1 /*
   2  * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "c1/c1_Compilation.hpp"
  28 #include "c1/c1_LIRAssembler.hpp"
  29 #include "c1/c1_MacroAssembler.hpp"
  30 #include "c1/c1_Runtime1.hpp"
  31 #include "c1/c1_ValueStack.hpp"
  32 #include "ci/ciArrayKlass.hpp"
  33 #include "ci/ciInstance.hpp"
  34 #include "gc/shared/collectedHeap.hpp"
  35 #include "gc/shared/barrierSet.hpp"
  36 #include "gc/shared/cardTableModRefBS.hpp"
  37 #include "nativeInst_s390.hpp"
  38 #include "oops/objArrayKlass.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "vmreg_s390.inline.hpp"
  41 
  42 #define __ _masm->
  43 
  44 #ifndef PRODUCT
  45 #undef __
  46 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm) : _masm)->
  47 #endif
  48 
  49 //------------------------------------------------------------
  50 
  51 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  52   // Not used on ZARCH_64
  53   ShouldNotCallThis();
  54   return false;
  55 }
  56 
  57 LIR_Opr LIR_Assembler::receiverOpr() {
  58   return FrameMap::Z_R2_oop_opr;
  59 }
  60 
  61 LIR_Opr LIR_Assembler::osrBufferPointer() {
  62   return FrameMap::Z_R2_opr;
  63 }
  64 
  65 int LIR_Assembler::initial_frame_size_in_bytes() const {
  66   return in_bytes(frame_map()->framesize_in_bytes());
  67 }
  68 
  69 // Inline cache check: done before the frame is built.
  70 // The inline cached class is in Z_inline_cache(Z_R9).
  71 // We fetch the class of the receiver and compare it with the cached class.
  72 // If they do not match we jump to the slow case.
  73 int LIR_Assembler::check_icache() {
  74   Register receiver = receiverOpr()->as_register();
  75   int offset = __ offset();
  76   __ inline_cache_check(receiver, Z_inline_cache);
  77   return offset;
  78 }
  79 
  80 void LIR_Assembler::osr_entry() {
  81   // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
  82   //
  83   //   1. Create a new compiled activation.
  84   //   2. Initialize local variables in the compiled activation. The expression stack must be empty
  85   //      at the osr_bci; it is not initialized.
  86   //   3. Jump to the continuation address in compiled code to resume execution.
  87 
  88   // OSR entry point
  89   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
  90   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
  91   ValueStack* entry_state = osr_entry->end()->state();
  92   int number_of_locks = entry_state->locks_size();
  93 
  94   // Create a frame for the compiled activation.
  95   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
  96 
  97   // OSR buffer is
  98   //
  99   // locals[nlocals-1..0]
 100   // monitors[number_of_locks-1..0]
 101   //
 102   // Locals is a direct copy of the interpreter frame so in the osr buffer
 103   // the first slot in the local array is the last local from the interpreter
 104   // and the last slot is local[0] (receiver) from the interpreter
 105   //
 106   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 107   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 108   // in the interpreter frame (the method lock if a sync method)
 109 
 110   // Initialize monitors in the compiled activation.
 111   //   I0: pointer to osr buffer
 112   //
 113   // All other registers are dead at this point and the locals will be
 114   // copied into place by code emitted in the IR.
 115 
 116   Register OSR_buf = osrBufferPointer()->as_register();
 117   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 118     int monitor_offset = BytesPerWord * method()->max_locals() +
 119       (2 * BytesPerWord) * (number_of_locks - 1);
 120     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 121     // the OSR buffer using 2 word entries: first the lock and then
 122     // the oop.
 123     for (int i = 0; i < number_of_locks; i++) {
 124       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 125       // Verify the interpreter's monitor has a non-null object.
 126       __ asm_assert_mem8_isnot_zero(slot_offset + 1*BytesPerWord, OSR_buf, "locked object is NULL", __LINE__);
 127       // Copy the lock field into the compiled activation.
 128       __ z_lg(Z_R1_scratch, slot_offset + 0, OSR_buf);
 129       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_lock(i));
 130       __ z_lg(Z_R1_scratch, slot_offset + 1*BytesPerWord, OSR_buf);
 131       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_object(i));
 132     }
 133   }
 134 }
 135 
 136 // --------------------------------------------------------------------------------------------
 137 
 138 address LIR_Assembler::emit_call_c(address a) {
 139   __ align_call_far_patchable(__ pc());
 140   address call_addr = __ call_c_opt(a);
 141   if (call_addr == NULL) {
 142     bailout("const section overflow");
 143   }
 144   return call_addr;
 145 }
 146 
 147 int LIR_Assembler::emit_exception_handler() {
 148   // If the last instruction is a call (typically to do a throw which
 149   // is coming at the end after block reordering) the return address
 150   // must still point into the code area in order to avoid assertion
 151   // failures when searching for the corresponding bci. => Add a nop.
 152   // (was bug 5/14/1999 - gri)
 153   __ nop();
 154 
 155   // Generate code for exception handler.
 156   address handler_base = __ start_a_stub(exception_handler_size());
 157   if (handler_base == NULL) {
 158     // Not enough space left for the handler.
 159     bailout("exception handler overflow");
 160     return -1;
 161   }
 162 
 163   int offset = code_offset();
 164 
 165   address a = Runtime1::entry_for (Runtime1::handle_exception_from_callee_id);
 166   address call_addr = emit_call_c(a);
 167   CHECK_BAILOUT_(-1);
 168   __ should_not_reach_here();
 169   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 170   __ end_a_stub();
 171 
 172   return offset;
 173 }
 174 
 175 // Emit the code to remove the frame from the stack in the exception
 176 // unwind path.
 177 int LIR_Assembler::emit_unwind_handler() {
 178 #ifndef PRODUCT
 179   if (CommentedAssembly) {
 180     _masm->block_comment("Unwind handler");
 181   }
 182 #endif
 183 
 184   int offset = code_offset();
 185   Register exception_oop_callee_saved = Z_R10; // Z_R10 is callee-saved.
 186   Register Rtmp1                      = Z_R11;
 187   Register Rtmp2                      = Z_R12;
 188 
 189   // Fetch the exception from TLS and clear out exception related thread state.
 190   Address exc_oop_addr = Address(Z_thread, JavaThread::exception_oop_offset());
 191   Address exc_pc_addr  = Address(Z_thread, JavaThread::exception_pc_offset());
 192   __ z_lg(Z_EXC_OOP, exc_oop_addr);
 193   __ clear_mem(exc_oop_addr, sizeof(oop));
 194   __ clear_mem(exc_pc_addr, sizeof(intptr_t));
 195 
 196   __ bind(_unwind_handler_entry);
 197   __ verify_not_null_oop(Z_EXC_OOP);
 198   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 199     __ lgr_if_needed(exception_oop_callee_saved, Z_EXC_OOP); // Preserve the exception.
 200   }
 201 
 202   // Preform needed unlocking.
 203   MonitorExitStub* stub = NULL;
 204   if (method()->is_synchronized()) {
 205     // Runtime1::monitorexit_id expects lock address in Z_R1_scratch.
 206     LIR_Opr lock = FrameMap::as_opr(Z_R1_scratch);
 207     monitor_address(0, lock);
 208     stub = new MonitorExitStub(lock, true, 0);
 209     __ unlock_object(Rtmp1, Rtmp2, lock->as_register(), *stub->entry());
 210     __ bind(*stub->continuation());
 211   }
 212 
 213   if (compilation()->env()->dtrace_method_probes()) {
 214     ShouldNotReachHere(); // Not supported.
 215 #if 0
 216     __ mov(rdi, r15_thread);
 217     __ mov_metadata(rsi, method()->constant_encoding());
 218     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
 219 #endif
 220   }
 221 
 222   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 223     __ lgr_if_needed(Z_EXC_OOP, exception_oop_callee_saved);  // Restore the exception.
 224   }
 225 
 226   // Remove the activation and dispatch to the unwind handler.
 227   __ pop_frame();
 228   __ z_lg(Z_EXC_PC, _z_abi16(return_pc), Z_SP);
 229 
 230   // Z_EXC_OOP: exception oop
 231   // Z_EXC_PC: exception pc
 232 
 233   // Dispatch to the unwind logic.
 234   __ load_const_optimized(Z_R5, Runtime1::entry_for (Runtime1::unwind_exception_id));
 235   __ z_br(Z_R5);
 236 
 237   // Emit the slow path assembly.
 238   if (stub != NULL) {
 239     stub->emit_code(this);
 240   }
 241 
 242   return offset;
 243 }
 244 
 245 int LIR_Assembler::emit_deopt_handler() {
 246   // If the last instruction is a call (typically to do a throw which
 247   // is coming at the end after block reordering) the return address
 248   // must still point into the code area in order to avoid assertion
 249   // failures when searching for the corresponding bci. => Add a nop.
 250   // (was bug 5/14/1999 - gri)
 251   __ nop();
 252 
 253   // Generate code for exception handler.
 254   address handler_base = __ start_a_stub(deopt_handler_size());
 255   if (handler_base == NULL) {
 256     // Not enough space left for the handler.
 257     bailout("deopt handler overflow");
 258     return -1;
 259   }  int offset = code_offset();
 260   // Size must be constant (see HandlerImpl::emit_deopt_handler).
 261   __ load_const(Z_R1_scratch, SharedRuntime::deopt_blob()->unpack());
 262   __ call(Z_R1_scratch);
 263   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 264   __ end_a_stub();
 265 
 266   return offset;
 267 }
 268 
 269 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 270   if (o == NULL) {
 271     __ clear_reg(reg, true/*64bit*/, false/*set cc*/); // Must not kill cc set by cmove.
 272   } else {
 273     AddressLiteral a = __ allocate_oop_address(o);
 274     bool success = __ load_oop_from_toc(reg, a, reg);
 275     if (!success) {
 276       bailout("const section overflow");
 277     }
 278   }
 279 }
 280 
 281 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 282   // Allocate a new index in table to hold the object once it's been patched.
 283   int oop_index = __ oop_recorder()->allocate_oop_index(NULL);
 284   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
 285 
 286   AddressLiteral addrlit((intptr_t)0, oop_Relocation::spec(oop_index));
 287   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 288   // The NULL will be dynamically patched later so the sequence to
 289   // load the address literal must not be optimized.
 290   __ load_const(reg, addrlit);
 291 
 292   patching_epilog(patch, lir_patch_normal, reg, info);
 293 }
 294 
 295 void LIR_Assembler::metadata2reg(Metadata* md, Register reg) {
 296   bool success = __ set_metadata_constant(md, reg);
 297   if (!success) {
 298     bailout("const section overflow");
 299     return;
 300   }
 301 }
 302 
 303 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
 304   // Allocate a new index in table to hold the klass once it's been patched.
 305   int index = __ oop_recorder()->allocate_metadata_index(NULL);
 306   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 307   AddressLiteral addrlit((intptr_t)0, metadata_Relocation::spec(index));
 308   assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
 309   // The NULL will be dynamically patched later so the sequence to
 310   // load the address literal must not be optimized.
 311   __ load_const(reg, addrlit);
 312 
 313   patching_epilog(patch, lir_patch_normal, reg, info);
 314 }
 315 
 316 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 317   switch (op->code()) {
 318     case lir_idiv:
 319     case lir_irem:
 320       arithmetic_idiv(op->code(),
 321                       op->in_opr1(),
 322                       op->in_opr2(),
 323                       op->in_opr3(),
 324                       op->result_opr(),
 325                       op->info());
 326       break;
 327     default: ShouldNotReachHere(); break;
 328   }
 329 }
 330 
 331 
 332 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 333 #ifdef ASSERT
 334   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 335   if (op->block() != NULL)  { _branch_target_blocks.append(op->block()); }
 336   if (op->ublock() != NULL) { _branch_target_blocks.append(op->ublock()); }
 337 #endif
 338 
 339   if (op->cond() == lir_cond_always) {
 340     if (op->info() != NULL) { add_debug_info_for_branch(op->info()); }
 341     __ branch_optimized(Assembler::bcondAlways, *(op->label()));
 342   } else {
 343     Assembler::branch_condition acond = Assembler::bcondZero;
 344     if (op->code() == lir_cond_float_branch) {
 345       assert(op->ublock() != NULL, "must have unordered successor");
 346       __ branch_optimized(Assembler::bcondNotOrdered, *(op->ublock()->label()));
 347     }
 348     switch (op->cond()) {
 349       case lir_cond_equal:        acond = Assembler::bcondEqual;     break;
 350       case lir_cond_notEqual:     acond = Assembler::bcondNotEqual;  break;
 351       case lir_cond_less:         acond = Assembler::bcondLow;       break;
 352       case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;   break;
 353       case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;    break;
 354       case lir_cond_greater:      acond = Assembler::bcondHigh;      break;
 355       case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;   break;
 356       case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;    break;
 357       default:                         ShouldNotReachHere();
 358     }
 359     __ branch_optimized(acond,*(op->label()));
 360   }
 361 }
 362 
 363 
 364 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 365   LIR_Opr src  = op->in_opr();
 366   LIR_Opr dest = op->result_opr();
 367 
 368   switch (op->bytecode()) {
 369     case Bytecodes::_i2l:
 370       __ move_reg_if_needed(dest->as_register_lo(), T_LONG, src->as_register(), T_INT);
 371       break;
 372 
 373     case Bytecodes::_l2i:
 374       __ move_reg_if_needed(dest->as_register(), T_INT, src->as_register_lo(), T_LONG);
 375       break;
 376 
 377     case Bytecodes::_i2b:
 378       __ move_reg_if_needed(dest->as_register(), T_BYTE, src->as_register(), T_INT);
 379       break;
 380 
 381     case Bytecodes::_i2c:
 382       __ move_reg_if_needed(dest->as_register(), T_CHAR, src->as_register(), T_INT);
 383       break;
 384 
 385     case Bytecodes::_i2s:
 386       __ move_reg_if_needed(dest->as_register(), T_SHORT, src->as_register(), T_INT);
 387       break;
 388 
 389     case Bytecodes::_f2d:
 390       assert(dest->is_double_fpu(), "check");
 391       __ move_freg_if_needed(dest->as_double_reg(), T_DOUBLE, src->as_float_reg(), T_FLOAT);
 392       break;
 393 
 394     case Bytecodes::_d2f:
 395       assert(dest->is_single_fpu(), "check");
 396       __ move_freg_if_needed(dest->as_float_reg(), T_FLOAT, src->as_double_reg(), T_DOUBLE);
 397       break;
 398 
 399     case Bytecodes::_i2f:
 400       __ z_cefbr(dest->as_float_reg(), src->as_register());
 401       break;
 402 
 403     case Bytecodes::_i2d:
 404       __ z_cdfbr(dest->as_double_reg(), src->as_register());
 405       break;
 406 
 407     case Bytecodes::_l2f:
 408       __ z_cegbr(dest->as_float_reg(), src->as_register_lo());
 409       break;
 410     case Bytecodes::_l2d:
 411       __ z_cdgbr(dest->as_double_reg(), src->as_register_lo());
 412       break;
 413 
 414     case Bytecodes::_f2i:
 415     case Bytecodes::_f2l: {
 416       Label done;
 417       FloatRegister Rsrc = src->as_float_reg();
 418       Register Rdst = (op->bytecode() == Bytecodes::_f2i ? dest->as_register() : dest->as_register_lo());
 419       __ clear_reg(Rdst, true, false);
 420       __ z_cebr(Rsrc, Rsrc);
 421       __ z_brno(done); // NaN -> 0
 422       if (op->bytecode() == Bytecodes::_f2i) {
 423         __ z_cfebr(Rdst, Rsrc, Assembler::to_zero);
 424       } else { // op->bytecode() == Bytecodes::_f2l
 425         __ z_cgebr(Rdst, Rsrc, Assembler::to_zero);
 426       }
 427       __ bind(done);
 428     }
 429     break;
 430 
 431     case Bytecodes::_d2i:
 432     case Bytecodes::_d2l: {
 433       Label done;
 434       FloatRegister Rsrc = src->as_double_reg();
 435       Register Rdst = (op->bytecode() == Bytecodes::_d2i ? dest->as_register() : dest->as_register_lo());
 436       __ clear_reg(Rdst, true, false);  // Don't set CC.
