1 /*
   2  * Copyright (c) 2005, 2016, 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 "c1/c1_Compilation.hpp"
  27 #include "c1/c1_FrameMap.hpp"
  28 #include "c1/c1_Instruction.hpp"
  29 #include "c1/c1_LIRAssembler.hpp"
  30 #include "c1/c1_LIRGenerator.hpp"
  31 #include "c1/c1_Runtime1.hpp"
  32 #include "c1/c1_ValueStack.hpp"
  33 #include "ci/ciArray.hpp"
  34 #include "ci/ciObjArrayKlass.hpp"
  35 #include "ci/ciTypeArrayKlass.hpp"
  36 #include "gc/shared/c1BarrierSetCodeGen.hpp"
  37 #include "runtime/sharedRuntime.hpp"
  38 #include "runtime/stubRoutines.hpp"
  39 #include "vmreg_x86.inline.hpp"
  40 
  41 #ifdef ASSERT
  42 #define __ gen()->lir(__FILE__, __LINE__)->
  43 #else
  44 #define __ gen()->lir()->
  45 #endif
  46 
  47 // Item will be loaded into a byte register; Intel only
  48 void LIRItem::load_byte_item() {
  49   load_item();
  50   LIR_Opr res = result();
  51 
  52   if (!res->is_virtual() || !_gen->is_vreg_flag_set(res, LIRGenerator::byte_reg)) {
  53     // make sure that it is a byte register
  54     assert(!value()->type()->is_float() && !value()->type()->is_double(),
  55            "can't load floats in byte register");
  56     LIR_Opr reg = _gen->rlock_byte(T_BYTE);
  57     __ move(res, reg);
  58 
  59     _result = reg;
  60   }
  61 }
  62 
  63 
  64 void LIRItem::load_nonconstant() {
  65   LIR_Opr r = value()->operand();
  66   if (r->is_constant()) {
  67     _result = r;
  68   } else {
  69     load_item();
  70   }
  71 }
  72 
  73 //--------------------------------------------------------------
  74 //               LIRGenerator
  75 //--------------------------------------------------------------
  76 
  77 
  78 LIR_Opr LIRGenerator::exceptionOopOpr() { return FrameMap::rax_oop_opr; }
  79 LIR_Opr LIRGenerator::exceptionPcOpr()  { return FrameMap::rdx_opr; }
  80 LIR_Opr LIRGenerator::divInOpr()        { return FrameMap::rax_opr; }
  81 LIR_Opr LIRGenerator::divOutOpr()       { return FrameMap::rax_opr; }
  82 LIR_Opr LIRGenerator::remOutOpr()       { return FrameMap::rdx_opr; }
  83 LIR_Opr LIRGenerator::shiftCountOpr()   { return FrameMap::rcx_opr; }
  84 LIR_Opr LIRGenerator::syncLockOpr()     { return new_register(T_INT); }
  85 LIR_Opr LIRGenerator::syncTempOpr()     { return FrameMap::rax_opr; }
  86 LIR_Opr LIRGenerator::getThreadTemp()   { return LIR_OprFact::illegalOpr; }
  87 
  88 
  89 LIR_Opr LIRGenerator::result_register_for(ValueType* type, bool callee) {
  90   LIR_Opr opr;
  91   switch (type->tag()) {
  92     case intTag:     opr = FrameMap::rax_opr;          break;
  93     case objectTag:  opr = FrameMap::rax_oop_opr;      break;
  94     case longTag:    opr = FrameMap::long0_opr;        break;
  95     case floatTag:   opr = UseSSE >= 1 ? FrameMap::xmm0_float_opr  : FrameMap::fpu0_float_opr;  break;
  96     case doubleTag:  opr = UseSSE >= 2 ? FrameMap::xmm0_double_opr : FrameMap::fpu0_double_opr;  break;
  97 
  98     case addressTag:
  99     default: ShouldNotReachHere(); return LIR_OprFact::illegalOpr;
 100   }
 101 
 102   assert(opr->type_field() == as_OprType(as_BasicType(type)), "type mismatch");
 103   return opr;
 104 }
 105 
 106 
 107 LIR_Opr LIRGenerator::rlock_byte(BasicType type) {
 108   LIR_Opr reg = new_register(T_INT);
 109   set_vreg_flag(reg, LIRGenerator::byte_reg);
 110   return reg;
 111 }
 112 
 113 
 114 //--------- loading items into registers --------------------------------
 115 
 116 
 117 // i486 instructions can inline constants
 118 bool LIRGenerator::can_store_as_constant(Value v, BasicType type) const {
 119   if (type == T_SHORT || type == T_CHAR) {
 120     // there is no immediate move of word values in asembler_i486.?pp
 121     return false;
 122   }
 123   Constant* c = v->as_Constant();
 124   if (c && c->state_before() == NULL) {
 125     // constants of any type can be stored directly, except for
 126     // unloaded object constants.
 127     return true;
 128   }
 129   return false;
 130 }
 131 
 132 
 133 bool LIRGenerator::can_inline_as_constant(Value v) const {
 134   if (v->type()->tag() == longTag) return false;
 135   return v->type()->tag() != objectTag ||
 136     (v->type()->is_constant() && v->type()->as_ObjectType()->constant_value()->is_null_object());
 137 }
 138 
 139 
 140 bool LIRGenerator::can_inline_as_constant(LIR_Const* c) const {
 141   if (c->type() == T_LONG) return false;
 142   return c->type() != T_OBJECT || c->as_jobject() == NULL;
 143 }
 144 
 145 
 146 LIR_Opr LIRGenerator::safepoint_poll_register() {
 147   return LIR_OprFact::illegalOpr;
 148 }
 149 
 150 
 151 LIR_Address* LIRGenerator::generate_address(LIR_Opr base, LIR_Opr index,
 152                                             int shift, int disp, BasicType type) {
 153   assert(base->is_register(), "must be");
 154   if (index->is_constant()) {
 155     LIR_Const *constant = index->as_constant_ptr();
 156 #ifdef _LP64
 157     jlong c;
 158     if (constant->type() == T_INT) {
 159       c = (jlong(index->as_jint()) << shift) + disp;
 160     } else {
 161       assert(constant->type() == T_LONG, "should be");
 162       c = (index->as_jlong() << shift) + disp;
 163     }
 164     if ((jlong)((jint)c) == c) {
 165       return new LIR_Address(base, (jint)c, type);
 166     } else {
 167       LIR_Opr tmp = new_register(T_LONG);
 168       __ move(index, tmp);
 169       return new LIR_Address(base, tmp, type);
 170     }
 171 #else
 172     return new LIR_Address(base,
 173                            ((intx)(constant->as_jint()) << shift) + disp,
 174                            type);
 175 #endif
 176   } else {
 177     return new LIR_Address(base, index, (LIR_Address::Scale)shift, disp, type);
 178   }
 179 }
 180 
 181 
 182 LIR_Address* LIRGenerator::emit_array_address(LIR_Opr array_opr, LIR_Opr index_opr,
 183                                               BasicType type, bool needs_card_mark) {
 184   int offset_in_bytes = arrayOopDesc::base_offset_in_bytes(type);
 185 
 186   LIR_Address* addr;
 187   if (index_opr->is_constant()) {
 188     int elem_size = type2aelembytes(type);
 189     addr = new LIR_Address(array_opr,
 190                            offset_in_bytes + (intx)(index_opr->as_jint()) * elem_size, type);
 191   } else {
 192 #ifdef _LP64
 193     if (index_opr->type() == T_INT) {
 194       LIR_Opr tmp = new_register(T_LONG);
 195       __ convert(Bytecodes::_i2l, index_opr, tmp);
 196       index_opr = tmp;
 197     }
 198 #endif // _LP64
 199     addr =  new LIR_Address(array_opr,
 200                             index_opr,
 201                             LIR_Address::scale(type),
 202                             offset_in_bytes, type);
 203   }
 204   if (needs_card_mark) {
 205     // This store will need a precise card mark, so go ahead and
 206     // compute the full adddres instead of computing once for the
 207     // store and again for the card mark.
