1 /*
   2  * Copyright (c) 2005, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #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/c1/barrierSetC1.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   NOT_LP64( if (SafepointMechanism::uses_thread_local_poll()) { return new_register(T_ADDRESS); } )
 148   return LIR_OprFact::illegalOpr;
 149 }
 150 
 151 
 152 LIR_Address* LIRGenerator::generate_address(LIR_Opr base, LIR_Opr index,
 153                                             int shift, int disp, BasicType type) {
 154   assert(base->is_register(), "must be");
 155   if (index->is_constant()) {
 156     LIR_Const *constant = index->as_constant_ptr();
 157 #ifdef _LP64
 158     jlong c;
 159     if (constant->type() == T_INT) {
 160       c = (jlong(index->as_jint()) << shift) + disp;
 161     } else {
 162       assert(constant->type() == T_LONG, "should be");
 163       c = (index->as_jlong() << shift) + disp;
 164     }
 165     if ((jlong)((jint)c) == c) {
 166       return new LIR_Address(base, (jint)c, type);
 167     } else {
 168       LIR_Opr tmp = new_register(T_LONG);
 169       __ move(index, tmp);
 170       return new LIR_Address(base, tmp, type);
 171     }
 172 #else
 173     return new LIR_Address(base,
 174                            ((intx)(constant->as_jint()) << shift) + disp,
 175                            type);
 176 #endif
 177   } else {
 178     return new LIR_Address(base, index, (LIR_Address::Scale)shift, disp, type);
 179   }
 180 }
 181 
 182 
 183 LIR_Address* LIRGenerator::emit_array_address(LIR_Opr array_opr, LIR_Opr index_opr,
 184                                               BasicType type) {
 185   int offset_in_bytes = arrayOopDesc::base_offset_in_bytes(type);
 186 
 187   LIR_Address* addr;
 188   if (index_opr->is_constant()) {
 189     int elem_size = type2aelembytes(type);
 190     addr = new LIR_Address(array_opr,
 191                            offset_in_bytes + (intx)(index_opr->as_jint()) * elem_size, type);
 192   } else {
 193 #ifdef _LP64
 194     if (index_opr->type() == T_INT) {
 195       LIR_Opr tmp = new_register(T_LONG);
 196       __ convert(Bytecodes::_i2l, index_opr, tmp);
 197       index_opr = tmp;
 198     }
 199 #endif // _LP64
 200     addr =  new LIR_Address(array_opr,
 201                             index_opr,
 202                             LIR_Address::scale(type),
 203                             offset_in_bytes, type);
 204   }
 205   return addr;
 206 }
 207 
 208 
 209 LIR_Opr LIRGenerator::load_immediate(int x, BasicType type) {
 210   LIR_Opr r = NULL;
 211   if (type == T_LONG) {
 212     r = LIR_OprFact::longConst(x);
 213   } else if (type == T_INT) {
 214     r = LIR_OprFact::intConst(x);
 215   } else {
 216     ShouldNotReachHere();
 217   }
 218   return r;
 219 }
 220 
 221 void LIRGenerator::increment_counter(address counter, BasicType type, int step) {
 222   LIR_Opr pointer = new_pointer_register();
 223   __ move(LIR_OprFact::intptrConst(counter), pointer);
 224   LIR_Address* addr = new LIR_Address(pointer, type);
 225   increment_counter(addr, step);
 226 }
 227 
 228 
 229 void LIRGenerator::increment_counter(LIR_Address* addr, int step) {
 230   __ add((LIR_Opr)addr, LIR_OprFact::intConst(step), (LIR_Opr)addr);
 231 }
 232 
 233 void LIRGenerator::cmp_mem_int(LIR_Condition condition, LIR_Opr base, int disp, int c, CodeEmitInfo* info) {
 234   __ cmp_mem_int(condition, base, disp, c, info);
 235 }
 236 
 237 
 238 void LIRGenerator::cmp_reg_mem(LIR_Condition condition, LIR_Opr reg, LIR_Opr base, int disp, BasicType type, CodeEmitInfo* info) {
 239   __ cmp_reg_mem(condition, reg, new LIR_Address(base, disp, type), info);
 240 }
 241 
 242 
 243 bool LIRGenerator::strength_reduce_multiply(LIR_Opr left, jint c, LIR_Opr result, LIR_Opr tmp) {
 244   if (tmp->is_valid() && c > 0 && c < max_jint) {
 245     if (is_power_of_2(c + 1)) {
 246       __ move(left, tmp);
 247       __ shift_left(left, log2_intptr(c + 1), left);
 248       __ sub(left, tmp, result);
 249       return true;
 250     } else if (is_power_of_2(c - 1)) {
 251       __ move(left, tmp);
 252       __ shift_left(left, log2_intptr(c - 1), left);
 253       __ add(left, tmp, result);
 254       return true;
 255     }
 256   }
 257   return false;
 258 }
 259 
 260 
 261 void LIRGenerator::store_stack_parameter (LIR_Opr item, ByteSize offset_from_sp) {
 262   BasicType type = item->type();
 263   __ store(item, new LIR_Address(FrameMap::rsp_opr, in_bytes(offset_from_sp), type));
 264 }
 265 
 266 void LIRGenerator::array_store_check(LIR_Opr value, LIR_Opr array, CodeEmitInfo* store_check_info, ciMethod* profiled_method, int profiled_bci) {
 267   LIR_Opr tmp1 = new_register(objectType);
 268   LIR_Opr tmp2 = new_register(objectType);
 269   LIR_Opr tmp3 = new_register(objectType);
 270   __ store_check(value, array, tmp1, tmp2, tmp3, store_check_info, profiled_method, profiled_bci);
 271 }
 272 
 273 //----------------------------------------------------------------------
 274 //             visitor functions
 275 //----------------------------------------------------------------------
 276 
 277 void LIRGenerator::do_MonitorEnter(MonitorEnter* x) {
 278   assert(x->is_pinned(),"");
 279   LIRItem obj(x->obj(), this);
 280   obj.load_item();
 281 
 282   set_no_result(x);
 283 
 284   // "lock" stores the address of the monitor stack slot, so this is not an oop
 285   LIR_Opr lock = new_register(T_INT);
 286   // Need a scratch register for biased locking on x86
 287   LIR_Opr scratch = LIR_OprFact::illegalOpr;
 288   if (UseBiasedLocking) {
 289     scratch = new_register(T_INT);
 290   }
 291 
 292   CodeEmitInfo* info_for_exception = NULL;
 293   if (x->needs_null_check()) {
 294     info_for_exception = state_for(x);
 295   }
 296   // this CodeEmitInfo must not have the xhandlers because here the
 297   // object is already locked (xhandlers expect object to be unlocked)
 298   CodeEmitInfo* info = state_for(x, x->state(), true);
 299   monitor_enter(obj.result(), lock, syncTempOpr(), scratch,
 300                         x->monitor_no(), info_for_exception, info);
 301 }
 302 
 303 
 304 void LIRGenerator::do_MonitorExit(MonitorExit* x) {
 305   assert(x->is_pinned(),"");
 306 
 307   LIRItem obj(x->obj(), this);
 308   obj.dont_load_item();
 309 
 310   LIR_Opr lock = new_register(T_INT);
 311   LIR_Opr obj_temp = new_register(T_INT);
 312   set_no_result(x);
 313   monitor_exit(obj_temp, lock, syncTempOpr(), LIR_OprFact::illegalOpr, x->monitor_no());
 314 }
 315 
 316 
 317 // _ineg, _lneg, _fneg, _dneg
 318 void LIRGenerator::do_NegateOp(NegateOp* x) {
 319   LIRItem value(x->x(), this);
 320   value.set_destroys_register();
 321   value.load_item();
 322   LIR_Opr reg = rlock(x);
 323   __ negate(value.result(), reg);
 324 