 437       __ z_cdbr(Rsrc, Rsrc);
 438       __ z_brno(done); // NaN -> 0
 439       if (op->bytecode() == Bytecodes::_d2i) {
 440         __ z_cfdbr(Rdst, Rsrc, Assembler::to_zero);
 441       } else { // Bytecodes::_d2l
 442         __ z_cgdbr(Rdst, Rsrc, Assembler::to_zero);
 443       }
 444       __ bind(done);
 445     }
 446     break;
 447 
 448     default: ShouldNotReachHere();
 449   }
 450 }
 451 
 452 void LIR_Assembler::align_call(LIR_Code code) {
 453   // End of call instruction must be 4 byte aligned.
 454   int offset = __ offset();
 455   switch (code) {
 456     case lir_icvirtual_call:
 457       offset += MacroAssembler::load_const_from_toc_size();
 458       // no break
 459     case lir_static_call:
 460     case lir_optvirtual_call:
 461     case lir_dynamic_call:
 462       offset += NativeCall::call_far_pcrelative_displacement_offset;
 463       break;
 464     case lir_virtual_call:   // currently, sparc-specific for niagara
 465     default: ShouldNotReachHere();
 466   }
 467   if ((offset & (NativeCall::call_far_pcrelative_displacement_alignment-1)) != 0) {
 468     __ nop();
 469   }
 470 }
 471 
 472 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 473   assert((__ offset() + NativeCall::call_far_pcrelative_displacement_offset) % NativeCall::call_far_pcrelative_displacement_alignment == 0,
 474          "must be aligned (offset=%d)", __ offset());
 475   assert(rtype == relocInfo::none ||
 476          rtype == relocInfo::opt_virtual_call_type ||
 477          rtype == relocInfo::static_call_type, "unexpected rtype");
 478   // Prepend each BRASL with a nop.
 479   __ relocate(rtype);
 480   __ z_nop();
 481   __ z_brasl(Z_R14, op->addr());
 482   add_call_info(code_offset(), op->info());
 483 }
 484 
 485 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 486   address virtual_call_oop_addr = NULL;
 487   AddressLiteral empty_ic((address) Universe::non_oop_word());
 488   virtual_call_oop_addr = __ pc();
 489   bool success = __ load_const_from_toc(Z_inline_cache, empty_ic);
 490   if (!success) {
 491     bailout("const section overflow");
 492     return;
 493   }
 494 
 495   // CALL to fixup routine. Fixup routine uses ScopeDesc info
 496   // to determine who we intended to call.
 497   __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
 498   call(op, relocInfo::none);
 499 }
 500 
 501 // not supported
 502 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
 503   ShouldNotReachHere();
 504 }
 505 
 506 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 507   if (from_reg != to_reg) __ z_lgr(to_reg, from_reg);
 508 }
 509 
 510 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 511   assert(src->is_constant(), "should not call otherwise");
 512   assert(dest->is_stack(), "should not call otherwise");
 513   LIR_Const* c = src->as_constant_ptr();
 514 
 515   unsigned int lmem = 0;
 516   unsigned int lcon = 0;
 517   int64_t cbits = 0;
 518   Address dest_addr;
 519   switch (c->type()) {
 520     case T_INT:  // fall through
 521     case T_FLOAT:
 522       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 523       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 524       break;
 525 
 526     case T_ADDRESS:
 527       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 528       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 529       break;
 530 
 531     case T_OBJECT:
 532       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 533       if (c->as_jobject() == NULL) {
 534         __ store_const(dest_addr, (int64_t)NULL_WORD, 8, 8);
 535       } else {
 536         jobject2reg(c->as_jobject(), Z_R1_scratch);
 537         __ reg2mem_opt(Z_R1_scratch, dest_addr, true);
 538       }
 539       return;
 540 
 541     case T_LONG:  // fall through
 542     case T_DOUBLE:
 543       dest_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 544       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 545       break;
 546 
 547     default:
 548       ShouldNotReachHere();
 549   }
 550 
 551   __ store_const(dest_addr, cbits, lmem, lcon);
 552 }
 553 
 554 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 555   assert(src->is_constant(), "should not call otherwise");
 556   assert(dest->is_address(), "should not call otherwise");
 557   // See special case in LIRGenerator::do_StoreIndexed.
 558   // T_BYTE: Special case for card mark store.
 559   assert(type == T_BYTE || !dest->as_address_ptr()->index()->is_valid(), "not supported");
 560   LIR_Const* c = src->as_constant_ptr();
 561   Address addr = as_Address(dest->as_address_ptr());
 562 
 563   int store_offset = -1;
 564   unsigned int lmem = 0;
 565   unsigned int lcon = 0;
 566   int64_t cbits = 0;
 567   switch (type) {
 568     case T_INT:    // fall through
 569     case T_FLOAT:
 570       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 571       break;
 572 
 573     case T_ADDRESS:
 574       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 575       break;
 576 
 577     case T_OBJECT:  // fall through
 578     case T_ARRAY:
 579       if (c->as_jobject() == NULL) {
 580         if (UseCompressedOops && !wide) {
 581           store_offset = __ store_const(addr, (int32_t)NULL_WORD, 4, 4);
 582         } else {
 583           store_offset = __ store_const(addr, (int64_t)NULL_WORD, 8, 8);
 584         }
 585       } else {
 586         jobject2reg(c->as_jobject(), Z_R1_scratch);
 587         if (UseCompressedOops && !wide) {
 588           __ encode_heap_oop(Z_R1_scratch);
 589           store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 590         } else {
 591           store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 592         }
 593       }
 594       assert(store_offset >= 0, "check");
 595       break;
 596 
 597     case T_LONG:    // fall through
 598     case T_DOUBLE:
 599       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 600       break;
 601 
 602     case T_BOOLEAN: // fall through
 603     case T_BYTE:
 604       lmem = 1; lcon = 1; cbits = (int8_t)(c->as_jint());
 605       break;
 606 
 607     case T_CHAR:    // fall through
 608     case T_SHORT:
 609       lmem = 2; lcon = 2; cbits = (int16_t)(c->as_jint());
 610       break;
 611 
 612     default:
 613       ShouldNotReachHere();
 614   };
 615 
 616   // Index register is normally not supported, but for
 617   // LIRGenerator::CardTableModRef_post_barrier we make an exception.
 618   if (type == T_BYTE && dest->as_address_ptr()->index()->is_valid()) {
 619     __ load_const_optimized(Z_R0_scratch, (int8_t)(c->as_jint()));
 620     store_offset = __ offset();
 621     if (Immediate::is_uimm12(addr.disp())) {
 622       __ z_stc(Z_R0_scratch, addr);
 623     } else {
 624       __ z_stcy(Z_R0_scratch, addr);
 625     }
 626   }
 627 
 628   if (store_offset == -1) {
 629     store_offset = __ store_const(addr, cbits, lmem, lcon);
 630     assert(store_offset >= 0, "check");
 631   }
 632 
 633   if (info != NULL) {
 634     add_debug_info_for_null_check(store_offset, info);
 635   }
 636 }
 637 
 638 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 639   assert(src->is_constant(), "should not call otherwise");
 640   assert(dest->is_register(), "should not call otherwise");
 641   LIR_Const* c = src->as_constant_ptr();
 642 
 643   switch (c->type()) {
 644     case T_INT: {
 645       assert(patch_code == lir_patch_none, "no patching handled here");
 646       __ load_const_optimized(dest->as_register(), c->as_jint());
 647       break;
 648     }
 649 
 650     case T_ADDRESS: {
 651       assert(patch_code == lir_patch_none, "no patching handled here");
 652       __ load_const_optimized(dest->as_register(), c->as_jint());
 653       break;
 654     }
 655 
 656     case T_LONG: {
 657       assert(patch_code == lir_patch_none, "no patching handled here");
 658       __ load_const_optimized(dest->as_register_lo(), (intptr_t)c->as_jlong());
 659       break;
 660     }
 661 
 662     case T_OBJECT: {
 663       if (patch_code != lir_patch_none) {
 664         jobject2reg_with_patching(dest->as_register(), info);
 665       } else {
 666         jobject2reg(c->as_jobject(), dest->as_register());
 667       }
 668       break;
 669     }
 670 
 671     case T_METADATA: {
 672       if (patch_code != lir_patch_none) {
 673         klass2reg_with_patching(dest->as_register(), info);
 674       } else {
 675         metadata2reg(c->as_metadata(), dest->as_register());
 676       }
 677       break;
 678     }
 679 
 680     case T_FLOAT: {
 681       Register toc_reg = Z_R1_scratch;
 682       __ load_toc(toc_reg);
 683       address const_addr = __ float_constant(c->as_jfloat());
 684       if (const_addr == NULL) {
 685         bailout("const section overflow");
 686         break;
 687       }
 688       int displ = const_addr - _masm->code()->consts()->start();
 689       if (dest->is_single_fpu()) {
 690         __ z_ley(dest->as_float_reg(), displ, toc_reg);
 691       } else {
 692         assert(dest->is_single_cpu(), "Must be a cpu register.");
 693         __ z_ly(dest->as_register(), displ, toc_reg);
 694       }
 695     }
 696     break;
 697 
 698     case T_DOUBLE: {
 699       Register toc_reg = Z_R1_scratch;
 700       __ load_toc(toc_reg);
 701       address const_addr = __ double_constant(c->as_jdouble());
 702       if (const_addr == NULL) {
 703         bailout("const section overflow");
 704         break;
 705       }
 706       int displ = const_addr - _masm->code()->consts()->start();
 707       if (dest->is_double_fpu()) {
 708         __ z_ldy(dest->as_double_reg(), displ, toc_reg);
 709       } else {
 710         assert(dest->is_double_cpu(), "Must be a long register.");
 711         __ z_lg(dest->as_register_lo(), displ, toc_reg);
 712       }
 713     }
 714     break;
 715 
 716     default:
 717       ShouldNotReachHere();
 718   }
 719 }
 720 
 721 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 722   if (addr->base()->is_illegal()) {
 723     Unimplemented();
 724   }
 725 
 726   Register base = addr->base()->as_pointer_register();
 727 
 728   if (addr->index()->is_illegal()) {
 729     return Address(base, addr->disp());
 730   } else if (addr->index()->is_cpu_register()) {
 731     Register index = addr->index()->as_pointer_register();
 732     return Address(base, index, addr->disp());
 733   } else if (addr->index()->is_constant()) {
 734     intptr_t addr_offset = addr->index()->as_constant_ptr()->as_jint() + addr->disp();
 735     return Address(base, addr_offset);
 736   } else {
 737     ShouldNotReachHere();
 738     return Address();
 739   }
 740 }
 741 
 742 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 743   switch (type) {
 744     case T_INT:
 745     case T_FLOAT: {
 746       Register tmp = Z_R1_scratch;
 747       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 748       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 749       __ mem2reg_opt(tmp, from, false);
 750       __ reg2mem_opt(tmp, to, false);
 751       break;
 752     }
 753     case T_ADDRESS:
 754     case T_OBJECT: {
 755       Register tmp = Z_R1_scratch;
 756       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 757       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 758       __ mem2reg_opt(tmp, from, true);
 759       __ reg2mem_opt(tmp, to, true);
 760       break;
 761     }
 762     case T_LONG:
 763     case T_DOUBLE: {
 764       Register tmp = Z_R1_scratch;
 765       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
 766       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
 767       __ mem2reg_opt(tmp, from, true);
 768       __ reg2mem_opt(tmp, to, true);
 769       break;
 770     }
 771 
 772     default:
 773       ShouldNotReachHere();
 774   }
 775 }
 776 
 777 // 4-byte accesses only! Don't use it to access 8 bytes!
 778 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 779   ShouldNotCallThis();
 780   return 0; // unused
 781 }
 782 
 783 // 4-byte accesses only! Don't use it to access 8 bytes!