 208     LIR_Opr tmp = new_pointer_register();
 209     __ leal(LIR_OprFact::address(addr), tmp);
 210     return new LIR_Address(tmp, type);
 211   } else {
 212     return addr;
 213   }
 214 }
 215 
 216 
 217 LIR_Opr LIRGenerator::load_immediate(int x, BasicType type) {
 218   LIR_Opr r = NULL;
 219   if (type == T_LONG) {
 220     r = LIR_OprFact::longConst(x);
 221   } else if (type == T_INT) {
 222     r = LIR_OprFact::intConst(x);
 223   } else {
 224     ShouldNotReachHere();
 225   }
 226   return r;
 227 }
 228 
 229 void LIRGenerator::increment_counter(address counter, BasicType type, int step) {
 230   LIR_Opr pointer = new_pointer_register();
 231   __ move(LIR_OprFact::intptrConst(counter), pointer);
 232   LIR_Address* addr = new LIR_Address(pointer, type);
 233   increment_counter(addr, step);
 234 }
 235 
 236 
 237 void LIRGenerator::increment_counter(LIR_Address* addr, int step) {
 238   __ add((LIR_Opr)addr, LIR_OprFact::intConst(step), (LIR_Opr)addr);
 239 }
 240 
 241 void LIRGenerator::cmp_mem_int(LIR_Condition condition, LIR_Opr base, int disp, int c, CodeEmitInfo* info) {
 242   __ cmp_mem_int(condition, base, disp, c, info);
 243 }
 244 
 245 
 246 void LIRGenerator::cmp_reg_mem(LIR_Condition condition, LIR_Opr reg, LIR_Opr base, int disp, BasicType type, CodeEmitInfo* info) {
 247   __ cmp_reg_mem(condition, reg, new LIR_Address(base, disp, type), info);
 248 }
 249 
 250 
 251 void LIRGenerator::cmp_reg_mem(LIR_Condition condition, LIR_Opr reg, LIR_Opr base, LIR_Opr disp, BasicType type, CodeEmitInfo* info) {
 252   __ cmp_reg_mem(condition, reg, new LIR_Address(base, disp, type), info);
 253 }
 254 
 255 
 256 bool LIRGenerator::strength_reduce_multiply(LIR_Opr left, int c, LIR_Opr result, LIR_Opr tmp) {
 257   if (tmp->is_valid()) {
 258     if (is_power_of_2(c + 1)) {
 259       __ move(left, tmp);
 260       __ shift_left(left, log2_intptr(c + 1), left);
 261       __ sub(left, tmp, result);
 262       return true;
 263     } else if (is_power_of_2(c - 1)) {
 264       __ move(left, tmp);
 265       __ shift_left(left, log2_intptr(c - 1), left);
 266       __ add(left, tmp, result);
 267       return true;
 268     }
 269   }
 270   return false;
 271 }
 272 
 273 
 274 void LIRGenerator::store_stack_parameter (LIR_Opr item, ByteSize offset_from_sp) {
 275   BasicType type = item->type();
 276   __ store(item, new LIR_Address(FrameMap::rsp_opr, in_bytes(offset_from_sp), type));
 277 }
 278 
 279 void LIRGenerator::array_store_check(LIR_Opr value, LIR_Opr array, CodeEmitInfo* store_check_info, ciMethod* profiled_method, int profiled_bci) {
 280   LIR_Opr tmp1 = new_register(objectType);
 281   LIR_Opr tmp2 = new_register(objectType);
 282   LIR_Opr tmp3 = new_register(objectType);
 283   __ store_check(value, array, tmp1, tmp2, tmp3, store_check_info, profiled_method, profiled_bci);
 284 }
 285 
 286 //----------------------------------------------------------------------
 287 //             visitor functions
 288 //----------------------------------------------------------------------
 289 
 290 void LIRGenerator::do_MonitorEnter(MonitorEnter* x) {
 291   assert(x->is_pinned(),"");
 292   LIRItem obj(x->obj(), this);
 293   obj.load_item();
 294 
 295   set_no_result(x);
 296 
 297   // "lock" stores the address of the monitor stack slot, so this is not an oop
 298   LIR_Opr lock = new_register(T_INT);
 299   // Need a scratch register for biased locking on x86
 300   LIR_Opr scratch = LIR_OprFact::illegalOpr;
 301   if (UseBiasedLocking) {
 302     scratch = new_register(T_INT);
 303   }
 304 
 305   CodeEmitInfo* info_for_exception = NULL;
 306   if (x->needs_null_check()) {
 307     info_for_exception = state_for(x);
 308   }
 309   // this CodeEmitInfo must not have the xhandlers because here the
 310   // object is already locked (xhandlers expect object to be unlocked)
 311   CodeEmitInfo* info = state_for(x, x->state(), true);
 312   monitor_enter(obj.result(), lock, syncTempOpr(), scratch,
 313                         x->monitor_no(), info_for_exception, info);
 314 }
 315 
 316 
 317 void LIRGenerator::do_MonitorExit(MonitorExit* x) {
 318   assert(x->is_pinned(),"");
 319 
 320   LIRItem obj(x->obj(), this);
 321   obj.dont_load_item();
 322 
 323   LIR_Opr lock = new_register(T_INT);
 324   LIR_Opr obj_temp = new_register(T_INT);
 325   set_no_result(x);
 326   monitor_exit(obj_temp, lock, syncTempOpr(), LIR_OprFact::illegalOpr, x->monitor_no());
 327 }
 328 
 329 
 330 // _ineg, _lneg, _fneg, _dneg
 331 void LIRGenerator::do_NegateOp(NegateOp* x) {
 332   LIRItem value(x->x(), this);
 333   value.set_destroys_register();
 334   value.load_item();
 335   LIR_Opr reg = rlock(x);
 336   __ negate(value.result(), reg);
 337 
 338   set_result(x, round_item(reg));
 339 }
 340 
 341 
 342 // for  _fadd, _fmul, _fsub, _fdiv, _frem
 343 //      _dadd, _dmul, _dsub, _ddiv, _drem
 344 void LIRGenerator::do_ArithmeticOp_FPU(ArithmeticOp* x) {
 345   LIRItem left(x->x(),  this);
 346   LIRItem right(x->y(), this);
 347   LIRItem* left_arg  = &left;
 348   LIRItem* right_arg = &right;
 349   assert(!left.is_stack() || !right.is_stack(), "can't both be memory operands");
 350   bool must_load_both = (x->op() == Bytecodes::_frem || x->op() == Bytecodes::_drem);
 351   if (left.is_register() || x->x()->type()->is_constant() || must_load_both) {
 352     left.load_item();
 353   } else {
 354     left.dont_load_item();
 355   }
 356 
 357   // do not load right operand if it is a constant.  only 0 and 1 are
 358   // loaded because there are special instructions for loading them
 359   // without memory access (not needed for SSE2 instructions)
 360   bool must_load_right = false;
 361   if (right.is_constant()) {
 362     LIR_Const* c = right.result()->as_constant_ptr();
 363     assert(c != NULL, "invalid constant");
 364     assert(c->type() == T_FLOAT || c->type() == T_DOUBLE, "invalid type");
 365 
 366     if (c->type() == T_FLOAT) {
 367       must_load_right = UseSSE < 1 && (c->is_one_float() || c->is_zero_float());
 368     } else {
 369       must_load_right = UseSSE < 2 && (c->is_one_double() || c->is_zero_double());
 370     }
 371   }
 372 
 373   if (must_load_both) {
 374     // frem and drem destroy also right operand, so move it to a new register
 375     right.set_destroys_register();
 376     right.load_item();
 377   } else if (right.is_register() || must_load_right) {
 378     right.load_item();
 379   } else {
 380     right.dont_load_item();
 381   }
 382   LIR_Opr reg = rlock(x);
 383   LIR_Opr tmp = LIR_OprFact::illegalOpr;
 384   if (x->is_strictfp() && (x->op() == Bytecodes::_dmul || x->op() == Bytecodes::_ddiv)) {
 385     tmp = new_register(T_DOUBLE);
 386   }
 387 
 388   if ((UseSSE >= 1 && x->op() == Bytecodes::_frem) || (UseSSE >= 2 && x->op() == Bytecodes::_drem)) {
 389     // special handling for frem and drem: no SSE instruction, so must use FPU with temporary fpu stack slots
 390     LIR_Opr fpu0, fpu1;
 391     if (x->op() == Bytecodes::_frem) {
 392       fpu0 = LIR_OprFact::single_fpu(0);
 393       fpu1 = LIR_OprFact::single_fpu(1);
 394     } else {
 395       fpu0 = LIR_OprFact::double_fpu(0);
 396       fpu1 = LIR_OprFact::double_fpu(1);
 397     }
 398     __ move(right.result(), fpu1); // order of left and right operand is important!