 325   set_result(x, round_item(reg));
 326 }
 327 
 328 void LIRGenerator::do_continuation_getFP(Intrinsic* x) {
 329   LIR_Opr result_reg = rlock_result(x);
 330   __ getfp(result_reg);
 331 }
 332 
 333 void LIRGenerator::do_continuation_getSP(Intrinsic* x) {
 334   LIR_Opr result_reg = rlock_result(x);
 335   __ getsp(result_reg);
 336 }
 337 
 338 void LIRGenerator::do_continuation_getPC(Intrinsic* x) {
 339   BasicTypeList signature(0);
 340   //signature.append(T_LONG);
 341   CallingConvention* cc = frame_map()->c_calling_convention(&signature);
 342 
 343   const LIR_Opr result_reg = result_register_for(x->type());
 344   address entry = StubRoutines::cont_getPC();
 345   LIR_Opr result = rlock_result(x);
 346   __ call_runtime_leaf(entry, getThreadTemp(), result_reg, cc->args());
 347   __ move(result_reg, result);
 348 }
 349 
 350 void LIRGenerator::do_continuation_doContinue(Intrinsic* x) {
 351   BasicTypeList signature(0);
 352   CallingConvention* cc = frame_map()->c_calling_convention(&signature);
 353 
 354   //const LIR_Opr result_reg = result_register_for(x->type());
 355   address entry = StubRoutines::cont_thaw();
 356   CodeEmitInfo* info = state_for(x, x->state());
 357   __ call_runtime(entry, getThreadTemp(), getThreadTemp(), cc->args(), info);
 358 }
 359 
 360 void LIRGenerator::do_continuation_doYield(Intrinsic* x) {
 361   BasicTypeList signature(1);
 362   signature.append(T_INT);
 363   CallingConvention* cc = frame_map()->c_calling_convention(&signature);
 364 
 365   LIRItem value(x->argument_at(0), this);
 366   value.load_item();
 367   __ move(value.result(), cc->at(0));
 368 
 369   address entry = StubRoutines::cont_doYield();
 370   CodeEmitInfo* info = state_for(x, x->state());
 371   __ call_runtime(entry, LIR_OprFact::illegalOpr, LIR_OprFact::illegalOpr, cc->args(), info);
 372 }
 373 
 374 // for  _fadd, _fmul, _fsub, _fdiv, _frem
 375 //      _dadd, _dmul, _dsub, _ddiv, _drem
 376 void LIRGenerator::do_ArithmeticOp_FPU(ArithmeticOp* x) {
 377   LIRItem left(x->x(),  this);
 378   LIRItem right(x->y(), this);
 379   LIRItem* left_arg  = &left;
 380   LIRItem* right_arg = &right;
 381   assert(!left.is_stack() || !right.is_stack(), "can't both be memory operands");
 382   bool must_load_both = (x->op() == Bytecodes::_frem || x->op() == Bytecodes::_drem);
 383   if (left.is_register() || x->x()->type()->is_constant() || must_load_both) {
 384     left.load_item();
 385   } else {
 386     left.dont_load_item();
 387   }
 388 
 389   // do not load right operand if it is a constant.  only 0 and 1 are
 390   // loaded because there are special instructions for loading them
 391   // without memory access (not needed for SSE2 instructions)
 392   bool must_load_right = false;
 393   if (right.is_constant()) {
 394     LIR_Const* c = right.result()->as_constant_ptr();
 395     assert(c != NULL, "invalid constant");
 396     assert(c->type() == T_FLOAT || c->type() == T_DOUBLE, "invalid type");
 397 
 398     if (c->type() == T_FLOAT) {
 399       must_load_right = UseSSE < 1 && (c->is_one_float() || c->is_zero_float());
 400     } else {
 401       must_load_right = UseSSE < 2 && (c->is_one_double() || c->is_zero_double());
 402     }
 403   }
 404 
 405   if (must_load_both) {
 406     // frem and drem destroy also right operand, so move it to a new register
 407     right.set_destroys_register();
 408     right.load_item();
 409   } else if (right.is_register() || must_load_right) {
 410     right.load_item();
 411   } else {
 412     right.dont_load_item();
 413   }
 414   LIR_Opr reg = rlock(x);
 415   LIR_Opr tmp = LIR_OprFact::illegalOpr;
 416   if (x->is_strictfp() && (x->op() == Bytecodes::_dmul || x->op() == Bytecodes::_ddiv)) {
 417     tmp = new_register(T_DOUBLE);
 418   }
 419 
 420   if ((UseSSE >= 1 && x->op() == Bytecodes::_frem) || (UseSSE >= 2 && x->op() == Bytecodes::_drem)) {
 421     // special handling for frem and drem: no SSE instruction, so must use FPU with temporary fpu stack slots
 422     LIR_Opr fpu0, fpu1;
 423     if (x->op() == Bytecodes::_frem) {
 424       fpu0 = LIR_OprFact::single_fpu(0);
 425       fpu1 = LIR_OprFact::single_fpu(1);
 426     } else {
 427       fpu0 = LIR_OprFact::double_fpu(0);
 428       fpu1 = LIR_OprFact::double_fpu(1);
 429     }
 430     __ move(right.result(), fpu1); // order of left and right operand is important!
 431     __ move(left.result(), fpu0);
 432     __ rem (fpu0, fpu1, fpu0);
 433     __ move(fpu0, reg);
 434 
 435   } else {
 436     arithmetic_op_fpu(x->op(), reg, left.result(), right.result(), x->is_strictfp(), tmp);
 437   }
 438 
 439   set_result(x, round_item(reg));
 440 }
 441 
 442 
 443 // for  _ladd, _lmul, _lsub, _ldiv, _lrem
 444 void LIRGenerator::do_ArithmeticOp_Long(ArithmeticOp* x) {
 445   if (x->op() == Bytecodes::_ldiv || x->op() == Bytecodes::_lrem ) {
 446     // long division is implemented as a direct call into the runtime
 447     LIRItem left(x->x(), this);
 448     LIRItem right(x->y(), this);
 449 
 450     // the check for division by zero destroys the right operand
 451     right.set_destroys_register();
 452 
 453     BasicTypeList signature(2);
 454     signature.append(T_LONG);
 455     signature.append(T_LONG);
 456     CallingConvention* cc = frame_map()->c_calling_convention(&signature);
 457 
 458     // check for division by zero (destroys registers of right operand!)
 459     CodeEmitInfo* info = state_for(x);
 460 
 461     const LIR_Opr result_reg = result_register_for(x->type());
 462     left.load_item_force(cc->at(1));
 463     right.load_item();
 464 
 465     __ move(right.result(), cc->at(0));
 466 
 467     __ cmp(lir_cond_equal, right.result(), LIR_OprFact::longConst(0));
 468     __ branch(lir_cond_equal, T_LONG, new DivByZeroStub(info));
 469 
 470     address entry = NULL;
 471     switch (x->op()) {
 472     case Bytecodes::_lrem:
 473       entry = CAST_FROM_FN_PTR(address, SharedRuntime::lrem);
 474       break; // check if dividend is 0 is done elsewhere
 475     case Bytecodes::_ldiv:
 476       entry = CAST_FROM_FN_PTR(address, SharedRuntime::ldiv);
 477       break; // check if dividend is 0 is done elsewhere
 478     case Bytecodes::_lmul:
 479       entry = CAST_FROM_FN_PTR(address, SharedRuntime::lmul);
 480       break;
 481     default:
 482       ShouldNotReachHere();
 483     }
 484 
 485     LIR_Opr result = rlock_result(x);
 486     __ call_runtime_leaf(entry, getThreadTemp(), result_reg, cc->args());
 487     __ move(result_reg, result);
 488   } else if (x->op() == Bytecodes::_lmul) {
 489     // missing test if instr is commutative and if we should swap
 490     LIRItem left(x->x(), this);
 491     LIRItem right(x->y(), this);
 492 
 493     // right register is destroyed by the long mul, so it must be
 494     // copied to a new register.