 784 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 785   ShouldNotCallThis();
 786   return 0; // unused
 787 }
 788 
 789 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code,
 790                             CodeEmitInfo* info, bool wide, bool unaligned) {
 791 
 792   assert(type != T_METADATA, "load of metadata ptr not supported");
 793   LIR_Address* addr = src_opr->as_address_ptr();
 794   LIR_Opr to_reg = dest;
 795 
 796   Register src = addr->base()->as_pointer_register();
 797   Register disp_reg = Z_R0;
 798   int disp_value = addr->disp();
 799   bool needs_patching = (patch_code != lir_patch_none);
 800 
 801   if (addr->base()->type() == T_OBJECT) {
 802     __ verify_oop(src);
 803   }
 804 
 805   PatchingStub* patch = NULL;
 806   if (needs_patching) {
 807     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 808     assert(!to_reg->is_double_cpu() ||
 809            patch_code == lir_patch_none ||
 810            patch_code == lir_patch_normal, "patching doesn't match register");
 811   }
 812 
 813   if (addr->index()->is_illegal()) {
 814     if (!Immediate::is_simm20(disp_value)) {
 815       if (needs_patching) {
 816         __ load_const(Z_R1_scratch, (intptr_t)0);
 817       } else {
 818         __ load_const_optimized(Z_R1_scratch, disp_value);
 819       }
 820       disp_reg = Z_R1_scratch;
 821       disp_value = 0;
 822     }
 823   } else {
 824     if (!Immediate::is_simm20(disp_value)) {
 825       __ load_const_optimized(Z_R1_scratch, disp_value);
 826       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
 827       disp_reg = Z_R1_scratch;
 828       disp_value = 0;
 829     }
 830     disp_reg = addr->index()->as_pointer_register();
 831   }
 832 
 833   // Remember the offset of the load. The patching_epilog must be done
 834   // before the call to add_debug_info, otherwise the PcDescs don't get
 835   // entered in increasing order.
 836   int offset = code_offset();
 837 
 838   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
 839 
 840   bool short_disp = Immediate::is_uimm12(disp_value);
 841 
 842   switch (type) {
 843     case T_BOOLEAN: // fall through
 844     case T_BYTE  :  __ z_lb(dest->as_register(),   disp_value, disp_reg, src); break;
 845     case T_CHAR  :  __ z_llgh(dest->as_register(), disp_value, disp_reg, src); break;
 846     case T_SHORT :
 847       if (short_disp) {
 848                     __ z_lh(dest->as_register(),   disp_value, disp_reg, src);
 849       } else {
 850                     __ z_lhy(dest->as_register(),  disp_value, disp_reg, src);
 851       }
 852       break;
 853     case T_INT   :
 854       if (short_disp) {
 855                     __ z_l(dest->as_register(),    disp_value, disp_reg, src);
 856       } else {
 857                     __ z_ly(dest->as_register(),   disp_value, disp_reg, src);
 858       }
 859       break;
 860     case T_ADDRESS:
 861       if (UseCompressedClassPointers && addr->disp() == oopDesc::klass_offset_in_bytes()) {
 862         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 863         __ decode_klass_not_null(dest->as_register());
 864       } else {
 865         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 866       }
 867       break;
 868     case T_ARRAY : // fall through
 869     case T_OBJECT:
 870     {
 871       if (UseCompressedOops && !wide) {
 872         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 873         __ oop_decoder(dest->as_register(), dest->as_register(), true);
 874       } else {
 875         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 876       }
 877       break;
 878     }
 879     case T_FLOAT:
 880       if (short_disp) {
 881                     __ z_le(dest->as_float_reg(),  disp_value, disp_reg, src);
 882       } else {
 883                     __ z_ley(dest->as_float_reg(), disp_value, disp_reg, src);
 884       }
 885       break;
 886     case T_DOUBLE:
 887       if (short_disp) {
 888                     __ z_ld(dest->as_double_reg(),  disp_value, disp_reg, src);
 889       } else {
 890                     __ z_ldy(dest->as_double_reg(), disp_value, disp_reg, src);
 891       }
 892       break;
 893     case T_LONG  :  __ z_lg(dest->as_register_lo(), disp_value, disp_reg, src); break;
 894     default      : ShouldNotReachHere();
 895   }
 896   if (type == T_ARRAY || type == T_OBJECT) {
 897     __ verify_oop(dest->as_register());
 898   }
 899 
 900   if (patch != NULL) {
 901     patching_epilog(patch, patch_code, src, info);
 902   }
 903   if (info != NULL) add_debug_info_for_null_check(offset, info);
 904 }
 905 
 906 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
 907   assert(src->is_stack(), "should not call otherwise");
 908   assert(dest->is_register(), "should not call otherwise");
 909 
 910   if (dest->is_single_cpu()) {
 911     if (type == T_ARRAY || type == T_OBJECT) {
 912       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 913       __ verify_oop(dest->as_register());
 914     } else if (type == T_METADATA) {
 915       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
 916     } else {
 917       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), false);
 918     }
 919   } else if (dest->is_double_cpu()) {
 920     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix());
 921     __ mem2reg_opt(dest->as_register_lo(), src_addr_LO, true);
 922   } else if (dest->is_single_fpu()) {
 923     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
 924     __ mem2freg_opt(dest->as_float_reg(), src_addr, false);
 925   } else if (dest->is_double_fpu()) {
 926     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
 927     __ mem2freg_opt(dest->as_double_reg(), src_addr, true);
 928   } else {
 929     ShouldNotReachHere();
 930   }
 931 }
 932 
 933 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
 934   assert(src->is_register(), "should not call otherwise");
 935   assert(dest->is_stack(), "should not call otherwise");
 936 
 937   if (src->is_single_cpu()) {
 938     const Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
 939     if (type == T_OBJECT || type == T_ARRAY) {
 940       __ verify_oop(src->as_register());
 941       __ reg2mem_opt(src->as_register(), dst, true);
 942     } else if (type == T_METADATA) {
 943       __ reg2mem_opt(src->as_register(), dst, true);
 944     } else {
 945       __ reg2mem_opt(src->as_register(), dst, false);
 946     }
 947   } else if (src->is_double_cpu()) {
 948     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix());
 949     __ reg2mem_opt(src->as_register_lo(), dstLO, true);
 950   } else if (src->is_single_fpu()) {
 951     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 952     __ freg2mem_opt(src->as_float_reg(), dst_addr, false);
 953   } else if (src->is_double_fpu()) {
 954     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 955     __ freg2mem_opt(src->as_double_reg(), dst_addr, true);
 956   } else {
 957     ShouldNotReachHere();
 958   }
 959 }
 960 
 961 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
 962   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
 963     if (from_reg->is_double_fpu()) {
 964       // double to double moves
 965       assert(to_reg->is_double_fpu(), "should match");
 966       __ z_ldr(to_reg->as_double_reg(), from_reg->as_double_reg());
 967     } else {
 968       // float to float moves
 969       assert(to_reg->is_single_fpu(), "should match");
 970       __ z_ler(to_reg->as_float_reg(), from_reg->as_float_reg());
 971     }
 972   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
 973     if (from_reg->is_double_cpu()) {
 974       __ z_lgr(to_reg->as_pointer_register(), from_reg->as_pointer_register());
 975     } else if (to_reg->is_double_cpu()) {
 976       // int to int moves
 977       __ z_lgr(to_reg->as_register_lo(), from_reg->as_register());
 978     } else {
 979       // int to int moves
 980       __ z_lgr(to_reg->as_register(), from_reg->as_register());
 981     }
 982   } else {
 983     ShouldNotReachHere();
 984   }
 985   if (to_reg->type() == T_OBJECT || to_reg->type() == T_ARRAY) {
 986     __ verify_oop(to_reg->as_register());
 987   }
 988 }
 989 
 990 void LIR_Assembler::reg2mem(LIR_Opr from, LIR_Opr dest_opr, BasicType type,
 991                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
 992                             bool wide, bool unaligned) {
 993   assert(type != T_METADATA, "store of metadata ptr not supported");
 994   LIR_Address* addr = dest_opr->as_address_ptr();
 995 
 996   Register dest = addr->base()->as_pointer_register();
 997   Register disp_reg = Z_R0;
 998   int disp_value = addr->disp();
 999   bool needs_patching = (patch_code != lir_patch_none);
1000 
1001   if (addr->base()->is_oop_register()) {
1002     __ verify_oop(dest);
1003   }
1004 
1005   PatchingStub* patch = NULL;
1006   if (needs_patching) {
1007     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1008     assert(!from->is_double_cpu() ||
1009            patch_code == lir_patch_none ||
1010            patch_code == lir_patch_normal, "patching doesn't match register");
1011   }
1012 
1013   assert(!needs_patching || (!Immediate::is_simm20(disp_value) && addr->index()->is_illegal()), "assumption");
1014   if (addr->index()->is_illegal()) {
1015     if (!Immediate::is_simm20(disp_value)) {
1016       if (needs_patching) {
1017         __ load_const(Z_R1_scratch, (intptr_t)0);
1018       } else {
1019         __ load_const_optimized(Z_R1_scratch, disp_value);
1020       }
1021       disp_reg = Z_R1_scratch;
1022       disp_value = 0;
1023     }
1024   } else {
1025     if (!Immediate::is_simm20(disp_value)) {
1026       __ load_const_optimized(Z_R1_scratch, disp_value);
1027       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
1028       disp_reg = Z_R1_scratch;
1029       disp_value = 0;
1030     }
1031     disp_reg = addr->index()->as_pointer_register();
1032   }
1033 
1034   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
1035 
1036   if (type == T_ARRAY || type == T_OBJECT) {
1037     __ verify_oop(from->as_register());
1038   }
1039 
1040   bool short_disp = Immediate::is_uimm12(disp_value);
1041 
1042   // Remember the offset of the store. The patching_epilog must be done
1043   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1044   // entered in increasing order.
1045   int offset = code_offset();
1046   switch (type) {
1047     case T_BOOLEAN: // fall through
1048     case T_BYTE  :
1049       if (short_disp) {
1050                     __ z_stc(from->as_register(),  disp_value, disp_reg, dest);
1051       } else {
1052                     __ z_stcy(from->as_register(), disp_value, disp_reg, dest);
1053       }
1054       break;
1055     case T_CHAR  : // fall through
1056     case T_SHORT :
1057       if (short_disp) {
1058                     __ z_sth(from->as_register(),  disp_value, disp_reg, dest);
1059       } else {
1060                     __ z_sthy(from->as_register(), disp_value, disp_reg, dest);
1061       }
1062       break;
1063     case T_INT   :
1064       if (short_disp) {
1065                     __ z_st(from->as_register(),  disp_value, disp_reg, dest);
1066       } else {
1067                     __ z_sty(from->as_register(), disp_value, disp_reg, dest);
1068       }
1069       break;
1070     case T_LONG  :  __ z_stg(from->as_register_lo(), disp_value, disp_reg, dest); break;
1071     case T_ADDRESS: __ z_stg(from->as_register(),    disp_value, disp_reg, dest); break;
1072       break;
1073     case T_ARRAY : // fall through
1074     case T_OBJECT:
1075       {
1076         if (UseCompressedOops && !wide) {
1077           Register compressed_src = Z_R14;
1078           __ z_lgr(compressed_src, from->as_register());
1079           __ encode_heap_oop(compressed_src);
1080           offset = code_offset();
1081           if (short_disp) {
1082             __ z_st(compressed_src,  disp_value, disp_reg, dest);
1083           } else {
1084             __ z_sty(compressed_src, disp_value, disp_reg, dest);
1085           }
1086         } else {
1087           __ z_stg(from->as_register(), disp_value, disp_reg, dest);
1088         }
1089         break;
1090       }
1091     case T_FLOAT :
1092       if (short_disp) {
1093                     __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
1094       } else {
1095                     __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
1096       }
1097       break;
1098     case T_DOUBLE:
1099       if (short_disp) {
1100                     __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
1101       } else {
1102                     __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
1103       }
1104       break;
1105     default: ShouldNotReachHere();
1106   }
1107 
1108   if (patch != NULL) {
1109     patching_epilog(patch, patch_code, dest, info);
1110   }
1111 
1112   if (info != NULL) add_debug_info_for_null_check(offset, info);
1113 }
1114 
1115 
1116 void LIR_Assembler::return_op(LIR_Opr result) {
1117   assert(result->is_illegal() ||
1118          (result->is_single_cpu() && result->as_register() == Z_R2) ||
1119          (result->is_double_cpu() && result->as_register_lo() == Z_R2) ||
1120          (result->is_single_fpu() && result->as_float_reg() == Z_F0) ||
1121          (result->is_double_fpu() && result->as_double_reg() == Z_F0), "convention");
1122 
1123   AddressLiteral pp(os::get_polling_page());
1124   __ load_const_optimized(Z_R1_scratch, pp);
1125 
1126   // Pop the frame before the safepoint code.
1127   int retPC_offset = initial_frame_size_in_bytes() + _z_abi16(return_pc);
1128   if (Displacement::is_validDisp(retPC_offset)) {
1129     __ z_lg(Z_R14, retPC_offset, Z_SP);
1130     __ add2reg(Z_SP, initial_frame_size_in_bytes());
1131   } else {
1132     __ add2reg(Z_SP, initial_frame_size_in_bytes());
1133     __ restore_return_pc();
1134   }
1135 
1136   // We need to mark the code position where the load from the safepoint
1137   // polling page was emitted as relocInfo::poll_return_type here.
1138   __ relocate(relocInfo::poll_return_type);
1139   __ load_from_polling_page(Z_R1_scratch);
1140 
1141   __ z_br(Z_R14); // Return to caller.
1142 }
1143 
1144 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1145   AddressLiteral pp(os::get_polling_page());
1146   __ load_const_optimized(tmp->as_register_lo(), pp);
1147   guarantee(info != NULL, "Shouldn't be NULL");
1148   add_debug_info_for_branch(info);
1149   int offset = __ offset();
1150   __ relocate(relocInfo::poll_type);
1151   __ load_from_polling_page(tmp->as_register_lo());
1152   return offset;
1153 }
1154 
1155 void LIR_Assembler::emit_static_call_stub() {
1156 
1157   // Stub is fixed up when the corresponding call is converted from calling
1158   // compiled code to calling interpreted code.