 399     __ move(left.result(), fpu0);
 400     __ rem (fpu0, fpu1, fpu0);
 401     __ move(fpu0, reg);
 402 
 403   } else {
 404     arithmetic_op_fpu(x->op(), reg, left.result(), right.result(), x->is_strictfp(), tmp);
 405   }
 406 
 407   set_result(x, round_item(reg));
 408 }
 409 
 410 
 411 // for  _ladd, _lmul, _lsub, _ldiv, _lrem
 412 void LIRGenerator::do_ArithmeticOp_Long(ArithmeticOp* x) {
 413   if (x->op() == Bytecodes::_ldiv || x->op() == Bytecodes::_lrem ) {
 414     // long division is implemented as a direct call into the runtime
 415     LIRItem left(x->x(), this);
 416     LIRItem right(x->y(), this);
 417 
 418     // the check for division by zero destroys the right operand
 419     right.set_destroys_register();
 420 
 421     BasicTypeList signature(2);
 422     signature.append(T_LONG);
 423     signature.append(T_LONG);
 424     CallingConvention* cc = frame_map()->c_calling_convention(&signature);
 425 
 426     // check for division by zero (destroys registers of right operand!)
 427     CodeEmitInfo* info = state_for(x);
 428 
 429     const LIR_Opr result_reg = result_register_for(x->type());
 430     left.load_item_force(cc->at(1));
 431     right.load_item();
 432 
 433     __ move(right.result(), cc->at(0));
 434 
 435     __ cmp(lir_cond_equal, right.result(), LIR_OprFact::longConst(0));
 436     __ branch(lir_cond_equal, T_LONG, new DivByZeroStub(info));
 437 
 438     address entry = NULL;
 439     switch (x->op()) {
 440     case Bytecodes::_lrem:
 441       entry = CAST_FROM_FN_PTR(address, SharedRuntime::lrem);
 442       break; // check if dividend is 0 is done elsewhere
 443     case Bytecodes::_ldiv:
 444       entry = CAST_FROM_FN_PTR(address, SharedRuntime::ldiv);
 445       break; // check if dividend is 0 is done elsewhere
 446     case Bytecodes::_lmul:
 447       entry = CAST_FROM_FN_PTR(address, SharedRuntime::lmul);
 448       break;
 449     default:
 450       ShouldNotReachHere();
 451     }
 452 
 453     LIR_Opr result = rlock_result(x);
 454     __ call_runtime_leaf(entry, getThreadTemp(), result_reg, cc->args());
 455     __ move(result_reg, result);
 456   } else if (x->op() == Bytecodes::_lmul) {
 457     // missing test if instr is commutative and if we should swap
 458     LIRItem left(x->x(), this);
 459     LIRItem right(x->y(), this);
 460 
 461     // right register is destroyed by the long mul, so it must be
 462     // copied to a new register.
 463     right.set_destroys_register();
 464 
 465     left.load_item();
 466     right.load_item();
 467 
 468     LIR_Opr reg = FrameMap::long0_opr;
 469     arithmetic_op_long(x->op(), reg, left.result(), right.result(), NULL);
 470     LIR_Opr result = rlock_result(x);
 471     __ move(reg, result);
 472   } else {
 473     // missing test if instr is commutative and if we should swap
 474     LIRItem left(x->x(), this);
 475     LIRItem right(x->y(), this);
 476 
 477     left.load_item();
 478     // don't load constants to save register
 479     right.load_nonconstant();
 480     rlock_result(x);
 481     arithmetic_op_long(x->op(), x->operand(), left.result(), right.result(), NULL);
 482   }
 483 }
 484 
 485 
 486 
 487 // for: _iadd, _imul, _isub, _idiv, _irem
 488 void LIRGenerator::do_ArithmeticOp_Int(ArithmeticOp* x) {
 489   if (x->op() == Bytecodes::_idiv || x->op() == Bytecodes::_irem) {
 490     // The requirements for division and modulo
 491     // input : rax,: dividend                         min_int
 492     //         reg: divisor   (may not be rax,/rdx)   -1
 493     //
 494     // output: rax,: quotient  (= rax, idiv reg)       min_int
 495     //         rdx: remainder (= rax, irem reg)       0
 496 
 497     // rax, and rdx will be destroyed
 498 
 499     // Note: does this invalidate the spec ???
 500     LIRItem right(x->y(), this);
 501     LIRItem left(x->x() , this);   // visit left second, so that the is_register test is valid
 502 
 503     // call state_for before load_item_force because state_for may
 504     // force the evaluation of other instructions that are needed for
 505     // correct debug info.  Otherwise the live range of the fix
 506     // register might be too long.
 507     CodeEmitInfo* info = state_for(x);
 508 
 509     left.load_item_force(divInOpr());
 510 
 511     right.load_item();
 512 
 513     LIR_Opr result = rlock_result(x);
 514     LIR_Opr result_reg;
 515     if (x->op() == Bytecodes::_idiv) {
 516       result_reg = divOutOpr();
 517     } else {
 518       result_reg = remOutOpr();
 519     }
 520 
 521     if (!ImplicitDiv0Checks) {
 522       __ cmp(lir_cond_equal, right.result(), LIR_OprFact::intConst(0));
 523       __ branch(lir_cond_equal, T_INT, new DivByZeroStub(info));
 524     }
 525     LIR_Opr tmp = FrameMap::rdx_opr; // idiv and irem use rdx in their implementation
 526     if (x->op() == Bytecodes::_irem) {
 527       __ irem(left.result(), right.result(), result_reg, tmp, info);
 528     } else if (x->op() == Bytecodes::_idiv) {
 529       __ idiv(left.result(), right.result(), result_reg, tmp, info);
 530     } else {
 531       ShouldNotReachHere();
 532     }
 533 
 534     __ move(result_reg, result);
 535   } else {
 536     // missing test if instr is commutative and if we should swap
 537     LIRItem left(x->x(),  this);
 538     LIRItem right(x->y(), this);
 539     LIRItem* left_arg = &left;
 540     LIRItem* right_arg = &right;
 541     if (x->is_commutative() && left.is_stack() && right.is_register()) {
 542       // swap them if left is real stack (or cached) and right is real register(not cached)
 543       left_arg = &right;
 544       right_arg = &left;
 545     }
 546 
 547     left_arg->load_item();
 548 
 549     // do not need to load right, as we can handle stack and constants
 550     if (x->op() == Bytecodes::_imul ) {
 551       // check if we can use shift instead
 552       bool use_constant = false;
 553       bool use_tmp = false;
 554       if (right_arg->is_constant()) {
 555         int iconst = right_arg->get_jint_constant();
 556         if (iconst > 0) {
 557           if (is_power_of_2(iconst)) {
 558             use_constant = true;
 559           } else if (is_power_of_2(iconst - 1) || is_power_of_2(iconst + 1)) {
 560             use_constant = true;
 561             use_tmp = true;
 562           }
 563         }
 564       }
 565       if (use_constant) {
 566         right_arg->dont_load_item();
 567       } else {
 568         right_arg->load_item();
 569       }
 570       LIR_Opr tmp = LIR_OprFact::illegalOpr;
 571       if (use_tmp) {
 572         tmp = new_register(T_INT);
 573       }
 574       rlock_result(x);
 575 
 576       arithmetic_op_int(x->op(), x->operand(), left_arg->result(), right_arg->result(), tmp);
 577     } else {
 578       right_arg->dont_load_item();
 579       rlock_result(x);
 580       LIR_Opr tmp = LIR_OprFact::illegalOpr;
 581       arithmetic_op_int(x->op(), x->operand(), left_arg->result(), right_arg->result(), tmp);
 582     }
 583   }
 584 }
 585 
 586 
 587 void LIRGenerator::do_ArithmeticOp(ArithmeticOp* x) {
 588   // when an operand with use count 1 is the left operand, then it is