 495     right.set_destroys_register();
 496 
 497     left.load_item();
 498     right.load_item();
 499 
 500     LIR_Opr reg = FrameMap::long0_opr;
 501     arithmetic_op_long(x->op(), reg, left.result(), right.result(), NULL);
 502     LIR_Opr result = rlock_result(x);
 503     __ move(reg, result);
 504   } else {
 505     // missing test if instr is commutative and if we should swap
 506     LIRItem left(x->x(), this);
 507     LIRItem right(x->y(), this);
 508 
 509     left.load_item();
 510     // don't load constants to save register
 511     right.load_nonconstant();
 512     rlock_result(x);
 513     arithmetic_op_long(x->op(), x->operand(), left.result(), right.result(), NULL);
 514   }
 515 }
 516 
 517 
 518 
 519 // for: _iadd, _imul, _isub, _idiv, _irem
 520 void LIRGenerator::do_ArithmeticOp_Int(ArithmeticOp* x) {
 521   if (x->op() == Bytecodes::_idiv || x->op() == Bytecodes::_irem) {
 522     // The requirements for division and modulo
 523     // input : rax,: dividend                         min_int
 524     //         reg: divisor   (may not be rax,/rdx)   -1
 525     //
 526     // output: rax,: quotient  (= rax, idiv reg)       min_int
 527     //         rdx: remainder (= rax, irem reg)       0
 528 
 529     // rax, and rdx will be destroyed
 530 
 531     // Note: does this invalidate the spec ???
 532     LIRItem right(x->y(), this);
 533     LIRItem left(x->x() , this);   // visit left second, so that the is_register test is valid
 534 
 535     // call state_for before load_item_force because state_for may
 536     // force the evaluation of other instructions that are needed for
 537     // correct debug info.  Otherwise the live range of the fix
 538     // register might be too long.
 539     CodeEmitInfo* info = state_for(x);
 540 
 541     left.load_item_force(divInOpr());
 542 
 543     right.load_item();
 544 
 545     LIR_Opr result = rlock_result(x);
 546     LIR_Opr result_reg;
 547     if (x->op() == Bytecodes::_idiv) {
 548       result_reg = divOutOpr();
 549     } else {
 550       result_reg = remOutOpr();
 551     }
 552 
 553     if (!ImplicitDiv0Checks) {
 554       __ cmp(lir_cond_equal, right.result(), LIR_OprFact::intConst(0));
 555       __ branch(lir_cond_equal, T_INT, new DivByZeroStub(info));
 556       // Idiv/irem cannot trap (passing info would generate an assertion).
 557       info = NULL;
 558     }
 559     LIR_Opr tmp = FrameMap::rdx_opr; // idiv and irem use rdx in their implementation
 560     if (x->op() == Bytecodes::_irem) {
 561       __ irem(left.result(), right.result(), result_reg, tmp, info);
 562     } else if (x->op() == Bytecodes::_idiv) {
 563       __ idiv(left.result(), right.result(), result_reg, tmp, info);
 564     } else {
 565       ShouldNotReachHere();
 566     }
 567 
 568     __ move(result_reg, result);
 569   } else {
 570     // missing test if instr is commutative and if we should swap
 571     LIRItem left(x->x(),  this);
 572     LIRItem right(x->y(), this);
 573     LIRItem* left_arg = &left;
 574     LIRItem* right_arg = &right;
 575     if (x->is_commutative() && left.is_stack() && right.is_register()) {
 576       // swap them if left is real stack (or cached) and right is real register(not cached)
 577       left_arg = &right;
 578       right_arg = &left;
 579     }
 580 
 581     left_arg->load_item();
 582 
 583     // do not need to load right, as we can handle stack and constants
 584     if (x->op() == Bytecodes::_imul ) {
 585       // check if we can use shift instead
 586       bool use_constant = false;
 587       bool use_tmp = false;
 588       if (right_arg->is_constant()) {
 589         jint iconst = right_arg->get_jint_constant();
 590         if (iconst > 0 && iconst < max_jint) {
 591           if (is_power_of_2(iconst)) {
 592             use_constant = true;
 593           } else if (is_power_of_2(iconst - 1) || is_power_of_2(iconst + 1)) {
 594             use_constant = true;
 595             use_tmp = true;
 596           }
 597         }
 598       }
 599       if (use_constant) {
 600         right_arg->dont_load_item();
 601       } else {
 602         right_arg->load_item();
 603       }
 604       LIR_Opr tmp = LIR_OprFact::illegalOpr;
 605       if (use_tmp) {
 606         tmp = new_register(T_INT);
 607       }
 608       rlock_result(x);
 609 
 610       arithmetic_op_int(x->op(), x->operand(), left_arg->result(), right_arg->result(), tmp);
 611     } else {
 612       right_arg->dont_load_item();
 613       rlock_result(x);
 614       LIR_Opr tmp = LIR_OprFact::illegalOpr;
 615       arithmetic_op_int(x->op(), x->operand(), left_arg->result(), right_arg->result(), tmp);
 616     }
 617   }
 618 }
 619 
 620 
 621 void LIRGenerator::do_ArithmeticOp(ArithmeticOp* x) {
 622   // when an operand with use count 1 is the left operand, then it is
 623   // likely that no move for 2-operand-LIR-form is necessary
 624   if (x->is_commutative() && x->y()->as_Constant() == NULL && x->x()->use_count() > x->y()->use_count()) {
 625     x->swap_operands();
 626   }
 627 
 628   ValueTag tag = x->type()->tag();
 629   assert(x->x()->type()->tag() == tag && x->y()->type()->tag() == tag, "wrong parameters");
 630   switch (tag) {
 631     case floatTag:
 632     case doubleTag:  do_ArithmeticOp_FPU(x);  return;
 633     case longTag:    do_ArithmeticOp_Long(x); return;
 634     case intTag:     do_ArithmeticOp_Int(x);  return;
 635     default:         ShouldNotReachHere();    return;
 636   }
 637 }
 638 
 639 
 640 // _ishl, _lshl, _ishr, _lshr, _iushr, _lushr
 641 void LIRGenerator::do_ShiftOp(ShiftOp* x) {
 642   // count must always be in rcx
 643   LIRItem value(x->x(), this);
 644   LIRItem count(x->y(), this);
 645 
 646   ValueTag elemType = x->type()->tag();
 647   bool must_load_count = !count.is_constant() || elemType == longTag;
 648   if (must_load_count) {
 649     // count for long must be in register
 650     count.load_item_force(shiftCountOpr());
 651   } else {
 652     count.dont_load_item();
 653   }
 654   value.load_item();
 655   LIR_Opr reg = rlock_result(x);
 656 
 657   shift_op(x->op(), reg, value.result(), count.result(), LIR_OprFact::illegalOpr);
 658 }
 659 
 660 
 661 // _iand, _land, _ior, _lor, _ixor, _lxor
 662 void LIRGenerator::do_LogicOp(LogicOp* x) {
 663   // when an operand with use count 1 is the left operand, then it is
 664   // likely that no move for 2-operand-LIR-form is necessary
