1 /*
   2  * Copyright (c) 2016, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2019, 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 "asm/macroAssembler.inline.hpp"
  28 #include "c1/c1_Compilation.hpp"
  29 #include "c1/c1_LIRAssembler.hpp"
  30 #include "c1/c1_MacroAssembler.hpp"
  31 #include "c1/c1_Runtime1.hpp"
  32 #include "c1/c1_ValueStack.hpp"
  33 #include "ci/ciArrayKlass.hpp"
  34 #include "ci/ciInstance.hpp"
  35 #include "gc/shared/collectedHeap.hpp"
  36 #include "gc/shared/barrierSet.hpp"
  37 #include "gc/shared/cardTableBarrierSet.hpp"
  38 #include "memory/universe.hpp"
  39 #include "nativeInst_s390.hpp"
  40 #include "oops/objArrayKlass.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/safepointMechanism.inline.hpp"
  43 #include "runtime/sharedRuntime.hpp"
  44 #include "vmreg_s390.inline.hpp"
  45 
  46 #define __ _masm->
  47 
  48 #ifndef PRODUCT
  49 #undef __
  50 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm) : _masm)->
  51 #endif
  52 
  53 //------------------------------------------------------------
  54 
  55 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
  56   // Not used on ZARCH_64
  57   ShouldNotCallThis();
  58   return false;
  59 }
  60 
  61 LIR_Opr LIR_Assembler::receiverOpr() {
  62   return FrameMap::Z_R2_oop_opr;
  63 }
  64 
  65 LIR_Opr LIR_Assembler::osrBufferPointer() {
  66   return FrameMap::Z_R2_opr;
  67 }
  68 
  69 int LIR_Assembler::initial_frame_size_in_bytes() const {
  70   return in_bytes(frame_map()->framesize_in_bytes());
  71 }
  72 
  73 // Inline cache check: done before the frame is built.
  74 // The inline cached class is in Z_inline_cache(Z_R9).
  75 // We fetch the class of the receiver and compare it with the cached class.
  76 // If they do not match we jump to the slow case.
  77 int LIR_Assembler::check_icache() {
  78   Register receiver = receiverOpr()->as_register();
  79   int offset = __ offset();
  80   __ inline_cache_check(receiver, Z_inline_cache);
  81   return offset;
  82 }
  83 
  84 void LIR_Assembler::clinit_barrier(ciMethod* method) {
  85   assert(!method->holder()->is_not_initialized(), "initialization should have been started");
  86 
  87   Label L_skip_barrier;
  88   Register klass = Z_R1_scratch;
  89 
  90   metadata2reg(method->holder()->constant_encoding(), klass);
  91   __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);
  92 
  93   __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
  94   __ z_br(klass);
  95 
  96   __ bind(L_skip_barrier);
  97 }
  98 
  99 void LIR_Assembler::osr_entry() {
 100   // On-stack-replacement entry sequence (interpreter frame layout described in interpreter_sparc.cpp):
 101   //
 102   //   1. Create a new compiled activation.
 103   //   2. Initialize local variables in the compiled activation. The expression stack must be empty
 104   //      at the osr_bci; it is not initialized.
 105   //   3. Jump to the continuation address in compiled code to resume execution.
 106 
 107   // OSR entry point
 108   offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
 109   BlockBegin* osr_entry = compilation()->hir()->osr_entry();
 110   ValueStack* entry_state = osr_entry->end()->state();
 111   int number_of_locks = entry_state->locks_size();
 112 
 113   // Create a frame for the compiled activation.
 114   __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
 115 
 116   // OSR buffer is
 117   //
 118   // locals[nlocals-1..0]
 119   // monitors[number_of_locks-1..0]
 120   //
 121   // Locals is a direct copy of the interpreter frame so in the osr buffer
 122   // the first slot in the local array is the last local from the interpreter
 123   // and the last slot is local[0] (receiver) from the interpreter
 124   //
 125   // Similarly with locks. The first lock slot in the osr buffer is the nth lock
 126   // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
 127   // in the interpreter frame (the method lock if a sync method)
 128 
 129   // Initialize monitors in the compiled activation.
 130   //   I0: pointer to osr buffer
 131   //
 132   // All other registers are dead at this point and the locals will be
 133   // copied into place by code emitted in the IR.
 134 
 135   Register OSR_buf = osrBufferPointer()->as_register();
 136   { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
 137     int monitor_offset = BytesPerWord * method()->max_locals() +
 138       (2 * BytesPerWord) * (number_of_locks - 1);
 139     // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
 140     // the OSR buffer using 2 word entries: first the lock and then
 141     // the oop.
 142     for (int i = 0; i < number_of_locks; i++) {
 143       int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
 144       // Verify the interpreter's monitor has a non-null object.
 145       __ asm_assert_mem8_isnot_zero(slot_offset + 1*BytesPerWord, OSR_buf, "locked object is NULL", __LINE__);
 146       // Copy the lock field into the compiled activation.
 147       __ z_lg(Z_R1_scratch, slot_offset + 0, OSR_buf);
 148       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_lock(i));
 149       __ z_lg(Z_R1_scratch, slot_offset + 1*BytesPerWord, OSR_buf);
 150       __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_object(i));
 151     }
 152   }
 153 }
 154 
 155 // --------------------------------------------------------------------------------------------
 156 
 157 address LIR_Assembler::emit_call_c(address a) {
 158   __ align_call_far_patchable(__ pc());
 159   address call_addr = __ call_c_opt(a);
 160   if (call_addr == NULL) {
 161     bailout("const section overflow");
 162   }
 163   return call_addr;
 164 }
 165 
 166 int LIR_Assembler::emit_exception_handler() {
 167   // If the last instruction is a call (typically to do a throw which
 168   // is coming at the end after block reordering) the return address
 169   // must still point into the code area in order to avoid assertion
 170   // failures when searching for the corresponding bci. => Add a nop.
 171   // (was bug 5/14/1999 - gri)
 172   __ nop();
 173 
 174   // Generate code for exception handler.
 175   address handler_base = __ start_a_stub(exception_handler_size());
 176   if (handler_base == NULL) {
 177     // Not enough space left for the handler.
 178     bailout("exception handler overflow");
 179     return -1;
 180   }
 181 
 182   int offset = code_offset();
 183 
 184   address a = Runtime1::entry_for (Runtime1::handle_exception_from_callee_id);
 185   address call_addr = emit_call_c(a);
 186   CHECK_BAILOUT_(-1);
 187   __ should_not_reach_here();
 188   guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
 189   __ end_a_stub();
 190 
 191   return offset;
 192 }
 193 
 194 // Emit the code to remove the frame from the stack in the exception
 195 // unwind path.
 196 int LIR_Assembler::emit_unwind_handler() {
 197 #ifndef PRODUCT
 198   if (CommentedAssembly) {
 199     _masm->block_comment("Unwind handler");
 200   }
 201 #endif
 202 
 203   int offset = code_offset();
 204   Register exception_oop_callee_saved = Z_R10; // Z_R10 is callee-saved.
 205   Register Rtmp1                      = Z_R11;
 206   Register Rtmp2                      = Z_R12;
 207 
 208   // Fetch the exception from TLS and clear out exception related thread state.
 209   Address exc_oop_addr = Address(Z_thread, JavaThread::exception_oop_offset());
 210   Address exc_pc_addr  = Address(Z_thread, JavaThread::exception_pc_offset());
 211   __ z_lg(Z_EXC_OOP, exc_oop_addr);
 212   __ clear_mem(exc_oop_addr, sizeof(oop));
 213   __ clear_mem(exc_pc_addr, sizeof(intptr_t));
 214 
 215   __ bind(_unwind_handler_entry);
 216   __ verify_not_null_oop(Z_EXC_OOP);
 217   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 218     __ lgr_if_needed(exception_oop_callee_saved, Z_EXC_OOP); // Preserve the exception.
 219   }
 220 
 221   // Preform needed unlocking.
 222   MonitorExitStub* stub = NULL;
 223   if (method()->is_synchronized()) {
 224     // Runtime1::monitorexit_id expects lock address in Z_R1_scratch.