 589   // likely that no move for 2-operand-LIR-form is necessary
 590   if (x->is_commutative() && x->y()->as_Constant() == NULL && x->x()->use_count() > x->y()->use_count()) {
 591     x->swap_operands();
 592   }
 593 
 594   ValueTag tag = x->type()->tag();
 595   assert(x->x()->type()->tag() == tag && x->y()->type()->tag() == tag, "wrong parameters");
 596   switch (tag) {
 597     case floatTag:
 598     case doubleTag:  do_ArithmeticOp_FPU(x);  return;
 599     case longTag:    do_ArithmeticOp_Long(x); return;
 600     case intTag:     do_ArithmeticOp_Int(x);  return;
 601   }
 602   ShouldNotReachHere();
 603 }
 604 
 605 
 606 // _ishl, _lshl, _ishr, _lshr, _iushr, _lushr
 607 void LIRGenerator::do_ShiftOp(ShiftOp* x) {
 608   // count must always be in rcx
 609   LIRItem value(x->x(), this);
 610   LIRItem count(x->y(), this);
 611 
 612   ValueTag elemType = x->type()->tag();
 613   bool must_load_count = !count.is_constant() || elemType == longTag;
 614   if (must_load_count) {
 615     // count for long must be in register
 616     count.load_item_force(shiftCountOpr());
 617   } else {
 618     count.dont_load_item();
 619   }
 620   value.load_item();
 621   LIR_Opr reg = rlock_result(x);
 622 
 623   shift_op(x->op(), reg, value.result(), count.result(), LIR_OprFact::illegalOpr);
 624 }
 625 
 626 
 627 // _iand, _land, _ior, _lor, _ixor, _lxor
 628 void LIRGenerator::do_LogicOp(LogicOp* x) {
 629   // when an operand with use count 1 is the left operand, then it is
 630   // likely that no move for 2-operand-LIR-form is necessary
 631   if (x->is_commutative() && x->y()->as_Constant() == NULL && x->x()->use_count() > x->y()->use_count()) {
 632     x->swap_operands();
 633   }
 634 
 635   LIRItem left(x->x(), this);
 636   LIRItem right(x->y(), this);
 637 
 638   left.load_item();
 639   right.load_nonconstant();
 640   LIR_Opr reg = rlock_result(x);
 641 
 642   logic_op(x->op(), reg, left.result(), right.result());
 643 }
 644 
 645 
 646 
 647 // _lcmp, _fcmpl, _fcmpg, _dcmpl, _dcmpg
 648 void LIRGenerator::do_CompareOp(CompareOp* x) {
 649   LIRItem left(x->x(), this);
 650   LIRItem right(x->y(), this);
 651   ValueTag tag = x->x()->type()->tag();
 652   if (tag == longTag) {
 653     left.set_destroys_register();
 654   }
 655   left.load_item();
 656   right.load_item();
 657   LIR_Opr reg = rlock_result(x);
 658 
 659   if (x->x()->type()->is_float_kind()) {
 660     Bytecodes::Code code = x->op();
 661     __ fcmp2int(left.result(), right.result(), reg, (code == Bytecodes::_fcmpl || code == Bytecodes::_dcmpl));
 662   } else if (x->x()->type()->tag() == longTag) {
 663     __ lcmp2int(left.result(), right.result(), reg);
 664   } else {
 665     Unimplemented();
 666   }
 667 }
 668 
 669 LIR_Opr LIRGenerator::cas(BasicType type, LIR_Opr addr, LIRItem& cmp_value, LIRItem& new_value) {
 670   LIR_Opr result = new_register(T_INT);
 671   LIR_Opr ill = LIR_OprFact::illegalOpr;  // for convenience
 672   if (type == T_OBJECT || type == T_ARRAY) {
 673     cmp_value.load_item_force(FrameMap::rax_oop_opr);
 674     new_value.load_item();
 675     __ cas_obj(addr, cmp_value.result(), new_value.result(), ill, ill);
 676   } else if (type == T_INT) {
 677     cmp_value.load_item_force(FrameMap::rax_opr);
 678     new_value.load_item();
 679     __ cas_int(addr, cmp_value.result(), new_value.result(), ill, ill);
 680   } else if (type == T_LONG) {
 681     cmp_value.load_item_force(FrameMap::long0_opr);
 682     new_value.load_item_force(FrameMap::long1_opr);
 683     __ cas_long(addr, cmp_value.result(), new_value.result(), ill, ill);
 684   } else {
 685     Unimplemented();
 686   }
 687   __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0),
 688            result, as_BasicType(new_value.type()));
 689   return result;
 690 }
 691 
 692 LIR_Opr LIRGenerator::swap(BasicType type, LIR_Opr addr, LIRItem& value) {
 693   bool is_oop = type == T_OBJECT || type == T_ARRAY;
 694   LIR_Opr result = new_register(type);
 695   value.load_item();
 696   // Because we want a 2-arg form of xchg and xadd
 697   __ move(value.result(), result);
 698   assert(type == T_INT || is_oop LP64_ONLY( || type == T_LONG ), "unexpected type");
 699   __ xchg(addr, result, result, LIR_OprFact::illegalOpr);
 700   return result;
 701 }
 702 
 703 LIR_Opr LIRGenerator::add(BasicType type, LIR_Opr addr, LIRItem& value) {
 704   LIR_Opr result = new_register(type);
 705   value.load_item();
 706   // Because we want a 2-arg form of xchg and xadd
 707   __ move(value.result(), result);
 708   assert(type == T_INT LP64_ONLY( || type == T_LONG ), "unexpected type");
 709   __ xadd(addr, result, result, LIR_OprFact::illegalOpr);
 710   return result;
 711 }
 712 
 713 void LIRGenerator::do_FmaIntrinsic(Intrinsic* x) {
 714   assert(x->number_of_arguments() == 3, "wrong type");
 715   assert(UseFMA, "Needs FMA instructions support.");
 716   LIRItem value(x->argument_at(0), this);
 717   LIRItem value1(x->argument_at(1), this);
 718   LIRItem value2(x->argument_at(2), this);
 719 
 720   value2.set_destroys_register();
 721 
 722   value.load_item();
 723   value1.load_item();
 724   value2.load_item();
 725 
 726   LIR_Opr calc_input = value.result();
 727   LIR_Opr calc_input1 = value1.result();
 728   LIR_Opr calc_input2 = value2.result();
 729   LIR_Opr calc_result = rlock_result(x);
 730 
 731   switch (x->id()) {
 732   case vmIntrinsics::_fmaD:   __ fmad(calc_input, calc_input1, calc_input2, calc_result); break;
 733   case vmIntrinsics::_fmaF:   __ fmaf(calc_input, calc_input1, calc_input2, calc_result); break;
 734   default:                    ShouldNotReachHere();
 735   }
 736 
 737 }
 738 
 739 
 740 void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
 741   assert(x->number_of_arguments() == 1 || (x->number_of_arguments() == 2 && x->id() == vmIntrinsics::_dpow), "wrong type");
 742 
 743   if (x->id() == vmIntrinsics::_dexp || x->id() == vmIntrinsics::_dlog ||
 744       x->id() == vmIntrinsics::_dpow || x->id() == vmIntrinsics::_dcos ||
 745       x->id() == vmIntrinsics::_dsin || x->id() == vmIntrinsics::_dtan ||
 746       x->id() == vmIntrinsics::_dlog10) {
 747     do_LibmIntrinsic(x);
 748     return;
 749   }
 750 
 751   LIRItem value(x->argument_at(0), this);
 752 
 753   bool use_fpu = false;
 754   if (UseSSE < 2) {
 755     value.set_destroys_register();
 756   }
 757   value.load_item();
 758 
 759   LIR_Opr calc_input = value.result();
 760   LIR_Opr calc_result = rlock_result(x);
 761 
 762   switch(x->id()) {
 763     case vmIntrinsics::_dabs:   __ abs  (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
 764     case vmIntrinsics::_dsqrt:  __ sqrt (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
 765     default:                    ShouldNotReachHere();
 766   }
 767 
 768   if (use_fpu) {
 769     __ move(calc_result, x->operand());
 770   }
 771 }
 772 
 773 void LIRGenerator::do_LibmIntrinsic(Intrinsic* x) {
 774   LIRItem value(x->argument_at(0), this);
 775   value.set_destroys_register();
 776 
 777   LIR_Opr calc_result = rlock_result(x);
 778   LIR_Opr result_reg = result_register_for(x->type());