1159 
1160   address call_pc = __ pc();
1161   address stub = __ start_a_stub(call_stub_size());
1162   if (stub == NULL) {
1163     bailout("static call stub overflow");
1164     return;
1165   }
1166 
1167   int start = __ offset();
1168 
1169   __ relocate(static_stub_Relocation::spec(call_pc));
1170 
1171   // See also Matcher::interpreter_method_oop_reg().
1172   AddressLiteral meta = __ allocate_metadata_address(NULL);
1173   bool success = __ load_const_from_toc(Z_method, meta);
1174 
1175   __ set_inst_mark();
1176   AddressLiteral a((address)-1);
1177   success = success && __ load_const_from_toc(Z_R1, a);
1178   if (!success) {
1179     bailout("const section overflow");
1180     return;
1181   }
1182 
1183   __ z_br(Z_R1);
1184   assert(__ offset() - start <= call_stub_size(), "stub too big");
1185   __ end_a_stub(); // Update current stubs pointer and restore insts_end.
1186 }
1187 
1188 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1189   bool unsigned_comp = condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual;
1190   if (opr1->is_single_cpu()) {
1191     Register reg1 = opr1->as_register();
1192     if (opr2->is_single_cpu()) {
1193       // cpu register - cpu register
1194       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1195         __ z_clgr(reg1, opr2->as_register());
1196       } else {
1197         assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
1198         if (unsigned_comp) {
1199           __ z_clr(reg1, opr2->as_register());
1200         } else {
1201           __ z_cr(reg1, opr2->as_register());
1202         }
1203       }
1204     } else if (opr2->is_stack()) {
1205       // cpu register - stack
1206       if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
1207         __ z_cg(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1208       } else {
1209         if (unsigned_comp) {
1210           __ z_cly(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1211         } else {
1212           __ z_cy(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1213         }
1214       }
1215     } else if (opr2->is_constant()) {
1216       // cpu register - constant
1217       LIR_Const* c = opr2->as_constant_ptr();
1218       if (c->type() == T_INT) {
1219         if (unsigned_comp) {
1220           __ z_clfi(reg1, c->as_jint());
1221         } else {
1222           __ z_cfi(reg1, c->as_jint());
1223         }
1224       } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
1225         // In 64bit oops are single register.
1226         jobject o = c->as_jobject();
1227         if (o == NULL) {
1228           __ z_ltgr(reg1, reg1);
1229         } else {
1230           jobject2reg(o, Z_R1_scratch);
1231           __ z_cgr(reg1, Z_R1_scratch);
1232         }
1233       } else {
1234         fatal("unexpected type: %s", basictype_to_str(c->type()));
1235       }
1236       // cpu register - address
1237     } else if (opr2->is_address()) {
1238       if (op->info() != NULL) {
1239         add_debug_info_for_null_check_here(op->info());
1240       }
1241       if (unsigned_comp) {
1242         __ z_cly(reg1, as_Address(opr2->as_address_ptr()));
1243       } else {
1244         __ z_cy(reg1, as_Address(opr2->as_address_ptr()));
1245       }
1246     } else {
1247       ShouldNotReachHere();
1248     }
1249 
1250   } else if (opr1->is_double_cpu()) {
1251     assert(!unsigned_comp, "unexpected");
1252     Register xlo = opr1->as_register_lo();
1253     Register xhi = opr1->as_register_hi();
1254     if (opr2->is_double_cpu()) {
1255       __ z_cgr(xlo, opr2->as_register_lo());
1256     } else if (opr2->is_constant()) {
1257       // cpu register - constant 0
1258       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
1259       __ z_ltgr(xlo, xlo);
1260     } else {
1261       ShouldNotReachHere();
1262     }
1263 
1264   } else if (opr1->is_single_fpu()) {
1265     if (opr2->is_single_fpu()) {
1266       __ z_cebr(opr1->as_float_reg(), opr2->as_float_reg());
1267     } else {
1268       // stack slot
1269       Address addr = frame_map()->address_for_slot(opr2->single_stack_ix());
1270       if (Immediate::is_uimm12(addr.disp())) {
1271         __ z_ceb(opr1->as_float_reg(), addr);
1272       } else {
1273         __ z_ley(Z_fscratch_1, addr);
1274         __ z_cebr(opr1->as_float_reg(), Z_fscratch_1);
1275       }
1276     }
1277   } else if (opr1->is_double_fpu()) {
1278     if (opr2->is_double_fpu()) {
1279     __ z_cdbr(opr1->as_double_reg(), opr2->as_double_reg());
1280     } else {
1281       // stack slot
1282       Address addr = frame_map()->address_for_slot(opr2->double_stack_ix());
1283       if (Immediate::is_uimm12(addr.disp())) {
1284         __ z_cdb(opr1->as_double_reg(), addr);
1285       } else {
1286         __ z_ldy(Z_fscratch_1, addr);
1287         __ z_cdbr(opr1->as_double_reg(), Z_fscratch_1);
1288       }
1289     }
1290   } else {
1291     ShouldNotReachHere();
1292   }
1293 }
1294 
1295 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1296   Label    done;
1297   Register dreg = dst->as_register();
1298 
1299   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1300     assert((left->is_single_fpu() && right->is_single_fpu()) ||
1301            (left->is_double_fpu() && right->is_double_fpu()), "unexpected operand types");
1302     bool is_single = left->is_single_fpu();
1303     bool is_unordered_less = (code == lir_ucmp_fd2i);
1304     FloatRegister lreg = is_single ? left->as_float_reg() : left->as_double_reg();
1305     FloatRegister rreg = is_single ? right->as_float_reg() : right->as_double_reg();
1306     if (is_single) {
1307       __ z_cebr(lreg, rreg);
1308     } else {
1309       __ z_cdbr(lreg, rreg);
1310     }
1311     if (VM_Version::has_LoadStoreConditional()) {
1312       Register one       = Z_R0_scratch;
1313       Register minus_one = Z_R1_scratch;
1314       __ z_lghi(minus_one, -1);
1315       __ z_lghi(one,  1);
1316       __ z_lghi(dreg, 0);
1317       __ z_locgr(dreg, one,       is_unordered_less ? Assembler::bcondHigh            : Assembler::bcondHighOrNotOrdered);
1318       __ z_locgr(dreg, minus_one, is_unordered_less ? Assembler::bcondLowOrNotOrdered : Assembler::bcondLow);
1319     } else {
1320       __ clear_reg(dreg, true, false);
1321       __ z_bre(done); // if (left == right) dst = 0
1322 
1323       // if (left > right || ((code ~= cmpg) && (left <> right)) dst := 1
1324       __ z_lhi(dreg, 1);
1325       __ z_brc(is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered, done);
1326 
1327       // if (left < right || ((code ~= cmpl) && (left <> right)) dst := -1
1328       __ z_lhi(dreg, -1);
1329     }
1330   } else {
1331     assert(code == lir_cmp_l2i, "check");
1332     if (VM_Version::has_LoadStoreConditional()) {
1333       Register one       = Z_R0_scratch;
1334       Register minus_one = Z_R1_scratch;
1335       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1336       __ z_lghi(minus_one, -1);
1337       __ z_lghi(one,  1);
1338       __ z_lghi(dreg, 0);
1339       __ z_locgr(dreg, one, Assembler::bcondHigh);
1340       __ z_locgr(dreg, minus_one, Assembler::bcondLow);
1341     } else {
1342       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1343       __ z_lghi(dreg,  0);     // eq value
1344       __ z_bre(done);
1345       __ z_lghi(dreg,  1);     // gt value
1346       __ z_brh(done);
1347       __ z_lghi(dreg, -1);     // lt value
1348     }
1349   }
1350   __ bind(done);
1351 }
1352 
1353 // result = condition ? opr1 : opr2
1354 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1355   Assembler::branch_condition acond = Assembler::bcondEqual, ncond = Assembler::bcondNotEqual;
1356   switch (condition) {
1357     case lir_cond_equal:        acond = Assembler::bcondEqual;    ncond = Assembler::bcondNotEqual; break;
1358     case lir_cond_notEqual:     acond = Assembler::bcondNotEqual; ncond = Assembler::bcondEqual;    break;
1359     case lir_cond_less:         acond = Assembler::bcondLow;      ncond = Assembler::bcondNotLow;   break;
1360     case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1361     case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1362     case lir_cond_greater:      acond = Assembler::bcondHigh;     ncond = Assembler::bcondNotHigh;  break;
1363     case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1364     case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1365     default:                    ShouldNotReachHere();
1366   }
1367 
1368   if (opr1->is_cpu_register()) {
1369     reg2reg(opr1, result);
1370   } else if (opr1->is_stack()) {
1371     stack2reg(opr1, result, result->type());
1372   } else if (opr1->is_constant()) {
1373     const2reg(opr1, result, lir_patch_none, NULL);
1374   } else {
1375     ShouldNotReachHere();
1376   }
1377 
1378   if (VM_Version::has_LoadStoreConditional() && !opr2->is_constant()) {
1379     // Optimized version that does not require a branch.
1380     if (opr2->is_single_cpu()) {
1381       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1382       __ z_locgr(result->as_register(), opr2->as_register(), ncond);
1383     } else if (opr2->is_double_cpu()) {
1384       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1385       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1386       __ z_locgr(result->as_register_lo(), opr2->as_register_lo(), ncond);
1387     } else if (opr2->is_single_stack()) {
1388       __ z_loc(result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()), ncond);
1389     } else if (opr2->is_double_stack()) {
1390       __ z_locg(result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix()), ncond);
1391     } else {
1392       ShouldNotReachHere();
1393     }
1394   } else {
1395     Label skip;
1396     __ z_brc(acond, skip);
1397     if (opr2->is_cpu_register()) {
1398       reg2reg(opr2, result);
1399     } else if (opr2->is_stack()) {
1400       stack2reg(opr2, result, result->type());
1401     } else if (opr2->is_constant()) {
1402       const2reg(opr2, result, lir_patch_none, NULL);
1403     } else {
1404       ShouldNotReachHere();
1405     }
1406     __ bind(skip);
1407   }
1408 }
1409 
1410 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1411                              CodeEmitInfo* info, bool pop_fpu_stack) {
1412   assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1413 
1414   if (left->is_single_cpu()) {
1415     assert(left == dest, "left and dest must be equal");
1416     Register lreg = left->as_register();
1417 
1418     if (right->is_single_cpu()) {
1419       // cpu register - cpu register
1420       Register rreg = right->as_register();
1421       switch (code) {
1422         case lir_add: __ z_ar (lreg, rreg); break;
1423         case lir_sub: __ z_sr (lreg, rreg); break;
1424         case lir_mul: __ z_msr(lreg, rreg); break;
1425         default: ShouldNotReachHere();
1426       }
1427 
1428     } else if (right->is_stack()) {
1429       // cpu register - stack
1430       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1431       switch (code) {
1432         case lir_add: __ z_ay(lreg, raddr); break;
1433         case lir_sub: __ z_sy(lreg, raddr); break;
1434         default: ShouldNotReachHere();
1435       }
1436 
1437     } else if (right->is_constant()) {
1438       // cpu register - constant
1439       jint c = right->as_constant_ptr()->as_jint();
1440       switch (code) {
1441         case lir_add: __ z_agfi(lreg, c);  break;
1442         case lir_sub: __ z_agfi(lreg, -c); break; // note: -min_jint == min_jint
1443         case lir_mul: __ z_msfi(lreg, c);  break;
1444         default: ShouldNotReachHere();
1445       }
1446 
1447     } else {
1448       ShouldNotReachHere();
1449     }
1450 
1451   } else if (left->is_double_cpu()) {
1452     assert(left == dest, "left and dest must be equal");
1453     Register lreg_lo = left->as_register_lo();
1454     Register lreg_hi = left->as_register_hi();
1455 
1456     if (right->is_double_cpu()) {
1457       // cpu register - cpu register
1458       Register rreg_lo = right->as_register_lo();
1459       Register rreg_hi = right->as_register_hi();
1460       assert_different_registers(lreg_lo, rreg_lo);
1461       switch (code) {
1462         case lir_add:
1463           __ z_agr(lreg_lo, rreg_lo);
1464           break;
1465         case lir_sub:
1466           __ z_sgr(lreg_lo, rreg_lo);
1467           break;
1468         case lir_mul:
1469           __ z_msgr(lreg_lo, rreg_lo);
1470           break;
1471         default:
1472           ShouldNotReachHere();
1473       }
1474 
1475     } else if (right->is_constant()) {
1476       // cpu register - constant
1477       jlong c = right->as_constant_ptr()->as_jlong_bits();
1478       switch (code) {
1479         case lir_add: __ z_agfi(lreg_lo, c); break;
1480         case lir_sub:
1481           if (c != min_jint) {
1482                       __ z_agfi(lreg_lo, -c);
1483           } else {
1484             // -min_jint cannot be represented as simm32 in z_agfi
1485             // min_jint sign extended:      0xffffffff80000000
1486             // -min_jint as 64 bit integer: 0x0000000080000000
1487             // 0x80000000 can be represented as uimm32 in z_algfi
1488             // lreg_lo := lreg_lo + -min_jint == lreg_lo + 0x80000000
1489                       __ z_algfi(lreg_lo, UCONST64(0x80000000));
1490           }
1491           break;
1492         case lir_mul: __ z_msgfi(lreg_lo, c); break;
1493         default:
1494           ShouldNotReachHere();
1495       }
1496 
1497     } else {
1498       ShouldNotReachHere();
1499     }
1500 
1501   } else if (left->is_single_fpu()) {
1502     assert(left == dest, "left and dest must be equal");
1503     FloatRegister lreg = left->as_float_reg();
1504     FloatRegister rreg = right->is_single_fpu() ? right->as_float_reg() : fnoreg;
1505     Address raddr;
1506 
1507     if (rreg == fnoreg) {
1508       assert(right->is_single_stack(), "constants should be loaded into register");
1509       raddr = frame_map()->address_for_slot(right->single_stack_ix());
1510       if (!Immediate::is_uimm12(raddr.disp())) {
1511         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, false);
1512       }
1513     }
1514 
1515     if (rreg != fnoreg) {
1516       switch (code) {
