptr(dest->as_register_lo(), c->as_jint_lo());
 581       __ movptr(dest->as_register_hi(), c->as_jint_hi());
 582 #endif // _LP64
 583       break;
 584     }
 585 
 586     case T_OBJECT: {
 587       if (patch_code != lir_patch_none) {
 588         jobject2reg_with_patching(dest->as_register(), info);
 589       } else {
 590         __ movoop(dest->as_register(), c->as_jobject());
 591       }
 592       break;
 593     }
 594 
 595     case T_METADATA: {
 596       if (patch_code != lir_patch_none) {
 597         klass2reg_with_patching(dest->as_register(), info);
 598       } else {
 599         __ mov_metadata(dest->as_register(), c->as_metadata());
 600       }
 601       break;
 602     }
 603 
 604     case T_FLOAT: {
 605       if (dest->is_single_xmm()) {
 606         if (LP64_ONLY(UseAVX <= 2 &&) c->is_zero_float()) {
 607           __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
 608         } else {
 609           __ movflt(dest->as_xmm_float_reg(),
 610                    InternalAddress(float_constant(c->as_jfloat())));
 611         }
 612       } else {
 613 #ifndef _LP64
 614         assert(dest->is_single_fpu(), "must be");
 615         assert(dest->fpu_regnr() == 0, "dest must be TOS");
 616         if (c->is_zero_float()) {
 617           __ fldz();
 618         } else if (c->is_one_float()) {
 619           __ fld1();
 620         } else {
 621           __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
 622         }
 623 #else
 624         ShouldNotReachHere();
 625 #endif // !_LP64
 626       }
 627       break;
 628     }
 629 
 630     case T_DOUBLE: {
 631       if (dest->is_double_xmm()) {
 632         if (LP64_ONLY(UseAVX <= 2 &&) c->is_zero_double()) {
 633           __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
 634         } else {
 635           __ movdbl(dest->as_xmm_double_reg(),
 636                     InternalAddress(double_constant(c->as_jdouble())));
 637         }
 638       } else {
 639 #ifndef _LP64
 640         assert(dest->is_double_fpu(), "must be");
 641         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
 642         if (c->is_zero_double()) {
 643           __ fldz();
 644         } else if (c->is_one_double()) {
 645           __ fld1();
 646         } else {
 647           __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
 648         }
 649 #else
 650         ShouldNotReachHere();
 651 #endif // !_LP64
 652       }
 653       break;
 654     }
 655 
 656     default:
 657       ShouldNotReachHere();
 658   }
 659 }
 660 
 661 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 662   assert(src->is_constant(), "should not call otherwise");
 663   assert(dest->is_stack(), "should not call otherwise");
 664   LIR_Const* c = src->as_constant_ptr();
 665 
 666   switch (c->type()) {
 667     case T_INT:  // fall through
 668     case T_FLOAT:
 669       __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
 670       break;
 671 
 672     case T_ADDRESS:
 673       __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
 674       break;
 675 
 676     case T_OBJECT:
 677       __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject(), rscratch1);
 678       break;
 679 
 680     case T_LONG:  // fall through
 681     case T_DOUBLE:
 682 #ifdef _LP64
 683       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
 684                                               lo_word_offset_in_bytes),
 685                 (intptr_t)c->as_jlong_bits(),
 686                 rscratch1);
 687 #else
 688       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
 689                                               lo_word_offset_in_bytes), c->as_jint_lo_bits());
 690       __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
 691                                               hi_word_offset_in_bytes), c->as_jint_hi_bits());
 692 #endif // _LP64
 693       break;
 694 
 695     default:
 696       ShouldNotReachHere();
 697   }
 698 }
 699 
 700 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 701   assert(src->is_constant(), "should not call otherwise");
 702   assert(dest->is_address(), "should not call otherwise");
 703   LIR_Const* c = src->as_constant_ptr();
 704   LIR_Address* addr = dest->as_address_ptr();
 705 
 706   int null_check_here = code_offset();
 707   switch (type) {
 708     case T_INT:    // fall through
 709     case T_FLOAT:
 710       __ movl(as_Address(addr), c->as_jint_bits());
 711       break;
 712 
 713     case T_ADDRESS:
 714       __ movptr(as_Address(addr), c->as_jint_bits());
 715       break;
 716 
 717     case T_OBJECT:  // fall through
 718     case T_ARRAY:
 719       if (c->as_jobject() == nullptr) {
 720         if (UseCompressedOops && !wide) {
 721           __ movl(as_Address(addr), NULL_WORD);
 722         } else {
 723 #ifdef _LP64
 724           __ xorptr(rscratch1, rscratch1);
 725           null_check_here = code_offset();
 726           __ movptr(as_Address(addr), rscratch1);
 727 #else
 728           __ movptr(as_Address(addr), NULL_WORD);
 729 #endif
 730         }
 731       } else {
 732         if (is_literal_address(addr)) {
 733           ShouldNotReachHere();
 734           __ movoop(as_Address(addr, noreg), c->as_jobject(), rscratch1);
 735         } else {
 736 #ifdef _LP64
 737           __ movoop(rscratch1, c->as_jobject());
 738           if (UseCompressedOops && !wide) {
 739             __ encode_heap_oop(rscratch1);
 740             null_check_here = code_offset();
 741             __ movl(as_Address_lo(addr), rscratch1);
 742           } else {
 743             null_check_here = code_offset();
 744             __ movptr(as_Address_lo(addr), rscratch1);
 745           }
 746 #else
 747           __ movoop(as_Address(addr), c->as_jobject(), noreg);
 748 #endif
 749         }
 750       }
 751       break;
 752 
 753     case T_LONG:    // fall through
 754     case T_DOUBLE:
 755 #ifdef _LP64
 756       if (is_literal_address(addr)) {
 757         ShouldNotReachHere();
 758         __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
 759       } else {
 760         __ movptr(r10, (intptr_t)c->as_jlong_bits());
 761         null_check_here = code_offset();
 762         __ movptr(as_Address_lo(addr), r10);
 763       }
 764 #else
 765       // Always reachable in 32bit so this doesn't produce useless move literal
 766       __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
 767       __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
 768 #endif // _LP64
 769       break;
 770 
 771     case T_BOOLEAN: // fall through
 772     case T_BYTE:
 773       __ movb(as_Address(addr), c->as_jint() & 0xFF);
 774       break;
 775 
 776     case T_CHAR:    // fall through
 777     case T_SHORT:
 778       __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
 779       break;
 780 
 781     default:
 782       ShouldNotReachHere();
 783   };
 784 
 785   if (info != nullptr) {
 786     add_debug_info_for_null_check(null_check_here, info);
 787   }
 788 }
 789 
 790 
 791 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
 792   assert(src->is_register(), "should not call otherwise");
 793   assert(dest->is_register(), "should not call otherwise");
 794 
 795   // move between cpu-registers
 796   if (dest->is_single_cpu()) {
 797 #ifdef _LP64
 798     if (src->type() == T_LONG) {
 799       // Can do LONG -> OBJECT
 800       move_regs(src->as_register_lo(), dest->as_register());
 801       return;
 802     }
 803 #endif
 804     assert(src->is_single_cpu(), "must match");
 805     if (src->type() == T_OBJECT) {
 806       __ verify_oop(src->as_register());
 807     }
 808     move_regs(src->as_register(), dest->as_register());
 809 
 810   } else if (dest->is_double_cpu()) {
 811 #ifdef _LP64
 812     if (is_reference_type(src->type())) {
 813       // Surprising to me but we can see move of a long to t_object
 814       __ verify_oop(src->as_register());
 815       move_regs(src->as_register(), dest->as_register_lo());
 816       return;
 817     }
 818 #endif
 819     assert(src->is_double_cpu(), "must match");
 820     Register f_lo = src->as_register_lo();
 821     Register f_hi = src->as_register_hi();
 822     Register t_lo = dest->as_register_lo();
 823     Register t_hi = dest->as_register_hi();
 824 #ifdef _LP64
 825     assert(f_hi == f_lo, "must be same");
 826     assert(t_hi == t_lo, "must be same");
 827     move_regs(f_lo, t_lo);
 828 #else
 829     assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
 830 
 831 
 832     if (f_lo == t_hi && f_hi == t_lo) {
 833       swap_reg(f_lo, f_hi);
 834     } else if (f_hi == t_lo) {
 835       assert(f_lo != t_hi, "overwriting register");
 836       move_regs(f_hi, t_hi);
 837       move_regs(f_lo, t_lo);
 838     } else {
 839       assert(f_hi != t_lo, "overwriting register");
 840       move_regs(f_lo, t_lo);
 841       move_regs(f_hi, t_hi);
 842     }
 843 #endif // LP64
 844 
 845 #ifndef _LP64
 846     // special moves from fpu-register to xmm-register
 847     // necessary for method results
 848   } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
 849     __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
 850     __ fld_s(Address(rsp, 0));
 851   } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
 852     __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
 853     __ fld_d(Address(rsp, 0));
 854   } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
 855     __ fstp_s(Address(rsp, 0));
 856     __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
 857   } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
 858     __ fstp_d(Address(rsp, 0));
 859     __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
 860 #endif // !_LP64
 861 
 862     // move between xmm-registers
 863   } else if (dest->is_single_xmm()) {
 864     assert(src->is_single_xmm(), "must match");
 865     __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
 866   } else if (dest->is_double_xmm()) {
 867     assert(src->is_double_xmm(), "must match");
 868     __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
 869 
 870 #ifndef _LP64
 871     // move between fpu-registers (no instruction necessary because of fpu-stack)
 872   } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
 873     assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
 874     assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
 875 #endif // !_LP64
 876 
 877   } else {
 878     ShouldNotReachHere();
 879   }
 880 }
 881 
 882 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
 883   assert(src->is_register(), "should not call otherwise");
 884   assert(dest->is_stack(), "should not call otherwise");
 885 
 886   if (src->is_single_cpu()) {
 887     Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
 888     if (is_reference_type(type)) {
 889       __ verify_oop(src->as_register());
 890       __ movptr (dst, src->as_register());
 891     } else if (type == T_METADATA || type == T_ADDRESS) {
 892       __ movptr (dst, src->as_register());
 893     } else {
 894       __ movl (dst, src->as_register());
 895     }
 896 
 897   } else if (src->is_double_cpu()) {
 898     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
 899     Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
 900     __ movptr (dstLO, src->as_register_lo());
 901     NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
 902 
 903   } else if (src->is_single_xmm()) {
 904     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 905     __ movflt(dst_addr, src->as_xmm_float_reg());
 906 
 907   } else if (src->is_double_xmm()) {
 908     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 909     __ movdbl(dst_addr, src->as_xmm_double_reg());
 910 
 911 #ifndef _LP64
 912   } else if (src->is_single_fpu()) {
 913     assert(src->fpu_regnr() == 0, "argument must be on TOS");
 914     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 915     if (pop_fpu_stack)     __ fstp_s (dst_addr);
 916     else                   __ fst_s  (dst_addr);
 917 
 918   } else if (src->is_double_fpu()) {
 919     assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
 920     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 921     if (pop_fpu_stack)     __ fstp_d (dst_addr);
 922     else                   __ fst_d  (dst_addr);
 923 #endif // !_LP64
 924 
 925   } else {
 926     ShouldNotReachHere();
 927   }
 928 }
 929 
 930 
 931 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool wide) {
 932   LIR_Address* to_addr = dest->as_address_ptr();
 933   PatchingStub* patch = nullptr;
 934   Register compressed_src = rscratch1;
 935 
 936   if (is_reference_type(type)) {
 937     __ verify_oop(src->as_register());
 938 #ifdef _LP64
 939     if (UseCompressedOops && !wide) {
 940       __ movptr(compressed_src, src->as_register());
 941       __ encode_heap_oop(compressed_src);
 942       if (patch_code != lir_patch_none) {
 943         info->oop_map()->set_narrowoop(compressed_src->as_VMReg());
 944       }
 945     }
 946 #endif
 947   }
 948 
 949   if (patch_code != lir_patch_none) {
 950     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 951     Address toa = as_Address(to_addr);
 952     assert(toa.disp() != 0, "must have");
 953   }
 954 
 955   int null_check_here = code_offset();
 956   switch (type) {
 957     case T_FLOAT: {
 958 #ifdef _LP64
 959       assert(src->is_single_xmm(), "not a float");
 960       __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
 961 #else
 962       if (src->is_single_xmm()) {
 963         __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
 964       } else {
 965         assert(src->is_single_fpu(), "must be");
 966         assert(src->fpu_regnr() == 0, "argument must be on TOS");
 967         if (pop_fpu_stack)      __ fstp_s(as_Address(to_addr));
 968         else                    __ fst_s (as_Address(to_addr));
 969       }
 970 #endif // _LP64
 971       break;
 972     }
 973 
 974     case T_DOUBLE: {
 975 #ifdef _LP64
 976       assert(src->is_double_xmm(), "not a double");
 977       __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
 978 #else
 979       if (src->is_double_xmm()) {
 980         __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
 981       } else {
 982         assert(src->is_double_fpu(), "must be");
 983         assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
 984         if (pop_fpu_stack)      __ fstp_d(as_Address(to_addr));
 985         else                    __ fst_d (as_Address(to_addr));
 986       }
 987 #endif // _LP64
 988       break;
 989     }
 990 
 991     case T_ARRAY:   // fall through
 992     case T_OBJECT:  // fall through
 993       if (UseCompressedOops && !wide) {
 994         __ movl(as_Address(to_addr), compressed_src);
 995       } else {
 996         __ movptr(as_Address(to_addr), src->as_register());
 997       }
 998       break;
 999     case T_METADATA:
1000       // We get here to store a method pointer to the stack to pass to
1001       // a dtrace runtime call. This can't work on 64 bit with
1002       // compressed klass ptrs: T_METADATA can be a compressed klass
1003       // ptr or a 64 bit method pointer.