 779 
 780   CallingConvention* cc = NULL;
 781 
 782   if (x->id() == vmIntrinsics::_dpow) {
 783     LIRItem value1(x->argument_at(1), this);
 784 
 785     value1.set_destroys_register();
 786 
 787     BasicTypeList signature(2);
 788     signature.append(T_DOUBLE);
 789     signature.append(T_DOUBLE);
 790     cc = frame_map()->c_calling_convention(&signature);
 791     value.load_item_force(cc->at(0));
 792     value1.load_item_force(cc->at(1));
 793   } else {
 794     BasicTypeList signature(1);
 795     signature.append(T_DOUBLE);
 796     cc = frame_map()->c_calling_convention(&signature);
 797     value.load_item_force(cc->at(0));
 798   }
 799 
 800 #ifndef _LP64
 801   LIR_Opr tmp = FrameMap::fpu0_double_opr;
 802   result_reg = tmp;
 803   switch(x->id()) {
 804     case vmIntrinsics::_dexp:
 805       if (StubRoutines::dexp() != NULL) {
 806         __ call_runtime_leaf(StubRoutines::dexp(), getThreadTemp(), result_reg, cc->args());
 807       } else {
 808         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dexp), getThreadTemp(), result_reg, cc->args());
 809       }
 810       break;
 811     case vmIntrinsics::_dlog:
 812       if (StubRoutines::dlog() != NULL) {
 813         __ call_runtime_leaf(StubRoutines::dlog(), getThreadTemp(), result_reg, cc->args());
 814       } else {
 815         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog), getThreadTemp(), result_reg, cc->args());
 816       }
 817       break;
 818     case vmIntrinsics::_dlog10:
 819       if (StubRoutines::dlog10() != NULL) {
 820        __ call_runtime_leaf(StubRoutines::dlog10(), getThreadTemp(), result_reg, cc->args());
 821       } else {
 822         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), getThreadTemp(), result_reg, cc->args());
 823       }
 824       break;
 825     case vmIntrinsics::_dpow:
 826       if (StubRoutines::dpow() != NULL) {
 827         __ call_runtime_leaf(StubRoutines::dpow(), getThreadTemp(), result_reg, cc->args());
 828       } else {
 829         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dpow), getThreadTemp(), result_reg, cc->args());
 830       }
 831       break;
 832     case vmIntrinsics::_dsin:
 833       if (VM_Version::supports_sse2() && StubRoutines::dsin() != NULL) {
 834         __ call_runtime_leaf(StubRoutines::dsin(), getThreadTemp(), result_reg, cc->args());
 835       } else {
 836         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), getThreadTemp(), result_reg, cc->args());
 837       }
 838       break;
 839     case vmIntrinsics::_dcos:
 840       if (VM_Version::supports_sse2() && StubRoutines::dcos() != NULL) {
 841         __ call_runtime_leaf(StubRoutines::dcos(), getThreadTemp(), result_reg, cc->args());
 842       } else {
 843         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), getThreadTemp(), result_reg, cc->args());
 844       }
 845       break;
 846     case vmIntrinsics::_dtan:
 847       if (StubRoutines::dtan() != NULL) {
 848         __ call_runtime_leaf(StubRoutines::dtan(), getThreadTemp(), result_reg, cc->args());
 849       } else {
 850         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), getThreadTemp(), result_reg, cc->args());
 851       }
 852       break;
 853     default:  ShouldNotReachHere();
 854   }
 855 #else
 856   switch (x->id()) {
 857     case vmIntrinsics::_dexp:
 858       if (StubRoutines::dexp() != NULL) {
 859         __ call_runtime_leaf(StubRoutines::dexp(), getThreadTemp(), result_reg, cc->args());
 860       } else {
 861         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dexp), getThreadTemp(), result_reg, cc->args());
 862       }
 863       break;
 864     case vmIntrinsics::_dlog:
 865       if (StubRoutines::dlog() != NULL) {
 866       __ call_runtime_leaf(StubRoutines::dlog(), getThreadTemp(), result_reg, cc->args());
 867       } else {
 868         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog), getThreadTemp(), result_reg, cc->args());
 869       }
 870       break;
 871     case vmIntrinsics::_dlog10:
 872       if (StubRoutines::dlog10() != NULL) {
 873       __ call_runtime_leaf(StubRoutines::dlog10(), getThreadTemp(), result_reg, cc->args());
 874       } else {
 875         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), getThreadTemp(), result_reg, cc->args());
 876       }
 877       break;
 878     case vmIntrinsics::_dpow:
 879        if (StubRoutines::dpow() != NULL) {
 880       __ call_runtime_leaf(StubRoutines::dpow(), getThreadTemp(), result_reg, cc->args());
 881       } else {
 882         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dpow), getThreadTemp(), result_reg, cc->args());
 883       }
 884       break;
 885     case vmIntrinsics::_dsin:
 886       if (StubRoutines::dsin() != NULL) {
 887         __ call_runtime_leaf(StubRoutines::dsin(), getThreadTemp(), result_reg, cc->args());
 888       } else {
 889         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), getThreadTemp(), result_reg, cc->args());
 890       }
 891       break;
 892     case vmIntrinsics::_dcos:
 893       if (StubRoutines::dcos() != NULL) {
 894         __ call_runtime_leaf(StubRoutines::dcos(), getThreadTemp(), result_reg, cc->args());
 895       } else {
 896         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), getThreadTemp(), result_reg, cc->args());
 897       }
 898       break;
 899     case vmIntrinsics::_dtan:
 900        if (StubRoutines::dtan() != NULL) {
 901       __ call_runtime_leaf(StubRoutines::dtan(), getThreadTemp(), result_reg, cc->args());
 902       } else {
 903         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), getThreadTemp(), result_reg, cc->args());
 904       }
 905       break;
 906     default:  ShouldNotReachHere();
 907   }
 908 #endif // _LP64
 909   __ move(result_reg, calc_result);
 910 }
 911 
 912 void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
 913   assert(x->number_of_arguments() == 5, "wrong type");
 914 
 915   // Make all state_for calls early since they can emit code
 916   CodeEmitInfo* info = state_for(x, x->state());
 917 
 918   LIRItem src(x->argument_at(0), this);
 919   LIRItem src_pos(x->argument_at(1), this);
 920   LIRItem dst(x->argument_at(2), this);
 921   LIRItem dst_pos(x->argument_at(3), this);
 922   LIRItem length(x->argument_at(4), this);
 923 
 924   // operands for arraycopy must use fixed registers, otherwise
 925   // LinearScan will fail allocation (because arraycopy always needs a
 926   // call)
 927 
 928 #ifndef _LP64
 929   src.load_item_force     (FrameMap::rcx_oop_opr);
 930   src_pos.load_item_force (FrameMap::rdx_opr);
 931   dst.load_item_force     (FrameMap::rax_oop_opr);
 932   dst_pos.load_item_force (FrameMap::rbx_opr);
 933   length.load_item_force  (FrameMap::rdi_opr);
 934   LIR_Opr tmp =           (FrameMap::rsi_opr);
 935 #else
 936 
 937   // The java calling convention will give us enough registers
 938   // so that on the stub side the args will be perfect already.
 939   // On the other slow/special case side we call C and the arg
 940   // positions are not similar enough to pick one as the best.