1517         case lir_add: __ z_aebr(lreg, rreg);  break;
1518         case lir_sub: __ z_sebr(lreg, rreg);  break;
1519         case lir_mul_strictfp: // fall through
1520         case lir_mul: __ z_meebr(lreg, rreg); break;
1521         case lir_div_strictfp: // fall through
1522         case lir_div: __ z_debr(lreg, rreg);  break;
1523         default: ShouldNotReachHere();
1524       }
1525     } else {
1526       switch (code) {
1527         case lir_add: __ z_aeb(lreg, raddr);  break;
1528         case lir_sub: __ z_seb(lreg, raddr);  break;
1529         case lir_mul_strictfp: // fall through
1530         case lir_mul: __ z_meeb(lreg, raddr);  break;
1531         case lir_div_strictfp: // fall through
1532         case lir_div: __ z_deb(lreg, raddr);  break;
1533         default: ShouldNotReachHere();
1534       }
1535     }
1536   } else if (left->is_double_fpu()) {
1537     assert(left == dest, "left and dest must be equal");
1538     FloatRegister lreg = left->as_double_reg();
1539     FloatRegister rreg = right->is_double_fpu() ? right->as_double_reg() : fnoreg;
1540     Address raddr;
1541 
1542     if (rreg == fnoreg) {
1543       assert(right->is_double_stack(), "constants should be loaded into register");
1544       raddr = frame_map()->address_for_slot(right->double_stack_ix());
1545       if (!Immediate::is_uimm12(raddr.disp())) {
1546         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, true);
1547       }
1548     }
1549 
1550     if (rreg != fnoreg) {
1551       switch (code) {
1552         case lir_add: __ z_adbr(lreg, rreg); break;
1553         case lir_sub: __ z_sdbr(lreg, rreg); break;
1554         case lir_mul_strictfp: // fall through
1555         case lir_mul: __ z_mdbr(lreg, rreg); break;
1556         case lir_div_strictfp: // fall through
1557         case lir_div: __ z_ddbr(lreg, rreg); break;
1558         default: ShouldNotReachHere();
1559       }
1560     } else {
1561       switch (code) {
1562         case lir_add: __ z_adb(lreg, raddr); break;
1563         case lir_sub: __ z_sdb(lreg, raddr); break;
1564         case lir_mul_strictfp: // fall through
1565         case lir_mul: __ z_mdb(lreg, raddr); break;
1566         case lir_div_strictfp: // fall through
1567         case lir_div: __ z_ddb(lreg, raddr); break;
1568         default: ShouldNotReachHere();
1569       }
1570     }
1571   } else if (left->is_address()) {
1572     assert(left == dest, "left and dest must be equal");
1573     assert(code == lir_add, "unsupported operation");
1574     assert(right->is_constant(), "unsupported operand");
1575     jint c = right->as_constant_ptr()->as_jint();
1576     LIR_Address* lir_addr = left->as_address_ptr();
1577     Address addr = as_Address(lir_addr);
1578     switch (lir_addr->type()) {
1579       case T_INT:
1580         __ add2mem_32(addr, c, Z_R1_scratch);
1581         break;
1582       case T_LONG:
1583         __ add2mem_64(addr, c, Z_R1_scratch);
1584         break;
1585       default:
1586         ShouldNotReachHere();
1587     }
1588   } else {
1589     ShouldNotReachHere();
1590   }
1591 }
1592 
1593 void LIR_Assembler::fpop() {
1594   // do nothing
1595 }
1596 
1597 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1598   switch (code) {
1599     case lir_sqrt: {
1600       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1601       FloatRegister src_reg = value->as_double_reg();
1602       FloatRegister dst_reg = dest->as_double_reg();
1603       __ z_sqdbr(dst_reg, src_reg);
1604       break;
1605     }
1606     case lir_abs: {
1607       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1608       FloatRegister src_reg = value->as_double_reg();
1609       FloatRegister dst_reg = dest->as_double_reg();
1610       __ z_lpdbr(dst_reg, src_reg);
1611       break;
1612     }
1613     default: {
1614       ShouldNotReachHere();
1615       break;
1616     }
1617   }
1618 }
1619 
1620 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1621   if (left->is_single_cpu()) {
1622     Register reg = left->as_register();
1623     if (right->is_constant()) {
1624       int val = right->as_constant_ptr()->as_jint();
1625       switch (code) {
1626         case lir_logic_and: __ z_nilf(reg, val); break;
1627         case lir_logic_or:  __ z_oilf(reg, val); break;
1628         case lir_logic_xor: __ z_xilf(reg, val); break;
1629         default: ShouldNotReachHere();
1630       }
1631     } else if (right->is_stack()) {
1632       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1633       switch (code) {
1634         case lir_logic_and: __ z_ny(reg, raddr); break;
1635         case lir_logic_or:  __ z_oy(reg, raddr); break;
1636         case lir_logic_xor: __ z_xy(reg, raddr); break;
1637         default: ShouldNotReachHere();
1638       }
1639     } else {
1640       Register rright = right->as_register();
1641       switch (code) {
1642         case lir_logic_and: __ z_nr(reg, rright); break;
1643         case lir_logic_or : __ z_or(reg, rright); break;
1644         case lir_logic_xor: __ z_xr(reg, rright); break;
1645         default: ShouldNotReachHere();
1646       }
1647     }
1648     move_regs(reg, dst->as_register());
1649   } else {
1650     Register l_lo = left->as_register_lo();
1651     if (right->is_constant()) {
1652       __ load_const_optimized(Z_R1_scratch, right->as_constant_ptr()->as_jlong());
1653       switch (code) {
1654         case lir_logic_and:
1655           __ z_ngr(l_lo, Z_R1_scratch);
1656           break;
1657         case lir_logic_or:
1658           __ z_ogr(l_lo, Z_R1_scratch);
1659           break;
1660         case lir_logic_xor:
1661           __ z_xgr(l_lo, Z_R1_scratch);
1662           break;
1663         default: ShouldNotReachHere();
1664       }
1665     } else {
1666       Register r_lo;
1667       if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
1668         r_lo = right->as_register();
1669       } else {
1670         r_lo = right->as_register_lo();
1671       }
1672       switch (code) {
1673         case lir_logic_and:
1674           __ z_ngr(l_lo, r_lo);
1675           break;
1676         case lir_logic_or:
1677           __ z_ogr(l_lo, r_lo);
1678           break;
1679         case lir_logic_xor:
1680           __ z_xgr(l_lo, r_lo);
1681           break;
1682         default: ShouldNotReachHere();
1683       }
1684     }
1685 
1686     Register dst_lo = dst->as_register_lo();
1687 
1688     move_regs(l_lo, dst_lo);
1689   }
1690 }
1691 
1692 // See operand selection in LIRGenerator::do_ArithmeticOp_Int().
1693 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
1694   if (left->is_double_cpu()) {
1695     // 64 bit integer case
1696     assert(left->is_double_cpu(), "left must be register");
1697     assert(right->is_double_cpu() || is_power_of_2_long(right->as_jlong()),
1698            "right must be register or power of 2 constant");
1699     assert(result->is_double_cpu(), "result must be register");
1700 
1701     Register lreg = left->as_register_lo();
1702     Register dreg = result->as_register_lo();
1703 
1704     if (right->is_constant()) {
1705       // Convert division by a power of two into some shifts and logical operations.
1706       Register treg1 = Z_R0_scratch;
1707       Register treg2 = Z_R1_scratch;
1708       jlong divisor = right->as_jlong();
1709       jlong log_divisor = log2_long(right->as_jlong());
1710 
1711       if (divisor == min_jlong) {
1712         // Min_jlong is special. Result is '0' except for min_jlong/min_jlong = 1.
1713         if (dreg == lreg) {
1714           NearLabel done;
1715           __ load_const_optimized(treg2, min_jlong);
1716           __ z_cgr(lreg, treg2);
1717           __ z_lghi(dreg, 0);           // Preserves condition code.
1718           __ z_brne(done);
1719           __ z_lghi(dreg, 1);           // min_jlong / min_jlong = 1
1720           __ bind(done);
1721         } else {
1722           assert_different_registers(dreg, lreg);
1723           NearLabel done;
1724           __ z_lghi(dreg, 0);
1725           __ compare64_and_branch(lreg, min_jlong, Assembler::bcondNotEqual, done);
1726           __ z_lghi(dreg, 1);
1727           __ bind(done);
1728         }
1729         return;
1730       }
1731       __ move_reg_if_needed(dreg, T_LONG, lreg, T_LONG);
1732       if (divisor == 2) {
1733         __ z_srlg(treg2, dreg, 63);     // dividend < 0 ? 1 : 0
1734       } else {
1735         __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1736         __ and_imm(treg2, divisor - 1, treg1, true);
1737       }
1738       if (code == lir_idiv) {
1739         __ z_agr(dreg, treg2);
1740         __ z_srag(dreg, dreg, log_divisor);
1741       } else {
1742         assert(code == lir_irem, "check");
1743         __ z_agr(treg2, dreg);
1744         __ and_imm(treg2, ~(divisor - 1), treg1, true);
1745         __ z_sgr(dreg, treg2);
1746       }
1747       return;
1748     }
1749 
1750     // Divisor is not a power of 2 constant.
1751     Register rreg = right->as_register_lo();
1752     Register treg = temp->as_register_lo();
1753     assert(right->is_double_cpu(), "right must be register");
1754     assert(lreg == Z_R11, "see ldivInOpr()");
1755     assert(rreg != lreg, "right register must not be same as left register");
1756     assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10) ||
1757            (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see ldivInOpr(), ldivOutOpr(), lremOutOpr()");
1758 
1759     Register R1 = lreg->predecessor();
1760     Register R2 = rreg;
1761     assert(code != lir_idiv || lreg==dreg, "see code below");
1762     if (code == lir_idiv) {
1763       __ z_lcgr(lreg, lreg);
1764     } else {
1765       __ clear_reg(dreg, true, false);
1766     }
1767     NearLabel done;
1768     __ compare64_and_branch(R2, -1, Assembler::bcondEqual, done);
1769     if (code == lir_idiv) {
1770       __ z_lcgr(lreg, lreg); // Revert lcgr above.
1771     }
1772     if (ImplicitDiv0Checks) {
1773       // No debug info because the idiv won't trap.
1774       // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1775       // which is unnecessary, too.
1776       add_debug_info_for_div0(__ offset(), info);
1777     }
1778     __ z_dsgr(R1, R2);
1779     __ bind(done);
1780     return;
1781   }
1782 
1783   // 32 bit integer case
1784 
1785   assert(left->is_single_cpu(), "left must be register");
1786   assert(right->is_single_cpu() || is_power_of_2(right->as_jint()), "right must be register or power of 2 constant");
1787   assert(result->is_single_cpu(), "result must be register");
1788 
1789   Register lreg = left->as_register();
1790   Register dreg = result->as_register();
1791 
1792   if (right->is_constant()) {
1793     // Convert division by a power of two into some shifts and logical operations.
1794     Register treg1 = Z_R0_scratch;
1795     Register treg2 = Z_R1_scratch;
1796     jlong divisor = right->as_jint();
1797     jlong log_divisor = log2_long(right->as_jint());
1798     __ move_reg_if_needed(dreg, T_LONG, lreg, T_INT); // sign extend
1799     if (divisor == 2) {
1800       __ z_srlg(treg2, dreg, 63);     // dividend < 0 ?  1 : 0
1801     } else {
1802       __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1803       __ and_imm(treg2, divisor - 1, treg1, true);
1804     }
1805     if (code == lir_idiv) {
1806       __ z_agr(dreg, treg2);
1807       __ z_srag(dreg, dreg, log_divisor);
1808     } else {
1809       assert(code == lir_irem, "check");
1810       __ z_agr(treg2, dreg);
1811       __ and_imm(treg2, ~(divisor - 1), treg1, true);
1812       __ z_sgr(dreg, treg2);
1813     }
1814     return;
1815   }
1816 
1817   // Divisor is not a power of 2 constant.
1818   Register rreg = right->as_register();
1819   Register treg = temp->as_register();
1820   assert(right->is_single_cpu(), "right must be register");
1821   assert(lreg == Z_R11, "left register must be rax,");
1822   assert(rreg != lreg, "right register must not be same as left register");
1823   assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10)
1824       || (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see divInOpr(), divOutOpr(), remOutOpr()");
1825 
1826   Register R1 = lreg->predecessor();
1827   Register R2 = rreg;
1828   __ move_reg_if_needed(lreg, T_LONG, lreg, T_INT); // sign extend
1829   if (ImplicitDiv0Checks) {
1830     // No debug info because the idiv won't trap.
1831     // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1832     // which is unnecessary, too.
1833     add_debug_info_for_div0(__ offset(), info);
1834   }
1835   __ z_dsgfr(R1, R2);
1836 }
1837 
1838 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1839   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1840   assert(exceptionPC->as_register() == Z_EXC_PC, "should match");
1841 
1842   // Exception object is not added to oop map by LinearScan
1843   // (LinearScan assumes that no oops are in fixed registers).
1844   info->add_register_oop(exceptionOop);
1845 
1846   // Reuse the debug info from the safepoint poll for the throw op itself.
1847   __ get_PC(Z_EXC_PC);
1848   add_call_info(__ offset(), info); // for exception handler
1849   address stub = Runtime1::entry_for (compilation()->has_fpu_code() ? Runtime1::handle_exception_id
1850                                                                     : Runtime1::handle_exception_nofpu_id);
1851   emit_call_c(stub);
1852 }
1853 
1854 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1855   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1856 
1857   __ branch_optimized(Assembler::bcondAlways, _unwind_handler_entry);
1858 }
1859 
1860 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1861   ciArrayKlass* default_type = op->expected_type();
1862   Register src = op->src()->as_register();
1863   Register dst = op->dst()->as_register();
1864   Register src_pos = op->src_pos()->as_register();
1865   Register dst_pos = op->dst_pos()->as_register();
1866   Register length  = op->length()->as_register();
1867   Register tmp = op->tmp()->as_register();
1868 
1869   CodeStub* stub = op->stub();
1870   int flags = op->flags();
1871   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
1872   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1873 
1874   // If we don't know anything, just go through the generic arraycopy.