1004       LP64_ONLY(ShouldNotReachHere());
1005       __ movptr(as_Address(to_addr), src->as_register());
1006       break;
1007     case T_ADDRESS:
1008       __ movptr(as_Address(to_addr), src->as_register());
1009       break;
1010     case T_INT:
1011       __ movl(as_Address(to_addr), src->as_register());
1012       break;
1013 
1014     case T_LONG: {
1015       Register from_lo = src->as_register_lo();
1016       Register from_hi = src->as_register_hi();
1017 #ifdef _LP64
1018       __ movptr(as_Address_lo(to_addr), from_lo);
1019 #else
1020       Register base = to_addr->base()->as_register();
1021       Register index = noreg;
1022       if (to_addr->index()->is_register()) {
1023         index = to_addr->index()->as_register();
1024       }
1025       if (base == from_lo || index == from_lo) {
1026         assert(base != from_hi, "can't be");
1027         assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1028         __ movl(as_Address_hi(to_addr), from_hi);
1029         if (patch != nullptr) {
1030           patching_epilog(patch, lir_patch_high, base, info);
1031           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1032           patch_code = lir_patch_low;
1033         }
1034         __ movl(as_Address_lo(to_addr), from_lo);
1035       } else {
1036         assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1037         __ movl(as_Address_lo(to_addr), from_lo);
1038         if (patch != nullptr) {
1039           patching_epilog(patch, lir_patch_low, base, info);
1040           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1041           patch_code = lir_patch_high;
1042         }
1043         __ movl(as_Address_hi(to_addr), from_hi);
1044       }
1045 #endif // _LP64
1046       break;
1047     }
1048 
1049     case T_BYTE:    // fall through
1050     case T_BOOLEAN: {
1051       Register src_reg = src->as_register();
1052       Address dst_addr = as_Address(to_addr);
1053       assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1054       __ movb(dst_addr, src_reg);
1055       break;
1056     }
1057 
1058     case T_CHAR:    // fall through
1059     case T_SHORT:
1060       __ movw(as_Address(to_addr), src->as_register());
1061       break;
1062 
1063     default:
1064       ShouldNotReachHere();
1065   }
1066   if (info != nullptr) {
1067     add_debug_info_for_null_check(null_check_here, info);
1068   }
1069 
1070   if (patch_code != lir_patch_none) {
1071     patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1072   }
1073 }
1074 
1075 
1076 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1077   assert(src->is_stack(), "should not call otherwise");
1078   assert(dest->is_register(), "should not call otherwise");
1079 
1080   if (dest->is_single_cpu()) {
1081     if (is_reference_type(type)) {
1082       __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1083       __ verify_oop(dest->as_register());
1084     } else if (type == T_METADATA || type == T_ADDRESS) {
1085       __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1086     } else {
1087       __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1088     }
1089 
1090   } else if (dest->is_double_cpu()) {
1091     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1092     Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1093     __ movptr(dest->as_register_lo(), src_addr_LO);
1094     NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1095 
1096   } else if (dest->is_single_xmm()) {
1097     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1098     __ movflt(dest->as_xmm_float_reg(), src_addr);
1099 
1100   } else if (dest->is_double_xmm()) {
1101     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1102     __ movdbl(dest->as_xmm_double_reg(), src_addr);
1103 
1104 #ifndef _LP64
1105   } else if (dest->is_single_fpu()) {
1106     assert(dest->fpu_regnr() == 0, "dest must be TOS");
1107     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1108     __ fld_s(src_addr);
1109 
1110   } else if (dest->is_double_fpu()) {
1111     assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1112     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1113     __ fld_d(src_addr);
1114 #endif // _LP64
1115 
1116   } else {
1117     ShouldNotReachHere();
1118   }
1119 }
1120 
1121 
1122 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1123   if (src->is_single_stack()) {
1124     if (is_reference_type(type)) {
1125       __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1126       __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1127     } else {
1128 #ifndef _LP64
1129       __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1130       __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1131 #else
1132       //no pushl on 64bits
1133       __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1134       __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1135 #endif
1136     }
1137 
1138   } else if (src->is_double_stack()) {
1139 #ifdef _LP64
1140     __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1141     __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1142 #else
1143     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1144     // push and pop the part at src + wordSize, adding wordSize for the previous push
1145     __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1146     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1147     __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1148 #endif // _LP64
1149 
1150   } else {
1151     ShouldNotReachHere();
1152   }
1153 }
1154 
1155 
1156 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide) {
1157   assert(src->is_address(), "should not call otherwise");
1158   assert(dest->is_register(), "should not call otherwise");
1159 
1160   LIR_Address* addr = src->as_address_ptr();
1161   Address from_addr = as_Address(addr);
1162 
1163   if (addr->base()->type() == T_OBJECT) {
1164     __ verify_oop(addr->base()->as_pointer_register());
1165   }
1166 
1167   switch (type) {
1168     case T_BOOLEAN: // fall through
1169     case T_BYTE:    // fall through
1170     case T_CHAR:    // fall through
1171     case T_SHORT:
1172       if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1173         // on pre P6 processors we may get partial register stalls
1174         // so blow away the value of to_rinfo before loading a
1175         // partial word into it.  Do it here so that it precedes
1176         // the potential patch point below.
1177         __ xorptr(dest->as_register(), dest->as_register());
1178       }
1179       break;
1180    default:
1181      break;
1182   }
1183 
1184   PatchingStub* patch = nullptr;
1185   if (patch_code != lir_patch_none) {
1186     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1187     assert(from_addr.disp() != 0, "must have");
1188   }
1189   if (info != nullptr) {
1190     add_debug_info_for_null_check_here(info);
1191   }
1192 
1193   switch (type) {
1194     case T_FLOAT: {
1195       if (dest->is_single_xmm()) {
1196         __ movflt(dest->as_xmm_float_reg(), from_addr);
1197       } else {
1198 #ifndef _LP64
1199         assert(dest->is_single_fpu(), "must be");
1200         assert(dest->fpu_regnr() == 0, "dest must be TOS");
1201         __ fld_s(from_addr);
1202 #else
1203         ShouldNotReachHere();
1204 #endif // !LP64
1205       }
1206       break;
1207     }
1208 
1209     case T_DOUBLE: {
1210       if (dest->is_double_xmm()) {
1211         __ movdbl(dest->as_xmm_double_reg(), from_addr);
1212       } else {
1213 #ifndef _LP64
1214         assert(dest->is_double_fpu(), "must be");
1215         assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1216         __ fld_d(from_addr);
1217 #else
1218         ShouldNotReachHere();
1219 #endif // !LP64
1220       }
1221       break;
1222     }
1223 
1224     case T_OBJECT:  // fall through
1225     case T_ARRAY:   // fall through
1226       if (UseCompressedOops && !wide) {
1227         __ movl(dest->as_register(), from_addr);
1228       } else {
1229         __ movptr(dest->as_register(), from_addr);
1230       }
1231       break;
1232 
1233     case T_ADDRESS:
1234       __ movptr(dest->as_register(), from_addr);
1235       break;
1236     case T_INT:
1237       __ movl(dest->as_register(), from_addr);
1238       break;
1239 
1240     case T_LONG: {
1241       Register to_lo = dest->as_register_lo();
1242       Register to_hi = dest->as_register_hi();
1243 #ifdef _LP64
1244       __ movptr(to_lo, as_Address_lo(addr));
1245 #else
1246       Register base = addr->base()->as_register();
1247       Register index = noreg;
1248       if (addr->index()->is_register()) {
1249         index = addr->index()->as_register();
1250       }
1251       if ((base == to_lo && index == to_hi) ||
1252           (base == to_hi && index == to_lo)) {
1253         // addresses with 2 registers are only formed as a result of
1254         // array access so this code will never have to deal with
1255         // patches or null checks.
1256         assert(info == nullptr && patch == nullptr, "must be");
1257         __ lea(to_hi, as_Address(addr));
1258         __ movl(to_lo, Address(to_hi, 0));
1259         __ movl(to_hi, Address(to_hi, BytesPerWord));
1260       } else if (base == to_lo || index == to_lo) {
1261         assert(base != to_hi, "can't be");
1262         assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1263         __ movl(to_hi, as_Address_hi(addr));
1264         if (patch != nullptr) {
1265           patching_epilog(patch, lir_patch_high, base, info);
1266           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1267           patch_code = lir_patch_low;
1268         }
1269         __ movl(to_lo, as_Address_lo(addr));
1270       } else {
1271         assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1272         __ movl(to_lo, as_Address_lo(addr));
1273         if (patch != nullptr) {
1274           patching_epilog(patch, lir_patch_low, base, info);
1275           patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1276           patch_code = lir_patch_high;
1277         }
1278         __ movl(to_hi, as_Address_hi(addr));
1279       }
1280 #endif // _LP64
1281       break;
1282     }
1283 
1284     case T_BOOLEAN: // fall through
1285     case T_BYTE: {
1286       Register dest_reg = dest->as_register();
1287       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1288       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1289         __ movsbl(dest_reg, from_addr);
1290       } else {
1291         __ movb(dest_reg, from_addr);
1292         __ shll(dest_reg, 24);
1293         __ sarl(dest_reg, 24);
1294       }
1295       break;
1296     }
1297 
1298     case T_CHAR: {
1299       Register dest_reg = dest->as_register();
1300       assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1301       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1302         __ movzwl(dest_reg, from_addr);
1303       } else {
1304         __ movw(dest_reg, from_addr);
1305       }
1306       break;
1307     }
1308 
1309     case T_SHORT: {
1310       Register dest_reg = dest->as_register();
1311       if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1312         __ movswl(dest_reg, from_addr);
1313       } else {
1314         __ movw(dest_reg, from_addr);
1315         __ shll(dest_reg, 16);
1316         __ sarl(dest_reg, 16);
1317       }
1318       break;
1319     }
1320 
1321     default:
1322       ShouldNotReachHere();
1323   }
1324 
1325   if (patch != nullptr) {
1326     patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1327   }
1328 
1329   if (is_reference_type(type)) {
1330 #ifdef _LP64
1331     if (UseCompressedOops && !wide) {
1332       __ decode_heap_oop(dest->as_register());
1333     }
1334 #endif
1335 
1336     if (!(UseZGC && !ZGenerational)) {
1337       // Load barrier has not yet been applied, so ZGC can't verify the oop here
1338       __ verify_oop(dest->as_register());
1339     }
1340   }
1341 }
1342 
1343 
1344 NEEDS_CLEANUP; // This could be static?
1345 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1346   int elem_size = type2aelembytes(type);
1347   switch (elem_size) {
1348     case 1: return Address::times_1;
1349     case 2: return Address::times_2;
1350     case 4: return Address::times_4;
1351     case 8: return Address::times_8;
1352   }
1353   ShouldNotReachHere();
1354   return Address::no_scale;
1355 }
1356 
1357 
1358 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1359   switch (op->code()) {
1360     case lir_idiv:
1361     case lir_irem:
1362       arithmetic_idiv(op->code(),
1363                       op->in_opr1(),
1364                       op->in_opr2(),
1365                       op->in_opr3(),
1366                       op->result_opr(),
1367                       op->info());
1368       break;
1369     case lir_fmad:
1370       __ fmad(op->result_opr()->as_xmm_double_reg(),
1371               op->in_opr1()->as_xmm_double_reg(),
1372               op->in_opr2()->as_xmm_double_reg(),
1373               op->in_opr3()->as_xmm_double_reg());
1374       break;
1375     case lir_fmaf:
1376       __ fmaf(op->result_opr()->as_xmm_float_reg(),
1377               op->in_opr1()->as_xmm_float_reg(),
1378               op->in_opr2()->as_xmm_float_reg(),
1379               op->in_opr3()->as_xmm_float_reg());
1380       break;
1381     default:      ShouldNotReachHere(); break;
1382   }
1383 }
1384 
1385 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1386 #ifdef ASSERT
1387   assert(op->block() == nullptr || op->block()->label() == op->label(), "wrong label");
1388   if (op->block() != nullptr)  _branch_target_blocks.append(op->block());
1389   if (op->ublock() != nullptr) _branch_target_blocks.append(op->ublock());
1390 #endif
1391 
1392   if (op->cond() == lir_cond_always) {
1393     if (op->info() != nullptr) add_debug_info_for_branch(op->info());
1394     __ jmp (*(op->label()));
1395   } else {
1396     Assembler::Condition acond = Assembler::zero;
1397     if (op->code() == lir_cond_float_branch) {
1398       assert(op->ublock() != nullptr, "must have unordered successor");
1399       __ jcc(Assembler::parity, *(op->ublock()->label()));
1400       switch(op->cond()) {
1401         case lir_cond_equal:        acond = Assembler::equal;      break;
1402         case lir_cond_notEqual:     acond = Assembler::notEqual;   break;
1403         case lir_cond_less:         acond = Assembler::below;      break;
1404         case lir_cond_lessEqual:    acond = Assembler::belowEqual; break;
1405         case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1406         case lir_cond_greater:      acond = Assembler::above;      break;
1407         default:                         ShouldNotReachHere();
1408       }
1409     } else {
1410       switch (op->cond()) {
1411         case lir_cond_equal:        acond = Assembler::equal;       break;
1412         case lir_cond_notEqual:     acond = Assembler::notEqual;    break;
1413         case lir_cond_less:         acond = Assembler::less;        break;
1414         case lir_cond_lessEqual:    acond = Assembler::lessEqual;   break;
1415         case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1416         case lir_cond_greater:      acond = Assembler::greater;     break;
1417         case lir_cond_belowEqual:   acond = Assembler::belowEqual;  break;
1418         case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;  break;
1419         default:                         ShouldNotReachHere();
1420       }
1421     }
1422     __ jcc(acond,*(op->label()));
1423   }
1424 }
1425 
1426 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1427   LIR_Opr src  = op->in_opr();
1428   LIR_Opr dest = op->result_opr();
1429 
1430   switch (op->bytecode()) {
1431     case Bytecodes::_i2l:
1432 #ifdef _LP64
1433       __ movl2ptr(dest->as_register_lo(), src->as_register());
1434 #else
1435       move_regs(src->as_register(), dest->as_register_lo());
1436       move_regs(src->as_register(), dest->as_register_hi());
1437       __ sarl(dest->as_register_hi(), 31);
1438 #endif // LP64
1439       break;
1440 
1441     case Bytecodes::_l2i:
1442 #ifdef _LP64
1443       __ movl(dest->as_register(), src->as_register_lo());
1444 #else
1445       move_regs(src->as_register_lo(), dest->as_register());
1446 #endif
1447       break;
1448 
1449     case Bytecodes::_i2b:
1450       move_regs(src->as_register(), dest->as_register());
1451       __ sign_extend_byte(dest->as_register());
1452       break;
1453 
1454     case Bytecodes::_i2c:
1455       move_regs(src->as_register(), dest->as_register());
1456       __ andl(dest->as_register(), 0xFFFF);
1457       break;
1458 
1459     case Bytecodes::_i2s:
1460       move_regs(src->as_register(), dest->as_register());
1461       __ sign_extend_short(dest->as_register());
1462       break;
1463 
1464 
1465 #ifdef _LP64
1466     case Bytecodes::_f2d:
1467       __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1468       break;
1469 
1470     case Bytecodes::_d2f:
1471       __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1472       break;
1473 
1474     case Bytecodes::_i2f:
1475       __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1476       break;
1477 
1478     case Bytecodes::_i2d:
1479       __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1480       break;
1481 
1482     case Bytecodes::_l2f:
1483       __ cvtsi2ssq(dest->as_xmm_float_reg(), src->as_register_lo());
1484       break;
1485 
1486     case Bytecodes::_l2d:
1487       __ cvtsi2sdq(dest->as_xmm_double_reg(), src->as_register_lo());
1488       break;
1489 
1490     case Bytecodes::_f2i:
1491       __ convert_f2i(dest->as_register(), src->as_xmm_float_reg());
1492       break;
1493 
1494     case Bytecodes::_d2i:
1495       __ convert_d2i(dest->as_register(), src->as_xmm_double_reg());