 941   // Also because the java calling convention is a "shifted" version
 942   // of the C convention we can process the java args trivially into C
 943   // args without worry of overwriting during the xfer
 944 
 945   src.load_item_force     (FrameMap::as_oop_opr(j_rarg0));
 946   src_pos.load_item_force (FrameMap::as_opr(j_rarg1));
 947   dst.load_item_force     (FrameMap::as_oop_opr(j_rarg2));
 948   dst_pos.load_item_force (FrameMap::as_opr(j_rarg3));
 949   length.load_item_force  (FrameMap::as_opr(j_rarg4));
 950 
 951   LIR_Opr tmp =           FrameMap::as_opr(j_rarg5);
 952 #endif // LP64
 953 
 954   set_no_result(x);
 955 
 956   int flags;
 957   ciArrayKlass* expected_type;
 958   arraycopy_helper(x, &flags, &expected_type);
 959 
 960   __ arraycopy(src.result(), src_pos.result(), dst.result(), dst_pos.result(), length.result(), tmp, expected_type, flags, info); // does add_safepoint
 961 }
 962 
 963 void LIRGenerator::do_update_CRC32(Intrinsic* x) {
 964   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
 965   // Make all state_for calls early since they can emit code
 966   LIR_Opr result = rlock_result(x);
 967   int flags = 0;
 968   switch (x->id()) {
 969     case vmIntrinsics::_updateCRC32: {
 970       LIRItem crc(x->argument_at(0), this);
 971       LIRItem val(x->argument_at(1), this);
 972       // val is destroyed by update_crc32
 973       val.set_destroys_register();
 974       crc.load_item();
 975       val.load_item();
 976       __ update_crc32(crc.result(), val.result(), result);
 977       break;
 978     }
 979     case vmIntrinsics::_updateBytesCRC32:
 980     case vmIntrinsics::_updateByteBufferCRC32: {
 981       bool is_updateBytes = (x->id() == vmIntrinsics::_updateBytesCRC32);
 982 
 983       LIRItem crc(x->argument_at(0), this);
 984       LIRItem buf(x->argument_at(1), this);
 985       LIRItem off(x->argument_at(2), this);
 986       LIRItem len(x->argument_at(3), this);
 987       buf.load_item();
 988       off.load_nonconstant();
 989 
 990       LIR_Opr index = off.result();
 991       int offset = is_updateBytes ? arrayOopDesc::base_offset_in_bytes(T_BYTE) : 0;
 992       if(off.result()->is_constant()) {
 993         index = LIR_OprFact::illegalOpr;
 994        offset += off.result()->as_jint();
 995       }
 996       LIR_Opr base_op = buf.result();
 997 
 998 #ifndef _LP64
 999       if (!is_updateBytes) { // long b raw address
1000          base_op = new_register(T_INT);
1001          __ convert(Bytecodes::_l2i, buf.result(), base_op);
1002       }
1003 #else
1004       if (index->is_valid()) {
1005         LIR_Opr tmp = new_register(T_LONG);
1006         __ convert(Bytecodes::_i2l, index, tmp);
1007         index = tmp;
1008       }
1009 #endif
1010 
1011       LIR_Address* a = new LIR_Address(base_op,
1012                                        index,
1013                                        offset,
1014                                        T_BYTE);
1015       BasicTypeList signature(3);
1016       signature.append(T_INT);
1017       signature.append(T_ADDRESS);
1018       signature.append(T_INT);
1019       CallingConvention* cc = frame_map()->c_calling_convention(&signature);
1020       const LIR_Opr result_reg = result_register_for(x->type());
1021 
1022       LIR_Opr addr = new_pointer_register();
1023       __ leal(LIR_OprFact::address(a), addr);
1024 
1025       crc.load_item_force(cc->at(0));
1026       __ move(addr, cc->at(1));
1027       len.load_item_force(cc->at(2));
1028 
1029       __ call_runtime_leaf(StubRoutines::updateBytesCRC32(), getThreadTemp(), result_reg, cc->args());
1030       __ move(result_reg, result);
1031 
1032       break;
1033     }
1034     default: {
1035       ShouldNotReachHere();
1036     }
1037   }
1038 }
1039 
1040 void LIRGenerator::do_update_CRC32C(Intrinsic* x) {
1041   Unimplemented();
1042 }
1043 
1044 void LIRGenerator::do_vectorizedMismatch(Intrinsic* x) {
1045   assert(UseVectorizedMismatchIntrinsic, "need AVX instruction support");
1046 
1047   // Make all state_for calls early since they can emit code
1048   LIR_Opr result = rlock_result(x);
1049 
1050   LIRItem a(x->argument_at(0), this); // Object
1051   LIRItem aOffset(x->argument_at(1), this); // long
1052   LIRItem b(x->argument_at(2), this); // Object
1053   LIRItem bOffset(x->argument_at(3), this); // long
1054   LIRItem length(x->argument_at(4), this); // int
1055   LIRItem log2ArrayIndexScale(x->argument_at(5), this); // int
1056 
1057   a.load_item();
1058   aOffset.load_nonconstant();
1059   b.load_item();
1060   bOffset.load_nonconstant();
1061 
1062   long constant_aOffset = 0;
1063   LIR_Opr result_aOffset = aOffset.result();
1064   if (result_aOffset->is_constant()) {
1065     constant_aOffset = result_aOffset->as_jlong();
1066     result_aOffset = LIR_OprFact::illegalOpr;
1067   }
1068   LIR_Opr result_a = a.result();
1069 
1070   long constant_bOffset = 0;
1071   LIR_Opr result_bOffset = bOffset.result();
1072   if (result_bOffset->is_constant()) {
1073     constant_bOffset = result_bOffset->as_jlong();
1074     result_bOffset = LIR_OprFact::illegalOpr;
1075   }
1076   LIR_Opr result_b = b.result();
1077 
1078 #ifndef _LP64
1079   result_a = new_register(T_INT);
1080   __ convert(Bytecodes::_l2i, a.result(), result_a);
1081   result_b = new_register(T_INT);
1082   __ convert(Bytecodes::_l2i, b.result(), result_b);
1083 #endif
1084 
1085 
1086   LIR_Address* addr_a = new LIR_Address(result_a,
1087                                         result_aOffset,
1088                                         constant_aOffset,
1089                                         T_BYTE);
1090 
1091   LIR_Address* addr_b = new LIR_Address(result_b,
1092                                         result_bOffset,
1093                                         constant_bOffset,
1094                                         T_BYTE);
1095 
1096   BasicTypeList signature(4);
1097   signature.append(T_ADDRESS);
1098   signature.append(T_ADDRESS);
1099   signature.append(T_INT);
1100   signature.append(T_INT);
1101   CallingConvention* cc = frame_map()->c_calling_convention(&signature);
1102   const LIR_Opr result_reg = result_register_for(x->type());
1103 
1104   LIR_Opr ptr_addr_a = new_pointer_register();
1105   __ leal(LIR_OprFact::address(addr_a), ptr_addr_a);
1106 
1107   LIR_Opr ptr_addr_b = new_pointer_register();
1108   __ leal(LIR_OprFact::address(addr_b), ptr_addr_b);
1109 
1110   __ move(ptr_addr_a, cc->at(0));
1111   __ move(ptr_addr_b, cc->at(1));
1112   length.load_item_force(cc->at(2));
1113   log2ArrayIndexScale.load_item_force(cc->at(3));
1114 
1115   __ call_runtime_leaf(StubRoutines::vectorizedMismatch(), getThreadTemp(), result_reg, cc->args());
1116   __ move(result_reg, result);
1117 }
1118 
1119 // _i2l, _i2f, _i2d, _l2i, _l2f, _l2d, _f2i, _f2l, _f2d, _d2i, _d2l, _d2f
1120 // _i2b, _i2c, _i2s
1121 LIR_Opr fixed_register_for(BasicType type) {
1122   switch (type) {
1123     case T_FLOAT:  return FrameMap::fpu0_float_opr;
1124     case T_DOUBLE: return FrameMap::fpu0_double_opr;
1125     case T_INT:    return FrameMap::rax_opr;