1875   if (default_type == NULL) {
1876     Label done;
1877     // Save outgoing arguments in callee saved registers (C convention) in case
1878     // a call to System.arraycopy is needed.
1879     Register callee_saved_src     = Z_R10;
1880     Register callee_saved_src_pos = Z_R11;
1881     Register callee_saved_dst     = Z_R12;
1882     Register callee_saved_dst_pos = Z_R13;
1883     Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
1884 
1885     __ lgr_if_needed(callee_saved_src, src);
1886     __ lgr_if_needed(callee_saved_src_pos, src_pos);
1887     __ lgr_if_needed(callee_saved_dst, dst);
1888     __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
1889     __ lgr_if_needed(callee_saved_length, length);
1890 
1891     // C function requires 64 bit values.
1892     __ z_lgfr(src_pos, src_pos);
1893     __ z_lgfr(dst_pos, dst_pos);
1894     __ z_lgfr(length, length);
1895 
1896     address C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
1897 
1898     address copyfunc_addr = StubRoutines::generic_arraycopy();
1899 
1900     // Pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint.
1901 
1902     // The arguments are in the corresponding registers.
1903     assert(Z_ARG1 == src,     "assumption");
1904     assert(Z_ARG2 == src_pos, "assumption");
1905     assert(Z_ARG3 == dst,     "assumption");
1906     assert(Z_ARG4 == dst_pos, "assumption");
1907     assert(Z_ARG5 == length,  "assumption");
1908     if (copyfunc_addr == NULL) { // Use C version if stub was not generated.
1909       emit_call_c(C_entry);
1910     } else {
1911 #ifndef PRODUCT
1912       if (PrintC1Statistics) {
1913         __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_generic_arraycopystub_cnt);
1914         __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
1915       }
1916 #endif
1917       emit_call_c(copyfunc_addr);
1918     }
1919     CHECK_BAILOUT();
1920 
1921     __ compare32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
1922 
1923     if (copyfunc_addr != NULL) {
1924       __ z_lgr(tmp, Z_RET);
1925       __ z_xilf(tmp, -1);
1926     }
1927 
1928     // Restore values from callee saved registers so they are where the stub
1929     // expects them.
1930     __ lgr_if_needed(src, callee_saved_src);
1931     __ lgr_if_needed(src_pos, callee_saved_src_pos);
1932     __ lgr_if_needed(dst, callee_saved_dst);
1933     __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
1934     __ lgr_if_needed(length, callee_saved_length);
1935 
1936     if (copyfunc_addr != NULL) {
1937       __ z_sr(length, tmp);
1938       __ z_ar(src_pos, tmp);
1939       __ z_ar(dst_pos, tmp);
1940     }
1941     __ branch_optimized(Assembler::bcondAlways, *stub->entry());
1942 
1943     __ bind(*stub->continuation());
1944     return;
1945   }
1946 
1947   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
1948 
1949   int elem_size = type2aelembytes(basic_type);
1950   int shift_amount;
1951 
1952   switch (elem_size) {
1953     case 1 :
1954       shift_amount = 0;
1955       break;
1956     case 2 :
1957       shift_amount = 1;
1958       break;
1959     case 4 :
1960       shift_amount = 2;
1961       break;
1962     case 8 :
1963       shift_amount = 3;
1964       break;
1965     default:
1966       shift_amount = -1;
1967       ShouldNotReachHere();
1968   }
1969 
1970   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
1971   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
1972   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
1973   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
1974 
1975   // Length and pos's are all sign extended at this point on 64bit.
1976 
1977   // test for NULL
1978   if (flags & LIR_OpArrayCopy::src_null_check) {
1979     __ compareU64_and_branch(src, (intptr_t)0, Assembler::bcondZero, *stub->entry());
1980   }
1981   if (flags & LIR_OpArrayCopy::dst_null_check) {
1982     __ compareU64_and_branch(dst, (intptr_t)0, Assembler::bcondZero, *stub->entry());
1983   }
1984 
1985   // Check if negative.
1986   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
1987     __ compare32_and_branch(src_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
1988   }
1989   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
1990     __ compare32_and_branch(dst_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
1991   }
1992 
1993   // If the compiler was not able to prove that exact type of the source or the destination
1994   // of the arraycopy is an array type, check at runtime if the source or the destination is
1995   // an instance type.
1996   if (flags & LIR_OpArrayCopy::type_check) {
1997     assert(Klass::_lh_neutral_value == 0, "or replace z_lt instructions");
1998 
1999     if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2000       __ load_klass(tmp, dst);
2001       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2002       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2003     }
2004 
2005     if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2006       __ load_klass(tmp, src);
2007       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2008       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2009     }
2010   }
2011 
2012   if (flags & LIR_OpArrayCopy::src_range_check) {
2013     __ z_la(tmp, Address(src_pos, length));
2014     __ z_cl(tmp, src_length_addr);
2015     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2016   }
2017   if (flags & LIR_OpArrayCopy::dst_range_check) {
2018     __ z_la(tmp, Address(dst_pos, length));
2019     __ z_cl(tmp, dst_length_addr);
2020     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2021   }
2022 
2023   if (flags & LIR_OpArrayCopy::length_positive_check) {
2024     __ z_ltr(length, length);
2025     __ branch_optimized(Assembler::bcondNegative, *stub->entry());
2026   }
2027 
2028   // Stubs require 64 bit values.
2029   __ z_lgfr(src_pos, src_pos); // int -> long
2030   __ z_lgfr(dst_pos, dst_pos); // int -> long
2031   __ z_lgfr(length, length);   // int -> long
2032 
2033   if (flags & LIR_OpArrayCopy::type_check) {
2034     // We don't know the array types are compatible.
2035     if (basic_type != T_OBJECT) {
2036       // Simple test for basic type arrays.
2037       if (UseCompressedClassPointers) {
2038         __ z_l(tmp, src_klass_addr);
2039         __ z_c(tmp, dst_klass_addr);
2040       } else {
2041         __ z_lg(tmp, src_klass_addr);
2042         __ z_cg(tmp, dst_klass_addr);
2043       }
2044       __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2045     } else {
2046       // For object arrays, if src is a sub class of dst then we can
2047       // safely do the copy.
2048       NearLabel cont, slow;
2049       Register src_klass = Z_R1_scratch;
2050       Register dst_klass = Z_R10;
2051 
2052       __ load_klass(src_klass, src);
2053       __ load_klass(dst_klass, dst);
2054 
2055       __ check_klass_subtype_fast_path(src_klass, dst_klass, tmp, &cont, &slow, NULL);
2056 
2057       store_parameter(src_klass, 0); // sub
2058       store_parameter(dst_klass, 1); // super
2059       emit_call_c(Runtime1::entry_for (Runtime1::slow_subtype_check_id));
2060       CHECK_BAILOUT();
2061       // Sets condition code 0 for match (2 otherwise).
2062       __ branch_optimized(Assembler::bcondEqual, cont);
2063 
2064       __ bind(slow);
2065 
2066       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2067       if (copyfunc_addr != NULL) { // use stub if available
2068         // Src is not a sub class of dst so we have to do a
2069         // per-element check.
2070 
2071         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2072         if ((flags & mask) != mask) {
2073           // Check that at least both of them object arrays.
2074           assert(flags & mask, "one of the two should be known to be an object array");
2075 
2076           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2077             __ load_klass(tmp, src);
2078           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2079             __ load_klass(tmp, dst);
2080           }
2081           Address klass_lh_addr(tmp, Klass::layout_helper_offset());
2082           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2083           __ load_const_optimized(Z_R1_scratch, objArray_lh);
2084           __ z_c(Z_R1_scratch, klass_lh_addr);
2085           __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2086         }
2087 
2088         // Save outgoing arguments in callee saved registers (C convention) in case
2089         // a call to System.arraycopy is needed.
2090         Register callee_saved_src     = Z_R10;
2091         Register callee_saved_src_pos = Z_R11;
2092         Register callee_saved_dst     = Z_R12;
2093         Register callee_saved_dst_pos = Z_R13;
2094         Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
2095 
2096         __ lgr_if_needed(callee_saved_src, src);
2097         __ lgr_if_needed(callee_saved_src_pos, src_pos);
2098         __ lgr_if_needed(callee_saved_dst, dst);
2099         __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
2100         __ lgr_if_needed(callee_saved_length, length);
2101 
2102         __ z_llgfr(length, length); // Higher 32bits must be null.
2103 
2104         __ z_sllg(Z_ARG1, src_pos, shift_amount); // index -> byte offset
2105         __ z_sllg(Z_ARG2, dst_pos, shift_amount); // index -> byte offset
2106 
2107         __ z_la(Z_ARG1, Address(src, Z_ARG1, arrayOopDesc::base_offset_in_bytes(basic_type)));
2108         assert_different_registers(Z_ARG1, dst, dst_pos, length);
2109         __ z_la(Z_ARG2, Address(dst, Z_ARG2, arrayOopDesc::base_offset_in_bytes(basic_type)));
2110         assert_different_registers(Z_ARG2, dst, length);
2111 
2112         __ z_lgr(Z_ARG3, length);
2113         assert_different_registers(Z_ARG3, dst);
2114 
2115         __ load_klass(Z_ARG5, dst);
2116         __ z_lg(Z_ARG5, Address(Z_ARG5, ObjArrayKlass::element_klass_offset()));
2117         __ z_lg(Z_ARG4, Address(Z_ARG5, Klass::super_check_offset_offset()));
2118         emit_call_c(copyfunc_addr);
2119         CHECK_BAILOUT();
2120 
2121 #ifndef PRODUCT
2122         if (PrintC1Statistics) {
2123           NearLabel failed;
2124           __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, failed);
2125           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_cnt);
2126           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2127           __ bind(failed);
2128         }
2129 #endif
2130 
2131         __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2132 
2133 #ifndef PRODUCT
2134         if (PrintC1Statistics) {
2135           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_attempt_cnt);
2136           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2137         }
2138 #endif
2139 
2140         __ z_lgr(tmp, Z_RET);
2141         __ z_xilf(tmp, -1);
2142 
2143         // Restore previously spilled arguments
2144         __ lgr_if_needed(src, callee_saved_src);
2145         __ lgr_if_needed(src_pos, callee_saved_src_pos);
2146         __ lgr_if_needed(dst, callee_saved_dst);
2147         __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2148         __ lgr_if_needed(length, callee_saved_length);
2149 
2150         __ z_sr(length, tmp);
2151         __ z_ar(src_pos, tmp);
2152         __ z_ar(dst_pos, tmp);
2153       }
2154 
2155       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2156 
2157       __ bind(cont);
2158     }
2159   }
2160 
2161 #ifdef ASSERT
2162   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2163     // Sanity check the known type with the incoming class. For the
2164     // primitive case the types must match exactly with src.klass and
2165     // dst.klass each exactly matching the default type. For the
2166     // object array case, if no type check is needed then either the
2167     // dst type is exactly the expected type and the src type is a
2168     // subtype which we can't check or src is the same array as dst
2169     // but not necessarily exactly of type default_type.
2170     NearLabel known_ok, halt;
2171     metadata2reg(default_type->constant_encoding(), tmp);
2172     if (UseCompressedClassPointers) {
2173       __ encode_klass_not_null(tmp);
2174     }
2175 
2176     if (basic_type != T_OBJECT) {
2177       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2178       else                                    { __ z_cg(tmp, dst_klass_addr); }
2179       __ branch_optimized(Assembler::bcondNotEqual, halt);
2180       if (UseCompressedClassPointers)         { __ z_c (tmp, src_klass_addr); }
2181       else                                    { __ z_cg(tmp, src_klass_addr); }
2182       __ branch_optimized(Assembler::bcondEqual, known_ok);
2183     } else {
2184       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2185       else                                    { __ z_cg(tmp, dst_klass_addr); }
2186       __ branch_optimized(Assembler::bcondEqual, known_ok);
2187       __ compareU64_and_branch(src, dst, Assembler::bcondEqual, known_ok);
2188     }
2189     __ bind(halt);
2190     __ stop("incorrect type information in arraycopy");
2191     __ bind(known_ok);
2192   }
2193 #endif
2194 
2195 #ifndef PRODUCT
2196   if (PrintC1Statistics) {
2197     __ load_const_optimized(Z_R1_scratch, Runtime1::arraycopy_count_address(basic_type));
2198     __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2199   }
2200 #endif
2201 
2202   __ z_sllg(tmp, src_pos, shift_amount); // index -> byte offset
2203   __ z_sllg(Z_R1_scratch, dst_pos, shift_amount); // index -> byte offset
2204 
2205   assert_different_registers(Z_ARG1, dst, dst_pos, length);
2206   __ z_la(Z_ARG1, Address(src, tmp, arrayOopDesc::base_offset_in_bytes(basic_type)));
2207   assert_different_registers(Z_ARG2, length);
2208   __ z_la(Z_ARG2, Address(dst, Z_R1_scratch, arrayOopDesc::base_offset_in_bytes(basic_type)));
2209   __ lgr_if_needed(Z_ARG3, length);
2210 
2211   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2212   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2213   const char *name;
2214   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2215   __ call_VM_leaf(entry);
2216 
2217   __ bind(*stub->continuation());
2218 }
2219 
2220 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2221   if (dest->is_single_cpu()) {
2222     if (left->type() == T_OBJECT) {
2223       switch (code) {
2224         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2225         case lir_shr:  __ z_srag (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2226         case lir_ushr: __ z_srlg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2227         default: ShouldNotReachHere();
2228       }
2229     } else {
2230       assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2231       Register masked_count = Z_R1_scratch;
2232       __ z_lr(masked_count, count->as_register());
2233       __ z_nill(masked_count, 31);
2234       switch (code) {
2235         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, masked_count); break;
2236         case lir_shr:  __ z_sra  (dest->as_register(), 0, masked_count); break;
2237         case lir_ushr: __ z_srl  (dest->as_register(), 0, masked_count); break;
2238         default: ShouldNotReachHere();
2239       }
2240     }
2241   } else {
2242     switch (code) {
2243       case lir_shl:  __ z_sllg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2244       case lir_shr:  __ z_srag (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2245       case lir_ushr: __ z_srlg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2246       default: ShouldNotReachHere();
2247     }
2248   }
2249 }
2250 
2251 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2252   if (left->type() == T_OBJECT) {
2253     count = count & 63;  // Shouldn't shift by more than sizeof(intptr_t).