 225     LIR_Opr lock = FrameMap::as_opr(Z_R1_scratch);
 226     monitor_address(0, lock);
 227     stub = new MonitorExitStub(lock, true, 0);
 228     __ unlock_object(Rtmp1, Rtmp2, lock->as_register(), *stub->entry());
 229     __ bind(*stub->continuation());
 230   }
 231 
 232   if (compilation()->env()->dtrace_method_probes()) {
 233     ShouldNotReachHere(); // Not supported.
 234 #if 0
 235     __ mov(rdi, r15_thread);
 236     __ mov_metadata(rsi, method()->constant_encoding());
 237     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
 238 #endif
 239   }
 240 
 241   if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
 242     __ lgr_if_needed(Z_EXC_OOP, exception_oop_callee_saved);  // Restore the exception.
 243   }
 244 
 245   // Remove the activation and dispatch to the unwind handler.
 246   __ pop_frame();
 247   __ z_lg(Z_EXC_PC, _z_abi16(return_pc), Z_SP);
 248 
 249   // Z_EXC_OOP: exception oop
 250   // Z_EXC_PC: exception pc
 251 
 252   // Dispatch to the unwind logic.
 253   __ load_const_optimized(Z_R5, Runtime1::entry_for (Runtime1::unwind_exception_id));
 254   __ z_br(Z_R5);
 255 
 256   // Emit the slow path assembly.
 257   if (stub != NULL) {
 258     stub->emit_code(this);
 259   }
 260 
 261   return offset;
 262 }
 263 
 264 int LIR_Assembler::emit_deopt_handler() {
 265   // If the last instruction is a call (typically to do a throw which
 266   // is coming at the end after block reordering) the return address
 267   // must still point into the code area in order to avoid assertion
 268   // failures when searching for the corresponding bci. => Add a nop.
 269   // (was bug 5/14/1999 - gri)
 270   __ nop();
 271 
 272   // Generate code for exception handler.
 273   address handler_base = __ start_a_stub(deopt_handler_size());
 274   if (handler_base == NULL) {
 275     // Not enough space left for the handler.
 276     bailout("deopt handler overflow");
 277     return -1;
 278   }  int offset = code_offset();
 279   // Size must be constant (see HandlerImpl::emit_deopt_handler).
 280   __ load_const(Z_R1_scratch, SharedRuntime::deopt_blob()->unpack());
 281   __ call(Z_R1_scratch);
 282   guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
 283   __ end_a_stub();
 284 
 285   return offset;
 286 }
 287 
 288 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
 289   if (o == NULL) {
 290     __ clear_reg(reg, true/*64bit*/, false/*set cc*/); // Must not kill cc set by cmove.
 291   } else {
 292     AddressLiteral a = __ allocate_oop_address(o);
 293     bool success = __ load_oop_from_toc(reg, a, reg);
 294     if (!success) {
 295       bailout("const section overflow");
 296     }
 297   }
 298 }
 299 
 300 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
 301   // Allocate a new index in table to hold the object once it's been patched.
 302   int oop_index = __ oop_recorder()->allocate_oop_index(NULL);
 303   PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
 304 
 305   AddressLiteral addrlit((intptr_t)0, oop_Relocation::spec(oop_index));
 306   assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
 307   // The NULL will be dynamically patched later so the sequence to
 308   // load the address literal must not be optimized.
 309   __ load_const(reg, addrlit);
 310 
 311   patching_epilog(patch, lir_patch_normal, reg, info);
 312 }
 313 
 314 void LIR_Assembler::metadata2reg(Metadata* md, Register reg) {
 315   bool success = __ set_metadata_constant(md, reg);
 316   if (!success) {
 317     bailout("const section overflow");
 318     return;
 319   }
 320 }
 321 
 322 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
 323   // Allocate a new index in table to hold the klass once it's been patched.
 324   int index = __ oop_recorder()->allocate_metadata_index(NULL);
 325   PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
 326   AddressLiteral addrlit((intptr_t)0, metadata_Relocation::spec(index));
 327   assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
 328   // The NULL will be dynamically patched later so the sequence to
 329   // load the address literal must not be optimized.
 330   __ load_const(reg, addrlit);
 331 
 332   patching_epilog(patch, lir_patch_normal, reg, info);
 333 }
 334 
 335 void LIR_Assembler::emit_op3(LIR_Op3* op) {
 336   switch (op->code()) {
 337     case lir_idiv:
 338     case lir_irem:
 339       arithmetic_idiv(op->code(),
 340                       op->in_opr1(),
 341                       op->in_opr2(),
 342                       op->in_opr3(),
 343                       op->result_opr(),
 344                       op->info());
 345       break;
 346     case lir_fmad: {
 347       const FloatRegister opr1 = op->in_opr1()->as_double_reg(),
 348                           opr2 = op->in_opr2()->as_double_reg(),
 349                           opr3 = op->in_opr3()->as_double_reg(),
 350                           res  = op->result_opr()->as_double_reg();
 351       __ z_madbr(opr3, opr1, opr2);
 352       if (res != opr3) { __ z_ldr(res, opr3); }
 353     } break;
 354     case lir_fmaf: {
 355       const FloatRegister opr1 = op->in_opr1()->as_float_reg(),
 356                           opr2 = op->in_opr2()->as_float_reg(),
 357                           opr3 = op->in_opr3()->as_float_reg(),
 358                           res  = op->result_opr()->as_float_reg();
 359       __ z_maebr(opr3, opr1, opr2);
 360       if (res != opr3) { __ z_ler(res, opr3); }
 361     } break;
 362     default: ShouldNotReachHere(); break;
 363   }
 364 }
 365 
 366 
 367 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
 368 #ifdef ASSERT
 369   assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
 370   if (op->block() != NULL)  { _branch_target_blocks.append(op->block()); }
 371   if (op->ublock() != NULL) { _branch_target_blocks.append(op->ublock()); }
 372 #endif
 373 
 374   if (op->cond() == lir_cond_always) {
 375     if (op->info() != NULL) { add_debug_info_for_branch(op->info()); }
 376     __ branch_optimized(Assembler::bcondAlways, *(op->label()));
 377   } else {
 378     Assembler::branch_condition acond = Assembler::bcondZero;
 379     if (op->code() == lir_cond_float_branch) {
 380       assert(op->ublock() != NULL, "must have unordered successor");
 381       __ branch_optimized(Assembler::bcondNotOrdered, *(op->ublock()->label()));
 382     }
 383     switch (op->cond()) {
 384       case lir_cond_equal:        acond = Assembler::bcondEqual;     break;
 385       case lir_cond_notEqual:     acond = Assembler::bcondNotEqual;  break;
 386       case lir_cond_less:         acond = Assembler::bcondLow;       break;
 387       case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;   break;
 388       case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;    break;
 389       case lir_cond_greater:      acond = Assembler::bcondHigh;      break;
 390       case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;   break;
 391       case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;    break;
 392       default:                         ShouldNotReachHere();
 393     }
 394     __ branch_optimized(acond,*(op->label()));
 395   }
 396 }
 397 
 398 
 399 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
 400   LIR_Opr src  = op->in_opr();
 401   LIR_Opr dest = op->result_opr();
 402 
 403   switch (op->bytecode()) {
 404     case Bytecodes::_i2l:
 405       __ move_reg_if_needed(dest->as_register_lo(), T_LONG, src->as_register(), T_INT);
 406       break;
 407 
 408     case Bytecodes::_l2i:
 409       __ move_reg_if_needed(dest->as_register(), T_INT, src->as_register_lo(), T_LONG);
 410       break;
 411 
 412     case Bytecodes::_i2b:
 413       __ move_reg_if_needed(dest->as_register(), T_BYTE, src->as_register(), T_INT);
 414       break;
 415 
 416     case Bytecodes::_i2c:
 417       __ move_reg_if_needed(dest->as_register(), T_CHAR, src->as_register(), T_INT);
 418       break;
 419 
 420     case Bytecodes::_i2s:
 421       __ move_reg_if_needed(dest->as_register(), T_SHORT, src->as_register(), T_INT);
 422       break;
 423 
 424     case Bytecodes::_f2d:
 425       assert(dest->is_double_fpu(), "check");
 426       __ move_freg_if_needed(dest->as_double_reg(), T_DOUBLE, src->as_float_reg(), T_FLOAT);
 427       break;
 428 
 429     case Bytecodes::_d2f:
 430       assert(dest->is_single_fpu(), "check");
 431       __ move_freg_if_needed(dest->as_float_reg(), T_FLOAT, src->as_double_reg(), T_DOUBLE);
 432       break;
 433 
 434     case Bytecodes::_i2f:
 435       __ z_cefbr(dest->as_float_reg(), src->as_register());
 436       break;
 437 
 438     case Bytecodes::_i2d:
 439       __ z_cdfbr(dest->as_double_reg(), src->as_register());
 440       break;
 441 
 442     case Bytecodes::_l2f:
 443       __ z_cegbr(dest->as_float_reg(), src->as_register_lo());
 444       break;
 445     case Bytecodes::_l2d:
 446       __ z_cdgbr(dest->as_double_reg(), src->as_register_lo());
 447       break;
 448 
 449     case Bytecodes::_f2i:
 450     case Bytecodes::_f2l: {
 451       Label done;
 452       FloatRegister Rsrc = src->as_float_reg();
 453       Register Rdst = (op->bytecode() == Bytecodes::_f2i ? dest->as_register() : dest->as_register_lo());
 454       __ clear_reg(Rdst, true, false);
 455       __ z_cebr(Rsrc, Rsrc);
 456       __ z_brno(done); // NaN -> 0
 457       if (op->bytecode() == Bytecodes::_f2i) {
 458         __ z_cfebr(Rdst, Rsrc, Assembler::to_zero);
 459       } else { // op->bytecode() == Bytecodes::_f2l
 460         __ z_cgebr(Rdst, Rsrc, Assembler::to_zero);
 461       }
 462       __ bind(done);
 463     }
 464     break;
 465 
 466     case Bytecodes::_d2i:
 467     case Bytecodes::_d2l: {
 468       Label done;
 469       FloatRegister Rsrc = src->as_double_reg();
 470       Register Rdst = (op->bytecode() == Bytecodes::_d2i ? dest->as_register() : dest->as_register_lo());
 471       __ clear_reg(Rdst, true, false);  // Don't set CC.