1126     case T_LONG:   return FrameMap::long0_opr;
1127     default:       ShouldNotReachHere(); return LIR_OprFact::illegalOpr;
1128   }
1129 }
1130 
1131 void LIRGenerator::do_Convert(Convert* x) {
1132   // flags that vary for the different operations and different SSE-settings
1133   bool fixed_input = false, fixed_result = false, round_result = false, needs_stub = false;
1134 
1135   switch (x->op()) {
1136     case Bytecodes::_i2l: // fall through
1137     case Bytecodes::_l2i: // fall through
1138     case Bytecodes::_i2b: // fall through
1139     case Bytecodes::_i2c: // fall through
1140     case Bytecodes::_i2s: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = false; break;
1141 
1142     case Bytecodes::_f2d: fixed_input = UseSSE == 1; fixed_result = false;       round_result = false;      needs_stub = false; break;
1143     case Bytecodes::_d2f: fixed_input = false;       fixed_result = UseSSE == 1; round_result = UseSSE < 1; needs_stub = false; break;
1144     case Bytecodes::_i2f: fixed_input = false;       fixed_result = false;       round_result = UseSSE < 1; needs_stub = false; break;
1145     case Bytecodes::_i2d: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = false; break;
1146     case Bytecodes::_f2i: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = true;  break;
1147     case Bytecodes::_d2i: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = true;  break;
1148     case Bytecodes::_l2f: fixed_input = false;       fixed_result = UseSSE >= 1; round_result = UseSSE < 1; needs_stub = false; break;
1149     case Bytecodes::_l2d: fixed_input = false;       fixed_result = UseSSE >= 2; round_result = UseSSE < 2; needs_stub = false; break;
1150     case Bytecodes::_f2l: fixed_input = true;        fixed_result = true;        round_result = false;      needs_stub = false; break;
1151     case Bytecodes::_d2l: fixed_input = true;        fixed_result = true;        round_result = false;      needs_stub = false; break;
1152     default: ShouldNotReachHere();
1153   }
1154 
1155   LIRItem value(x->value(), this);
1156   value.load_item();
1157   LIR_Opr input = value.result();
1158   LIR_Opr result = rlock(x);
1159 
1160   // arguments of lir_convert
1161   LIR_Opr conv_input = input;
1162   LIR_Opr conv_result = result;
1163   ConversionStub* stub = NULL;
1164 
1165   if (fixed_input) {
1166     conv_input = fixed_register_for(input->type());
1167     __ move(input, conv_input);
1168   }
1169 
1170   assert(fixed_result == false || round_result == false, "cannot set both");
1171   if (fixed_result) {
1172     conv_result = fixed_register_for(result->type());
1173   } else if (round_result) {
1174     result = new_register(result->type());
1175     set_vreg_flag(result, must_start_in_memory);
1176   }
1177 
1178   if (needs_stub) {
1179     stub = new ConversionStub(x->op(), conv_input, conv_result);
1180   }
1181 
1182   __ convert(x->op(), conv_input, conv_result, stub);
1183 
1184   if (result != conv_result) {
1185     __ move(conv_result, result);
1186   }
1187 
1188   assert(result->is_virtual(), "result must be virtual register");
1189   set_result(x, result);
1190 }
1191 
1192 
1193 void LIRGenerator::do_NewInstance(NewInstance* x) {
1194   print_if_not_loaded(x);
1195 
1196   CodeEmitInfo* info = state_for(x, x->state());
1197   LIR_Opr reg = result_register_for(x->type());
1198   new_instance(reg, x->klass(), x->is_unresolved(),
1199                        FrameMap::rcx_oop_opr,
1200                        FrameMap::rdi_oop_opr,
1201                        FrameMap::rsi_oop_opr,
1202                        LIR_OprFact::illegalOpr,
1203                        FrameMap::rdx_metadata_opr, info);
1204   LIR_Opr result = rlock_result(x);
1205   __ move(reg, result);
1206 }
1207 
1208 
1209 void LIRGenerator::do_NewTypeArray(NewTypeArray* x) {
1210   CodeEmitInfo* info = state_for(x, x->state());
1211 
1212   LIRItem length(x->length(), this);
1213   length.load_item_force(FrameMap::rbx_opr);
1214 
1215   LIR_Opr reg = result_register_for(x->type());
1216   LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
1217   LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
1218   LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
1219   LIR_Opr tmp4 = reg;
1220   LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
1221   LIR_Opr len = length.result();
1222   BasicType elem_type = x->elt_type();
1223 
1224   __ metadata2reg(ciTypeArrayKlass::make(elem_type)->constant_encoding(), klass_reg);
1225 
1226   CodeStub* slow_path = new NewTypeArrayStub(klass_reg, len, reg, info);
1227   __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, elem_type, klass_reg, slow_path);
1228 
1229   LIR_Opr result = rlock_result(x);
1230   __ move(reg, result);
1231 }
1232 
1233 
1234 void LIRGenerator::do_NewObjectArray(NewObjectArray* x) {
1235   LIRItem length(x->length(), this);
1236   // in case of patching (i.e., object class is not yet loaded), we need to reexecute the instruction
1237   // and therefore provide the state before the parameters have been consumed
1238   CodeEmitInfo* patching_info = NULL;
1239   if (!x->klass()->is_loaded() || PatchALot) {
1240     patching_info =  state_for(x, x->state_before());
1241   }
1242 
1243   CodeEmitInfo* info = state_for(x, x->state());
1244 
1245   const LIR_Opr reg = result_register_for(x->type());
1246   LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
1247   LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
1248   LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
1249   LIR_Opr tmp4 = reg;
1250   LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
1251 
1252   length.load_item_force(FrameMap::rbx_opr);
1253   LIR_Opr len = length.result();
1254 
1255   CodeStub* slow_path = new NewObjectArrayStub(klass_reg, len, reg, info);
1256   ciKlass* obj = (ciKlass*) ciObjArrayKlass::make(x->klass());
1257   if (obj == ciEnv::unloaded_ciobjarrayklass()) {
1258     BAILOUT("encountered unloaded_ciobjarrayklass due to out of memory error");
1259   }
1260   klass2reg_with_patching(klass_reg, obj, patching_info);
1261   __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, T_OBJECT, klass_reg, slow_path);
1262 
1263   LIR_Opr result = rlock_result(x);
1264   __ move(reg, result);
1265 }
1266 
1267 
1268 void LIRGenerator::do_NewMultiArray(NewMultiArray* x) {
1269   Values* dims = x->dims();
1270   int i = dims->length();
1271   LIRItemList* items = new LIRItemList(i, i, NULL);
1272   while (i-- > 0) {
1273     LIRItem* size = new LIRItem(dims->at(i), this);
1274     items->at_put(i, size);
1275   }
1276 
1277   // Evaluate state_for early since it may emit code.
1278   CodeEmitInfo* patching_info = NULL;
1279   if (!x->klass()->is_loaded() || PatchALot) {
1280     patching_info = state_for(x, x->state_before());
1281 
1282     // Cannot re-use same xhandlers for multiple CodeEmitInfos, so
1283     // clone all handlers (NOTE: Usually this is handled transparently
1284     // by the CodeEmitInfo cloning logic in CodeStub constructors but
1285     // is done explicitly here because a stub isn't being used).