2254     Register l = left->as_register();
2255     Register d = dest->as_register_lo();
2256     switch (code) {
2257       case lir_shl:  __ z_sllg (d, l, count); break;
2258       case lir_shr:  __ z_srag (d, l, count); break;
2259       case lir_ushr: __ z_srlg (d, l, count); break;
2260       default: ShouldNotReachHere();
2261     }
2262     return;
2263   }
2264   if (dest->is_single_cpu()) {
2265     assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2266     count = count & 0x1F; // Java spec
2267     switch (code) {
2268       case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), count); break;
2269       case lir_shr:  __ z_sra  (dest->as_register(), count); break;
2270       case lir_ushr: __ z_srl  (dest->as_register(), count); break;
2271       default: ShouldNotReachHere();
2272     }
2273   } else if (dest->is_double_cpu()) {
2274     count = count & 63; // Java spec
2275     Register l = left->as_pointer_register();
2276     Register d = dest->as_pointer_register();
2277     switch (code) {
2278       case lir_shl:  __ z_sllg (d, l, count); break;
2279       case lir_shr:  __ z_srag (d, l, count); break;
2280       case lir_ushr: __ z_srlg (d, l, count); break;
2281       default: ShouldNotReachHere();
2282     }
2283   } else {
2284     ShouldNotReachHere();
2285   }
2286 }
2287 
2288 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2289   if (op->init_check()) {
2290     // Make sure klass is initialized & doesn't have finalizer.
2291     const int state_offset = in_bytes(InstanceKlass::init_state_offset());
2292     Register iklass = op->klass()->as_register();
2293     add_debug_info_for_null_check_here(op->stub()->info());
2294     if (Immediate::is_uimm12(state_offset)) {
2295       __ z_cli(state_offset, iklass, InstanceKlass::fully_initialized);
2296     } else {
2297       __ z_cliy(state_offset, iklass, InstanceKlass::fully_initialized);
2298     }
2299     __ branch_optimized(Assembler::bcondNotEqual, *op->stub()->entry()); // Use long branch, because slow_case might be far.
2300   }
2301   __ allocate_object(op->obj()->as_register(),
2302                      op->tmp1()->as_register(),
2303                      op->tmp2()->as_register(),
2304                      op->header_size(),
2305                      op->object_size(),
2306                      op->klass()->as_register(),
2307                      *op->stub()->entry());
2308   __ bind(*op->stub()->continuation());
2309   __ verify_oop(op->obj()->as_register());
2310 }
2311 
2312 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2313   Register len = op->len()->as_register();
2314   __ move_reg_if_needed(len, T_LONG, len, T_INT); // sign extend
2315 
2316   if (UseSlowPath ||
2317       (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
2318       (!UseFastNewTypeArray   && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
2319     __ z_brul(*op->stub()->entry());
2320   } else {
2321     __ allocate_array(op->obj()->as_register(),
2322                       op->len()->as_register(),
2323                       op->tmp1()->as_register(),
2324                       op->tmp2()->as_register(),
2325                       arrayOopDesc::header_size(op->type()),
2326                       type2aelembytes(op->type()),
2327                       op->klass()->as_register(),
2328                       *op->stub()->entry());
2329   }
2330   __ bind(*op->stub()->continuation());
2331 }
2332 
2333 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
2334                                         Register recv, Register tmp1, Label* update_done) {
2335   uint i;
2336   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2337     Label next_test;
2338     // See if the receiver is receiver[n].
2339     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2340     __ z_cg(recv, receiver_addr);
2341     __ z_brne(next_test);
2342     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
2343     __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2344     __ branch_optimized(Assembler::bcondAlways, *update_done);
2345     __ bind(next_test);
2346   }
2347 
2348   // Didn't find receiver; find next empty slot and fill it in.
2349   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2350     Label next_test;
2351     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2352     __ z_ltg(Z_R0_scratch, recv_addr);
2353     __ z_brne(next_test);
2354     __ z_stg(recv, recv_addr);
2355     __ load_const_optimized(tmp1, DataLayout::counter_increment);
2356     __ z_stg(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)), mdo);
2357     __ branch_optimized(Assembler::bcondAlways, *update_done);
2358     __ bind(next_test);
2359   }
2360 }
2361 
2362 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2363                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2364   Unimplemented();
2365 }
2366 
2367 void LIR_Assembler::store_parameter(Register r, int param_num) {
2368   assert(param_num >= 0, "invalid num");
2369   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2370   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2371   __ z_stg(r, offset_in_bytes, Z_SP);
2372 }
2373 
2374 void LIR_Assembler::store_parameter(jint c, int param_num) {
2375   assert(param_num >= 0, "invalid num");
2376   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2377   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2378   __ store_const(Address(Z_SP, offset_in_bytes), c, Z_R1_scratch, true);
2379 }
2380 
2381 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2382   // We always need a stub for the failure case.
2383   CodeStub* stub = op->stub();
2384   Register obj = op->object()->as_register();
2385   Register k_RInfo = op->tmp1()->as_register();
2386   Register klass_RInfo = op->tmp2()->as_register();
2387   Register dst = op->result_opr()->as_register();
2388   Register Rtmp1 = Z_R1_scratch;
2389   ciKlass* k = op->klass();
2390 
2391   assert(!op->tmp3()->is_valid(), "tmp3's not needed");
2392 
2393   // Check if it needs to be profiled.
2394   ciMethodData* md = NULL;
2395   ciProfileData* data = NULL;
2396 
2397   if (op->should_profile()) {
2398     ciMethod* method = op->profiled_method();
2399     assert(method != NULL, "Should have method");
2400     int bci = op->profiled_bci();
2401     md = method->method_data_or_null();
2402     assert(md != NULL, "Sanity");
2403     data = md->bci_to_data(bci);
2404     assert(data != NULL,                "need data for type check");
2405     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2406   }
2407 
2408   // Temp operands do not overlap with inputs, if this is their last
2409   // use (end of range is exclusive), so a register conflict is possible.
2410   if (obj == k_RInfo) {
2411     k_RInfo = dst;
2412   } else if (obj == klass_RInfo) {
2413     klass_RInfo = dst;
2414   }
2415   assert_different_registers(obj, k_RInfo, klass_RInfo);
2416 
2417   if (op->should_profile()) {
2418     NearLabel not_null;
2419     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2420     // Object is null; update MDO and exit.
2421     Register mdo = klass_RInfo;
2422     metadata2reg(md->constant_encoding(), mdo);
2423     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2424     int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2425     __ or2mem_8(data_addr, header_bits);
2426     __ branch_optimized(Assembler::bcondAlways, *obj_is_null);
2427     __ bind(not_null);
2428   } else {
2429     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondEqual, *obj_is_null);
2430   }
2431 
2432   NearLabel profile_cast_failure, profile_cast_success;
2433   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2434   Label *success_target = op->should_profile() ? &profile_cast_success : success;
2435 
2436   // Patching may screw with our temporaries on sparc,
2437   // so let's do it before loading the class.
2438   if (k->is_loaded()) {
2439     metadata2reg(k->constant_encoding(), k_RInfo);
2440   } else {
2441     klass2reg_with_patching(k_RInfo, op->info_for_patch());
2442   }
2443   assert(obj != k_RInfo, "must be different");
2444 
2445   __ verify_oop(obj);
2446 
2447   // Get object class.
2448   // Not a safepoint as obj null check happens earlier.
2449   if (op->fast_check()) {
2450     if (UseCompressedClassPointers) {
2451       __ load_klass(klass_RInfo, obj);
2452       __ compareU64_and_branch(k_RInfo, klass_RInfo, Assembler::bcondNotEqual, *failure_target);
2453     } else {
2454       __ z_cg(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
2455       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2456     }
2457     // Successful cast, fall through to profile or jump.
2458   } else {
2459     bool need_slow_path = !k->is_loaded() ||
2460                           ((int) k->super_check_offset() == in_bytes(Klass::secondary_super_cache_offset()));
2461     intptr_t super_check_offset = k->is_loaded() ? k->super_check_offset() : -1L;
2462     __ load_klass(klass_RInfo, obj);
2463     // Perform the fast part of the checking logic.
2464     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1,
2465                                      (need_slow_path ? success_target : NULL),
2466                                      failure_target, NULL,
2467                                      RegisterOrConstant(super_check_offset));
2468     if (need_slow_path) {
2469       // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2470       address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2471       store_parameter(klass_RInfo, 0); // sub
2472       store_parameter(k_RInfo, 1);     // super
2473       emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2474       CHECK_BAILOUT();
2475       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2476       // Fall through to success case.
2477     }
2478   }
2479 
2480   if (op->should_profile()) {
2481     Register mdo = klass_RInfo, recv = k_RInfo;
2482     assert_different_registers(obj, mdo, recv);
2483     __ bind(profile_cast_success);
2484     metadata2reg(md->constant_encoding(), mdo);
2485     __ load_klass(recv, obj);
2486     type_profile_helper(mdo, md, data, recv, Rtmp1, success);
2487     __ branch_optimized(Assembler::bcondAlways, *success);
2488 
2489     __ bind(profile_cast_failure);
2490     metadata2reg(md->constant_encoding(), mdo);
2491     __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2492     __ branch_optimized(Assembler::bcondAlways, *failure);
2493   } else {
2494     __ branch_optimized(Assembler::bcondAlways, *success);
2495   }
2496 }
2497 
2498 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2499   LIR_Code code = op->code();
2500   if (code == lir_store_check) {
2501     Register value = op->object()->as_register();
2502     Register array = op->array()->as_register();
2503     Register k_RInfo = op->tmp1()->as_register();
2504     Register klass_RInfo = op->tmp2()->as_register();
2505     Register Rtmp1 = Z_R1_scratch;
2506 
2507     CodeStub* stub = op->stub();
2508 
2509     // Check if it needs to be profiled.
2510     ciMethodData* md = NULL;
2511     ciProfileData* data = NULL;
2512 
2513     assert_different_registers(value, k_RInfo, klass_RInfo);
2514 
2515     if (op->should_profile()) {
2516       ciMethod* method = op->profiled_method();
2517       assert(method != NULL, "Should have method");
2518       int bci = op->profiled_bci();
2519       md = method->method_data_or_null();
2520       assert(md != NULL, "Sanity");
2521       data = md->bci_to_data(bci);
2522       assert(data != NULL,                "need data for type check");
2523       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2524     }
2525     NearLabel profile_cast_success, profile_cast_failure, done;
2526     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2527     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2528 
2529     if (op->should_profile()) {
2530       NearLabel not_null;
2531       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2532       // Object is null; update MDO and exit.
2533       Register mdo = klass_RInfo;
2534       metadata2reg(md->constant_encoding(), mdo);
2535       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2536       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2537       __ or2mem_8(data_addr, header_bits);
2538       __ branch_optimized(Assembler::bcondAlways, done);
2539       __ bind(not_null);
2540     } else {
2541       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondEqual, done);
2542     }
2543 
2544     add_debug_info_for_null_check_here(op->info_for_exception());
2545     __ load_klass(k_RInfo, array);
2546     __ load_klass(klass_RInfo, value);
2547 
2548     // Get instance klass (it's already uncompressed).
2549     __ z_lg(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
2550     // Perform the fast part of the checking logic.
2551     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
2552     // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2553     address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2554     store_parameter(klass_RInfo, 0); // sub
2555     store_parameter(k_RInfo, 1);     // super
2556     emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2557     CHECK_BAILOUT();
2558     __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2559     // Fall through to success case.
2560 
2561     if (op->should_profile()) {
2562       Register mdo = klass_RInfo, recv = k_RInfo;
2563       assert_different_registers(value, mdo, recv);
2564       __ bind(profile_cast_success);
2565       metadata2reg(md->constant_encoding(), mdo);
2566       __ load_klass(recv, value);
2567       type_profile_helper(mdo, md, data, recv, Rtmp1, &done);
2568       __ branch_optimized(Assembler::bcondAlways, done);
2569 
2570       __ bind(profile_cast_failure);
2571       metadata2reg(md->constant_encoding(), mdo);
2572       __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2573       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2574     }
2575 
2576     __ bind(done);
2577   } else {
2578     if (code == lir_checkcast) {
2579       Register obj = op->object()->as_register();
2580       Register dst = op->result_opr()->as_register();
2581       NearLabel success;
2582       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2583       __ bind(success);
2584       __ lgr_if_needed(dst, obj);
2585     } else {
2586       if (code == lir_instanceof) {
2587         Register obj = op->object()->as_register();
2588         Register dst = op->result_opr()->as_register();
2589         NearLabel success, failure, done;
2590         emit_typecheck_helper(op, &success, &failure, &failure);
2591         __ bind(failure);
2592         __ clear_reg(dst);
2593         __ branch_optimized(Assembler::bcondAlways, done);
2594         __ bind(success);
2595         __ load_const_optimized(dst, 1);
2596         __ bind(done);
2597       } else {
2598         ShouldNotReachHere();
2599       }
2600     }
2601   }
2602 }
2603 
2604 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2605   Register addr = op->addr()->as_pointer_register();
2606   Register t1_cmp = Z_R1_scratch;
2607   if (op->code() == lir_cas_long) {
2608     assert(VM_Version::supports_cx8(), "wrong machine");
2609     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2610     Register new_value_lo = op->new_value()->as_register_lo();
2611     __ z_lgr(t1_cmp, cmp_value_lo);
2612     // Perform the compare and swap operation.