 472       __ z_cdbr(Rsrc, Rsrc);
 473       __ z_brno(done); // NaN -> 0
 474       if (op->bytecode() == Bytecodes::_d2i) {
 475         __ z_cfdbr(Rdst, Rsrc, Assembler::to_zero);
 476       } else { // Bytecodes::_d2l
 477         __ z_cgdbr(Rdst, Rsrc, Assembler::to_zero);
 478       }
 479       __ bind(done);
 480     }
 481     break;
 482 
 483     default: ShouldNotReachHere();
 484   }
 485 }
 486 
 487 void LIR_Assembler::align_call(LIR_Code code) {
 488   // End of call instruction must be 4 byte aligned.
 489   int offset = __ offset();
 490   switch (code) {
 491     case lir_icvirtual_call:
 492       offset += MacroAssembler::load_const_from_toc_size();
 493       // no break
 494     case lir_static_call:
 495     case lir_optvirtual_call:
 496     case lir_dynamic_call:
 497       offset += NativeCall::call_far_pcrelative_displacement_offset;
 498       break;
 499     case lir_virtual_call:   // currently, sparc-specific for niagara
 500     default: ShouldNotReachHere();
 501   }
 502   if ((offset & (NativeCall::call_far_pcrelative_displacement_alignment-1)) != 0) {
 503     __ nop();
 504   }
 505 }
 506 
 507 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
 508   assert((__ offset() + NativeCall::call_far_pcrelative_displacement_offset) % NativeCall::call_far_pcrelative_displacement_alignment == 0,
 509          "must be aligned (offset=%d)", __ offset());
 510   assert(rtype == relocInfo::none ||
 511          rtype == relocInfo::opt_virtual_call_type ||
 512          rtype == relocInfo::static_call_type, "unexpected rtype");
 513   // Prepend each BRASL with a nop.
 514   __ relocate(rtype);
 515   __ z_nop();
 516   __ z_brasl(Z_R14, op->addr());
 517   add_call_info(code_offset(), op->info());
 518 }
 519 
 520 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
 521   address virtual_call_oop_addr = NULL;
 522   AddressLiteral empty_ic((address) Universe::non_oop_word());
 523   virtual_call_oop_addr = __ pc();
 524   bool success = __ load_const_from_toc(Z_inline_cache, empty_ic);
 525   if (!success) {
 526     bailout("const section overflow");
 527     return;
 528   }
 529 
 530   // CALL to fixup routine. Fixup routine uses ScopeDesc info
 531   // to determine who we intended to call.
 532   __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
 533   call(op, relocInfo::none);
 534 }
 535 
 536 // not supported
 537 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
 538   ShouldNotReachHere();
 539 }
 540 
 541 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
 542   if (from_reg != to_reg) __ z_lgr(to_reg, from_reg);
 543 }
 544 
 545 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
 546   assert(src->is_constant(), "should not call otherwise");
 547   assert(dest->is_stack(), "should not call otherwise");
 548   LIR_Const* c = src->as_constant_ptr();
 549 
 550   unsigned int lmem = 0;
 551   unsigned int lcon = 0;
 552   int64_t cbits = 0;
 553   Address dest_addr;
 554   switch (c->type()) {
 555     case T_INT:  // fall through
 556     case T_FLOAT:
 557       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 558       lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 559       break;
 560 
 561     case T_ADDRESS:
 562       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 563       lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 564       break;
 565 
 566     case T_OBJECT:
 567       dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
 568       if (c->as_jobject() == NULL) {
 569         __ store_const(dest_addr, (int64_t)NULL_WORD, 8, 8);
 570       } else {
 571         jobject2reg(c->as_jobject(), Z_R1_scratch);
 572         __ reg2mem_opt(Z_R1_scratch, dest_addr, true);
 573       }
 574       return;
 575 
 576     case T_LONG:  // fall through
 577     case T_DOUBLE:
 578       dest_addr = frame_map()->address_for_slot(dest->double_stack_ix());
 579       lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 580       break;
 581 
 582     default:
 583       ShouldNotReachHere();
 584   }
 585 
 586   __ store_const(dest_addr, cbits, lmem, lcon);
 587 }
 588 
 589 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
 590   assert(src->is_constant(), "should not call otherwise");
 591   assert(dest->is_address(), "should not call otherwise");
 592 
 593   LIR_Const* c = src->as_constant_ptr();
 594   Address addr = as_Address(dest->as_address_ptr());
 595 
 596   int store_offset = -1;
 597 
 598   if (dest->as_address_ptr()->index()->is_valid()) {
 599     switch (type) {
 600       case T_INT:    // fall through
 601       case T_FLOAT:
 602         __ load_const_optimized(Z_R0_scratch, c->as_jint_bits());
 603         store_offset = __ offset();
 604         if (Immediate::is_uimm12(addr.disp())) {
 605           __ z_st(Z_R0_scratch, addr);
 606         } else {
 607           __ z_sty(Z_R0_scratch, addr);
 608         }
 609         break;
 610 
 611       case T_ADDRESS:
 612         __ load_const_optimized(Z_R1_scratch, c->as_jint_bits());
 613         store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 614         break;
 615 
 616       case T_OBJECT:  // fall through
 617       case T_ARRAY:
 618         if (c->as_jobject() == NULL) {
 619           if (UseCompressedOops && !wide) {
 620             __ clear_reg(Z_R1_scratch, false);
 621             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 622           } else {
 623             __ clear_reg(Z_R1_scratch, true);
 624             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 625           }
 626         } else {
 627           jobject2reg(c->as_jobject(), Z_R1_scratch);
 628           if (UseCompressedOops && !wide) {
 629             __ encode_heap_oop(Z_R1_scratch);
 630             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 631           } else {
 632             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 633           }
 634         }
 635         assert(store_offset >= 0, "check");
 636         break;
 637 
 638       case T_LONG:    // fall through
 639       case T_DOUBLE:
 640         __ load_const_optimized(Z_R1_scratch, (int64_t)(c->as_jlong_bits()));
 641         store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 642         break;
 643 
 644       case T_BOOLEAN: // fall through
 645       case T_BYTE:
 646         __ load_const_optimized(Z_R0_scratch, (int8_t)(c->as_jint()));
 647         store_offset = __ offset();
 648         if (Immediate::is_uimm12(addr.disp())) {
 649           __ z_stc(Z_R0_scratch, addr);
 650         } else {
 651           __ z_stcy(Z_R0_scratch, addr);
 652         }
 653         break;
 654 
 655       case T_CHAR:    // fall through
 656       case T_SHORT:
 657         __ load_const_optimized(Z_R0_scratch, (int16_t)(c->as_jint()));
 658         store_offset = __ offset();
 659         if (Immediate::is_uimm12(addr.disp())) {
 660           __ z_sth(Z_R0_scratch, addr);
 661         } else {
 662           __ z_sthy(Z_R0_scratch, addr);
 663         }