 665   if (x->is_commutative() && x->y()->as_Constant() == NULL && x->x()->use_count() > x->y()->use_count()) {
 666     x->swap_operands();
 667   }
 668 
 669   LIRItem left(x->x(), this);
 670   LIRItem right(x->y(), this);
 671 
 672   left.load_item();
 673   right.load_nonconstant();
 674   LIR_Opr reg = rlock_result(x);
 675 
 676   logic_op(x->op(), reg, left.result(), right.result());
 677 }
 678 
 679 
 680 
 681 // _lcmp, _fcmpl, _fcmpg, _dcmpl, _dcmpg
 682 void LIRGenerator::do_CompareOp(CompareOp* x) {
 683   LIRItem left(x->x(), this);
 684   LIRItem right(x->y(), this);
 685   ValueTag tag = x->x()->type()->tag();
 686   if (tag == longTag) {
 687     left.set_destroys_register();
 688   }
 689   left.load_item();
 690   right.load_item();
 691   LIR_Opr reg = rlock_result(x);
 692 
 693   if (x->x()->type()->is_float_kind()) {
 694     Bytecodes::Code code = x->op();
 695     __ fcmp2int(left.result(), right.result(), reg, (code == Bytecodes::_fcmpl || code == Bytecodes::_dcmpl));
 696   } else if (x->x()->type()->tag() == longTag) {
 697     __ lcmp2int(left.result(), right.result(), reg);
 698   } else {
 699     Unimplemented();
 700   }
 701 }
 702 
 703 LIR_Opr LIRGenerator::atomic_cmpxchg(BasicType type, LIR_Opr addr, LIRItem& cmp_value, LIRItem& new_value) {
 704   LIR_Opr ill = LIR_OprFact::illegalOpr;  // for convenience
 705   if (type == T_OBJECT || type == T_ARRAY) {
 706     cmp_value.load_item_force(FrameMap::rax_oop_opr);
 707     new_value.load_item();
 708     __ cas_obj(addr->as_address_ptr()->base(), cmp_value.result(), new_value.result(), ill, ill);
 709   } else if (type == T_INT) {
 710     cmp_value.load_item_force(FrameMap::rax_opr);
 711     new_value.load_item();
 712     __ cas_int(addr->as_address_ptr()->base(), cmp_value.result(), new_value.result(), ill, ill);
 713   } else if (type == T_LONG) {
 714     cmp_value.load_item_force(FrameMap::long0_opr);
 715     new_value.load_item_force(FrameMap::long1_opr);
 716     __ cas_long(addr->as_address_ptr()->base(), cmp_value.result(), new_value.result(), ill, ill);
 717   } else {
 718     Unimplemented();
 719   }
 720   LIR_Opr result = new_register(T_INT);
 721   __ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0),
 722            result, type);
 723   return result;
 724 }
 725 
 726 LIR_Opr LIRGenerator::atomic_xchg(BasicType type, LIR_Opr addr, LIRItem& value) {
 727   bool is_oop = type == T_OBJECT || type == T_ARRAY;
 728   LIR_Opr result = new_register(type);
 729   value.load_item();
 730   // Because we want a 2-arg form of xchg and xadd
 731   __ move(value.result(), result);
 732   assert(type == T_INT || is_oop LP64_ONLY( || type == T_LONG ), "unexpected type");
 733   __ xchg(addr, result, result, LIR_OprFact::illegalOpr);
 734   return result;
 735 }
 736 
 737 LIR_Opr LIRGenerator::atomic_add(BasicType type, LIR_Opr addr, LIRItem& value) {
 738   LIR_Opr result = new_register(type);
 739   value.load_item();
 740   // Because we want a 2-arg form of xchg and xadd
 741   __ move(value.result(), result);
 742   assert(type == T_INT LP64_ONLY( || type == T_LONG ), "unexpected type");
 743   __ xadd(addr, result, result, LIR_OprFact::illegalOpr);
 744   return result;
 745 }
 746 
 747 void LIRGenerator::do_FmaIntrinsic(Intrinsic* x) {
 748   assert(x->number_of_arguments() == 3, "wrong type");
 749   assert(UseFMA, "Needs FMA instructions support.");
 750   LIRItem value(x->argument_at(0), this);
 751   LIRItem value1(x->argument_at(1), this);
 752   LIRItem value2(x->argument_at(2), this);
 753 
 754   value2.set_destroys_register();
 755 
 756   value.load_item();
 757   value1.load_item();
 758   value2.load_item();
 759 
 760   LIR_Opr calc_input = value.result();
 761   LIR_Opr calc_input1 = value1.result();
 762   LIR_Opr calc_input2 = value2.result();
 763   LIR_Opr calc_result = rlock_result(x);
 764 
 765   switch (x->id()) {
 766   case vmIntrinsics::_fmaD:   __ fmad(calc_input, calc_input1, calc_input2, calc_result); break;
 767   case vmIntrinsics::_fmaF:   __ fmaf(calc_input, calc_input1, calc_input2, calc_result); break;
 768   default:                    ShouldNotReachHere();
 769   }
 770 
 771 }
 772 
 773 
 774 void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
 775   assert(x->number_of_arguments() == 1 || (x->number_of_arguments() == 2 && x->id() == vmIntrinsics::_dpow), "wrong type");
 776 
 777   if (x->id() == vmIntrinsics::_dexp || x->id() == vmIntrinsics::_dlog ||
 778       x->id() == vmIntrinsics::_dpow || x->id() == vmIntrinsics::_dcos ||
 779       x->id() == vmIntrinsics::_dsin || x->id() == vmIntrinsics::_dtan ||
 780       x->id() == vmIntrinsics::_dlog10) {
 781     do_LibmIntrinsic(x);
 782     return;
 783   }
 784 
 785   LIRItem value(x->argument_at(0), this);
 786 
 787   bool use_fpu = false;
 788   if (UseSSE < 2) {
 789     value.set_destroys_register();
 790   }
 791   value.load_item();
 792 
 793   LIR_Opr calc_input = value.result();
 794   LIR_Opr calc_result = rlock_result(x);
 795 
 796   switch(x->id()) {
 797     case vmIntrinsics::_dabs:   __ abs  (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
 798     case vmIntrinsics::_dsqrt:  __ sqrt (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
 799     default:                    ShouldNotReachHere();
 800   }
 801 
 802   if (use_fpu) {
 803     __ move(calc_result, x->operand());
 804   }
 805 }
 806 
 807 void LIRGenerator::do_LibmIntrinsic(Intrinsic* x) {
 808   LIRItem value(x->argument_at(0), this);
 809   value.set_destroys_register();
 810 
 811   LIR_Opr calc_result = rlock_result(x);
 812   LIR_Opr result_reg = result_register_for(x->type());
 813 
 814   CallingConvention* cc = NULL;
 815 
 816   if (x->id() == vmIntrinsics::_dpow) {
 817     LIRItem value1(x->argument_at(1), this);
 818 
 819     value1.set_destroys_register();
 820 
 821     BasicTypeList signature(2);
 822     signature.append(T_DOUBLE);
 823     signature.append(T_DOUBLE);
 824     cc = frame_map()->c_calling_convention(&signature);
 825     value.load_item_force(cc->at(0));
 826     value1.load_item_force(cc->at(1));
 827   } else {
 828     BasicTypeList signature(1);
 829     signature.append(T_DOUBLE);
 830     cc = frame_map()->c_calling_convention(&signature);
 831     value.load_item_force(cc->at(0));
 832   }
 833 
 834 #ifndef _LP64
 835   LIR_Opr tmp = FrameMap::fpu0_double_opr;