1496       break;
1497 
1498     case Bytecodes::_f2l:
1499       __ convert_f2l(dest->as_register_lo(), src->as_xmm_float_reg());
1500       break;
1501 
1502     case Bytecodes::_d2l:
1503       __ convert_d2l(dest->as_register_lo(), src->as_xmm_double_reg());
1504       break;
1505 #else
1506     case Bytecodes::_f2d:
1507     case Bytecodes::_d2f:
1508       if (dest->is_single_xmm()) {
1509         __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1510       } else if (dest->is_double_xmm()) {
1511         __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1512       } else {
1513         assert(src->fpu() == dest->fpu(), "register must be equal");
1514         // do nothing (float result is rounded later through spilling)
1515       }
1516       break;
1517 
1518     case Bytecodes::_i2f:
1519     case Bytecodes::_i2d:
1520       if (dest->is_single_xmm()) {
1521         __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1522       } else if (dest->is_double_xmm()) {
1523         __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1524       } else {
1525         assert(dest->fpu() == 0, "result must be on TOS");
1526         __ movl(Address(rsp, 0), src->as_register());
1527         __ fild_s(Address(rsp, 0));
1528       }
1529       break;
1530 
1531     case Bytecodes::_l2f:
1532     case Bytecodes::_l2d:
1533       assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1534       assert(dest->fpu() == 0, "result must be on TOS");
1535       __ movptr(Address(rsp, 0),          src->as_register_lo());
1536       __ movl(Address(rsp, BytesPerWord), src->as_register_hi());
1537       __ fild_d(Address(rsp, 0));
1538       // float result is rounded later through spilling
1539       break;
1540 
1541     case Bytecodes::_f2i:
1542     case Bytecodes::_d2i:
1543       if (src->is_single_xmm()) {
1544         __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1545       } else if (src->is_double_xmm()) {
1546         __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1547       } else {
1548         assert(src->fpu() == 0, "input must be on TOS");
1549         __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_trunc()));
1550         __ fist_s(Address(rsp, 0));
1551         __ movl(dest->as_register(), Address(rsp, 0));
1552         __ fldcw(ExternalAddress(StubRoutines::x86::addr_fpu_cntrl_wrd_std()));
1553       }
1554       // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1555       assert(op->stub() != nullptr, "stub required");
1556       __ cmpl(dest->as_register(), 0x80000000);
1557       __ jcc(Assembler::equal, *op->stub()->entry());
1558       __ bind(*op->stub()->continuation());
1559       break;
1560 
1561     case Bytecodes::_f2l:
1562     case Bytecodes::_d2l:
1563       assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1564       assert(src->fpu() == 0, "input must be on TOS");
1565       assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1566 
1567       // instruction sequence too long to inline it here
1568       {
1569         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1570       }
1571       break;
1572 #endif // _LP64
1573 
1574     default: ShouldNotReachHere();
1575   }
1576 }
1577 
1578 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1579   if (op->init_check()) {
1580     add_debug_info_for_null_check_here(op->stub()->info());
1581     __ cmpb(Address(op->klass()->as_register(),
1582                     InstanceKlass::init_state_offset()),
1583                     InstanceKlass::fully_initialized);
1584     __ jcc(Assembler::notEqual, *op->stub()->entry());
1585   }
1586   __ allocate_object(op->obj()->as_register(),
1587                      op->tmp1()->as_register(),
1588                      op->tmp2()->as_register(),
1589                      op->header_size(),
1590                      op->object_size(),
1591                      op->klass()->as_register(),
1592                      *op->stub()->entry());
1593   __ bind(*op->stub()->continuation());
1594 }
1595 
1596 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1597   Register len =  op->len()->as_register();
1598   LP64_ONLY( __ movslq(len, len); )
1599 
1600   if (UseSlowPath ||
1601       (!UseFastNewObjectArray && is_reference_type(op->type())) ||
1602       (!UseFastNewTypeArray   && !is_reference_type(op->type()))) {
1603     __ jmp(*op->stub()->entry());
1604   } else {
1605     Register tmp1 = op->tmp1()->as_register();
1606     Register tmp2 = op->tmp2()->as_register();
1607     Register tmp3 = op->tmp3()->as_register();
1608     if (len == tmp1) {
1609       tmp1 = tmp3;
1610     } else if (len == tmp2) {
1611       tmp2 = tmp3;
1612     } else if (len == tmp3) {
1613       // everything is ok
1614     } else {
1615       __ mov(tmp3, len);
1616     }
1617     __ allocate_array(op->obj()->as_register(),
1618                       len,
1619                       tmp1,
1620                       tmp2,
1621                       arrayOopDesc::base_offset_in_bytes(op->type()),
1622                       array_element_size(op->type()),
1623                       op->klass()->as_register(),
1624                       *op->stub()->entry(),
1625                       op->zero_array());
1626   }
1627   __ bind(*op->stub()->continuation());
1628 }
1629 
1630 void LIR_Assembler::type_profile_helper(Register mdo,
1631                                         ciMethodData *md, ciProfileData *data,
1632                                         Register recv, Label* update_done) {
1633   for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1634     Label next_test;
1635     // See if the receiver is receiver[n].
1636     __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1637     __ jccb(Assembler::notEqual, next_test);
1638     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1639     __ addptr(data_addr, DataLayout::counter_increment);
1640     __ jmp(*update_done);
1641     __ bind(next_test);
1642   }
1643 
1644   // Didn't find receiver; find next empty slot and fill it in
1645   for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1646     Label next_test;
1647     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1648     __ cmpptr(recv_addr, NULL_WORD);
1649     __ jccb(Assembler::notEqual, next_test);
1650     __ movptr(recv_addr, recv);
1651     __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1652     __ jmp(*update_done);
1653     __ bind(next_test);
1654   }
1655 }
1656 
1657 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1658   // we always need a stub for the failure case.
1659   CodeStub* stub = op->stub();
1660   Register obj = op->object()->as_register();
1661   Register k_RInfo = op->tmp1()->as_register();
1662   Register klass_RInfo = op->tmp2()->as_register();
1663   Register dst = op->result_opr()->as_register();
1664   ciKlass* k = op->klass();
1665   Register Rtmp1 = noreg;
1666   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1667 
1668   // check if it needs to be profiled
1669   ciMethodData* md = nullptr;
1670   ciProfileData* data = nullptr;
1671 
1672   if (op->should_profile()) {
1673     ciMethod* method = op->profiled_method();
1674     assert(method != nullptr, "Should have method");
1675     int bci = op->profiled_bci();
1676     md = method->method_data_or_null();
1677     assert(md != nullptr, "Sanity");
1678     data = md->bci_to_data(bci);
1679     assert(data != nullptr,                "need data for type check");
1680     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1681   }
1682   Label* success_target = success;
1683   Label* failure_target = failure;
1684 
1685   if (obj == k_RInfo) {
1686     k_RInfo = dst;
1687   } else if (obj == klass_RInfo) {
1688     klass_RInfo = dst;
1689   }
1690   if (k->is_loaded() && !UseCompressedClassPointers) {
1691     select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1692   } else {
1693     Rtmp1 = op->tmp3()->as_register();
1694     select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1695   }
1696 
1697   assert_different_registers(obj, k_RInfo, klass_RInfo);
1698 
1699   __ testptr(obj, obj);
1700   if (op->should_profile()) {
1701     Label not_null;
1702     Register mdo  = klass_RInfo;
1703     __ mov_metadata(mdo, md->constant_encoding());
1704     __ jccb(Assembler::notEqual, not_null);
1705     // Object is null; update MDO and exit
1706     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()));
1707     int header_bits = BitData::null_seen_byte_constant();
1708     __ orb(data_addr, header_bits);
1709     __ jmp(*obj_is_null);
1710     __ bind(not_null);
1711 
1712     Label update_done;
1713     Register recv = k_RInfo;
1714     __ load_klass(recv, obj, tmp_load_klass);
1715     type_profile_helper(mdo, md, data, recv, &update_done);
1716 
1717     Address nonprofiled_receiver_count_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1718     __ addptr(nonprofiled_receiver_count_addr, DataLayout::counter_increment);
1719 
1720     __ bind(update_done);
1721   } else {
1722     __ jcc(Assembler::equal, *obj_is_null);
1723   }
1724 
1725   if (!k->is_loaded()) {
1726     klass2reg_with_patching(k_RInfo, op->info_for_patch());
1727   } else {
1728 #ifdef _LP64
1729     __ mov_metadata(k_RInfo, k->constant_encoding());
1730 #endif // _LP64
1731   }
1732   __ verify_oop(obj);
1733 
1734   if (op->fast_check()) {
1735     // get object class
1736     // not a safepoint as obj null check happens earlier
1737 #ifdef _LP64
1738     if (UseCompressedClassPointers) {
1739       __ load_klass(Rtmp1, obj, tmp_load_klass);
1740       __ cmpptr(k_RInfo, Rtmp1);
1741     } else {
1742       __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1743     }
1744 #else
1745     if (k->is_loaded()) {
1746       __ cmpklass(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1747     } else {
1748       __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1749     }
1750 #endif
1751     __ jcc(Assembler::notEqual, *failure_target);
1752     // successful cast, fall through to profile or jump
1753   } else {
1754     // get object class
1755     // not a safepoint as obj null check happens earlier
1756     __ load_klass(klass_RInfo, obj, tmp_load_klass);
1757     if (k->is_loaded()) {
1758       // See if we get an immediate positive hit
1759 #ifdef _LP64
1760       __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1761 #else
1762       __ cmpklass(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1763 #endif // _LP64
1764       if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
1765         __ jcc(Assembler::notEqual, *failure_target);
1766         // successful cast, fall through to profile or jump
1767       } else {
1768         // See if we get an immediate positive hit
1769         __ jcc(Assembler::equal, *success_target);
1770         // check for self
1771 #ifdef _LP64
1772         __ cmpptr(klass_RInfo, k_RInfo);
1773 #else
1774         __ cmpklass(klass_RInfo, k->constant_encoding());
1775 #endif // _LP64
1776         __ jcc(Assembler::equal, *success_target);
1777 
1778         __ push(klass_RInfo);
1779 #ifdef _LP64
1780         __ push(k_RInfo);
1781 #else
1782         __ pushklass(k->constant_encoding(), noreg);
1783 #endif // _LP64
1784         __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1785         __ pop(klass_RInfo);
1786         __ pop(klass_RInfo);
1787         // result is a boolean
1788         __ testl(klass_RInfo, klass_RInfo);
1789         __ jcc(Assembler::equal, *failure_target);
1790         // successful cast, fall through to profile or jump
1791       }
1792     } else {
1793       // perform the fast part of the checking logic
1794       __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, nullptr);
1795       // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1796       __ push(klass_RInfo);
1797       __ push(k_RInfo);
1798       __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1799       __ pop(klass_RInfo);
1800       __ pop(k_RInfo);
1801       // result is a boolean
1802       __ testl(k_RInfo, k_RInfo);
1803       __ jcc(Assembler::equal, *failure_target);
1804       // successful cast, fall through to profile or jump
1805     }
1806   }
1807   __ jmp(*success);
1808 }
1809 
1810 
1811 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1812   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1813   LIR_Code code = op->code();
1814   if (code == lir_store_check) {
1815     Register value = op->object()->as_register();
1816     Register array = op->array()->as_register();
1817     Register k_RInfo = op->tmp1()->as_register();
1818     Register klass_RInfo = op->tmp2()->as_register();
1819     Register Rtmp1 = op->tmp3()->as_register();
1820 
1821     CodeStub* stub = op->stub();
1822 
1823     // check if it needs to be profiled
1824     ciMethodData* md = nullptr;
1825     ciProfileData* data = nullptr;
1826 
1827     if (op->should_profile()) {
1828       ciMethod* method = op->profiled_method();
1829       assert(method != nullptr, "Should have method");
1830       int bci = op->profiled_bci();
1831       md = method->method_data_or_null();
1832       assert(md != nullptr, "Sanity");
1833       data = md->bci_to_data(bci);
1834       assert(data != nullptr,                "need data for type check");
1835       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1836     }
1837     Label done;
1838     Label* success_target = &done;
1839     Label* failure_target = stub->entry();
1840 
1841     __ testptr(value, value);
1842     if (op->should_profile()) {
1843       Label not_null;
1844       Register mdo  = klass_RInfo;
1845       __ mov_metadata(mdo, md->constant_encoding());
1846       __ jccb(Assembler::notEqual, not_null);
1847       // Object is null; update MDO and exit
1848       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::flags_offset()));
1849       int header_bits = BitData::null_seen_byte_constant();
1850       __ orb(data_addr, header_bits);
1851       __ jmp(done);
1852       __ bind(not_null);
1853 
1854       Label update_done;
1855       Register recv = k_RInfo;
1856       __ load_klass(recv, value, tmp_load_klass);
1857       type_profile_helper(mdo, md, data, recv, &update_done);
1858 
1859       Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1860       __ addptr(counter_addr, DataLayout::counter_increment);
1861       __ bind(update_done);
1862     } else {
1863       __ jcc(Assembler::equal, done);
1864     }
1865 
1866     add_debug_info_for_null_check_here(op->info_for_exception());
1867     __ load_klass(k_RInfo, array, tmp_load_klass);
1868     __ load_klass(klass_RInfo, value, tmp_load_klass);
1869 
1870     // get instance klass (it's already uncompressed)
1871     __ movptr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
1872     // perform the fast part of the checking logic
1873     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, nullptr);
1874     // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1875     __ push(klass_RInfo);
1876     __ push(k_RInfo);
1877     __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1878     __ pop(klass_RInfo);
1879     __ pop(k_RInfo);
1880     // result is a boolean
1881     __ testl(k_RInfo, k_RInfo);
1882     __ jcc(Assembler::equal, *failure_target);
1883     // fall through to the success case
1884 
1885     __ bind(done);
1886   } else
1887     if (code == lir_checkcast) {
1888       Register obj = op->object()->as_register();
1889       Register dst = op->result_opr()->as_register();
1890       Label success;
1891       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1892       __ bind(success);
1893       if (dst != obj) {
1894         __ mov(dst, obj);
1895       }
1896     } else
1897       if (code == lir_instanceof) {
1898         Register obj = op->object()->as_register();
1899         Register dst = op->result_opr()->as_register();
1900         Label success, failure, done;
1901         emit_typecheck_helper(op, &success, &failure, &failure);
1902         __ bind(failure);
1903         __ xorptr(dst, dst);
1904         __ jmpb(done);
1905         __ bind(success);
1906         __ movptr(dst, 1);
1907         __ bind(done);
1908       } else {
1909         ShouldNotReachHere();
1910       }
1911 
1912 }
1913 
1914 
1915 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1916   if (LP64_ONLY(false &&) op->code() == lir_cas_long) {
1917     assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1918     assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1919     assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1920     assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1921     Register addr = op->addr()->as_register();
1922     __ lock();
1923     NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1924 
1925   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1926     NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1927     Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1928     Register newval = op->new_value()->as_register();
1929     Register cmpval = op->cmp_value()->as_register();
1930     assert(cmpval == rax, "wrong register");
1931     assert(newval != noreg, "new val must be register");
1932     assert(cmpval != newval, "cmp and new values must be in different registers");
1933     assert(cmpval != addr, "cmp and addr must be in different registers");
1934     assert(newval != addr, "new value and addr must be in different registers");
1935 
1936     if ( op->code() == lir_cas_obj) {
1937 #ifdef _LP64
1938       if (UseCompressedOops) {
1939         __ encode_heap_oop(cmpval);
1940         __ mov(rscratch1, newval);
1941         __ encode_heap_oop(rscratch1);
1942         __ lock();
1943         // cmpval (rax) is implicitly used by this instruction
1944         __ cmpxchgl(rscratch1, Address(addr, 0));