 664         break;
 665 
 666       default:
 667         ShouldNotReachHere();
 668     }
 669 
 670   } else { // no index
 671 
 672     unsigned int lmem = 0;
 673     unsigned int lcon = 0;
 674     int64_t cbits = 0;
 675 
 676     switch (type) {
 677       case T_INT:    // fall through
 678       case T_FLOAT:
 679         lmem = 4; lcon = 4; cbits = c->as_jint_bits();
 680         break;
 681 
 682       case T_ADDRESS:
 683         lmem = 8; lcon = 4; cbits = c->as_jint_bits();
 684         break;
 685 
 686       case T_OBJECT:  // fall through
 687       case T_ARRAY:
 688         if (c->as_jobject() == NULL) {
 689           if (UseCompressedOops && !wide) {
 690             store_offset = __ store_const(addr, (int32_t)NULL_WORD, 4, 4);
 691           } else {
 692             store_offset = __ store_const(addr, (int64_t)NULL_WORD, 8, 8);
 693           }
 694         } else {
 695           jobject2reg(c->as_jobject(), Z_R1_scratch);
 696           if (UseCompressedOops && !wide) {
 697             __ encode_heap_oop(Z_R1_scratch);
 698             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
 699           } else {
 700             store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
 701           }
 702         }
 703         assert(store_offset >= 0, "check");
 704         break;
 705 
 706       case T_LONG:    // fall through
 707       case T_DOUBLE:
 708         lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
 709         break;
 710 
 711       case T_BOOLEAN: // fall through
 712       case T_BYTE:
 713         lmem = 1; lcon = 1; cbits = (int8_t)(c->as_jint());
 714         break;
 715 
 716       case T_CHAR:    // fall through
 717       case T_SHORT:
 718         lmem = 2; lcon = 2; cbits = (int16_t)(c->as_jint());
 719         break;
 720 
 721       default:
 722         ShouldNotReachHere();
 723     }
 724 
 725     if (store_offset == -1) {
 726       store_offset = __ store_const(addr, cbits, lmem, lcon);
 727       assert(store_offset >= 0, "check");
 728     }
 729   }
 730 
 731   if (info != NULL) {
 732     add_debug_info_for_null_check(store_offset, info);
 733   }
 734 }
 735 
 736 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
 737   assert(src->is_constant(), "should not call otherwise");
 738   assert(dest->is_register(), "should not call otherwise");
 739   LIR_Const* c = src->as_constant_ptr();
 740 
 741   switch (c->type()) {
 742     case T_INT: {
 743       assert(patch_code == lir_patch_none, "no patching handled here");
 744       __ load_const_optimized(dest->as_register(), c->as_jint());
 745       break;
 746     }
 747 
 748     case T_ADDRESS: {
 749       assert(patch_code == lir_patch_none, "no patching handled here");
 750       __ load_const_optimized(dest->as_register(), c->as_jint());
 751       break;
 752     }
 753 
 754     case T_LONG: {
 755       assert(patch_code == lir_patch_none, "no patching handled here");
 756       __ load_const_optimized(dest->as_register_lo(), (intptr_t)c->as_jlong());
 757       break;
 758     }
 759 
 760     case T_OBJECT: {
 761       if (patch_code != lir_patch_none) {
 762         jobject2reg_with_patching(dest->as_register(), info);
 763       } else {
 764         jobject2reg(c->as_jobject(), dest->as_register());
 765       }
 766       break;
 767     }
 768 
 769     case T_METADATA: {
 770       if (patch_code != lir_patch_none) {
 771         klass2reg_with_patching(dest->as_register(), info);
 772       } else {
 773         metadata2reg(c->as_metadata(), dest->as_register());
 774       }
 775       break;
 776     }
 777 
 778     case T_FLOAT: {
 779       Register toc_reg = Z_R1_scratch;
 780       __ load_toc(toc_reg);
 781       address const_addr = __ float_constant(c->as_jfloat());
 782       if (const_addr == NULL) {
 783         bailout("const section overflow");
 784         break;
 785       }
 786       int displ = const_addr - _masm->code()->consts()->start();
 787       if (dest->is_single_fpu()) {
 788         __ z_ley(dest->as_float_reg(), displ, toc_reg);
 789       } else {
 790         assert(dest->is_single_cpu(), "Must be a cpu register.");
 791         __ z_ly(dest->as_register(), displ, toc_reg);
 792       }
 793     }
 794     break;
 795 
 796     case T_DOUBLE: {
 797       Register toc_reg = Z_R1_scratch;
 798       __ load_toc(toc_reg);
 799       address const_addr = __ double_constant(c->as_jdouble());
 800       if (const_addr == NULL) {
 801         bailout("const section overflow");
 802         break;
 803       }
 804       int displ = const_addr - _masm->code()->consts()->start();
 805       if (dest->is_double_fpu()) {
 806         __ z_ldy(dest->as_double_reg(), displ, toc_reg);
 807       } else {
 808         assert(dest->is_double_cpu(), "Must be a long register.");
 809         __ z_lg(dest->as_register_lo(), displ, toc_reg);
 810       }
 811     }
 812     break;
 813 
 814     default:
 815       ShouldNotReachHere();
 816   }
 817 }
 818 
 819 Address LIR_Assembler::as_Address(LIR_Address* addr) {
 820   if (addr->base()->is_illegal()) {
 821     Unimplemented();
 822   }
 823 
 824   Register base = addr->base()->as_pointer_register();
 825 
 826   if (addr->index()->is_illegal()) {
 827     return Address(base, addr->disp());
 828   } else if (addr->index()->is_cpu_register()) {
 829     Register index = addr->index()->as_pointer_register();
 830     return Address(base, index, addr->disp());
 831   } else if (addr->index()->is_constant()) {
 832     intptr_t addr_offset = addr->index()->as_constant_ptr()->as_jint() + addr->disp();
 833     return Address(base, addr_offset);
 834   } else {
 835     ShouldNotReachHere();
 836     return Address();
 837   }
 838 }
 839 
 840 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
 841   switch (type) {
 842     case T_INT:
 843     case T_FLOAT: {
 844       Register tmp = Z_R1_scratch;
 845       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 846       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 847       __ mem2reg_opt(tmp, from, false);
 848       __ reg2mem_opt(tmp, to, false);
 849       break;
 850     }
 851     case T_ADDRESS:
 852     case T_OBJECT: {
 853       Register tmp = Z_R1_scratch;
 854       Address from = frame_map()->address_for_slot(src->single_stack_ix());
 855       Address to   = frame_map()->address_for_slot(dest->single_stack_ix());
 856       __ mem2reg_opt(tmp, from, true);
 857       __ reg2mem_opt(tmp, to, true);
 858       break;
 859     }
 860     case T_LONG:
 861     case T_DOUBLE: {
 862       Register tmp = Z_R1_scratch;
 863       Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
 864       Address to   = frame_map()->address_for_double_slot(dest->double_stack_ix());
 865       __ mem2reg_opt(tmp, from, true);
 866       __ reg2mem_opt(tmp, to, true);
 867       break;
 868     }
 869 
 870     default:
 871       ShouldNotReachHere();
 872   }
 873 }
 874 
 875 // 4-byte accesses only! Don't use it to access 8 bytes!