 836   result_reg = tmp;
 837   switch(x->id()) {
 838     case vmIntrinsics::_dexp:
 839       if (StubRoutines::dexp() != NULL) {
 840         __ call_runtime_leaf(StubRoutines::dexp(), getThreadTemp(), result_reg, cc->args());
 841       } else {
 842         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dexp), getThreadTemp(), result_reg, cc->args());
 843       }
 844       break;
 845     case vmIntrinsics::_dlog:
 846       if (StubRoutines::dlog() != NULL) {
 847         __ call_runtime_leaf(StubRoutines::dlog(), getThreadTemp(), result_reg, cc->args());
 848       } else {
 849         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog), getThreadTemp(), result_reg, cc->args());
 850       }
 851       break;
 852     case vmIntrinsics::_dlog10:
 853       if (StubRoutines::dlog10() != NULL) {
 854        __ call_runtime_leaf(StubRoutines::dlog10(), getThreadTemp(), result_reg, cc->args());
 855       } else {
 856         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), getThreadTemp(), result_reg, cc->args());
 857       }
 858       break;
 859     case vmIntrinsics::_dpow:
 860       if (StubRoutines::dpow() != NULL) {
 861         __ call_runtime_leaf(StubRoutines::dpow(), getThreadTemp(), result_reg, cc->args());
 862       } else {
 863         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dpow), getThreadTemp(), result_reg, cc->args());
 864       }
 865       break;
 866     case vmIntrinsics::_dsin:
 867       if (VM_Version::supports_sse2() && StubRoutines::dsin() != NULL) {
 868         __ call_runtime_leaf(StubRoutines::dsin(), getThreadTemp(), result_reg, cc->args());
 869       } else {
 870         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), getThreadTemp(), result_reg, cc->args());
 871       }
 872       break;
 873     case vmIntrinsics::_dcos:
 874       if (VM_Version::supports_sse2() && StubRoutines::dcos() != NULL) {
 875         __ call_runtime_leaf(StubRoutines::dcos(), getThreadTemp(), result_reg, cc->args());
 876       } else {
 877         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), getThreadTemp(), result_reg, cc->args());
 878       }
 879       break;
 880     case vmIntrinsics::_dtan:
 881       if (StubRoutines::dtan() != NULL) {
 882         __ call_runtime_leaf(StubRoutines::dtan(), getThreadTemp(), result_reg, cc->args());
 883       } else {
 884         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), getThreadTemp(), result_reg, cc->args());
 885       }
 886       break;
 887     default:  ShouldNotReachHere();
 888   }
 889 #else
 890   switch (x->id()) {
 891     case vmIntrinsics::_dexp:
 892       if (StubRoutines::dexp() != NULL) {
 893         __ call_runtime_leaf(StubRoutines::dexp(), getThreadTemp(), result_reg, cc->args());
 894       } else {
 895         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dexp), getThreadTemp(), result_reg, cc->args());
 896       }
 897       break;
 898     case vmIntrinsics::_dlog:
 899       if (StubRoutines::dlog() != NULL) {
 900       __ call_runtime_leaf(StubRoutines::dlog(), getThreadTemp(), result_reg, cc->args());
 901       } else {
 902         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog), getThreadTemp(), result_reg, cc->args());
 903       }
 904       break;
 905     case vmIntrinsics::_dlog10:
 906       if (StubRoutines::dlog10() != NULL) {
 907       __ call_runtime_leaf(StubRoutines::dlog10(), getThreadTemp(), result_reg, cc->args());
 908       } else {
 909         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), getThreadTemp(), result_reg, cc->args());
 910       }
 911       break;
 912     case vmIntrinsics::_dpow:
 913        if (StubRoutines::dpow() != NULL) {
 914       __ call_runtime_leaf(StubRoutines::dpow(), getThreadTemp(), result_reg, cc->args());
 915       } else {
 916         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dpow), getThreadTemp(), result_reg, cc->args());
 917       }
 918       break;
 919     case vmIntrinsics::_dsin:
 920       if (StubRoutines::dsin() != NULL) {
 921         __ call_runtime_leaf(StubRoutines::dsin(), getThreadTemp(), result_reg, cc->args());
 922       } else {
 923         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), getThreadTemp(), result_reg, cc->args());
 924       }
 925       break;
 926     case vmIntrinsics::_dcos:
 927       if (StubRoutines::dcos() != NULL) {
 928         __ call_runtime_leaf(StubRoutines::dcos(), getThreadTemp(), result_reg, cc->args());
 929       } else {
 930         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), getThreadTemp(), result_reg, cc->args());
 931       }
 932       break;
 933     case vmIntrinsics::_dtan:
 934        if (StubRoutines::dtan() != NULL) {
 935       __ call_runtime_leaf(StubRoutines::dtan(), getThreadTemp(), result_reg, cc->args());
 936       } else {
 937         __ call_runtime_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), getThreadTemp(), result_reg, cc->args());
 938       }
 939       break;
 940     default:  ShouldNotReachHere();
 941   }
 942 #endif // _LP64
 943   __ move(result_reg, calc_result);
 944 }
 945 
 946 void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
 947   assert(x->number_of_arguments() == 5, "wrong type");
 948 
 949   // Make all state_for calls early since they can emit code
 950   CodeEmitInfo* info = state_for(x, x->state());
 951 
 952   LIRItem src(x->argument_at(0), this);
 953   LIRItem src_pos(x->argument_at(1), this);
 954   LIRItem dst(x->argument_at(2), this);
 955   LIRItem dst_pos(x->argument_at(3), this);
 956   LIRItem length(x->argument_at(4), this);
 957 
 958   // operands for arraycopy must use fixed registers, otherwise
 959   // LinearScan will fail allocation (because arraycopy always needs a
 960   // call)
 961 
 962 #ifndef _LP64
 963   src.load_item_force     (FrameMap::rcx_oop_opr);
 964   src_pos.load_item_force (FrameMap::rdx_opr);
 965   dst.load_item_force     (FrameMap::rax_oop_opr);
 966   dst_pos.load_item_force (FrameMap::rbx_opr);
 967   length.load_item_force  (FrameMap::rdi_opr);
 968   LIR_Opr tmp =           (FrameMap::rsi_opr);
 969 #else
 970 
 971   // The java calling convention will give us enough registers
 972   // so that on the stub side the args will be perfect already.
 973   // On the other slow/special case side we call C and the arg
 974   // positions are not similar enough to pick one as the best.