1945       } else
1946 #endif
1947       {
1948         __ lock();
1949         __ cmpxchgptr(newval, Address(addr, 0));
1950       }
1951     } else {
1952       assert(op->code() == lir_cas_int, "lir_cas_int expected");
1953       __ lock();
1954       __ cmpxchgl(newval, Address(addr, 0));
1955     }
1956 #ifdef _LP64
1957   } else if (op->code() == lir_cas_long) {
1958     Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1959     Register newval = op->new_value()->as_register_lo();
1960     Register cmpval = op->cmp_value()->as_register_lo();
1961     assert(cmpval == rax, "wrong register");
1962     assert(newval != noreg, "new val must be register");
1963     assert(cmpval != newval, "cmp and new values must be in different registers");
1964     assert(cmpval != addr, "cmp and addr must be in different registers");
1965     assert(newval != addr, "new value and addr must be in different registers");
1966     __ lock();
1967     __ cmpxchgq(newval, Address(addr, 0));
1968 #endif // _LP64
1969   } else {
1970     Unimplemented();
1971   }
1972 }
1973 
1974 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1975                           LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1976   assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on x86");
1977 
1978   Assembler::Condition acond, ncond;
1979   switch (condition) {
1980     case lir_cond_equal:        acond = Assembler::equal;        ncond = Assembler::notEqual;     break;
1981     case lir_cond_notEqual:     acond = Assembler::notEqual;     ncond = Assembler::equal;        break;
1982     case lir_cond_less:         acond = Assembler::less;         ncond = Assembler::greaterEqual; break;
1983     case lir_cond_lessEqual:    acond = Assembler::lessEqual;    ncond = Assembler::greater;      break;
1984     case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less;         break;
1985     case lir_cond_greater:      acond = Assembler::greater;      ncond = Assembler::lessEqual;    break;
1986     case lir_cond_belowEqual:   acond = Assembler::belowEqual;   ncond = Assembler::above;        break;
1987     case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;   ncond = Assembler::below;        break;
1988     default:                    acond = Assembler::equal;        ncond = Assembler::notEqual;
1989                                 ShouldNotReachHere();
1990   }
1991 
1992   if (opr1->is_cpu_register()) {
1993     reg2reg(opr1, result);
1994   } else if (opr1->is_stack()) {
1995     stack2reg(opr1, result, result->type());
1996   } else if (opr1->is_constant()) {
1997     const2reg(opr1, result, lir_patch_none, nullptr);
1998   } else {
1999     ShouldNotReachHere();
2000   }
2001 
2002   if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2003     // optimized version that does not require a branch
2004     if (opr2->is_single_cpu()) {
2005       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2006       __ cmov(ncond, result->as_register(), opr2->as_register());
2007     } else if (opr2->is_double_cpu()) {
2008       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2009       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2010       __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2011       NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2012     } else if (opr2->is_single_stack()) {
2013       __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2014     } else if (opr2->is_double_stack()) {
2015       __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2016       NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2017     } else {
2018       ShouldNotReachHere();
2019     }
2020 
2021   } else {
2022     Label skip;
2023     __ jccb(acond, skip);
2024     if (opr2->is_cpu_register()) {
2025       reg2reg(opr2, result);
2026     } else if (opr2->is_stack()) {
2027       stack2reg(opr2, result, result->type());
2028     } else if (opr2->is_constant()) {
2029       const2reg(opr2, result, lir_patch_none, nullptr);
2030     } else {
2031       ShouldNotReachHere();
2032     }
2033     __ bind(skip);
2034   }
2035 }
2036 
2037 
2038 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2039   assert(info == nullptr, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2040 
2041   if (left->is_single_cpu()) {
2042     assert(left == dest, "left and dest must be equal");
2043     Register lreg = left->as_register();
2044 
2045     if (right->is_single_cpu()) {
2046       // cpu register - cpu register
2047       Register rreg = right->as_register();
2048       switch (code) {
2049         case lir_add: __ addl (lreg, rreg); break;
2050         case lir_sub: __ subl (lreg, rreg); break;
2051         case lir_mul: __ imull(lreg, rreg); break;
2052         default:      ShouldNotReachHere();
2053       }
2054 
2055     } else if (right->is_stack()) {
2056       // cpu register - stack
2057       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2058       switch (code) {
2059         case lir_add: __ addl(lreg, raddr); break;
2060         case lir_sub: __ subl(lreg, raddr); break;
2061         default:      ShouldNotReachHere();
2062       }
2063 
2064     } else if (right->is_constant()) {
2065       // cpu register - constant
2066       jint c = right->as_constant_ptr()->as_jint();
2067       switch (code) {
2068         case lir_add: {
2069           __ incrementl(lreg, c);
2070           break;
2071         }
2072         case lir_sub: {
2073           __ decrementl(lreg, c);
2074           break;
2075         }
2076         default: ShouldNotReachHere();
2077       }
2078 
2079     } else {
2080       ShouldNotReachHere();
2081     }
2082 
2083   } else if (left->is_double_cpu()) {
2084     assert(left == dest, "left and dest must be equal");
2085     Register lreg_lo = left->as_register_lo();
2086     Register lreg_hi = left->as_register_hi();
2087 
2088     if (right->is_double_cpu()) {
2089       // cpu register - cpu register
2090       Register rreg_lo = right->as_register_lo();
2091       Register rreg_hi = right->as_register_hi();
2092       NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2093       LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2094       switch (code) {
2095         case lir_add:
2096           __ addptr(lreg_lo, rreg_lo);
2097           NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2098           break;
2099         case lir_sub:
2100           __ subptr(lreg_lo, rreg_lo);
2101           NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2102           break;
2103         case lir_mul:
2104 #ifdef _LP64
2105           __ imulq(lreg_lo, rreg_lo);
2106 #else
2107           assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2108           __ imull(lreg_hi, rreg_lo);
2109           __ imull(rreg_hi, lreg_lo);
2110           __ addl (rreg_hi, lreg_hi);
2111           __ mull (rreg_lo);
2112           __ addl (lreg_hi, rreg_hi);
2113 #endif // _LP64
2114           break;
2115         default:
2116           ShouldNotReachHere();
2117       }
2118 
2119     } else if (right->is_constant()) {
2120       // cpu register - constant
2121 #ifdef _LP64
2122       jlong c = right->as_constant_ptr()->as_jlong_bits();
2123       __ movptr(r10, (intptr_t) c);
2124       switch (code) {
2125         case lir_add:
2126           __ addptr(lreg_lo, r10);
2127           break;
2128         case lir_sub:
2129           __ subptr(lreg_lo, r10);
2130           break;
2131         default:
2132           ShouldNotReachHere();
2133       }
2134 #else
2135       jint c_lo = right->as_constant_ptr()->as_jint_lo();
2136       jint c_hi = right->as_constant_ptr()->as_jint_hi();
2137       switch (code) {
2138         case lir_add:
2139           __ addptr(lreg_lo, c_lo);
2140           __ adcl(lreg_hi, c_hi);
2141           break;
2142         case lir_sub:
2143           __ subptr(lreg_lo, c_lo);
2144           __ sbbl(lreg_hi, c_hi);
2145           break;
2146         default:
2147           ShouldNotReachHere();
2148       }
2149 #endif // _LP64
2150 
2151     } else {
2152       ShouldNotReachHere();
2153     }
2154 
2155   } else if (left->is_single_xmm()) {
2156     assert(left == dest, "left and dest must be equal");
2157     XMMRegister lreg = left->as_xmm_float_reg();
2158 
2159     if (right->is_single_xmm()) {
2160       XMMRegister rreg = right->as_xmm_float_reg();
2161       switch (code) {
2162         case lir_add: __ addss(lreg, rreg);  break;
2163         case lir_sub: __ subss(lreg, rreg);  break;
2164         case lir_mul: __ mulss(lreg, rreg);  break;
2165         case lir_div: __ divss(lreg, rreg);  break;
2166         default: ShouldNotReachHere();
2167       }
2168     } else {
2169       Address raddr;
2170       if (right->is_single_stack()) {
2171         raddr = frame_map()->address_for_slot(right->single_stack_ix());
2172       } else if (right->is_constant()) {
2173         // hack for now
2174         raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2175       } else {
2176         ShouldNotReachHere();
2177       }
2178       switch (code) {
2179         case lir_add: __ addss(lreg, raddr);  break;
2180         case lir_sub: __ subss(lreg, raddr);  break;
2181         case lir_mul: __ mulss(lreg, raddr);  break;
2182         case lir_div: __ divss(lreg, raddr);  break;
2183         default: ShouldNotReachHere();
2184       }
2185     }
2186 
2187   } else if (left->is_double_xmm()) {
2188     assert(left == dest, "left and dest must be equal");
2189 
2190     XMMRegister lreg = left->as_xmm_double_reg();
2191     if (right->is_double_xmm()) {
2192       XMMRegister rreg = right->as_xmm_double_reg();
2193       switch (code) {
2194         case lir_add: __ addsd(lreg, rreg);  break;
2195         case lir_sub: __ subsd(lreg, rreg);  break;
2196         case lir_mul: __ mulsd(lreg, rreg);  break;
2197         case lir_div: __ divsd(lreg, rreg);  break;
2198         default: ShouldNotReachHere();
2199       }
2200     } else {
2201       Address raddr;
2202       if (right->is_double_stack()) {
2203         raddr = frame_map()->address_for_slot(right->double_stack_ix());
2204       } else if (right->is_constant()) {
2205         // hack for now
2206         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2207       } else {
2208         ShouldNotReachHere();
2209       }
2210       switch (code) {
2211         case lir_add: __ addsd(lreg, raddr);  break;
2212         case lir_sub: __ subsd(lreg, raddr);  break;
2213         case lir_mul: __ mulsd(lreg, raddr);  break;
2214         case lir_div: __ divsd(lreg, raddr);  break;
2215         default: ShouldNotReachHere();
2216       }
2217     }
2218 
2219 #ifndef _LP64
2220   } else if (left->is_single_fpu()) {
2221     assert(dest->is_single_fpu(),  "fpu stack allocation required");
2222 
2223     if (right->is_single_fpu()) {
2224       arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2225 
2226     } else {
2227       assert(left->fpu_regnr() == 0, "left must be on TOS");
2228       assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2229 
2230       Address raddr;
2231       if (right->is_single_stack()) {
2232         raddr = frame_map()->address_for_slot(right->single_stack_ix());
2233       } else if (right->is_constant()) {
2234         address const_addr = float_constant(right->as_jfloat());
2235         assert(const_addr != nullptr, "incorrect float/double constant maintenance");
2236         // hack for now
2237         raddr = __ as_Address(InternalAddress(const_addr));
2238       } else {
2239         ShouldNotReachHere();
2240       }
2241 
2242       switch (code) {
2243         case lir_add: __ fadd_s(raddr); break;
2244         case lir_sub: __ fsub_s(raddr); break;
2245         case lir_mul: __ fmul_s(raddr); break;
2246         case lir_div: __ fdiv_s(raddr); break;
2247         default:      ShouldNotReachHere();
2248       }
2249     }
2250 
2251   } else if (left->is_double_fpu()) {
2252     assert(dest->is_double_fpu(),  "fpu stack allocation required");
2253 
2254     if (code == lir_mul || code == lir_div) {
2255       // Double values require special handling for strictfp mul/div on x86
2256       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias1()));
2257       __ fmulp(left->fpu_regnrLo() + 1);
2258     }
2259 
2260     if (right->is_double_fpu()) {
2261       arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2262 
2263     } else {
2264       assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2265       assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2266 
2267       Address raddr;
2268       if (right->is_double_stack()) {
2269         raddr = frame_map()->address_for_slot(right->double_stack_ix());
2270       } else if (right->is_constant()) {
2271         // hack for now
2272         raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2273       } else {
2274         ShouldNotReachHere();
2275       }
2276 
2277       switch (code) {
2278         case lir_add: __ fadd_d(raddr); break;
2279         case lir_sub: __ fsub_d(raddr); break;
2280         case lir_mul: __ fmul_d(raddr); break;
2281         case lir_div: __ fdiv_d(raddr); break;
2282         default: ShouldNotReachHere();
2283       }
2284     }
2285 
2286     if (code == lir_mul || code == lir_div) {
2287       // Double values require special handling for strictfp mul/div on x86
2288       __ fld_x(ExternalAddress(StubRoutines::x86::addr_fpu_subnormal_bias2()));
2289       __ fmulp(dest->fpu_regnrLo() + 1);
2290     }
2291 #endif // !_LP64
2292 
2293   } else if (left->is_single_stack() || left->is_address()) {
2294     assert(left == dest, "left and dest must be equal");
2295 
2296     Address laddr;
2297     if (left->is_single_stack()) {
2298       laddr = frame_map()->address_for_slot(left->single_stack_ix());
2299     } else if (left->is_address()) {
2300       laddr = as_Address(left->as_address_ptr());
2301     } else {
2302       ShouldNotReachHere();
2303     }
2304 
2305     if (right->is_single_cpu()) {
2306       Register rreg = right->as_register();
2307       switch (code) {
2308         case lir_add: __ addl(laddr, rreg); break;
2309         case lir_sub: __ subl(laddr, rreg); break;
2310         default:      ShouldNotReachHere();
2311       }
2312     } else if (right->is_constant()) {
2313       jint c = right->as_constant_ptr()->as_jint();
2314       switch (code) {
2315         case lir_add: {
2316           __ incrementl(laddr, c);
2317           break;
2318         }
2319         case lir_sub: {
2320           __ decrementl(laddr, c);
2321           break;
2322         }
2323         default: ShouldNotReachHere();
2324       }
2325     } else {
2326       ShouldNotReachHere();
2327     }
2328 
2329   } else {
2330     ShouldNotReachHere();
2331   }
2332 }
2333 
2334 #ifndef _LP64
2335 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2336   assert(pop_fpu_stack  || (left_index     == dest_index || right_index     == dest_index), "invalid LIR");
2337   assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2338   assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2339 
2340   bool left_is_tos = (left_index == 0);
2341   bool dest_is_tos = (dest_index == 0);
2342   int non_tos_index = (left_is_tos ? right_index : left_index);
2343 
2344   switch (code) {
2345     case lir_add:
2346       if (pop_fpu_stack)       __ faddp(non_tos_index);
2347       else if (dest_is_tos)    __ fadd (non_tos_index);
2348       else                     __ fadda(non_tos_index);
2349       break;
2350 
2351     case lir_sub:
2352       if (left_is_tos) {
2353         if (pop_fpu_stack)     __ fsubrp(non_tos_index);
2354         else if (dest_is_tos)  __ fsub  (non_tos_index);
2355         else                   __ fsubra(non_tos_index);
2356       } else {
2357         if (pop_fpu_stack)     __ fsubp (non_tos_index);
2358         else if (dest_is_tos)  __ fsubr (non_tos_index);
2359         else                   __ fsuba (non_tos_index);
2360       }
2361       break;
2362 
2363     case lir_mul:
2364       if (pop_fpu_stack)       __ fmulp(non_tos_index);
2365       else if (dest_is_tos)    __ fmul (non_tos_index);
2366       else                     __ fmula(non_tos_index);
2367       break;
2368 
2369     case lir_div:
2370       if (left_is_tos) {
2371         if (pop_fpu_stack)     __ fdivrp(non_tos_index);
2372         else if (dest_is_tos)  __ fdiv  (non_tos_index);
2373         else                   __ fdivra(non_tos_index);
2374       } else {
2375         if (pop_fpu_stack)     __ fdivp (non_tos_index);
2376         else if (dest_is_tos)  __ fdivr (non_tos_index);
2377         else                   __ fdiva (non_tos_index);
2378       }
2379       break;
2380 
2381     case lir_rem:
2382       assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2383       __ fremr(noreg);
2384       break;
2385 
2386     default:
2387       ShouldNotReachHere();
2388   }
2389 }
2390 #endif // _LP64
2391 
2392 
2393 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr tmp, LIR_Opr dest, LIR_Op* op) {
2394   if (value->is_double_xmm()) {
2395     switch(code) {
2396       case lir_abs :
2397         {
2398 #ifdef _LP64
2399           if (UseAVX > 2 && !VM_Version::supports_avx512vl()) {
2400             assert(tmp->is_valid(), "need temporary");
2401             __ vpandn(dest->as_xmm_double_reg(), tmp->as_xmm_double_reg(), value->as_xmm_double_reg(), 2);
2402           } else
2403 #endif
2404           {
2405             if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2406               __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2407             }
2408             assert(!tmp->is_valid(), "do not need temporary");
2409             __ andpd(dest->as_xmm_double_reg(),
2410                      ExternalAddress((address)double_signmask_pool),