 876 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
 877   ShouldNotCallThis();
 878   return 0; // unused
 879 }
 880 
 881 // 4-byte accesses only! Don't use it to access 8 bytes!
 882 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
 883   ShouldNotCallThis();
 884   return 0; // unused
 885 }
 886 
 887 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code,
 888                             CodeEmitInfo* info, bool wide, bool unaligned) {
 889 
 890   assert(type != T_METADATA, "load of metadata ptr not supported");
 891   LIR_Address* addr = src_opr->as_address_ptr();
 892   LIR_Opr to_reg = dest;
 893 
 894   Register src = addr->base()->as_pointer_register();
 895   Register disp_reg = Z_R0;
 896   int disp_value = addr->disp();
 897   bool needs_patching = (patch_code != lir_patch_none);
 898 
 899   if (addr->base()->type() == T_OBJECT) {
 900     __ verify_oop(src);
 901   }
 902 
 903   PatchingStub* patch = NULL;
 904   if (needs_patching) {
 905     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
 906     assert(!to_reg->is_double_cpu() ||
 907            patch_code == lir_patch_none ||
 908            patch_code == lir_patch_normal, "patching doesn't match register");
 909   }
 910 
 911   if (addr->index()->is_illegal()) {
 912     if (!Immediate::is_simm20(disp_value)) {
 913       if (needs_patching) {
 914         __ load_const(Z_R1_scratch, (intptr_t)0);
 915       } else {
 916         __ load_const_optimized(Z_R1_scratch, disp_value);
 917       }
 918       disp_reg = Z_R1_scratch;
 919       disp_value = 0;
 920     }
 921   } else {
 922     if (!Immediate::is_simm20(disp_value)) {
 923       __ load_const_optimized(Z_R1_scratch, disp_value);
 924       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
 925       disp_reg = Z_R1_scratch;
 926       disp_value = 0;
 927     }
 928     disp_reg = addr->index()->as_pointer_register();
 929   }
 930 
 931   // Remember the offset of the load. The patching_epilog must be done
 932   // before the call to add_debug_info, otherwise the PcDescs don't get
 933   // entered in increasing order.
 934   int offset = code_offset();
 935 
 936   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
 937 
 938   bool short_disp = Immediate::is_uimm12(disp_value);
 939 
 940   switch (type) {
 941     case T_BOOLEAN: // fall through
 942     case T_BYTE  :  __ z_lb(dest->as_register(),   disp_value, disp_reg, src); break;
 943     case T_CHAR  :  __ z_llgh(dest->as_register(), disp_value, disp_reg, src); break;
 944     case T_SHORT :
 945       if (short_disp) {
 946                     __ z_lh(dest->as_register(),   disp_value, disp_reg, src);
 947       } else {
 948                     __ z_lhy(dest->as_register(),  disp_value, disp_reg, src);
 949       }
 950       break;
 951     case T_INT   :
 952       if (short_disp) {
 953                     __ z_l(dest->as_register(),    disp_value, disp_reg, src);
 954       } else {
 955                     __ z_ly(dest->as_register(),   disp_value, disp_reg, src);
 956       }
 957       break;
 958     case T_ADDRESS:
 959       if (UseCompressedClassPointers && addr->disp() == oopDesc::klass_offset_in_bytes()) {
 960         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 961         __ decode_klass_not_null(dest->as_register());
 962       } else {
 963         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 964       }
 965       break;
 966     case T_ARRAY : // fall through
 967     case T_OBJECT:
 968     {
 969       if (UseCompressedOops && !wide) {
 970         __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
 971         __ oop_decoder(dest->as_register(), dest->as_register(), true);
 972       } else {
 973         __ z_lg(dest->as_register(), disp_value, disp_reg, src);
 974       }
 975       __ verify_oop(dest->as_register());
 976       break;
 977     }
 978     case T_FLOAT:
 979       if (short_disp) {
 980                     __ z_le(dest->as_float_reg(),  disp_value, disp_reg, src);
 981       } else {
 982                     __ z_ley(dest->as_float_reg(), disp_value, disp_reg, src);
 983       }
 984       break;
 985     case T_DOUBLE:
 986       if (short_disp) {
 987                     __ z_ld(dest->as_double_reg(),  disp_value, disp_reg, src);
 988       } else {
 989                     __ z_ldy(dest->as_double_reg(), disp_value, disp_reg, src);
 990       }
 991       break;
 992     case T_LONG  :  __ z_lg(dest->as_register_lo(), disp_value, disp_reg, src); break;
 993     default      : ShouldNotReachHere();
 994   }
 995 
 996   if (patch != NULL) {
 997     patching_epilog(patch, patch_code, src, info);
 998   }
 999   if (info != NULL) add_debug_info_for_null_check(offset, info);
1000 }
1001 
1002 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1003   assert(src->is_stack(), "should not call otherwise");
1004   assert(dest->is_register(), "should not call otherwise");
1005 
1006   if (dest->is_single_cpu()) {
1007     if (is_reference_type(type)) {
1008       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
1009       __ verify_oop(dest->as_register());
1010     } else if (type == T_METADATA) {
1011       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
1012     } else {
1013       __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), false);
1014     }
1015   } else if (dest->is_double_cpu()) {
1016     Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix());
1017     __ mem2reg_opt(dest->as_register_lo(), src_addr_LO, true);
1018   } else if (dest->is_single_fpu()) {
1019     Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1020     __ mem2freg_opt(dest->as_float_reg(), src_addr, false);
1021   } else if (dest->is_double_fpu()) {
1022     Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1023     __ mem2freg_opt(dest->as_double_reg(), src_addr, true);
1024   } else {
1025     ShouldNotReachHere();
1026   }
1027 }
1028 
1029 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1030   assert(src->is_register(), "should not call otherwise");
1031   assert(dest->is_stack(), "should not call otherwise");
1032 
1033   if (src->is_single_cpu()) {
1034     const Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
1035     if (is_reference_type(type)) {
1036       __ verify_oop(src->as_register());
1037       __ reg2mem_opt(src->as_register(), dst, true);
1038     } else if (type == T_METADATA) {
1039       __ reg2mem_opt(src->as_register(), dst, true);
1040     } else {
1041       __ reg2mem_opt(src->as_register(), dst, false);
1042     }
1043   } else if (src->is_double_cpu()) {
1044     Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix());
1045     __ reg2mem_opt(src->as_register_lo(), dstLO, true);
1046   } else if (src->is_single_fpu()) {
1047     Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1048     __ freg2mem_opt(src->as_float_reg(), dst_addr, false);
1049   } else if (src->is_double_fpu()) {
1050     Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1051     __ freg2mem_opt(src->as_double_reg(), dst_addr, true);
1052   } else {
1053     ShouldNotReachHere();
1054   }
1055 }
1056 
1057 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1058   if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1059     if (from_reg->is_double_fpu()) {
1060       // double to double moves
1061       assert(to_reg->is_double_fpu(), "should match");
1062       __ z_ldr(to_reg->as_double_reg(), from_reg->as_double_reg());
1063     } else {
1064       // float to float moves
1065       assert(to_reg->is_single_fpu(), "should match");
1066       __ z_ler(to_reg->as_float_reg(), from_reg->as_float_reg());
1067     }
1068   } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1069     if (from_reg->is_double_cpu()) {
1070       __ z_lgr(to_reg->as_pointer_register(), from_reg->as_pointer_register());
1071     } else if (to_reg->is_double_cpu()) {
1072       // int to int moves
1073       __ z_lgr(to_reg->as_register_lo(), from_reg->as_register());
1074     } else {
1075       // int to int moves
1076       __ z_lgr(to_reg->as_register(), from_reg->as_register());
1077     }
1078   } else {
1079     ShouldNotReachHere();
1080   }
1081   if (is_reference_type(to_reg->type())) {
1082     __ verify_oop(to_reg->as_register());
1083   }
1084 }
1085 
1086 void LIR_Assembler::reg2mem(LIR_Opr from, LIR_Opr dest_opr, BasicType type,
1087                             LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1088                             bool wide, bool unaligned) {
1089   assert(type != T_METADATA, "store of metadata ptr not supported");
1090   LIR_Address* addr = dest_opr->as_address_ptr();
1091 
1092   Register dest = addr->base()->as_pointer_register();
1093   Register disp_reg = Z_R0;
1094   int disp_value = addr->disp();
1095   bool needs_patching = (patch_code != lir_patch_none);
1096 
1097   if (addr->base()->is_oop_register()) {
1098     __ verify_oop(dest);
1099   }
1100 
1101   PatchingStub* patch = NULL;
1102   if (needs_patching) {
1103     patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1104     assert(!from->is_double_cpu() ||
1105            patch_code == lir_patch_none ||
1106            patch_code == lir_patch_normal, "patching doesn't match register");
1107   }
1108 
1109   assert(!needs_patching || (!Immediate::is_simm20(disp_value) && addr->index()->is_illegal()), "assumption");
1110   if (addr->index()->is_illegal()) {
1111     if (!Immediate::is_simm20(disp_value)) {
1112       if (needs_patching) {
1113         __ load_const(Z_R1_scratch, (intptr_t)0);
1114       } else {
1115         __ load_const_optimized(Z_R1_scratch, disp_value);
1116       }
1117       disp_reg = Z_R1_scratch;
1118       disp_value = 0;
1119     }
1120   } else {
1121     if (!Immediate::is_simm20(disp_value)) {
1122       __ load_const_optimized(Z_R1_scratch, disp_value);
1123       __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
1124       disp_reg = Z_R1_scratch;
1125       disp_value = 0;
1126     }
1127     disp_reg = addr->index()->as_pointer_register();
1128   }
1129 
1130   assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
1131 
1132   if (is_reference_type(type)) {
1133     __ verify_oop(from->as_register());
1134   }
1135 
1136   bool short_disp = Immediate::is_uimm12(disp_value);
1137 
1138   // Remember the offset of the store. The patching_epilog must be done
1139   // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1140   // entered in increasing order.