 975   // Also because the java calling convention is a "shifted" version
 976   // of the C convention we can process the java args trivially into C
 977   // args without worry of overwriting during the xfer
 978 
 979   src.load_item_force     (FrameMap::as_oop_opr(j_rarg0));
 980   src_pos.load_item_force (FrameMap::as_opr(j_rarg1));
 981   dst.load_item_force     (FrameMap::as_oop_opr(j_rarg2));
 982   dst_pos.load_item_force (FrameMap::as_opr(j_rarg3));
 983   length.load_item_force  (FrameMap::as_opr(j_rarg4));
 984 
 985   LIR_Opr tmp =           FrameMap::as_opr(j_rarg5);
 986 #endif // LP64
 987 
 988   set_no_result(x);
 989 
 990   int flags;
 991   ciArrayKlass* expected_type;
 992   arraycopy_helper(x, &flags, &expected_type);
 993 
 994   __ arraycopy(src.result(), src_pos.result(), dst.result(), dst_pos.result(), length.result(), tmp, expected_type, flags, info); // does add_safepoint
 995 }
 996 
 997 void LIRGenerator::do_update_CRC32(Intrinsic* x) {
 998   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
 999   // Make all state_for calls early since they can emit code
1000   LIR_Opr result = rlock_result(x);
1001   int flags = 0;
1002   switch (x->id()) {
1003     case vmIntrinsics::_updateCRC32: {
1004       LIRItem crc(x->argument_at(0), this);
1005       LIRItem val(x->argument_at(1), this);
1006       // val is destroyed by update_crc32
1007       val.set_destroys_register();
1008       crc.load_item();
1009       val.load_item();
1010       __ update_crc32(crc.result(), val.result(), result);
1011       break;
1012     }
1013     case vmIntrinsics::_updateBytesCRC32:
1014     case vmIntrinsics::_updateByteBufferCRC32: {
1015       bool is_updateBytes = (x->id() == vmIntrinsics::_updateBytesCRC32);
1016 
1017       LIRItem crc(x->argument_at(0), this);
1018       LIRItem buf(x->argument_at(1), this);
1019       LIRItem off(x->argument_at(2), this);
1020       LIRItem len(x->argument_at(3), this);
1021       buf.load_item();
1022       off.load_nonconstant();
1023 
1024       LIR_Opr index = off.result();
1025       int offset = is_updateBytes ? arrayOopDesc::base_offset_in_bytes(T_BYTE) : 0;
1026       if(off.result()->is_constant()) {
1027         index = LIR_OprFact::illegalOpr;
1028        offset += off.result()->as_jint();
1029       }
1030       LIR_Opr base_op = buf.result();
1031 
1032 #ifndef _LP64
1033       if (!is_updateBytes) { // long b raw address
1034          base_op = new_register(T_INT);
1035          __ convert(Bytecodes::_l2i, buf.result(), base_op);
1036       }
1037 #else
1038       if (index->is_valid()) {
1039         LIR_Opr tmp = new_register(T_LONG);
1040         __ convert(Bytecodes::_i2l, index, tmp);
1041         index = tmp;
1042       }
1043 #endif
1044 
1045       LIR_Address* a = new LIR_Address(base_op,
1046                                        index,
1047                                        offset,
1048                                        T_BYTE);
1049       BasicTypeList signature(3);
1050       signature.append(T_INT);
1051       signature.append(T_ADDRESS);
1052       signature.append(T_INT);
1053       CallingConvention* cc = frame_map()->c_calling_convention(&signature);
1054       const LIR_Opr result_reg = result_register_for(x->type());
1055 
1056       LIR_Opr addr = new_pointer_register();
1057       __ leal(LIR_OprFact::address(a), addr);
1058 
1059       crc.load_item_force(cc->at(0));
1060       __ move(addr, cc->at(1));
1061       len.load_item_force(cc->at(2));
1062 
1063       __ call_runtime_leaf(StubRoutines::updateBytesCRC32(), getThreadTemp(), result_reg, cc->args());
1064       __ move(result_reg, result);
1065 
1066       break;
1067     }
1068     default: {
1069       ShouldNotReachHere();
1070     }
1071   }
1072 }
1073 
1074 void LIRGenerator::do_update_CRC32C(Intrinsic* x) {
1075   Unimplemented();
1076 }
1077 
1078 void LIRGenerator::do_vectorizedMismatch(Intrinsic* x) {
1079   assert(UseVectorizedMismatchIntrinsic, "need AVX instruction support");
1080 
1081   // Make all state_for calls early since they can emit code
1082   LIR_Opr result = rlock_result(x);
1083 
1084   LIRItem a(x->argument_at(0), this); // Object
1085   LIRItem aOffset(x->argument_at(1), this); // long
1086   LIRItem b(x->argument_at(2), this); // Object
1087   LIRItem bOffset(x->argument_at(3), this); // long
1088   LIRItem length(x->argument_at(4), this); // int
1089   LIRItem log2ArrayIndexScale(x->argument_at(5), this); // int
1090 
1091   a.load_item();
1092   aOffset.load_nonconstant();
1093   b.load_item();
1094   bOffset.load_nonconstant();
1095 
1096   long constant_aOffset = 0;
1097   LIR_Opr result_aOffset = aOffset.result();
1098   if (result_aOffset->is_constant()) {
1099     constant_aOffset = result_aOffset->as_jlong();
1100     result_aOffset = LIR_OprFact::illegalOpr;
1101   }
1102   LIR_Opr result_a = a.result();
1103 
1104   long constant_bOffset = 0;
1105   LIR_Opr result_bOffset = bOffset.result();
1106   if (result_bOffset->is_constant()) {
1107     constant_bOffset = result_bOffset->as_jlong();
1108     result_bOffset = LIR_OprFact::illegalOpr;
1109   }
1110   LIR_Opr result_b = b.result();
1111 
1112 #ifndef _LP64
1113   result_a = new_register(T_INT);
1114   __ convert(Bytecodes::_l2i, a.result(), result_a);
1115   result_b = new_register(T_INT);
1116   __ convert(Bytecodes::_l2i, b.result(), result_b);
1117 #endif
1118 
1119 
1120   LIR_Address* addr_a = new LIR_Address(result_a,
1121                                         result_aOffset,
1122                                         constant_aOffset,
1123                                         T_BYTE);
1124 
1125   LIR_Address* addr_b = new LIR_Address(result_b,
1126                                         result_bOffset,
1127                                         constant_bOffset,
1128                                         T_BYTE);
1129 
1130   BasicTypeList signature(4);
1131   signature.append(T_ADDRESS);
1132   signature.append(T_ADDRESS);
1133   signature.append(T_INT);
1134   signature.append(T_INT);
1135   CallingConvention* cc = frame_map()->c_calling_convention(&signature);
1136   const LIR_Opr result_reg = result_register_for(x->type());
1137 
1138   LIR_Opr ptr_addr_a = new_pointer_register();
1139   __ leal(LIR_OprFact::address(addr_a), ptr_addr_a);
1140 
1141   LIR_Opr ptr_addr_b = new_pointer_register();
1142   __ leal(LIR_OprFact::address(addr_b), ptr_addr_b);
1143 
1144   __ move(ptr_addr_a, cc->at(0));
1145   __ move(ptr_addr_b, cc->at(1));
1146   length.load_item_force(cc->at(2));
1147   log2ArrayIndexScale.load_item_force(cc->at(3));
1148 
1149   __ call_runtime_leaf(StubRoutines::vectorizedMismatch(), getThreadTemp(), result_reg, cc->args());
1150   __ move(result_reg, result);
1151 }
1152 
1153 // _i2l, _i2f, _i2d, _l2i, _l2f, _l2d, _f2i, _f2l, _f2d, _d2i, _d2l, _d2f
1154 // _i2b, _i2c, _i2s
1155 LIR_Opr fixed_register_for(BasicType type) {
1156   switch (type) {
1157     case T_FLOAT:  return FrameMap::fpu0_float_opr;
1158     case T_DOUBLE: return FrameMap::fpu0_double_opr;
1159     case T_INT:    return FrameMap::rax_opr;