2411                      rscratch1);
2412           }
2413         }
2414         break;
2415 
2416       case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2417       // all other intrinsics are not available in the SSE instruction set, so FPU is used
2418       default      : ShouldNotReachHere();
2419     }
2420 
2421 #ifndef _LP64
2422   } else if (value->is_double_fpu()) {
2423     assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2424     switch(code) {
2425       case lir_abs   : __ fabs() ; break;
2426       case lir_sqrt  : __ fsqrt(); break;
2427       default      : ShouldNotReachHere();
2428     }
2429 #endif // !_LP64
2430   } else if (code == lir_f2hf) {
2431     __ flt_to_flt16(dest->as_register(), value->as_xmm_float_reg(), tmp->as_xmm_float_reg());
2432   } else if (code == lir_hf2f) {
2433     __ flt16_to_flt(dest->as_xmm_float_reg(), value->as_register());
2434   } else {
2435     Unimplemented();
2436   }
2437 }
2438 
2439 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2440   // assert(left->destroys_register(), "check");
2441   if (left->is_single_cpu()) {
2442     Register reg = left->as_register();
2443     if (right->is_constant()) {
2444       int val = right->as_constant_ptr()->as_jint();
2445       switch (code) {
2446         case lir_logic_and: __ andl (reg, val); break;
2447         case lir_logic_or:  __ orl  (reg, val); break;
2448         case lir_logic_xor: __ xorl (reg, val); break;
2449         default: ShouldNotReachHere();
2450       }
2451     } else if (right->is_stack()) {
2452       // added support for stack operands
2453       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2454       switch (code) {
2455         case lir_logic_and: __ andl (reg, raddr); break;
2456         case lir_logic_or:  __ orl  (reg, raddr); break;
2457         case lir_logic_xor: __ xorl (reg, raddr); break;
2458         default: ShouldNotReachHere();
2459       }
2460     } else {
2461       Register rright = right->as_register();
2462       switch (code) {
2463         case lir_logic_and: __ andptr (reg, rright); break;
2464         case lir_logic_or : __ orptr  (reg, rright); break;
2465         case lir_logic_xor: __ xorptr (reg, rright); break;
2466         default: ShouldNotReachHere();
2467       }
2468     }
2469     move_regs(reg, dst->as_register());
2470   } else {
2471     Register l_lo = left->as_register_lo();
2472     Register l_hi = left->as_register_hi();
2473     if (right->is_constant()) {
2474 #ifdef _LP64
2475       __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2476       switch (code) {
2477         case lir_logic_and:
2478           __ andq(l_lo, rscratch1);
2479           break;
2480         case lir_logic_or:
2481           __ orq(l_lo, rscratch1);
2482           break;
2483         case lir_logic_xor:
2484           __ xorq(l_lo, rscratch1);
2485           break;
2486         default: ShouldNotReachHere();
2487       }
2488 #else
2489       int r_lo = right->as_constant_ptr()->as_jint_lo();
2490       int r_hi = right->as_constant_ptr()->as_jint_hi();
2491       switch (code) {
2492         case lir_logic_and:
2493           __ andl(l_lo, r_lo);
2494           __ andl(l_hi, r_hi);
2495           break;
2496         case lir_logic_or:
2497           __ orl(l_lo, r_lo);
2498           __ orl(l_hi, r_hi);
2499           break;
2500         case lir_logic_xor:
2501           __ xorl(l_lo, r_lo);
2502           __ xorl(l_hi, r_hi);
2503           break;
2504         default: ShouldNotReachHere();
2505       }
2506 #endif // _LP64
2507     } else {
2508 #ifdef _LP64
2509       Register r_lo;
2510       if (is_reference_type(right->type())) {
2511         r_lo = right->as_register();
2512       } else {
2513         r_lo = right->as_register_lo();
2514       }
2515 #else
2516       Register r_lo = right->as_register_lo();
2517       Register r_hi = right->as_register_hi();
2518       assert(l_lo != r_hi, "overwriting registers");
2519 #endif
2520       switch (code) {
2521         case lir_logic_and:
2522           __ andptr(l_lo, r_lo);
2523           NOT_LP64(__ andptr(l_hi, r_hi);)
2524           break;
2525         case lir_logic_or:
2526           __ orptr(l_lo, r_lo);
2527           NOT_LP64(__ orptr(l_hi, r_hi);)
2528           break;
2529         case lir_logic_xor:
2530           __ xorptr(l_lo, r_lo);
2531           NOT_LP64(__ xorptr(l_hi, r_hi);)
2532           break;
2533         default: ShouldNotReachHere();
2534       }
2535     }
2536 
2537     Register dst_lo = dst->as_register_lo();
2538     Register dst_hi = dst->as_register_hi();
2539 
2540 #ifdef _LP64
2541     move_regs(l_lo, dst_lo);
2542 #else
2543     if (dst_lo == l_hi) {
2544       assert(dst_hi != l_lo, "overwriting registers");
2545       move_regs(l_hi, dst_hi);
2546       move_regs(l_lo, dst_lo);
2547     } else {
2548       assert(dst_lo != l_hi, "overwriting registers");
2549       move_regs(l_lo, dst_lo);
2550       move_regs(l_hi, dst_hi);
2551     }
2552 #endif // _LP64
2553   }
2554 }
2555 
2556 
2557 // we assume that rax, and rdx can be overwritten
2558 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2559 
2560   assert(left->is_single_cpu(),   "left must be register");
2561   assert(right->is_single_cpu() || right->is_constant(),  "right must be register or constant");
2562   assert(result->is_single_cpu(), "result must be register");
2563 
2564   //  assert(left->destroys_register(), "check");
2565   //  assert(right->destroys_register(), "check");
2566 
2567   Register lreg = left->as_register();
2568   Register dreg = result->as_register();
2569 
2570   if (right->is_constant()) {
2571     jint divisor = right->as_constant_ptr()->as_jint();
2572     assert(divisor > 0 && is_power_of_2(divisor), "must be");
2573     if (code == lir_idiv) {
2574       assert(lreg == rax, "must be rax,");
2575       assert(temp->as_register() == rdx, "tmp register must be rdx");
2576       __ cdql(); // sign extend into rdx:rax
2577       if (divisor == 2) {
2578         __ subl(lreg, rdx);
2579       } else {
2580         __ andl(rdx, divisor - 1);
2581         __ addl(lreg, rdx);
2582       }
2583       __ sarl(lreg, log2i_exact(divisor));
2584       move_regs(lreg, dreg);
2585     } else if (code == lir_irem) {
2586       Label done;
2587       __ mov(dreg, lreg);
2588       __ andl(dreg, 0x80000000 | (divisor - 1));
2589       __ jcc(Assembler::positive, done);
2590       __ decrement(dreg);
2591       __ orl(dreg, ~(divisor - 1));
2592       __ increment(dreg);
2593       __ bind(done);
2594     } else {
2595       ShouldNotReachHere();
2596     }
2597   } else {
2598     Register rreg = right->as_register();
2599     assert(lreg == rax, "left register must be rax,");
2600     assert(rreg != rdx, "right register must not be rdx");
2601     assert(temp->as_register() == rdx, "tmp register must be rdx");
2602 
2603     move_regs(lreg, rax);
2604 
2605     int idivl_offset = __ corrected_idivl(rreg);
2606     if (ImplicitDiv0Checks) {
2607       add_debug_info_for_div0(idivl_offset, info);
2608     }
2609     if (code == lir_irem) {
2610       move_regs(rdx, dreg); // result is in rdx
2611     } else {
2612       move_regs(rax, dreg);
2613     }
2614   }
2615 }
2616 
2617 
2618 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2619   if (opr1->is_single_cpu()) {
2620     Register reg1 = opr1->as_register();
2621     if (opr2->is_single_cpu()) {
2622       // cpu register - cpu register
2623       if (is_reference_type(opr1->type())) {
2624         __ cmpoop(reg1, opr2->as_register());
2625       } else {
2626         assert(!is_reference_type(opr2->type()), "cmp int, oop?");
2627         __ cmpl(reg1, opr2->as_register());
2628       }
2629     } else if (opr2->is_stack()) {
2630       // cpu register - stack
2631       if (is_reference_type(opr1->type())) {
2632         __ cmpoop(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2633       } else {
2634         __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2635       }
2636     } else if (opr2->is_constant()) {
2637       // cpu register - constant
2638       LIR_Const* c = opr2->as_constant_ptr();
2639       if (c->type() == T_INT) {
2640         jint i = c->as_jint();
2641         if (i == 0) {
2642           __ testl(reg1, reg1);
2643         } else {
2644           __ cmpl(reg1, i);
2645         }
2646       } else if (c->type() == T_METADATA) {
2647         // All we need for now is a comparison with null for equality.
2648         assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
2649         Metadata* m = c->as_metadata();
2650         if (m == nullptr) {
2651           __ testptr(reg1, reg1);
2652         } else {
2653           ShouldNotReachHere();
2654         }
2655       } else if (is_reference_type(c->type())) {
2656         // In 64bit oops are single register
2657         jobject o = c->as_jobject();
2658         if (o == nullptr) {
2659           __ testptr(reg1, reg1);
2660         } else {
2661           __ cmpoop(reg1, o, rscratch1);
2662         }
2663       } else {
2664         fatal("unexpected type: %s", basictype_to_str(c->type()));
2665       }
2666       // cpu register - address
2667     } else if (opr2->is_address()) {
2668       if (op->info() != nullptr) {
2669         add_debug_info_for_null_check_here(op->info());
2670       }
2671       __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2672     } else {
2673       ShouldNotReachHere();
2674     }
2675 
2676   } else if(opr1->is_double_cpu()) {
2677     Register xlo = opr1->as_register_lo();
2678     Register xhi = opr1->as_register_hi();
2679     if (opr2->is_double_cpu()) {
2680 #ifdef _LP64
2681       __ cmpptr(xlo, opr2->as_register_lo());
2682 #else
2683       // cpu register - cpu register
2684       Register ylo = opr2->as_register_lo();
2685       Register yhi = opr2->as_register_hi();
2686       __ subl(xlo, ylo);
2687       __ sbbl(xhi, yhi);
2688       if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2689         __ orl(xhi, xlo);
2690       }
2691 #endif // _LP64
2692     } else if (opr2->is_constant()) {
2693       // cpu register - constant 0
2694       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2695 #ifdef _LP64
2696       __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2697 #else
2698       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2699       __ orl(xhi, xlo);
2700 #endif // _LP64
2701     } else {
2702       ShouldNotReachHere();
2703     }
2704 
2705   } else if (opr1->is_single_xmm()) {
2706     XMMRegister reg1 = opr1->as_xmm_float_reg();
2707     if (opr2->is_single_xmm()) {
2708       // xmm register - xmm register
2709       __ ucomiss(reg1, opr2->as_xmm_float_reg());
2710     } else if (opr2->is_stack()) {
2711       // xmm register - stack
2712       __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2713     } else if (opr2->is_constant()) {
2714       // xmm register - constant
2715       __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2716     } else if (opr2->is_address()) {
2717       // xmm register - address
2718       if (op->info() != nullptr) {
2719         add_debug_info_for_null_check_here(op->info());
2720       }
2721       __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2722     } else {
2723       ShouldNotReachHere();
2724     }
2725 
2726   } else if (opr1->is_double_xmm()) {
2727     XMMRegister reg1 = opr1->as_xmm_double_reg();
2728     if (opr2->is_double_xmm()) {
2729       // xmm register - xmm register
2730       __ ucomisd(reg1, opr2->as_xmm_double_reg());
2731     } else if (opr2->is_stack()) {
2732       // xmm register - stack
2733       __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2734     } else if (opr2->is_constant()) {
2735       // xmm register - constant
2736       __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2737     } else if (opr2->is_address()) {
2738       // xmm register - address
2739       if (op->info() != nullptr) {
2740         add_debug_info_for_null_check_here(op->info());
2741       }
2742       __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2743     } else {
2744       ShouldNotReachHere();
2745     }
2746 
2747 #ifndef _LP64
2748   } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2749     assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2750     assert(opr2->is_fpu_register(), "both must be registers");
2751     __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2752 #endif // LP64
2753 
2754   } else if (opr1->is_address() && opr2->is_constant()) {
2755     LIR_Const* c = opr2->as_constant_ptr();
2756 #ifdef _LP64
2757     if (is_reference_type(c->type())) {
2758       assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2759       __ movoop(rscratch1, c->as_jobject());
2760     }
2761 #endif // LP64
2762     if (op->info() != nullptr) {
2763       add_debug_info_for_null_check_here(op->info());
2764     }
2765     // special case: address - constant
2766     LIR_Address* addr = opr1->as_address_ptr();
2767     if (c->type() == T_INT) {
2768       __ cmpl(as_Address(addr), c->as_jint());
2769     } else if (is_reference_type(c->type())) {
2770 #ifdef _LP64
2771       // %%% Make this explode if addr isn't reachable until we figure out a
2772       // better strategy by giving noreg as the temp for as_Address
2773       __ cmpoop(rscratch1, as_Address(addr, noreg));
2774 #else
2775       __ cmpoop(as_Address(addr), c->as_jobject());
2776 #endif // _LP64
2777     } else {
2778       ShouldNotReachHere();
2779     }
2780 
2781   } else {
2782     ShouldNotReachHere();
2783   }
2784 }
2785 
2786 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2787   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2788     if (left->is_single_xmm()) {
2789       assert(right->is_single_xmm(), "must match");
2790       __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2791     } else if (left->is_double_xmm()) {
2792       assert(right->is_double_xmm(), "must match");
2793       __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2794 
2795     } else {
2796 #ifdef _LP64
2797       ShouldNotReachHere();
2798 #else
2799       assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2800       assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2801 
2802       assert(left->fpu() == 0, "left must be on TOS");
2803       __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2804                   op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2805 #endif // LP64
2806     }
2807   } else {
2808     assert(code == lir_cmp_l2i, "check");
2809 #ifdef _LP64
2810     Label done;
2811     Register dest = dst->as_register();
2812     __ cmpptr(left->as_register_lo(), right->as_register_lo());
2813     __ movl(dest, -1);
2814     __ jccb(Assembler::less, done);
2815     __ setb(Assembler::notZero, dest);
2816     __ movzbl(dest, dest);
2817     __ bind(done);
2818 #else
2819     __ lcmp2int(left->as_register_hi(),
2820                 left->as_register_lo(),
2821                 right->as_register_hi(),
2822                 right->as_register_lo());
2823     move_regs(left->as_register_hi(), dst->as_register());
2824 #endif // _LP64
2825   }
2826 }
2827 
2828 
2829 void LIR_Assembler::align_call(LIR_Code code) {
2830   // make sure that the displacement word of the call ends up word aligned
2831   int offset = __ offset();
2832   switch (code) {
2833   case lir_static_call:
2834   case lir_optvirtual_call:
2835   case lir_dynamic_call:
2836     offset += NativeCall::displacement_offset;
2837     break;
2838   case lir_icvirtual_call:
2839     offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2840     break;
2841   default: ShouldNotReachHere();
2842   }
2843   __ align(BytesPerWord, offset);
2844 }
2845 
2846 
2847 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2848   assert((__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2849          "must be aligned");
2850   __ call(AddressLiteral(op->addr(), rtype));
2851   add_call_info(code_offset(), op->info());
2852   __ post_call_nop();
2853 }
2854 
2855 
2856 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2857   __ ic_call(op->addr());
2858   add_call_info(code_offset(), op->info());
2859   assert((__ offset() - NativeCall::instruction_size + NativeCall::displacement_offset) % BytesPerWord == 0,
2860          "must be aligned");
2861   __ post_call_nop();
2862 }
2863 
2864 
2865 void LIR_Assembler::emit_static_call_stub() {
2866   address call_pc = __ pc();
2867   address stub = __ start_a_stub(call_stub_size());
2868   if (stub == nullptr) {
2869     bailout("static call stub overflow");
2870     return;
2871   }
2872 
2873   int start = __ offset();
2874 
2875   // make sure that the displacement word of the call ends up word aligned
2876   __ align(BytesPerWord, __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset);
2877   __ relocate(static_stub_Relocation::spec(call_pc));
2878   __ mov_metadata(rbx, (Metadata*)nullptr);
2879   // must be set to -1 at code generation time
2880   assert(((__ offset() + 1) % BytesPerWord) == 0, "must be aligned");
2881   // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2882   __ jump(RuntimeAddress(__ pc()));
2883 
2884   assert(__ offset() - start <= call_stub_size(), "stub too big");
2885   __ end_a_stub();
2886 }
2887 
2888 
2889 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2890   assert(exceptionOop->as_register() == rax, "must match");
2891   assert(exceptionPC->as_register() == rdx, "must match");
2892 
2893   // exception object is not added to oop map by LinearScan
2894   // (LinearScan assumes that no oops are in fixed registers)
2895   info->add_register_oop(exceptionOop);
2896   Runtime1::StubID unwind_id;
2897 
2898   // get current pc information
2899   // pc is only needed if the method has an exception handler, the unwind code does not need it.