1141   int offset = code_offset();
1142   switch (type) {
1143     case T_BOOLEAN: // fall through
1144     case T_BYTE  :
1145       if (short_disp) {
1146                     __ z_stc(from->as_register(),  disp_value, disp_reg, dest);
1147       } else {
1148                     __ z_stcy(from->as_register(), disp_value, disp_reg, dest);
1149       }
1150       break;
1151     case T_CHAR  : // fall through
1152     case T_SHORT :
1153       if (short_disp) {
1154                     __ z_sth(from->as_register(),  disp_value, disp_reg, dest);
1155       } else {
1156                     __ z_sthy(from->as_register(), disp_value, disp_reg, dest);
1157       }
1158       break;
1159     case T_INT   :
1160       if (short_disp) {
1161                     __ z_st(from->as_register(),  disp_value, disp_reg, dest);
1162       } else {
1163                     __ z_sty(from->as_register(), disp_value, disp_reg, dest);
1164       }
1165       break;
1166     case T_LONG  :  __ z_stg(from->as_register_lo(), disp_value, disp_reg, dest); break;
1167     case T_ADDRESS: __ z_stg(from->as_register(),    disp_value, disp_reg, dest); break;
1168       break;
1169     case T_ARRAY : // fall through
1170     case T_OBJECT:
1171       {
1172         if (UseCompressedOops && !wide) {
1173           Register compressed_src = Z_R14;
1174           __ oop_encoder(compressed_src, from->as_register(), true, (disp_reg != Z_R1) ? Z_R1 : Z_R0, -1, true);
1175           offset = code_offset();
1176           if (short_disp) {
1177             __ z_st(compressed_src,  disp_value, disp_reg, dest);
1178           } else {
1179             __ z_sty(compressed_src, disp_value, disp_reg, dest);
1180           }
1181         } else {
1182           __ z_stg(from->as_register(), disp_value, disp_reg, dest);
1183         }
1184         break;
1185       }
1186     case T_FLOAT :
1187       if (short_disp) {
1188         __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
1189       } else {
1190         __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
1191       }
1192       break;
1193     case T_DOUBLE:
1194       if (short_disp) {
1195         __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
1196       } else {
1197         __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
1198       }
1199       break;
1200     default: ShouldNotReachHere();
1201   }
1202 
1203   if (patch != NULL) {
1204     patching_epilog(patch, patch_code, dest, info);
1205   }
1206 
1207   if (info != NULL) add_debug_info_for_null_check(offset, info);
1208 }
1209 
1210 
1211 void LIR_Assembler::return_op(LIR_Opr result) {
1212   assert(result->is_illegal() ||
1213          (result->is_single_cpu() && result->as_register() == Z_R2) ||
1214          (result->is_double_cpu() && result->as_register_lo() == Z_R2) ||
1215          (result->is_single_fpu() && result->as_float_reg() == Z_F0) ||
1216          (result->is_double_fpu() && result->as_double_reg() == Z_F0), "convention");
1217 
1218   if (SafepointMechanism::uses_thread_local_poll()) {
1219     __ z_lg(Z_R1_scratch, Address(Z_thread, Thread::polling_page_offset()));
1220   } else {
1221     AddressLiteral pp(os::get_polling_page());
1222     __ load_const_optimized(Z_R1_scratch, pp);
1223   }
1224 
1225   // Pop the frame before the safepoint code.
1226   __ pop_frame_restore_retPC(initial_frame_size_in_bytes());
1227 
1228   if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
1229     __ reserved_stack_check(Z_R14);
1230   }
1231 
1232   // We need to mark the code position where the load from the safepoint
1233   // polling page was emitted as relocInfo::poll_return_type here.
1234   __ relocate(relocInfo::poll_return_type);
1235   __ load_from_polling_page(Z_R1_scratch);
1236 
1237   __ z_br(Z_R14); // Return to caller.
1238 }
1239 
1240 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1241   const Register poll_addr = tmp->as_register_lo();
1242   if (SafepointMechanism::uses_thread_local_poll()) {
1243     __ z_lg(poll_addr, Address(Z_thread, Thread::polling_page_offset()));
1244   } else {
1245     AddressLiteral pp(os::get_polling_page());
1246     __ load_const_optimized(poll_addr, pp);
1247   }
1248   guarantee(info != NULL, "Shouldn't be NULL");
1249   add_debug_info_for_branch(info);
1250   int offset = __ offset();
1251   __ relocate(relocInfo::poll_type);
1252   __ load_from_polling_page(poll_addr);
1253   return offset;
1254 }
1255 
1256 void LIR_Assembler::emit_static_call_stub() {
1257 
1258   // Stub is fixed up when the corresponding call is converted from calling
1259   // compiled code to calling interpreted code.
1260 
1261   address call_pc = __ pc();
1262   address stub = __ start_a_stub(call_stub_size());
1263   if (stub == NULL) {
1264     bailout("static call stub overflow");
1265     return;
1266   }
1267 
1268   int start = __ offset();
1269 
1270   __ relocate(static_stub_Relocation::spec(call_pc));
1271 
1272   // See also Matcher::interpreter_method_oop_reg().
1273   AddressLiteral meta = __ allocate_metadata_address(NULL);
1274   bool success = __ load_const_from_toc(Z_method, meta);
1275 
1276   __ set_inst_mark();
1277   AddressLiteral a((address)-1);
1278   success = success && __ load_const_from_toc(Z_R1, a);
1279   if (!success) {
1280     bailout("const section overflow");
1281     return;
1282   }
1283 
1284   __ z_br(Z_R1);
1285   assert(__ offset() - start <= call_stub_size(), "stub too big");
1286   __ end_a_stub(); // Update current stubs pointer and restore insts_end.