1286     x->set_exception_handlers(new XHandlers(x->exception_handlers()));
1287   }
1288   CodeEmitInfo* info = state_for(x, x->state());
1289 
1290   i = dims->length();
1291   while (i-- > 0) {
1292     LIRItem* size = items->at(i);
1293     size->load_nonconstant();
1294 
1295     store_stack_parameter(size->result(), in_ByteSize(i*4));
1296   }
1297 
1298   LIR_Opr klass_reg = FrameMap::rax_metadata_opr;
1299   klass2reg_with_patching(klass_reg, x->klass(), patching_info);
1300 
1301   LIR_Opr rank = FrameMap::rbx_opr;
1302   __ move(LIR_OprFact::intConst(x->rank()), rank);
1303   LIR_Opr varargs = FrameMap::rcx_opr;
1304   __ move(FrameMap::rsp_opr, varargs);
1305   LIR_OprList* args = new LIR_OprList(3);
1306   args->append(klass_reg);
1307   args->append(rank);
1308   args->append(varargs);
1309   LIR_Opr reg = result_register_for(x->type());
1310   __ call_runtime(Runtime1::entry_for(Runtime1::new_multi_array_id),
1311                   LIR_OprFact::illegalOpr,
1312                   reg, args, info);
1313 
1314   LIR_Opr result = rlock_result(x);
1315   __ move(reg, result);
1316 }
1317 
1318 
1319 void LIRGenerator::do_BlockBegin(BlockBegin* x) {
1320   // nothing to do for now
1321 }
1322 
1323 
1324 void LIRGenerator::do_CheckCast(CheckCast* x) {
1325   LIRItem obj(x->obj(), this);
1326 
1327   CodeEmitInfo* patching_info = NULL;
1328   if (!x->klass()->is_loaded() || (PatchALot && !x->is_incompatible_class_change_check())) {
1329     // must do this before locking the destination register as an oop register,
1330     // and before the obj is loaded (the latter is for deoptimization)
1331     patching_info = state_for(x, x->state_before());
1332   }
1333   obj.load_item();
1334 
1335   // info for exceptions
1336   CodeEmitInfo* info_for_exception = state_for(x);
1337 
1338   CodeStub* stub;
1339   if (x->is_incompatible_class_change_check()) {
1340     assert(patching_info == NULL, "can't patch this");
1341     stub = new SimpleExceptionStub(Runtime1::throw_incompatible_class_change_error_id, LIR_OprFact::illegalOpr, info_for_exception);
1342   } else {
1343     stub = new SimpleExceptionStub(Runtime1::throw_class_cast_exception_id, obj.result(), info_for_exception);
1344   }
1345   LIR_Opr reg = rlock_result(x);
1346   LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
1347   if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
1348     tmp3 = new_register(objectType);
1349   }
1350   __ checkcast(reg, obj.result(), x->klass(),
1351                new_register(objectType), new_register(objectType), tmp3,
1352                x->direct_compare(), info_for_exception, patching_info, stub,
1353                x->profiled_method(), x->profiled_bci());
1354 }
1355 
1356 
1357 void LIRGenerator::do_InstanceOf(InstanceOf* x) {
1358   LIRItem obj(x->obj(), this);
1359 
1360   // result and test object may not be in same register
1361   LIR_Opr reg = rlock_result(x);
1362   CodeEmitInfo* patching_info = NULL;
1363   if ((!x->klass()->is_loaded() || PatchALot)) {
1364     // must do this before locking the destination register as an oop register
1365     patching_info = state_for(x, x->state_before());
1366   }
1367   obj.load_item();
1368   LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
1369   if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
1370     tmp3 = new_register(objectType);
1371   }
1372   __ instanceof(reg, obj.result(), x->klass(),
1373                 new_register(objectType), new_register(objectType), tmp3,
1374                 x->direct_compare(), patching_info, x->profiled_method(), x->profiled_bci());
1375 }
1376 
1377 
1378 void LIRGenerator::do_If(If* x) {
1379   assert(x->number_of_sux() == 2, "inconsistency");
1380   ValueTag tag = x->x()->type()->tag();
1381   bool is_safepoint = x->is_safepoint();
1382 
1383   If::Condition cond = x->cond();
1384 
1385   LIRItem xitem(x->x(), this);
1386   LIRItem yitem(x->y(), this);
1387   LIRItem* xin = &xitem;
1388   LIRItem* yin = &yitem;
1389 
1390   if (tag == longTag) {
1391     // for longs, only conditions "eql", "neq", "lss", "geq" are valid;
1392     // mirror for other conditions
1393     if (cond == If::gtr || cond == If::leq) {
1394       cond = Instruction::mirror(cond);
1395       xin = &yitem;
1396       yin = &xitem;
1397     }
1398     xin->set_destroys_register();
1399   }
1400   xin->load_item();
1401   if (tag == longTag && yin->is_constant() && yin->get_jlong_constant() == 0 && (cond == If::eql || cond == If::neq)) {
1402     // inline long zero
1403     yin->dont_load_item();
1404   } else if (tag == longTag || tag == floatTag || tag == doubleTag) {
1405     // longs cannot handle constants at right side
1406     yin->load_item();
1407   } else {
1408     yin->dont_load_item();
1409   }
1410 
1411   // add safepoint before generating condition code so it can be recomputed
1412   if (x->is_safepoint()) {
1413     // increment backedge counter if needed
1414     increment_backedge_counter(state_for(x, x->state_before()), x->profiled_bci());
1415     __ safepoint(LIR_OprFact::illegalOpr, state_for(x, x->state_before()));
1416   }
1417   set_no_result(x);
1418 
1419   LIR_Opr left = xin->result();
1420   LIR_Opr right = yin->result();
1421   __ cmp(lir_cond(cond), left, right);
1422   // Generate branch profiling. Profiling code doesn't kill flags.
1423   profile_branch(x, cond);
1424   move_to_phi(x->state());
1425   if (x->x()->type()->is_float_kind()) {
1426     __ branch(lir_cond(cond), right->type(), x->tsux(), x->usux());
1427   } else {
1428     __ branch(lir_cond(cond), right->type(), x->tsux());
1429   }
1430   assert(x->default_sux() == x->fsux(), "wrong destination above");
1431   __ jump(x->default_sux());
1432 }
1433 
1434 
1435 LIR_Opr LIRGenerator::getThreadPointer() {
1436 #ifdef _LP64
1437   return FrameMap::as_pointer_opr(r15_thread);
1438 #else
1439   LIR_Opr result = new_register(T_INT);
1440   __ get_thread(result);
1441   return result;
1442 #endif //
1443 }
1444 
1445 void LIRGenerator::trace_block_entry(BlockBegin* block) {
1446   store_stack_parameter(LIR_OprFact::intConst(block->block_id()), in_ByteSize(0));
1447   LIR_OprList* args = new LIR_OprList();
1448   address func = CAST_FROM_FN_PTR(address, Runtime1::trace_block_entry);
1449   __ call_runtime_leaf(func, LIR_OprFact::illegalOpr, LIR_OprFact::illegalOpr, args);
1450 }
1451 
1452 
1453 void LIRGenerator::volatile_field_store(LIR_Opr value, LIR_Address* address,
1454                                         CodeEmitInfo* info) {
1455   if (address->type() == T_LONG) {
1456     address = new LIR_Address(address->base(),
1457                               address->index(), address->scale(),
1458                               address->disp(), T_DOUBLE);
1459     // Transfer the value atomically by using FP moves.  This means
1460     // the value has to be moved between CPU and FPU registers.  It
1461     // always has to be moved through spill slot since there's no
1462     // quick way to pack the value into an SSE register.
1463     LIR_Opr temp_double = new_register(T_DOUBLE);
1464     LIR_Opr spill = new_register(T_LONG);
1465     set_vreg_flag(spill, must_start_in_memory);
1466     __ move(value, spill);
1467     __ volatile_move(spill, temp_double, T_LONG);
1468     __ volatile_move(temp_double, LIR_OprFact::address(address), T_LONG, info);
1469   } else {
1470     __ store(value, address, info);
1471   }
1472 }
1473 
1474 void LIRGenerator::volatile_field_load(LIR_Address* address, LIR_Opr result,
1475                                        CodeEmitInfo* info) {
1476   if (address->type() == T_LONG) {
1477     address = new LIR_Address(address->base(),
1478                               address->index(), address->scale(),
1479                               address->disp(), T_DOUBLE);
1480     // Transfer the value atomically by using FP moves.  This means
1481     // the value has to be moved between CPU and FPU registers.  In
1482     // SSE0 and SSE1 mode it has to be moved through spill slot but in
1483     // SSE2+ mode it can be moved directly.
1484     LIR_Opr temp_double = new_register(T_DOUBLE);
1485     __ volatile_move(LIR_OprFact::address(address), temp_double, T_LONG, info);
1486     __ volatile_move(temp_double, result, T_LONG);
1487     if (UseSSE < 2) {
1488       // no spill slot needed in SSE2 mode because xmm->cpu register move is possible
1489       set_vreg_flag(result, must_start_in_memory);
1490     }
1491   } else {
1492     __ load(address, result, info);
1493   }
1494 }