2900   int pc_for_athrow_offset = __ offset();
2901   InternalAddress pc_for_athrow(__ pc());
2902   __ lea(exceptionPC->as_register(), pc_for_athrow);
2903   add_call_info(pc_for_athrow_offset, info); // for exception handler
2904 
2905   __ verify_not_null_oop(rax);
2906   // search an exception handler (rax: exception oop, rdx: throwing pc)
2907   if (compilation()->has_fpu_code()) {
2908     unwind_id = Runtime1::handle_exception_id;
2909   } else {
2910     unwind_id = Runtime1::handle_exception_nofpu_id;
2911   }
2912   __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2913 
2914   // enough room for two byte trap
2915   __ nop();
2916 }
2917 
2918 
2919 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2920   assert(exceptionOop->as_register() == rax, "must match");
2921 
2922   __ jmp(_unwind_handler_entry);
2923 }
2924 
2925 
2926 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2927 
2928   // optimized version for linear scan:
2929   // * count must be already in ECX (guaranteed by LinearScan)
2930   // * left and dest must be equal
2931   // * tmp must be unused
2932   assert(count->as_register() == SHIFT_count, "count must be in ECX");
2933   assert(left == dest, "left and dest must be equal");
2934   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2935 
2936   if (left->is_single_cpu()) {
2937     Register value = left->as_register();
2938     assert(value != SHIFT_count, "left cannot be ECX");
2939 
2940     switch (code) {
2941       case lir_shl:  __ shll(value); break;
2942       case lir_shr:  __ sarl(value); break;
2943       case lir_ushr: __ shrl(value); break;
2944       default: ShouldNotReachHere();
2945     }
2946   } else if (left->is_double_cpu()) {
2947     Register lo = left->as_register_lo();
2948     Register hi = left->as_register_hi();
2949     assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
2950 #ifdef _LP64
2951     switch (code) {
2952       case lir_shl:  __ shlptr(lo);        break;
2953       case lir_shr:  __ sarptr(lo);        break;
2954       case lir_ushr: __ shrptr(lo);        break;
2955       default: ShouldNotReachHere();
2956     }
2957 #else
2958 
2959     switch (code) {
2960       case lir_shl:  __ lshl(hi, lo);        break;
2961       case lir_shr:  __ lshr(hi, lo, true);  break;
2962       case lir_ushr: __ lshr(hi, lo, false); break;
2963       default: ShouldNotReachHere();
2964     }
2965 #endif // LP64
2966   } else {
2967     ShouldNotReachHere();
2968   }
2969 }
2970 
2971 
2972 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2973   if (dest->is_single_cpu()) {
2974     // first move left into dest so that left is not destroyed by the shift
2975     Register value = dest->as_register();
2976     count = count & 0x1F; // Java spec
2977 
2978     move_regs(left->as_register(), value);
2979     switch (code) {
2980       case lir_shl:  __ shll(value, count); break;
2981       case lir_shr:  __ sarl(value, count); break;
2982       case lir_ushr: __ shrl(value, count); break;
2983       default: ShouldNotReachHere();
2984     }
2985   } else if (dest->is_double_cpu()) {
2986 #ifndef _LP64
2987     Unimplemented();
2988 #else
2989     // first move left into dest so that left is not destroyed by the shift
2990     Register value = dest->as_register_lo();
2991     count = count & 0x1F; // Java spec
2992 
2993     move_regs(left->as_register_lo(), value);
2994     switch (code) {
2995       case lir_shl:  __ shlptr(value, count); break;
2996       case lir_shr:  __ sarptr(value, count); break;
2997       case lir_ushr: __ shrptr(value, count); break;
2998       default: ShouldNotReachHere();
2999     }
3000 #endif // _LP64
3001   } else {
3002     ShouldNotReachHere();
3003   }
3004 }
3005 
3006 
3007 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3008   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3009   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3010   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3011   __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3012 }
3013 
3014 
3015 void LIR_Assembler::store_parameter(jint c,     int offset_from_rsp_in_words) {
3016   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3017   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3018   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3019   __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3020 }
3021 
3022 
3023 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3024   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3025   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3026   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3027   __ movoop(Address(rsp, offset_from_rsp_in_bytes), o, rscratch1);
3028 }
3029 
3030 
3031 void LIR_Assembler::store_parameter(Metadata* m, int offset_from_rsp_in_words) {
3032   assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3033   int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3034   assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3035   __ mov_metadata(Address(rsp, offset_from_rsp_in_bytes), m, rscratch1);
3036 }
3037 
3038 
3039 // This code replaces a call to arraycopy; no exception may
3040 // be thrown in this code, they must be thrown in the System.arraycopy
3041 // activation frame; we could save some checks if this would not be the case
3042 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3043   ciArrayKlass* default_type = op->expected_type();
3044   Register src = op->src()->as_register();
3045   Register dst = op->dst()->as_register();
3046   Register src_pos = op->src_pos()->as_register();
3047   Register dst_pos = op->dst_pos()->as_register();
3048   Register length  = op->length()->as_register();
3049   Register tmp = op->tmp()->as_register();
3050   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
3051   Register tmp2 = UseCompactObjectHeaders ? rscratch2 : noreg;
3052 
3053   CodeStub* stub = op->stub();
3054   int flags = op->flags();
3055   BasicType basic_type = default_type != nullptr ? default_type->element_type()->basic_type() : T_ILLEGAL;
3056   if (is_reference_type(basic_type)) basic_type = T_OBJECT;
3057 
3058   // if we don't know anything, just go through the generic arraycopy
3059   if (default_type == nullptr) {
3060     // save outgoing arguments on stack in case call to System.arraycopy is needed
3061     // HACK ALERT. This code used to push the parameters in a hardwired fashion
3062     // for interpreter calling conventions. Now we have to do it in new style conventions.
3063     // For the moment until C1 gets the new register allocator I just force all the
3064     // args to the right place (except the register args) and then on the back side
3065     // reload the register args properly if we go slow path. Yuck
3066 
3067     // These are proper for the calling convention
3068     store_parameter(length, 2);
3069     store_parameter(dst_pos, 1);
3070     store_parameter(dst, 0);
3071 
3072     // these are just temporary placements until we need to reload
3073     store_parameter(src_pos, 3);
3074     store_parameter(src, 4);
3075     NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3076 
3077     address copyfunc_addr = StubRoutines::generic_arraycopy();
3078     assert(copyfunc_addr != nullptr, "generic arraycopy stub required");
3079 
3080     // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3081 #ifdef _LP64
3082     // The arguments are in java calling convention so we can trivially shift them to C
3083     // convention
3084     assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3085     __ mov(c_rarg0, j_rarg0);
3086     assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3087     __ mov(c_rarg1, j_rarg1);
3088     assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3089     __ mov(c_rarg2, j_rarg2);
3090     assert_different_registers(c_rarg3, j_rarg4);
3091     __ mov(c_rarg3, j_rarg3);
3092 #ifdef _WIN64
3093     // Allocate abi space for args but be sure to keep stack aligned
3094     __ subptr(rsp, 6*wordSize);
3095     store_parameter(j_rarg4, 4);
3096 #ifndef PRODUCT
3097     if (PrintC1Statistics) {
3098       __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt), rscratch1);
3099     }
3100 #endif
3101     __ call(RuntimeAddress(copyfunc_addr));
3102     __ addptr(rsp, 6*wordSize);
3103 #else
3104     __ mov(c_rarg4, j_rarg4);
3105 #ifndef PRODUCT
3106     if (PrintC1Statistics) {
3107       __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt), rscratch1);
3108     }
3109 #endif
3110     __ call(RuntimeAddress(copyfunc_addr));
3111 #endif // _WIN64
3112 #else
3113     __ push(length);
3114     __ push(dst_pos);
3115     __ push(dst);
3116     __ push(src_pos);
3117     __ push(src);
3118 
3119 #ifndef PRODUCT
3120     if (PrintC1Statistics) {
3121       __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt), rscratch1);
3122     }
3123 #endif
3124     __ call_VM_leaf(copyfunc_addr, 5); // removes pushed parameter from the stack
3125 
3126 #endif // _LP64
3127 
3128     __ testl(rax, rax);
3129     __ jcc(Assembler::equal, *stub->continuation());
3130 
3131     __ mov(tmp, rax);
3132     __ xorl(tmp, -1);
3133 
3134     // Reload values from the stack so they are where the stub
3135     // expects them.
3136     __ movptr   (dst,     Address(rsp, 0*BytesPerWord));
3137     __ movptr   (dst_pos, Address(rsp, 1*BytesPerWord));
3138     __ movptr   (length,  Address(rsp, 2*BytesPerWord));
3139     __ movptr   (src_pos, Address(rsp, 3*BytesPerWord));
3140     __ movptr   (src,     Address(rsp, 4*BytesPerWord));
3141 
3142     __ subl(length, tmp);
3143     __ addl(src_pos, tmp);
3144     __ addl(dst_pos, tmp);
3145     __ jmp(*stub->entry());
3146 
3147     __ bind(*stub->continuation());
3148     return;
3149   }
3150 
3151   assert(default_type != nullptr && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3152 
3153   int elem_size = type2aelembytes(basic_type);
3154   Address::ScaleFactor scale;
3155 
3156   switch (elem_size) {
3157     case 1 :
3158       scale = Address::times_1;
3159       break;
3160     case 2 :
3161       scale = Address::times_2;
3162       break;
3163     case 4 :
3164       scale = Address::times_4;
3165       break;
3166     case 8 :
3167       scale = Address::times_8;
3168       break;
3169     default:
3170       scale = Address::no_scale;
3171       ShouldNotReachHere();
3172   }
3173 
3174   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3175   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3176 
3177   // length and pos's are all sign extended at this point on 64bit
3178 
3179   // test for null
3180   if (flags & LIR_OpArrayCopy::src_null_check) {
3181     __ testptr(src, src);
3182     __ jcc(Assembler::zero, *stub->entry());
3183   }
3184   if (flags & LIR_OpArrayCopy::dst_null_check) {
3185     __ testptr(dst, dst);
3186     __ jcc(Assembler::zero, *stub->entry());
3187   }
3188 
3189   // If the compiler was not able to prove that exact type of the source or the destination
3190   // of the arraycopy is an array type, check at runtime if the source or the destination is
3191   // an instance type.
3192   if (flags & LIR_OpArrayCopy::type_check) {
3193     if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
3194       __ load_klass(tmp, dst, tmp_load_klass);
3195       __ cmpl(Address(tmp, in_bytes(Klass::layout_helper_offset())), Klass::_lh_neutral_value);
3196       __ jcc(Assembler::greaterEqual, *stub->entry());
3197     }
3198 
3199     if (!(flags & LIR_OpArrayCopy::src_objarray)) {
3200       __ load_klass(tmp, src, tmp_load_klass);
3201       __ cmpl(Address(tmp, in_bytes(Klass::layout_helper_offset())), Klass::_lh_neutral_value);
3202       __ jcc(Assembler::greaterEqual, *stub->entry());
3203     }
3204   }
3205 
3206   // check if negative
3207   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3208     __ testl(src_pos, src_pos);
3209     __ jcc(Assembler::less, *stub->entry());
3210   }
3211   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3212     __ testl(dst_pos, dst_pos);
3213     __ jcc(Assembler::less, *stub->entry());
3214   }
3215 
3216   if (flags & LIR_OpArrayCopy::src_range_check) {
3217     __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3218     __ cmpl(tmp, src_length_addr);
3219     __ jcc(Assembler::above, *stub->entry());
3220   }
3221   if (flags & LIR_OpArrayCopy::dst_range_check) {
3222     __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3223     __ cmpl(tmp, dst_length_addr);
3224     __ jcc(Assembler::above, *stub->entry());
3225   }
3226 
3227   if (flags & LIR_OpArrayCopy::length_positive_check) {
3228     __ testl(length, length);
3229     __ jcc(Assembler::less, *stub->entry());
3230   }
3231 
3232 #ifdef _LP64
3233   __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3234   __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3235 #endif
3236 
3237   if (flags & LIR_OpArrayCopy::type_check) {
3238     // We don't know the array types are compatible
3239     if (basic_type != T_OBJECT) {
3240       // Simple test for basic type arrays
3241       __ cmp_klass(src, dst, tmp, tmp2);
3242       __ jcc(Assembler::notEqual, *stub->entry());
3243     } else {
3244       // For object arrays, if src is a sub class of dst then we can
3245       // safely do the copy.