1287 }
1288 
1289 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1290   bool unsigned_comp = condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual;
1291   if (opr1->is_single_cpu()) {
1292     Register reg1 = opr1->as_register();
1293     if (opr2->is_single_cpu()) {
1294       // cpu register - cpu register
1295       if (is_reference_type(opr1->type())) {
1296         __ z_clgr(reg1, opr2->as_register());
1297       } else {
1298         assert(! is_reference_type(opr2->type()), "cmp int, oop?");
1299         if (unsigned_comp) {
1300           __ z_clr(reg1, opr2->as_register());
1301         } else {
1302           __ z_cr(reg1, opr2->as_register());
1303         }
1304       }
1305     } else if (opr2->is_stack()) {
1306       // cpu register - stack
1307       if (is_reference_type(opr1->type())) {
1308         __ z_cg(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1309       } else {
1310         if (unsigned_comp) {
1311           __ z_cly(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1312         } else {
1313           __ z_cy(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1314         }
1315       }
1316     } else if (opr2->is_constant()) {
1317       // cpu register - constant
1318       LIR_Const* c = opr2->as_constant_ptr();
1319       if (c->type() == T_INT) {
1320         if (unsigned_comp) {
1321           __ z_clfi(reg1, c->as_jint());
1322         } else {
1323           __ z_cfi(reg1, c->as_jint());
1324         }
1325       } else if (is_reference_type(c->type())) {
1326         // In 64bit oops are single register.
1327         jobject o = c->as_jobject();
1328         if (o == NULL) {
1329           __ z_ltgr(reg1, reg1);
1330         } else {
1331           jobject2reg(o, Z_R1_scratch);
1332           __ z_cgr(reg1, Z_R1_scratch);
1333         }
1334       } else {
1335         fatal("unexpected type: %s", basictype_to_str(c->type()));
1336       }
1337       // cpu register - address
1338     } else if (opr2->is_address()) {
1339       if (op->info() != NULL) {
1340         add_debug_info_for_null_check_here(op->info());
1341       }
1342       if (unsigned_comp) {
1343         __ z_cly(reg1, as_Address(opr2->as_address_ptr()));
1344       } else {
1345         __ z_cy(reg1, as_Address(opr2->as_address_ptr()));
1346       }
1347     } else {
1348       ShouldNotReachHere();
1349     }
1350 
1351   } else if (opr1->is_double_cpu()) {
1352     assert(!unsigned_comp, "unexpected");
1353     Register xlo = opr1->as_register_lo();
1354     Register xhi = opr1->as_register_hi();
1355     if (opr2->is_double_cpu()) {
1356       __ z_cgr(xlo, opr2->as_register_lo());
1357     } else if (opr2->is_constant()) {
1358       // cpu register - constant 0
1359       assert(opr2->as_jlong() == (jlong)0, "only handles zero");
1360       __ z_ltgr(xlo, xlo);
1361     } else {
1362       ShouldNotReachHere();
1363     }
1364 
1365   } else if (opr1->is_single_fpu()) {
1366     if (opr2->is_single_fpu()) {
1367       __ z_cebr(opr1->as_float_reg(), opr2->as_float_reg());
1368     } else {
1369       // stack slot
1370       Address addr = frame_map()->address_for_slot(opr2->single_stack_ix());
1371       if (Immediate::is_uimm12(addr.disp())) {
1372         __ z_ceb(opr1->as_float_reg(), addr);
1373       } else {
1374         __ z_ley(Z_fscratch_1, addr);
1375         __ z_cebr(opr1->as_float_reg(), Z_fscratch_1);
1376       }
1377     }
1378   } else if (opr1->is_double_fpu()) {
1379     if (opr2->is_double_fpu()) {
1380     __ z_cdbr(opr1->as_double_reg(), opr2->as_double_reg());
1381     } else {
1382       // stack slot
1383       Address addr = frame_map()->address_for_slot(opr2->double_stack_ix());
1384       if (Immediate::is_uimm12(addr.disp())) {
1385         __ z_cdb(opr1->as_double_reg(), addr);
1386       } else {
1387         __ z_ldy(Z_fscratch_1, addr);
1388         __ z_cdbr(opr1->as_double_reg(), Z_fscratch_1);
1389       }
1390     }
1391   } else {
1392     ShouldNotReachHere();
1393   }
1394 }
1395 
1396 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1397   Label    done;
1398   Register dreg = dst->as_register();
1399 
1400   if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1401     assert((left->is_single_fpu() && right->is_single_fpu()) ||
1402            (left->is_double_fpu() && right->is_double_fpu()), "unexpected operand types");
1403     bool is_single = left->is_single_fpu();
1404     bool is_unordered_less = (code == lir_ucmp_fd2i);
1405     FloatRegister lreg = is_single ? left->as_float_reg() : left->as_double_reg();
1406     FloatRegister rreg = is_single ? right->as_float_reg() : right->as_double_reg();
1407     if (is_single) {
1408       __ z_cebr(lreg, rreg);
1409     } else {
1410       __ z_cdbr(lreg, rreg);
1411     }
1412     if (VM_Version::has_LoadStoreConditional()) {
1413       Register one       = Z_R0_scratch;
1414       Register minus_one = Z_R1_scratch;
1415       __ z_lghi(minus_one, -1);
1416       __ z_lghi(one,  1);
1417       __ z_lghi(dreg, 0);
1418       __ z_locgr(dreg, one,       is_unordered_less ? Assembler::bcondHigh            : Assembler::bcondHighOrNotOrdered);
1419       __ z_locgr(dreg, minus_one, is_unordered_less ? Assembler::bcondLowOrNotOrdered : Assembler::bcondLow);
1420     } else {
1421       __ clear_reg(dreg, true, false);
1422       __ z_bre(done); // if (left == right) dst = 0
1423 
1424       // if (left > right || ((code ~= cmpg) && (left <> right)) dst := 1
1425       __ z_lhi(dreg, 1);
1426       __ z_brc(is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered, done);
1427 
1428       // if (left < right || ((code ~= cmpl) && (left <> right)) dst := -1
1429       __ z_lhi(dreg, -1);
1430     }
1431   } else {
1432     assert(code == lir_cmp_l2i, "check");
1433     if (VM_Version::has_LoadStoreConditional()) {
1434       Register one       = Z_R0_scratch;
1435       Register minus_one = Z_R1_scratch;
1436       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1437       __ z_lghi(minus_one, -1);
1438       __ z_lghi(one,  1);
1439       __ z_lghi(dreg, 0);
1440       __ z_locgr(dreg, one, Assembler::bcondHigh);
1441       __ z_locgr(dreg, minus_one, Assembler::bcondLow);
1442     } else {
1443       __ z_cgr(left->as_register_lo(), right->as_register_lo());
1444       __ z_lghi(dreg,  0);     // eq value
1445       __ z_bre(done);
1446       __ z_lghi(dreg,  1);     // gt value
1447       __ z_brh(done);
1448       __ z_lghi(dreg, -1);     // lt value
1449     }
1450   }
1451   __ bind(done);
1452 }
1453 
1454 // result = condition ? opr1 : opr2
1455 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1456   Assembler::branch_condition acond = Assembler::bcondEqual, ncond = Assembler::bcondNotEqual;
1457   switch (condition) {
1458     case lir_cond_equal:        acond = Assembler::bcondEqual;    ncond = Assembler::bcondNotEqual; break;
1459     case lir_cond_notEqual:     acond = Assembler::bcondNotEqual; ncond = Assembler::bcondEqual;    break;
1460     case lir_cond_less:         acond = Assembler::bcondLow;      ncond = Assembler::bcondNotLow;   break;
1461     case lir_cond_lessEqual:    acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1462     case lir_cond_greaterEqual: acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1463     case lir_cond_greater:      acond = Assembler::bcondHigh;     ncond = Assembler::bcondNotHigh;  break;
1464     case lir_cond_belowEqual:   acond = Assembler::bcondNotHigh;  ncond = Assembler::bcondHigh;     break;
1465     case lir_cond_aboveEqual:   acond = Assembler::bcondNotLow;   ncond = Assembler::bcondLow;      break;
1466     default:                    ShouldNotReachHere();
1467   }
1468 
1469   if (opr1->is_cpu_register()) {
1470     reg2reg(opr1, result);
1471   } else if (opr1->is_stack()) {
1472     stack2reg(opr1, result, result->type());
1473   } else if (opr1->is_constant()) {
1474     const2reg(opr1, result, lir_patch_none, NULL);
1475   } else {
1476     ShouldNotReachHere();
1477   }
1478 
1479   if (VM_Version::has_LoadStoreConditional() && !opr2->is_constant()) {
1480     // Optimized version that does not require a branch.