1160     case T_LONG:   return FrameMap::long0_opr;
1161     default:       ShouldNotReachHere(); return LIR_OprFact::illegalOpr;
1162   }
1163 }
1164 
1165 void LIRGenerator::do_Convert(Convert* x) {
1166   // flags that vary for the different operations and different SSE-settings
1167   bool fixed_input = false, fixed_result = false, round_result = false, needs_stub = false;
1168 
1169   switch (x->op()) {
1170     case Bytecodes::_i2l: // fall through
1171     case Bytecodes::_l2i: // fall through
1172     case Bytecodes::_i2b: // fall through
1173     case Bytecodes::_i2c: // fall through
1174     case Bytecodes::_i2s: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = false; break;
1175 
1176     case Bytecodes::_f2d: fixed_input = UseSSE == 1; fixed_result = false;       round_result = false;      needs_stub = false; break;
1177     case Bytecodes::_d2f: fixed_input = false;       fixed_result = UseSSE == 1; round_result = UseSSE < 1; needs_stub = false; break;
1178     case Bytecodes::_i2f: fixed_input = false;       fixed_result = false;       round_result = UseSSE < 1; needs_stub = false; break;
1179     case Bytecodes::_i2d: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = false; break;
1180     case Bytecodes::_f2i: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = true;  break;
1181     case Bytecodes::_d2i: fixed_input = false;       fixed_result = false;       round_result = false;      needs_stub = true;  break;
1182     case Bytecodes::_l2f: fixed_input = false;       fixed_result = UseSSE >= 1; round_result = UseSSE < 1; needs_stub = false; break;
1183     case Bytecodes::_l2d: fixed_input = false;       fixed_result = UseSSE >= 2; round_result = UseSSE < 2; needs_stub = false; break;
1184     case Bytecodes::_f2l: fixed_input = true;        fixed_result = true;        round_result = false;      needs_stub = false; break;
1185     case Bytecodes::_d2l: fixed_input = true;        fixed_result = true;        round_result = false;      needs_stub = false; break;
1186     default: ShouldNotReachHere();
1187   }
1188 
1189   LIRItem value(x->value(), this);
1190   value.load_item();
1191   LIR_Opr input = value.result();
1192   LIR_Opr result = rlock(x);
1193 
1194   // arguments of lir_convert
1195   LIR_Opr conv_input = input;
1196   LIR_Opr conv_result = result;
1197   ConversionStub* stub = NULL;
1198 
1199   if (fixed_input) {
1200     conv_input = fixed_register_for(input->type());
1201     __ move(input, conv_input);
1202   }
1203 
1204   assert(fixed_result == false || round_result == false, "cannot set both");
1205   if (fixed_result) {
1206     conv_result = fixed_register_for(result->type());
1207   } else if (round_result) {
1208     result = new_register(result->type());
1209     set_vreg_flag(result, must_start_in_memory);
1210   }
1211 
1212   if (needs_stub) {
1213     stub = new ConversionStub(x->op(), conv_input, conv_result);
1214   }
1215 
1216   __ convert(x->op(), conv_input, conv_result, stub);
1217 
1218   if (result != conv_result) {
1219     __ move(conv_result, result);
1220   }
1221 
1222   assert(result->is_virtual(), "result must be virtual register");
1223   set_result(x, result);
1224 }
1225 
1226 
1227 void LIRGenerator::do_NewInstance(NewInstance* x) {
1228   print_if_not_loaded(x);
1229 
1230   CodeEmitInfo* info = state_for(x, x->state());
1231   LIR_Opr reg = result_register_for(x->type());
1232   new_instance(reg, x->klass(), x->is_unresolved(),
1233                        FrameMap::rcx_oop_opr,
1234                        FrameMap::rdi_oop_opr,
1235                        FrameMap::rsi_oop_opr,
1236                        LIR_OprFact::illegalOpr,
1237                        FrameMap::rdx_metadata_opr, info);
1238   LIR_Opr result = rlock_result(x);
1239   __ move(reg, result);
1240 }
1241 
1242 
1243 void LIRGenerator::do_NewTypeArray(NewTypeArray* x) {
1244   CodeEmitInfo* info = state_for(x, x->state());
1245 
1246   LIRItem length(x->length(), this);
1247   length.load_item_force(FrameMap::rbx_opr);
1248 
1249   LIR_Opr reg = result_register_for(x->type());
1250   LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
1251   LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
1252   LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
1253   LIR_Opr tmp4 = reg;
1254   LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
1255   LIR_Opr len = length.result();
1256   BasicType elem_type = x->elt_type();
1257 
1258   __ metadata2reg(ciTypeArrayKlass::make(elem_type)->constant_encoding(), klass_reg);
1259 
1260   CodeStub* slow_path = new NewTypeArrayStub(klass_reg, len, reg, info);
1261   __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, elem_type, klass_reg, slow_path);
1262 
1263   LIR_Opr result = rlock_result(x);
1264   __ move(reg, result);
1265 }
1266 
1267 
1268 void LIRGenerator::do_NewObjectArray(NewObjectArray* x) {
1269   LIRItem length(x->length(), this);
1270   // in case of patching (i.e., object class is not yet loaded), we need to reexecute the instruction
1271   // and therefore provide the state before the parameters have been consumed
1272   CodeEmitInfo* patching_info = NULL;
1273   if (!x->klass()->is_loaded() || PatchALot) {
1274     patching_info =  state_for(x, x->state_before());
1275   }
1276 
1277   CodeEmitInfo* info = state_for(x, x->state());
1278 
1279   const LIR_Opr reg = result_register_for(x->type());
1280   LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
1281   LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
1282   LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
1283   LIR_Opr tmp4 = reg;
1284   LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
1285 
1286   length.load_item_force(FrameMap::rbx_opr);
1287   LIR_Opr len = length.result();
1288 
1289   CodeStub* slow_path = new NewObjectArrayStub(klass_reg, len, reg, info);
1290   ciKlass* obj = (ciKlass*) ciObjArrayKlass::make(x->klass());
1291   if (obj == ciEnv::unloaded_ciobjarrayklass()) {
1292     BAILOUT("encountered unloaded_ciobjarrayklass due to out of memory error");
1293   }
1294   klass2reg_with_patching(klass_reg, obj, patching_info);
1295   __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, T_OBJECT, klass_reg, slow_path);
1296 
1297   LIR_Opr result = rlock_result(x);
1298   __ move(reg, result);
1299 }
1300 
1301 
1302 void LIRGenerator::do_NewMultiArray(NewMultiArray* x) {
1303   Values* dims = x->dims();
1304   int i = dims->length();
1305   LIRItemList* items = new LIRItemList(i, i, NULL);
1306   while (i-- > 0) {
1307     LIRItem* size = new LIRItem(dims->at(i), this);
1308     items->at_put(i, size);
1309   }
1310 
1311   // Evaluate state_for early since it may emit code.
1312   CodeEmitInfo* patching_info = NULL;
1313   if (!x->klass()->is_loaded() || PatchALot) {
1314     patching_info = state_for(x, x->state_before());
1315 
1316     // Cannot re-use same xhandlers for multiple CodeEmitInfos, so
1317     // clone all handlers (NOTE: Usually this is handled transparently
1318     // by the CodeEmitInfo cloning logic in CodeStub constructors but
1319     // is done explicitly here because a stub isn't being used).