3246       Label cont, slow;
3247 
3248       __ push(src);
3249       __ push(dst);
3250 
3251       __ load_klass(src, src, tmp_load_klass);
3252       __ load_klass(dst, dst, tmp_load_klass);
3253 
3254       __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, nullptr);
3255 
3256       __ push(src);
3257       __ push(dst);
3258       __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
3259       __ pop(dst);
3260       __ pop(src);
3261 
3262       __ testl(src, src);
3263       __ jcc(Assembler::notEqual, cont);
3264 
3265       __ bind(slow);
3266       __ pop(dst);
3267       __ pop(src);
3268 
3269       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
3270       if (copyfunc_addr != nullptr) { // use stub if available
3271         // src is not a sub class of dst so we have to do a
3272         // per-element check.
3273 
3274         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
3275         if ((flags & mask) != mask) {
3276           // Check that at least both of them object arrays.
3277           assert(flags & mask, "one of the two should be known to be an object array");
3278 
3279           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
3280             __ load_klass(tmp, src, tmp_load_klass);
3281           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
3282             __ load_klass(tmp, dst, tmp_load_klass);
3283           }
3284           int lh_offset = in_bytes(Klass::layout_helper_offset());
3285           Address klass_lh_addr(tmp, lh_offset);
3286           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
3287           __ cmpl(klass_lh_addr, objArray_lh);
3288           __ jcc(Assembler::notEqual, *stub->entry());
3289         }
3290 
3291        // Spill because stubs can use any register they like and it's
3292        // easier to restore just those that we care about.
3293        store_parameter(dst, 0);
3294        store_parameter(dst_pos, 1);
3295        store_parameter(length, 2);
3296        store_parameter(src_pos, 3);
3297        store_parameter(src, 4);
3298 
3299 #ifndef _LP64
3300        Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3301         __ movptr(tmp, dst_klass_addr);
3302         __ movptr(tmp, Address(tmp, ObjArrayKlass::element_klass_offset()));
3303         __ push(tmp);
3304         __ movl(tmp, Address(tmp, Klass::super_check_offset_offset()));
3305         __ push(tmp);
3306         __ push(length);
3307         __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3308         __ push(tmp);
3309         __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3310         __ push(tmp);
3311 
3312         __ call_VM_leaf(copyfunc_addr, 5);
3313 #else
3314         __ movl2ptr(length, length); //higher 32bits must be null
3315 
3316         __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3317         assert_different_registers(c_rarg0, dst, dst_pos, length);
3318         __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3319         assert_different_registers(c_rarg1, dst, length);
3320 
3321         __ mov(c_rarg2, length);
3322         assert_different_registers(c_rarg2, dst);
3323 
3324 #ifdef _WIN64
3325         // Allocate abi space for args but be sure to keep stack aligned
3326         __ subptr(rsp, 6*wordSize);
3327         __ load_klass(c_rarg3, dst, tmp_load_klass);
3328         __ movptr(c_rarg3, Address(c_rarg3, ObjArrayKlass::element_klass_offset()));
3329         store_parameter(c_rarg3, 4);
3330         __ movl(c_rarg3, Address(c_rarg3, Klass::super_check_offset_offset()));
3331         __ call(RuntimeAddress(copyfunc_addr));
3332         __ addptr(rsp, 6*wordSize);
3333 #else
3334         __ load_klass(c_rarg4, dst, tmp_load_klass);
3335         __ movptr(c_rarg4, Address(c_rarg4, ObjArrayKlass::element_klass_offset()));
3336         __ movl(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset()));
3337         __ call(RuntimeAddress(copyfunc_addr));
3338 #endif
3339 
3340 #endif
3341 
3342 #ifndef PRODUCT
3343         if (PrintC1Statistics) {
3344           Label failed;
3345           __ testl(rax, rax);
3346           __ jcc(Assembler::notZero, failed);
3347           __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt), rscratch1);
3348           __ bind(failed);
3349         }
3350 #endif
3351 
3352         __ testl(rax, rax);
3353         __ jcc(Assembler::zero, *stub->continuation());
3354 
3355 #ifndef PRODUCT
3356         if (PrintC1Statistics) {
3357           __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt), rscratch1);
3358         }
3359 #endif
3360 
3361         __ mov(tmp, rax);
3362 
3363         __ xorl(tmp, -1);
3364 
3365         // Restore previously spilled arguments
3366         __ movptr   (dst,     Address(rsp, 0*BytesPerWord));
3367         __ movptr   (dst_pos, Address(rsp, 1*BytesPerWord));
3368         __ movptr   (length,  Address(rsp, 2*BytesPerWord));
3369         __ movptr   (src_pos, Address(rsp, 3*BytesPerWord));
3370         __ movptr   (src,     Address(rsp, 4*BytesPerWord));
3371 
3372 
3373         __ subl(length, tmp);
3374         __ addl(src_pos, tmp);
3375         __ addl(dst_pos, tmp);
3376       }
3377 
3378       __ jmp(*stub->entry());
3379 
3380       __ bind(cont);
3381       __ pop(dst);
3382       __ pop(src);
3383     }
3384   }
3385 
3386 #ifdef ASSERT
3387   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3388     // Sanity check the known type with the incoming class.  For the
3389     // primitive case the types must match exactly with src.klass and
3390     // dst.klass each exactly matching the default type.  For the
3391     // object array case, if no type check is needed then either the
3392     // dst type is exactly the expected type and the src type is a
3393     // subtype which we can't check or src is the same array as dst
3394     // but not necessarily exactly of type default_type.
3395     Label known_ok, halt;
3396     __ mov_metadata(tmp, default_type->constant_encoding());
3397 #ifdef _LP64
3398     if (UseCompressedClassPointers) {
3399       __ encode_klass_not_null(tmp, rscratch1);
3400     }
3401 #endif
3402 
3403     if (basic_type != T_OBJECT) {
3404       __ cmp_klass(tmp, dst, tmp2);
3405       __ jcc(Assembler::notEqual, halt);
3406       __ cmp_klass(tmp, src, tmp2);
3407       __ jcc(Assembler::equal, known_ok);
3408     } else {
3409       __ cmp_klass(tmp, dst, tmp2);
3410       __ jcc(Assembler::equal, known_ok);
3411       __ cmpptr(src, dst);
3412       __ jcc(Assembler::equal, known_ok);
3413     }
3414     __ bind(halt);
3415     __ stop("incorrect type information in arraycopy");
3416     __ bind(known_ok);
3417   }
3418 #endif
3419 
3420 #ifndef PRODUCT
3421   if (PrintC1Statistics) {
3422     __ incrementl(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)), rscratch1);
3423   }
3424 #endif
3425 
3426 #ifdef _LP64
3427   assert_different_registers(c_rarg0, dst, dst_pos, length);
3428   __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3429   assert_different_registers(c_rarg1, length);
3430   __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3431   __ mov(c_rarg2, length);
3432 
3433 #else
3434   __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3435   store_parameter(tmp, 0);
3436   __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3437   store_parameter(tmp, 1);
3438   store_parameter(length, 2);
3439 #endif // _LP64
3440 
3441   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
3442   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
3443   const char *name;
3444   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
3445   __ call_VM_leaf(entry, 0);
3446 
3447   if (stub != nullptr) {
3448     __ bind(*stub->continuation());
3449   }
3450 }
3451 
3452 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3453   assert(op->crc()->is_single_cpu(),  "crc must be register");
3454   assert(op->val()->is_single_cpu(),  "byte value must be register");
3455   assert(op->result_opr()->is_single_cpu(), "result must be register");
3456   Register crc = op->crc()->as_register();
3457   Register val = op->val()->as_register();
3458   Register res = op->result_opr()->as_register();
3459 
3460   assert_different_registers(val, crc, res);
3461 
3462   __ lea(res, ExternalAddress(StubRoutines::crc_table_addr()));
3463   __ notl(crc); // ~crc
3464   __ update_byte_crc32(crc, val, res);
3465   __ notl(crc); // ~crc
3466   __ mov(res, crc);
3467 }
3468 
3469 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3470   Register obj = op->obj_opr()->as_register();  // may not be an oop
3471   Register hdr = op->hdr_opr()->as_register();
3472   Register lock = op->lock_opr()->as_register();
3473   if (LockingMode == LM_MONITOR) {
3474     if (op->info() != nullptr) {
3475       add_debug_info_for_null_check_here(op->info());
3476       __ null_check(obj);
3477     }
3478     __ jmp(*op->stub()->entry());
3479   } else if (op->code() == lir_lock) {
3480     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3481     Register tmp = LockingMode == LM_LIGHTWEIGHT ? op->scratch_opr()->as_register() : noreg;
3482     // add debug info for NullPointerException only if one is possible
3483     int null_check_offset = __ lock_object(hdr, obj, lock, tmp, *op->stub()->entry());
3484     if (op->info() != nullptr) {
3485       add_debug_info_for_null_check(null_check_offset, op->info());
3486     }
3487     // done
3488   } else if (op->code() == lir_unlock) {
3489     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3490     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3491   } else {
3492     Unimplemented();
3493   }
3494   __ bind(*op->stub()->continuation());
3495 }
3496 
3497 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
3498   Register obj = op->obj()->as_pointer_register();
3499   Register result = op->result_opr()->as_pointer_register();
3500 
3501   CodeEmitInfo* info = op->info();
3502   if (info != nullptr) {
3503     add_debug_info_for_null_check_here(info);
3504   }
3505 
3506 #ifdef _LP64
3507   if (UseCompactObjectHeaders) {
3508     Register tmp = rscratch1;
3509     assert_different_registers(tmp, obj);
3510     assert_different_registers(tmp, result);
3511 
3512     __ movq(result, Address(obj, oopDesc::mark_offset_in_bytes()));
3513     __ shrq(result, markWord::klass_shift);
3514     __ decode_klass_not_null(result, tmp);
3515   } else if (UseCompressedClassPointers) {
3516     __ movl(result, Address(obj, oopDesc::klass_offset_in_bytes()));
3517     __ decode_klass_not_null(result, rscratch1);
3518   } else
3519 #endif
3520   {
3521     __ movptr(result, Address(obj, oopDesc::klass_offset_in_bytes()));
3522   }
3523 }
3524 
3525 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3526   ciMethod* method = op->profiled_method();
3527   int bci          = op->profiled_bci();
3528   ciMethod* callee = op->profiled_callee();
3529   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
3530 
3531   // Update counter for all call types
3532   ciMethodData* md = method->method_data_or_null();
3533   assert(md != nullptr, "Sanity");
3534   ciProfileData* data = md->bci_to_data(bci);
3535   assert(data != nullptr && data->is_CounterData(), "need CounterData for calls");
3536   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
3537   Register mdo  = op->mdo()->as_register();
3538   __ mov_metadata(mdo, md->constant_encoding());
3539   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3540   // Perform additional virtual call profiling for invokevirtual and
3541   // invokeinterface bytecodes
3542   if (op->should_profile_receiver_type()) {
3543     assert(op->recv()->is_single_cpu(), "recv must be allocated");
3544     Register recv = op->recv()->as_register();
3545     assert_different_registers(mdo, recv);
3546     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3547     ciKlass* known_klass = op->known_holder();
3548     if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
3549       // We know the type that will be seen at this call site; we can
3550       // statically update the MethodData* rather than needing to do
3551       // dynamic tests on the receiver type
3552 
3553       // NOTE: we should probably put a lock around this search to
3554       // avoid collisions by concurrent compilations
3555       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3556       uint i;
3557       for (i = 0; i < VirtualCallData::row_limit(); i++) {
3558         ciKlass* receiver = vc_data->receiver(i);
3559         if (known_klass->equals(receiver)) {
3560           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3561           __ addptr(data_addr, DataLayout::counter_increment);
3562           return;
3563         }
3564       }
3565 
3566       // Receiver type not found in profile data; select an empty slot
3567 
3568       // Note that this is less efficient than it should be because it
3569       // always does a write to the receiver part of the
3570       // VirtualCallData rather than just the first time
3571       for (i = 0; i < VirtualCallData::row_limit(); i++) {
3572         ciKlass* receiver = vc_data->receiver(i);
3573         if (receiver == nullptr) {
3574           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3575           __ mov_metadata(recv_addr, known_klass->constant_encoding(), rscratch1);
3576           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3577           __ addptr(data_addr, DataLayout::counter_increment);
3578           return;
3579         }
3580       }
3581     } else {
3582       __ load_klass(recv, recv, tmp_load_klass);
3583       Label update_done;
3584       type_profile_helper(mdo, md, data, recv, &update_done);
3585       // Receiver did not match any saved receiver and there is no empty row for it.
3586       // Increment total counter to indicate polymorphic case.
3587       __ addptr(counter_addr, DataLayout::counter_increment);
3588 
3589       __ bind(update_done);
3590     }
3591   } else {
3592     // Static call
3593     __ addptr(counter_addr, DataLayout::counter_increment);
3594   }
3595 }
3596 
3597 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
3598   Register obj = op->obj()->as_register();
3599   Register tmp = op->tmp()->as_pointer_register();
3600   Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
3601   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
3602   ciKlass* exact_klass = op->exact_klass();
3603   intptr_t current_klass = op->current_klass();
3604   bool not_null = op->not_null();
3605   bool no_conflict = op->no_conflict();
3606 
3607   Label update, next, none;
3608 
3609   bool do_null = !not_null;
3610   bool exact_klass_set = exact_klass != nullptr && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
3611   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
3612 
3613   assert(do_null || do_update, "why are we here?");
3614   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
3615 
3616   __ verify_oop(obj);
3617 
3618 #ifdef ASSERT
3619   if (obj == tmp) {
3620 #ifdef _LP64
3621     assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
3622 #else
3623     assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
3624 #endif
3625   } else {
3626 #ifdef _LP64
3627     assert_different_registers(obj, tmp, rscratch1, mdo_addr.base(), mdo_addr.index());
3628 #else
3629     assert_different_registers(obj, tmp, mdo_addr.base(), mdo_addr.index());
3630 #endif
3631   }
3632 #endif
3633   if (do_null) {
3634     __ testptr(obj, obj);
3635     __ jccb(Assembler::notZero, update);
3636     if (!TypeEntries::was_null_seen(current_klass)) {
3637       __ testptr(mdo_addr, TypeEntries::null_seen);
3638 #ifndef ASSERT
3639       __ jccb(Assembler::notZero, next); // already set
3640 #else
3641       __ jcc(Assembler::notZero, next); // already set
3642 #endif
3643       // atomic update to prevent overwriting Klass* with 0
3644       __ lock();
3645       __ orptr(mdo_addr, TypeEntries::null_seen);
3646     }
3647     if (do_update) {
3648 #ifndef ASSERT
3649       __ jmpb(next);
3650     }
3651 #else
3652       __ jmp(next);
3653     }
3654   } else {
3655     __ testptr(obj, obj);
3656     __ jcc(Assembler::notZero, update);
3657     __ stop("unexpected null obj");
3658 #endif
3659   }
3660 
3661   __ bind(update);
3662 
3663   if (do_update) {
3664 #ifdef ASSERT
3665     if (exact_klass != nullptr) {
3666       Label ok;
3667       __ load_klass(tmp, obj, tmp_load_klass);
3668       __ push(tmp);
3669       __ mov_metadata(tmp, exact_klass->constant_encoding());
3670       __ cmpptr(tmp, Address(rsp, 0));
3671       __ jcc(Assembler::equal, ok);
3672       __ stop("exact klass and actual klass differ");
3673       __ bind(ok);
3674       __ pop(tmp);
3675     }
3676 #endif
3677     if (!no_conflict) {
3678       if (exact_klass == nullptr || TypeEntries::is_type_none(current_klass)) {
3679         if (exact_klass != nullptr) {
3680           __ mov_metadata(tmp, exact_klass->constant_encoding());
3681         } else {
3682           __ load_klass(tmp, obj, tmp_load_klass);
3683         }
3684 #ifdef _LP64
3685         __ mov(rscratch1, tmp); // save original value before XOR
3686 #endif
3687         __ xorptr(tmp, mdo_addr);
3688         __ testptr(tmp, TypeEntries::type_klass_mask);
3689         // klass seen before, nothing to do. The unknown bit may have been
3690         // set already but no need to check.