2613     __ z_csg(t1_cmp, new_value_lo, 0, addr);
2614   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2615     Register cmp_value = op->cmp_value()->as_register();
2616     Register new_value = op->new_value()->as_register();
2617     if (op->code() == lir_cas_obj) {
2618       if (UseCompressedOops) {
2619                  t1_cmp = op->tmp1()->as_register();
2620         Register t2_new = op->tmp2()->as_register();
2621         assert_different_registers(cmp_value, new_value, addr, t1_cmp, t2_new);
2622         __ oop_encoder(t1_cmp, cmp_value, true /*maybe null*/);
2623         __ oop_encoder(t2_new, new_value, true /*maybe null*/);
2624         __ z_cs(t1_cmp, t2_new, 0, addr);
2625       } else {
2626         __ z_lgr(t1_cmp, cmp_value);
2627         __ z_csg(t1_cmp, new_value, 0, addr);
2628       }
2629     } else {
2630       __ z_lr(t1_cmp, cmp_value);
2631       __ z_cs(t1_cmp, new_value, 0, addr);
2632     }
2633   } else {
2634     ShouldNotReachHere(); // new lir_cas_??
2635   }
2636 }
2637 
2638 void LIR_Assembler::set_24bit_FPU() {
2639   ShouldNotCallThis(); // x86 only
2640 }
2641 
2642 void LIR_Assembler::reset_FPU() {
2643   ShouldNotCallThis(); // x86 only
2644 }
2645 
2646 void LIR_Assembler::breakpoint() {
2647   Unimplemented();
2648   //  __ breakpoint_trap();
2649 }
2650 
2651 void LIR_Assembler::push(LIR_Opr opr) {
2652   ShouldNotCallThis(); // unused
2653 }
2654 
2655 void LIR_Assembler::pop(LIR_Opr opr) {
2656   ShouldNotCallThis(); // unused
2657 }
2658 
2659 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2660   Address addr = frame_map()->address_for_monitor_lock(monitor_no);
2661   __ add2reg(dst_opr->as_register(), addr.disp(), addr.base());
2662 }
2663 
2664 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2665   Register obj = op->obj_opr()->as_register();  // May not be an oop.
2666   Register hdr = op->hdr_opr()->as_register();
2667   Register lock = op->lock_opr()->as_register();
2668   if (!UseFastLocking) {
2669     __ branch_optimized(Assembler::bcondAlways, *op->stub()->entry());
2670   } else if (op->code() == lir_lock) {
2671     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2672     // Add debug info for NullPointerException only if one is possible.
2673     if (op->info() != NULL) {
2674       add_debug_info_for_null_check_here(op->info());
2675     }
2676     __ lock_object(hdr, obj, lock, *op->stub()->entry());
2677     // done
2678   } else if (op->code() == lir_unlock) {
2679     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2680     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2681   } else {
2682     ShouldNotReachHere();
2683   }
2684   __ bind(*op->stub()->continuation());
2685 }
2686 
2687 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2688   ciMethod* method = op->profiled_method();
2689   int bci          = op->profiled_bci();
2690   ciMethod* callee = op->profiled_callee();
2691 
2692   // Update counter for all call types.
2693   ciMethodData* md = method->method_data_or_null();
2694   assert(md != NULL, "Sanity");
2695   ciProfileData* data = md->bci_to_data(bci);
2696   assert(data->is_CounterData(), "need CounterData for calls");
2697   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2698   Register mdo  = op->mdo()->as_register();
2699   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2700   Register tmp1 = op->tmp1()->as_register_lo();
2701   metadata2reg(md->constant_encoding(), mdo);
2702 
2703   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2704   Bytecodes::Code bc = method->java_code_at_bci(bci);
2705   const bool callee_is_static = callee->is_loaded() && callee->is_static();
2706   // Perform additional virtual call profiling for invokevirtual and
2707   // invokeinterface bytecodes.
2708   if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
2709       !callee_is_static &&  // Required for optimized MH invokes.
2710       C1ProfileVirtualCalls) {
2711     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2712     Register recv = op->recv()->as_register();
2713     assert_different_registers(mdo, tmp1, recv);
2714     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2715     ciKlass* known_klass = op->known_holder();
2716     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2717       // We know the type that will be seen at this call site; we can
2718       // statically update the MethodData* rather than needing to do
2719       // dynamic tests on the receiver type.
2720 
2721       // NOTE: we should probably put a lock around this search to
2722       // avoid collisions by concurrent compilations.
2723       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2724       uint i;
2725       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2726         ciKlass* receiver = vc_data->receiver(i);
2727         if (known_klass->equals(receiver)) {
2728           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2729           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2730           return;
2731         }
2732       }
2733 
2734       // Receiver type not found in profile data. Select an empty slot.
2735 
2736       // Note that this is less efficient than it should be because it
2737       // always does a write to the receiver part of the
2738       // VirtualCallData rather than just the first time.
2739       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2740         ciKlass* receiver = vc_data->receiver(i);
2741         if (receiver == NULL) {
2742           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
2743           metadata2reg(known_klass->constant_encoding(), tmp1);
2744           __ z_stg(tmp1, recv_addr);
2745           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2746           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2747           return;
2748         }
2749       }
2750     } else {
2751       __ load_klass(recv, recv);
2752       NearLabel update_done;
2753       type_profile_helper(mdo, md, data, recv, tmp1, &update_done);
2754       // Receiver did not match any saved receiver and there is no empty row for it.
2755       // Increment total counter to indicate polymorphic case.
2756       __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2757       __ bind(update_done);
2758     }
2759   } else {
2760     // static call
2761     __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2762   }
2763 }
2764 
2765 void LIR_Assembler::align_backward_branch_target() {
2766   __ align(OptoLoopAlignment);
2767 }
2768 
2769 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2770   ShouldNotCallThis(); // There are no delay slots on ZARCH_64.
2771 }
2772 
2773 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
2774   assert(left->is_register(), "can only handle registers");
2775 
2776   if (left->is_single_cpu()) {
2777     __ z_lcr(dest->as_register(), left->as_register());
2778   } else if (left->is_single_fpu()) {
2779     __ z_lcebr(dest->as_float_reg(), left->as_float_reg());
2780   } else if (left->is_double_fpu()) {
2781     __ z_lcdbr(dest->as_double_reg(), left->as_double_reg());
2782   } else {
2783     assert(left->is_double_cpu(), "Must be a long");
2784     __ z_lcgr(dest->as_register_lo(), left->as_register_lo());
2785   }
2786 }
2787 
2788 void LIR_Assembler::fxch(int i) {
2789   ShouldNotCallThis(); // x86 only
2790 }
2791 
2792 void LIR_Assembler::fld(int i) {
2793   ShouldNotCallThis(); // x86 only
2794 }
2795 
2796 void LIR_Assembler::ffree(int i) {
2797   ShouldNotCallThis(); // x86 only
2798 }
2799 
2800 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2801                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2802   assert(!tmp->is_valid(), "don't need temporary");
2803   emit_call_c(dest);
2804   CHECK_BAILOUT();
2805   if (info != NULL) {
2806     add_call_info_here(info);
2807   }
2808 }
2809 
2810 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2811   ShouldNotCallThis(); // not needed on ZARCH_64
2812 }
2813 
2814 void LIR_Assembler::membar() {
2815   __ z_fence();
2816 }
2817 
2818 void LIR_Assembler::membar_acquire() {
2819   __ z_acquire();
2820 }
2821 
2822 void LIR_Assembler::membar_release() {
2823   __ z_release();
2824 }
2825 
2826 void LIR_Assembler::membar_loadload() {
2827   __ z_acquire();
2828 }
2829 
2830 void LIR_Assembler::membar_storestore() {
2831   __ z_release();
2832 }
2833 
2834 void LIR_Assembler::membar_loadstore() {
2835   __ z_acquire();
2836 }
2837 
2838 void LIR_Assembler::membar_storeload() {
2839   __ z_fence();
2840 }
2841 
2842 void LIR_Assembler::on_spin_wait() {
2843   Unimplemented();
2844 }
2845 
2846 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest) {
2847   LIR_Address* addr = addr_opr->as_address_ptr();
2848   assert(addr->scale() == LIR_Address::times_1, "scaling unsupported");
2849   __ load_address(dest->as_pointer_register(), as_Address(addr));
2850 }
2851 
2852 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2853   ShouldNotCallThis(); // unused
2854 }
2855 
2856 #ifdef ASSERT
2857 // Emit run-time assertion.
2858 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2859   Unimplemented();
2860 }
2861 #endif
2862 
2863 void LIR_Assembler::peephole(LIR_List*) {
2864   // Do nothing for now.
2865 }
2866 
2867 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2868   assert(code == lir_xadd, "lir_xchg not supported");
2869   Address src_addr = as_Address(src->as_address_ptr());
2870   Register base = src_addr.base();
2871   intptr_t disp = src_addr.disp();
2872   if (src_addr.index()->is_valid()) {
2873     // LAA and LAAG do not support index register.
2874     __ load_address(Z_R1_scratch, src_addr);
2875     base = Z_R1_scratch;
2876     disp = 0;
2877   }
2878   if (data->type() == T_INT) {
2879     __ z_laa(dest->as_register(), data->as_register(), disp, base);
2880   } else if (data->type() == T_LONG) {
2881     assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
2882     __ z_laag(dest->as_register_lo(), data->as_register_lo(), disp, base);
2883   } else {
2884     ShouldNotReachHere();
2885   }
2886 }
2887 
2888 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2889   Register obj = op->obj()->as_register();
2890   Register tmp1 = op->tmp()->as_pointer_register();
2891   Register tmp2 = Z_R1_scratch;
2892   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2893   ciKlass* exact_klass = op->exact_klass();
2894   intptr_t current_klass = op->current_klass();
2895   bool not_null = op->not_null();
2896   bool no_conflict = op->no_conflict();
2897 
2898   Label update, next, none, null_seen, init_klass;
2899 
2900   bool do_null = !not_null;
2901   bool exact_klass_set = exact_klass != NULL && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2902   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2903 
2904   assert(do_null || do_update, "why are we here?");
2905   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2906 
2907   __ verify_oop(obj);
2908 
2909   if (do_null || tmp1 != obj DEBUG_ONLY(|| true)) {
2910     __ z_ltgr(tmp1, obj);
2911   }
2912   if (do_null) {
2913     __ z_brnz(update);
2914     if (!TypeEntries::was_null_seen(current_klass)) {
2915       __ z_lg(tmp1, mdo_addr);
2916       __ z_oill(tmp1, TypeEntries::null_seen);
2917       __ z_stg(tmp1, mdo_addr);
2918     }
2919     if (do_update) {
2920       __ z_bru(next);
2921     }
2922   } else {
2923     __ asm_assert_ne("unexpect null obj", __LINE__);
2924   }
2925 
2926   __ bind(update);
2927 
2928   if (do_update) {
2929 #ifdef ASSERT
2930     if (exact_klass != NULL) {
2931       __ load_klass(tmp1, tmp1);
2932       metadata2reg(exact_klass->constant_encoding(), tmp2);
2933       __ z_cgr(tmp1, tmp2);
2934       __ asm_assert_eq("exact klass and actual klass differ", __LINE__);
2935     }
2936 #endif
2937 
2938     Label do_update;
2939     __ z_lg(tmp2, mdo_addr);
2940 
2941     if (!no_conflict) {
2942       if (exact_klass == NULL || TypeEntries::is_type_none(current_klass)) {
2943         if (exact_klass != NULL) {
2944           metadata2reg(exact_klass->constant_encoding(), tmp1);
2945         } else {
2946           __ load_klass(tmp1, tmp1);
2947         }
2948 
2949         // Klass seen before: nothing to do (regardless of unknown bit).
2950         __ z_lgr(Z_R0_scratch, tmp2);
2951         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
2952         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
2953         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
2954 
2955         // Already unknown: Nothing to do anymore.
2956         __ z_tmll(tmp2, TypeEntries::type_unknown);
2957         __ z_brc(Assembler::bcondAllOne, next);
2958 
2959         if (TypeEntries::is_type_none(current_klass)) {
2960           __ z_lgr(Z_R0_scratch, tmp2);
2961           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
2962           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
2963           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, init_klass);
2964         }
2965       } else {
2966         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
2967                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
2968 
2969         // Already unknown: Nothing to do anymore.
2970         __ z_tmll(tmp2, TypeEntries::type_unknown);
2971         __ z_brc(Assembler::bcondAllOne, next);
2972       }
2973 
2974       // Different than before. Cannot keep accurate profile.
2975       __ z_oill(tmp2, TypeEntries::type_unknown);
2976       __ z_bru(do_update);
2977     } else {
2978       // There's a single possible klass at this profile point.
2979       assert(exact_klass != NULL, "should be");
2980       if (TypeEntries::is_type_none(current_klass)) {
2981         metadata2reg(exact_klass->constant_encoding(), tmp1);
2982         __ z_lgr(Z_R0_scratch, tmp2);
2983         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
2984         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
2985         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
2986 #ifdef ASSERT
2987         {
2988           Label ok;
2989           __ z_lgr(Z_R0_scratch, tmp2);
2990           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
2991           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
2992           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, ok);
2993           __ stop("unexpected profiling mismatch");
2994           __ bind(ok);
2995         }
2996 #endif
2997 
2998       } else {
2999         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3000                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3001 
3002         // Already unknown: Nothing to do anymore.
3003         __ z_tmll(tmp2, TypeEntries::type_unknown);
3004         __ z_brc(Assembler::bcondAllOne, next);
3005         __ z_oill(tmp2, TypeEntries::type_unknown);
3006         __ z_bru(do_update);
3007       }
3008     }
3009 
3010     __ bind(init_klass);
3011     // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3012     __ z_ogr(tmp2, tmp1);
3013 
3014     __ bind(do_update);
3015     __ z_stg(tmp2, mdo_addr);
3016 
3017     __ bind(next);
3018   }
3019 }
3020 
3021 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3022   assert(op->crc()->is_single_cpu(), "crc must be register");
3023   assert(op->val()->is_single_cpu(), "byte value must be register");
3024   assert(op->result_opr()->is_single_cpu(), "result must be register");
3025   Register crc = op->crc()->as_register();
3026   Register val = op->val()->as_register();
3027   Register res = op->result_opr()->as_register();
3028 
3029   assert_different_registers(val, crc, res);
3030 
3031   __ load_const_optimized(res, StubRoutines::crc_table_addr());
3032   __ not_(crc, noreg, false); // ~crc
3033   __ update_byte_crc32(crc, val, res);
3034   __ not_(res, crc, false); // ~crc
3035 }
3036 
3037 #undef __