1320     x->set_exception_handlers(new XHandlers(x->exception_handlers()));
1321   }
1322   CodeEmitInfo* info = state_for(x, x->state());
1323 
1324   i = dims->length();
1325   while (i-- > 0) {
1326     LIRItem* size = items->at(i);
1327     size->load_nonconstant();
1328 
1329     store_stack_parameter(size->result(), in_ByteSize(i*4));
1330   }
1331 
1332   LIR_Opr klass_reg = FrameMap::rax_metadata_opr;
1333   klass2reg_with_patching(klass_reg, x->klass(), patching_info);
1334 
1335   LIR_Opr rank = FrameMap::rbx_opr;
1336   __ move(LIR_OprFact::intConst(x->rank()), rank);
1337   LIR_Opr varargs = FrameMap::rcx_opr;
1338   __ move(FrameMap::rsp_opr, varargs);
1339   LIR_OprList* args = new LIR_OprList(3);
1340   args->append(klass_reg);
1341   args->append(rank);
1342   args->append(varargs);
1343   LIR_Opr reg = result_register_for(x->type());
1344   __ call_runtime(Runtime1::entry_for(Runtime1::new_multi_array_id),
1345                   LIR_OprFact::illegalOpr,
1346                   reg, args, info);
1347 
1348   LIR_Opr result = rlock_result(x);
1349   __ move(reg, result);
1350 }
1351 
1352 
1353 void LIRGenerator::do_BlockBegin(BlockBegin* x) {
1354   // nothing to do for now
1355 }
1356 
1357 
1358 void LIRGenerator::do_CheckCast(CheckCast* x) {
1359   LIRItem obj(x->obj(), this);
1360 
1361   CodeEmitInfo* patching_info = NULL;
1362   if (!x->klass()->is_loaded() || (PatchALot && !x->is_incompatible_class_change_check() && !x->is_invokespecial_receiver_check())) {
1363     // must do this before locking the destination register as an oop register,
1364     // and before the obj is loaded (the latter is for deoptimization)
1365     patching_info = state_for(x, x->state_before());
1366   }
1367   obj.load_item();
1368 
1369   // info for exceptions
1370   CodeEmitInfo* info_for_exception =
1371       (x->needs_exception_state() ? state_for(x) :
1372                                     state_for(x, x->state_before(), true /*ignore_xhandler*/));
1373 
1374   CodeStub* stub;
1375   if (x->is_incompatible_class_change_check()) {
1376     assert(patching_info == NULL, "can't patch this");
1377     stub = new SimpleExceptionStub(Runtime1::throw_incompatible_class_change_error_id, LIR_OprFact::illegalOpr, info_for_exception);
1378   } else if (x->is_invokespecial_receiver_check()) {
1379     assert(patching_info == NULL, "can't patch this");
1380     stub = new DeoptimizeStub(info_for_exception, Deoptimization::Reason_class_check, Deoptimization::Action_none);
1381   } else {
1382     stub = new SimpleExceptionStub(Runtime1::throw_class_cast_exception_id, obj.result(), info_for_exception);
1383   }
1384   LIR_Opr reg = rlock_result(x);
1385   LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
1386   if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
1387     tmp3 = new_register(objectType);
1388   }
1389   __ checkcast(reg, obj.result(), x->klass(),
1390                new_register(objectType), new_register(objectType), tmp3,
1391                x->direct_compare(), info_for_exception, patching_info, stub,
1392                x->profiled_method(), x->profiled_bci());
1393 }
1394 
1395 
1396 void LIRGenerator::do_InstanceOf(InstanceOf* x) {
1397   LIRItem obj(x->obj(), this);
1398 
1399   // result and test object may not be in same register
1400   LIR_Opr reg = rlock_result(x);
1401   CodeEmitInfo* patching_info = NULL;
1402   if ((!x->klass()->is_loaded() || PatchALot)) {
1403     // must do this before locking the destination register as an oop register
1404     patching_info = state_for(x, x->state_before());
1405   }
1406   obj.load_item();
1407   LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
1408   if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
1409     tmp3 = new_register(objectType);
1410   }
1411   __ instanceof(reg, obj.result(), x->klass(),
1412                 new_register(objectType), new_register(objectType), tmp3,
1413                 x->direct_compare(), patching_info, x->profiled_method(), x->profiled_bci());
1414 }
1415 
1416 
1417 void LIRGenerator::do_If(If* x) {
1418   assert(x->number_of_sux() == 2, "inconsistency");
1419   ValueTag tag = x->x()->type()->tag();
1420   bool is_safepoint = x->is_safepoint();
1421 
1422   If::Condition cond = x->cond();
1423 
1424   LIRItem xitem(x->x(), this);
1425   LIRItem yitem(x->y(), this);
1426   LIRItem* xin = &xitem;
1427   LIRItem* yin = &yitem;
1428 
1429   if (tag == longTag) {
1430     // for longs, only conditions "eql", "neq", "lss", "geq" are valid;
1431     // mirror for other conditions
1432     if (cond == If::gtr || cond == If::leq) {
1433       cond = Instruction::mirror(cond);
1434       xin = &yitem;
1435       yin = &xitem;
1436     }
1437     xin->set_destroys_register();
1438   }
1439   xin->load_item();
1440   if (tag == longTag && yin->is_constant() && yin->get_jlong_constant() == 0 && (cond == If::eql || cond == If::neq)) {
1441     // inline long zero
1442     yin->dont_load_item();
1443   } else if (tag == longTag || tag == floatTag || tag == doubleTag) {
1444     // longs cannot handle constants at right side
1445     yin->load_item();
1446   } else {
1447     yin->dont_load_item();
1448   }
1449 
1450   LIR_Opr left = xin->result();
1451   LIR_Opr right = yin->result();
1452 
1453   set_no_result(x);
1454 
1455   // add safepoint before generating condition code so it can be recomputed
1456   if (x->is_safepoint()) {
1457     // increment backedge counter if needed
1458     increment_backedge_counter_conditionally(lir_cond(cond), left, right, state_for(x, x->state_before()),
1459         x->tsux()->bci(), x->fsux()->bci(), x->profiled_bci());
1460     __ safepoint(safepoint_poll_register(), state_for(x, x->state_before()));
1461   }
1462 
1463   __ cmp(lir_cond(cond), left, right);
1464   // Generate branch profiling. Profiling code doesn't kill flags.
1465   profile_branch(x, cond);
1466   move_to_phi(x->state());
1467   if (x->x()->type()->is_float_kind()) {
1468     __ branch(lir_cond(cond), right->type(), x->tsux(), x->usux());
1469   } else {
1470     __ branch(lir_cond(cond), right->type(), x->tsux());
1471   }
1472   assert(x->default_sux() == x->fsux(), "wrong destination above");
1473   __ jump(x->default_sux());
1474 }
1475 
1476 
1477 LIR_Opr LIRGenerator::getThreadPointer() {
1478 #ifdef _LP64
1479   return FrameMap::as_pointer_opr(r15_thread);
1480 #else
1481   LIR_Opr result = new_register(T_INT);
1482   __ get_thread(result);
1483   return result;
1484 #endif //
1485 }
1486 
1487 void LIRGenerator::trace_block_entry(BlockBegin* block) {
1488   store_stack_parameter(LIR_OprFact::intConst(block->block_id()), in_ByteSize(0));
1489   LIR_OprList* args = new LIR_OprList();
1490   address func = CAST_FROM_FN_PTR(address, Runtime1::trace_block_entry);
1491   __ call_runtime_leaf(func, LIR_OprFact::illegalOpr, LIR_OprFact::illegalOpr, args);
1492 }
1493 
1494 
1495 void LIRGenerator::volatile_field_store(LIR_Opr value, LIR_Address* address,
1496                                         CodeEmitInfo* info) {
1497   if (address->type() == T_LONG) {
1498     address = new LIR_Address(address->base(),
1499                               address->index(), address->scale(),
1500                               address->disp(), T_DOUBLE);
1501     // Transfer the value atomically by using FP moves.  This means
1502     // the value has to be moved between CPU and FPU registers.  It
1503     // always has to be moved through spill slot since there's no
1504     // quick way to pack the value into an SSE register.
1505     LIR_Opr temp_double = new_register(T_DOUBLE);
1506     LIR_Opr spill = new_register(T_LONG);
1507     set_vreg_flag(spill, must_start_in_memory);
1508     __ move(value, spill);
1509     __ volatile_move(spill, temp_double, T_LONG);
1510     __ volatile_move(temp_double, LIR_OprFact::address(address), T_LONG, info);
1511   } else {
1512     __ store(value, address, info);
1513   }
1514 }
1515 
1516 void LIRGenerator::volatile_field_load(LIR_Address* address, LIR_Opr result,
1517                                        CodeEmitInfo* info) {
1518   if (address->type() == T_LONG) {
1519     address = new LIR_Address(address->base(),
1520                               address->index(), address->scale(),
1521                               address->disp(), T_DOUBLE);
1522     // Transfer the value atomically by using FP moves.  This means
1523     // the value has to be moved between CPU and FPU registers.  In
1524     // SSE0 and SSE1 mode it has to be moved through spill slot but in
1525     // SSE2+ mode it can be moved directly.
1526     LIR_Opr temp_double = new_register(T_DOUBLE);
1527     __ volatile_move(LIR_OprFact::address(address), temp_double, T_LONG, info);
1528     __ volatile_move(temp_double, result, T_LONG);
1529     if (UseSSE < 2) {
1530       // no spill slot needed in SSE2 mode because xmm->cpu register move is possible
1531       set_vreg_flag(result, must_start_in_memory);
1532     }
1533   } else {
1534     __ load(address, result, info);
1535   }
1536 }