3691         __ jccb(Assembler::zero, next);
3692 
3693         __ testptr(tmp, TypeEntries::type_unknown);
3694         __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3695 
3696         if (TypeEntries::is_type_none(current_klass)) {
3697           __ testptr(mdo_addr, TypeEntries::type_mask);
3698           __ jccb(Assembler::zero, none);
3699 #ifdef _LP64
3700           // There is a chance that the checks above (re-reading profiling
3701           // data from memory) fail if another thread has just set the
3702           // profiling to this obj's klass
3703           __ mov(tmp, rscratch1); // get back original value before XOR
3704           __ xorptr(tmp, mdo_addr);
3705           __ testptr(tmp, TypeEntries::type_klass_mask);
3706           __ jccb(Assembler::zero, next);
3707 #endif
3708         }
3709       } else {
3710         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3711                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3712 
3713         __ testptr(mdo_addr, TypeEntries::type_unknown);
3714         __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3715       }
3716 
3717       // different than before. Cannot keep accurate profile.
3718       __ orptr(mdo_addr, TypeEntries::type_unknown);
3719 
3720       if (TypeEntries::is_type_none(current_klass)) {
3721         __ jmpb(next);
3722 
3723         __ bind(none);
3724         // first time here. Set profile type.
3725         __ movptr(mdo_addr, tmp);
3726 #ifdef ASSERT
3727         __ andptr(tmp, TypeEntries::type_klass_mask);
3728         __ verify_klass_ptr(tmp);
3729 #endif
3730       }
3731     } else {
3732       // There's a single possible klass at this profile point
3733       assert(exact_klass != nullptr, "should be");
3734       if (TypeEntries::is_type_none(current_klass)) {
3735         __ mov_metadata(tmp, exact_klass->constant_encoding());
3736         __ xorptr(tmp, mdo_addr);
3737         __ testptr(tmp, TypeEntries::type_klass_mask);
3738 #ifdef ASSERT
3739         __ jcc(Assembler::zero, next);
3740 
3741         {
3742           Label ok;
3743           __ push(tmp);
3744           __ testptr(mdo_addr, TypeEntries::type_mask);
3745           __ jcc(Assembler::zero, ok);
3746           // may have been set by another thread
3747           __ mov_metadata(tmp, exact_klass->constant_encoding());
3748           __ xorptr(tmp, mdo_addr);
3749           __ testptr(tmp, TypeEntries::type_mask);
3750           __ jcc(Assembler::zero, ok);
3751 
3752           __ stop("unexpected profiling mismatch");
3753           __ bind(ok);
3754           __ pop(tmp);
3755         }
3756 #else
3757         __ jccb(Assembler::zero, next);
3758 #endif
3759         // first time here. Set profile type.
3760         __ movptr(mdo_addr, tmp);
3761 #ifdef ASSERT
3762         __ andptr(tmp, TypeEntries::type_klass_mask);
3763         __ verify_klass_ptr(tmp);
3764 #endif
3765       } else {
3766         assert(ciTypeEntries::valid_ciklass(current_klass) != nullptr &&
3767                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3768 
3769         __ testptr(mdo_addr, TypeEntries::type_unknown);
3770         __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3771 
3772         __ orptr(mdo_addr, TypeEntries::type_unknown);
3773       }
3774     }
3775   }
3776   __ bind(next);
3777 }
3778 
3779 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3780   Unimplemented();
3781 }
3782 
3783 
3784 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3785   __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3786 }
3787 
3788 
3789 void LIR_Assembler::align_backward_branch_target() {
3790   __ align(BytesPerWord);
3791 }
3792 
3793 
3794 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
3795   if (left->is_single_cpu()) {
3796     __ negl(left->as_register());
3797     move_regs(left->as_register(), dest->as_register());
3798 
3799   } else if (left->is_double_cpu()) {
3800     Register lo = left->as_register_lo();
3801 #ifdef _LP64
3802     Register dst = dest->as_register_lo();
3803     __ movptr(dst, lo);
3804     __ negptr(dst);
3805 #else
3806     Register hi = left->as_register_hi();
3807     __ lneg(hi, lo);
3808     if (dest->as_register_lo() == hi) {
3809       assert(dest->as_register_hi() != lo, "destroying register");
3810       move_regs(hi, dest->as_register_hi());
3811       move_regs(lo, dest->as_register_lo());
3812     } else {
3813       move_regs(lo, dest->as_register_lo());
3814       move_regs(hi, dest->as_register_hi());
3815     }
3816 #endif // _LP64
3817 
3818   } else if (dest->is_single_xmm()) {
3819 #ifdef _LP64
3820     if (UseAVX > 2 && !VM_Version::supports_avx512vl()) {
3821       assert(tmp->is_valid(), "need temporary");
3822       assert_different_registers(left->as_xmm_float_reg(), tmp->as_xmm_float_reg());
3823       __ vpxor(dest->as_xmm_float_reg(), tmp->as_xmm_float_reg(), left->as_xmm_float_reg(), 2);
3824     }
3825     else
3826 #endif
3827     {
3828       assert(!tmp->is_valid(), "do not need temporary");
3829       if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3830         __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3831       }
3832       __ xorps(dest->as_xmm_float_reg(),
3833                ExternalAddress((address)float_signflip_pool),
3834                rscratch1);
3835     }
3836   } else if (dest->is_double_xmm()) {
3837 #ifdef _LP64
3838     if (UseAVX > 2 && !VM_Version::supports_avx512vl()) {
3839       assert(tmp->is_valid(), "need temporary");
3840       assert_different_registers(left->as_xmm_double_reg(), tmp->as_xmm_double_reg());
3841       __ vpxor(dest->as_xmm_double_reg(), tmp->as_xmm_double_reg(), left->as_xmm_double_reg(), 2);
3842     }
3843     else
3844 #endif
3845     {
3846       assert(!tmp->is_valid(), "do not need temporary");
3847       if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3848         __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3849       }
3850       __ xorpd(dest->as_xmm_double_reg(),
3851                ExternalAddress((address)double_signflip_pool),
3852                rscratch1);
3853     }
3854 #ifndef _LP64
3855   } else if (left->is_single_fpu() || left->is_double_fpu()) {
3856     assert(left->fpu() == 0, "arg must be on TOS");
3857     assert(dest->fpu() == 0, "dest must be TOS");
3858     __ fchs();
3859 #endif // !_LP64
3860 
3861   } else {
3862     ShouldNotReachHere();
3863   }
3864 }
3865 
3866 
3867 void LIR_Assembler::leal(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
3868   assert(src->is_address(), "must be an address");
3869   assert(dest->is_register(), "must be a register");
3870 
3871   PatchingStub* patch = nullptr;
3872   if (patch_code != lir_patch_none) {
3873     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
3874   }
3875 
3876   Register reg = dest->as_pointer_register();
3877   LIR_Address* addr = src->as_address_ptr();
3878   __ lea(reg, as_Address(addr));
3879 
3880   if (patch != nullptr) {
3881     patching_epilog(patch, patch_code, addr->base()->as_register(), info);
3882   }
3883 }
3884 
3885 
3886 
3887 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3888   assert(!tmp->is_valid(), "don't need temporary");
3889   __ call(RuntimeAddress(dest));
3890   if (info != nullptr) {
3891     add_call_info_here(info);
3892   }
3893   __ post_call_nop();
3894 }
3895 
3896 
3897 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3898   assert(type == T_LONG, "only for volatile long fields");
3899 
3900   if (info != nullptr) {
3901     add_debug_info_for_null_check_here(info);
3902   }
3903 
3904   if (src->is_double_xmm()) {
3905     if (dest->is_double_cpu()) {
3906 #ifdef _LP64
3907       __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3908 #else
3909       __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3910       __ psrlq(src->as_xmm_double_reg(), 32);
3911       __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3912 #endif // _LP64
3913     } else if (dest->is_double_stack()) {
3914       __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3915     } else if (dest->is_address()) {
3916       __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3917     } else {
3918       ShouldNotReachHere();
3919     }
3920 
3921   } else if (dest->is_double_xmm()) {
3922     if (src->is_double_stack()) {
3923       __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3924     } else if (src->is_address()) {
3925       __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3926     } else {
3927       ShouldNotReachHere();
3928     }
3929 
3930 #ifndef _LP64
3931   } else if (src->is_double_fpu()) {
3932     assert(src->fpu_regnrLo() == 0, "must be TOS");
3933     if (dest->is_double_stack()) {
3934       __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3935     } else if (dest->is_address()) {
3936       __ fistp_d(as_Address(dest->as_address_ptr()));
3937     } else {
3938       ShouldNotReachHere();
3939     }
3940 
3941   } else if (dest->is_double_fpu()) {
3942     assert(dest->fpu_regnrLo() == 0, "must be TOS");
3943     if (src->is_double_stack()) {
3944       __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3945     } else if (src->is_address()) {
3946       __ fild_d(as_Address(src->as_address_ptr()));
3947     } else {
3948       ShouldNotReachHere();
3949     }
3950 #endif // !_LP64
3951 
3952   } else {
3953     ShouldNotReachHere();
3954   }
3955 }
3956 
3957 #ifdef ASSERT
3958 // emit run-time assertion
3959 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
3960   assert(op->code() == lir_assert, "must be");
3961 
3962   if (op->in_opr1()->is_valid()) {
3963     assert(op->in_opr2()->is_valid(), "both operands must be valid");
3964     comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
3965   } else {
3966     assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
3967     assert(op->condition() == lir_cond_always, "no other conditions allowed");
3968   }
3969 
3970   Label ok;
3971   if (op->condition() != lir_cond_always) {
3972     Assembler::Condition acond = Assembler::zero;
3973     switch (op->condition()) {
3974       case lir_cond_equal:        acond = Assembler::equal;       break;
3975       case lir_cond_notEqual:     acond = Assembler::notEqual;    break;
3976       case lir_cond_less:         acond = Assembler::less;        break;
3977       case lir_cond_lessEqual:    acond = Assembler::lessEqual;   break;
3978       case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
3979       case lir_cond_greater:      acond = Assembler::greater;     break;
3980       case lir_cond_belowEqual:   acond = Assembler::belowEqual;  break;
3981       case lir_cond_aboveEqual:   acond = Assembler::aboveEqual;  break;
3982       default:                    ShouldNotReachHere();
3983     }
3984     __ jcc(acond, ok);
3985   }
3986   if (op->halt()) {
3987     const char* str = __ code_string(op->msg());
3988     __ stop(str);
3989   } else {
3990     breakpoint();
3991   }
3992   __ bind(ok);
3993 }
3994 #endif
3995 
3996 void LIR_Assembler::membar() {
3997   // QQQ sparc TSO uses this,
3998   __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3999 }
4000 
4001 void LIR_Assembler::membar_acquire() {
4002   // No x86 machines currently require load fences
4003 }
4004 
4005 void LIR_Assembler::membar_release() {
4006   // No x86 machines currently require store fences
4007 }
4008 
4009 void LIR_Assembler::membar_loadload() {
4010   // no-op
4011   //__ membar(Assembler::Membar_mask_bits(Assembler::loadload));
4012 }
4013 
4014 void LIR_Assembler::membar_storestore() {
4015   // no-op
4016   //__ membar(Assembler::Membar_mask_bits(Assembler::storestore));
4017 }
4018 
4019 void LIR_Assembler::membar_loadstore() {
4020   // no-op
4021   //__ membar(Assembler::Membar_mask_bits(Assembler::loadstore));
4022 }
4023 
4024 void LIR_Assembler::membar_storeload() {
4025   __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
4026 }
4027 
4028 void LIR_Assembler::on_spin_wait() {
4029   __ pause ();
4030 }
4031 
4032 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
4033   assert(result_reg->is_register(), "check");
4034 #ifdef _LP64
4035   // __ get_thread(result_reg->as_register_lo());
4036   __ mov(result_reg->as_register(), r15_thread);
4037 #else
4038   __ get_thread(result_reg->as_register());
4039 #endif // _LP64
4040 }
4041 
4042 
4043 void LIR_Assembler::peephole(LIR_List*) {
4044   // do nothing for now
4045 }
4046 
4047 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
4048   assert(data == dest, "xchg/xadd uses only 2 operands");
4049 
4050   if (data->type() == T_INT) {
4051     if (code == lir_xadd) {
4052       __ lock();
4053       __ xaddl(as_Address(src->as_address_ptr()), data->as_register());
4054     } else {
4055       __ xchgl(data->as_register(), as_Address(src->as_address_ptr()));
4056     }
4057   } else if (data->is_oop()) {
4058     assert (code == lir_xchg, "xadd for oops");
4059     Register obj = data->as_register();
4060 #ifdef _LP64
4061     if (UseCompressedOops) {
4062       __ encode_heap_oop(obj);
4063       __ xchgl(obj, as_Address(src->as_address_ptr()));
4064       __ decode_heap_oop(obj);
4065     } else {
4066       __ xchgptr(obj, as_Address(src->as_address_ptr()));
4067     }
4068 #else
4069     __ xchgl(obj, as_Address(src->as_address_ptr()));
4070 #endif
4071   } else if (data->type() == T_LONG) {
4072 #ifdef _LP64
4073     assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
4074     if (code == lir_xadd) {
4075       __ lock();
4076       __ xaddq(as_Address(src->as_address_ptr()), data->as_register_lo());
4077     } else {
4078       __ xchgq(data->as_register_lo(), as_Address(src->as_address_ptr()));
4079     }
4080 #else
4081     ShouldNotReachHere();
4082 #endif
4083   } else {
4084     ShouldNotReachHere();
4085   }
4086 }
4087 
4088 #undef __