1481     if (opr2->is_single_cpu()) {
1482       assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1483       __ z_locgr(result->as_register(), opr2->as_register(), ncond);
1484     } else if (opr2->is_double_cpu()) {
1485       assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1486       assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1487       __ z_locgr(result->as_register_lo(), opr2->as_register_lo(), ncond);
1488     } else if (opr2->is_single_stack()) {
1489       __ z_loc(result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()), ncond);
1490     } else if (opr2->is_double_stack()) {
1491       __ z_locg(result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix()), ncond);
1492     } else {
1493       ShouldNotReachHere();
1494     }
1495   } else {
1496     Label skip;
1497     __ z_brc(acond, skip);
1498     if (opr2->is_cpu_register()) {
1499       reg2reg(opr2, result);
1500     } else if (opr2->is_stack()) {
1501       stack2reg(opr2, result, result->type());
1502     } else if (opr2->is_constant()) {
1503       const2reg(opr2, result, lir_patch_none, NULL);
1504     } else {
1505       ShouldNotReachHere();
1506     }
1507     __ bind(skip);
1508   }
1509 }
1510 
1511 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1512                              CodeEmitInfo* info, bool pop_fpu_stack) {
1513   assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1514 
1515   if (left->is_single_cpu()) {
1516     assert(left == dest, "left and dest must be equal");
1517     Register lreg = left->as_register();
1518 
1519     if (right->is_single_cpu()) {
1520       // cpu register - cpu register
1521       Register rreg = right->as_register();
1522       switch (code) {
1523         case lir_add: __ z_ar (lreg, rreg); break;
1524         case lir_sub: __ z_sr (lreg, rreg); break;
1525         case lir_mul: __ z_msr(lreg, rreg); break;
1526         default: ShouldNotReachHere();
1527       }
1528 
1529     } else if (right->is_stack()) {
1530       // cpu register - stack
1531       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1532       switch (code) {
1533         case lir_add: __ z_ay(lreg, raddr); break;
1534         case lir_sub: __ z_sy(lreg, raddr); break;
1535         default: ShouldNotReachHere();
1536       }
1537 
1538     } else if (right->is_constant()) {
1539       // cpu register - constant
1540       jint c = right->as_constant_ptr()->as_jint();
1541       switch (code) {
1542         case lir_add: __ z_agfi(lreg, c);  break;
1543         case lir_sub: __ z_agfi(lreg, -c); break; // note: -min_jint == min_jint
1544         case lir_mul: __ z_msfi(lreg, c);  break;
1545         default: ShouldNotReachHere();
1546       }
1547 
1548     } else {
1549       ShouldNotReachHere();
1550     }
1551 
1552   } else if (left->is_double_cpu()) {
1553     assert(left == dest, "left and dest must be equal");
1554     Register lreg_lo = left->as_register_lo();
1555     Register lreg_hi = left->as_register_hi();
1556 
1557     if (right->is_double_cpu()) {
1558       // cpu register - cpu register
1559       Register rreg_lo = right->as_register_lo();
1560       Register rreg_hi = right->as_register_hi();
1561       assert_different_registers(lreg_lo, rreg_lo);
1562       switch (code) {
1563         case lir_add:
1564           __ z_agr(lreg_lo, rreg_lo);
1565           break;
1566         case lir_sub:
1567           __ z_sgr(lreg_lo, rreg_lo);
1568           break;
1569         case lir_mul:
1570           __ z_msgr(lreg_lo, rreg_lo);
1571           break;
1572         default:
1573           ShouldNotReachHere();
1574       }
1575 
1576     } else if (right->is_constant()) {
1577       // cpu register - constant
1578       jlong c = right->as_constant_ptr()->as_jlong_bits();
1579       switch (code) {
1580         case lir_add: __ z_agfi(lreg_lo, c); break;
1581         case lir_sub:
1582           if (c != min_jint) {
1583                       __ z_agfi(lreg_lo, -c);
1584           } else {
1585             // -min_jint cannot be represented as simm32 in z_agfi
1586             // min_jint sign extended:      0xffffffff80000000
1587             // -min_jint as 64 bit integer: 0x0000000080000000
1588             // 0x80000000 can be represented as uimm32 in z_algfi
1589             // lreg_lo := lreg_lo + -min_jint == lreg_lo + 0x80000000
1590                       __ z_algfi(lreg_lo, UCONST64(0x80000000));
1591           }
1592           break;
1593         case lir_mul: __ z_msgfi(lreg_lo, c); break;
1594         default:
1595           ShouldNotReachHere();
1596       }
1597 
1598     } else {
1599       ShouldNotReachHere();
1600     }
1601 
1602   } else if (left->is_single_fpu()) {
1603     assert(left == dest, "left and dest must be equal");
1604     FloatRegister lreg = left->as_float_reg();
1605     FloatRegister rreg = right->is_single_fpu() ? right->as_float_reg() : fnoreg;
1606     Address raddr;
1607 
1608     if (rreg == fnoreg) {
1609       assert(right->is_single_stack(), "constants should be loaded into register");
1610       raddr = frame_map()->address_for_slot(right->single_stack_ix());
1611       if (!Immediate::is_uimm12(raddr.disp())) {
1612         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, false);
1613       }
1614     }
1615 
1616     if (rreg != fnoreg) {
1617       switch (code) {
1618         case lir_add: __ z_aebr(lreg, rreg);  break;
1619         case lir_sub: __ z_sebr(lreg, rreg);  break;
1620         case lir_mul_strictfp: // fall through
1621         case lir_mul: __ z_meebr(lreg, rreg); break;
1622         case lir_div_strictfp: // fall through
1623         case lir_div: __ z_debr(lreg, rreg);  break;
1624         default: ShouldNotReachHere();
1625       }
1626     } else {
1627       switch (code) {
1628         case lir_add: __ z_aeb(lreg, raddr);  break;
1629         case lir_sub: __ z_seb(lreg, raddr);  break;
1630         case lir_mul_strictfp: // fall through
1631         case lir_mul: __ z_meeb(lreg, raddr);  break;
1632         case lir_div_strictfp: // fall through
1633         case lir_div: __ z_deb(lreg, raddr);  break;
1634         default: ShouldNotReachHere();
1635       }
1636     }
1637   } else if (left->is_double_fpu()) {
1638     assert(left == dest, "left and dest must be equal");
1639     FloatRegister lreg = left->as_double_reg();
1640     FloatRegister rreg = right->is_double_fpu() ? right->as_double_reg() : fnoreg;
1641     Address raddr;
1642 
1643     if (rreg == fnoreg) {
1644       assert(right->is_double_stack(), "constants should be loaded into register");
1645       raddr = frame_map()->address_for_slot(right->double_stack_ix());
1646       if (!Immediate::is_uimm12(raddr.disp())) {
1647         __ mem2freg_opt(rreg = Z_fscratch_1, raddr, true);
1648       }
1649     }
1650 
1651     if (rreg != fnoreg) {
1652       switch (code) {
1653         case lir_add: __ z_adbr(lreg, rreg); break;
1654         case lir_sub: __ z_sdbr(lreg, rreg); break;
1655         case lir_mul_strictfp: // fall through
1656         case lir_mul: __ z_mdbr(lreg, rreg); break;
1657         case lir_div_strictfp: // fall through
1658         case lir_div: __ z_ddbr(lreg, rreg); break;
1659         default: ShouldNotReachHere();
1660       }
1661     } else {
1662       switch (code) {
1663         case lir_add: __ z_adb(lreg, raddr); break;
1664         case lir_sub: __ z_sdb(lreg, raddr); break;
1665         case lir_mul_strictfp: // fall through
1666         case lir_mul: __ z_mdb(lreg, raddr); break;
1667         case lir_div_strictfp: // fall through
1668         case lir_div: __ z_ddb(lreg, raddr); break;
1669         default: ShouldNotReachHere();
1670       }
1671     }
1672   } else if (left->is_address()) {
1673     assert(left == dest, "left and dest must be equal");
1674     assert(code == lir_add, "unsupported operation");
1675     assert(right->is_constant(), "unsupported operand");
1676     jint c = right->as_constant_ptr()->as_jint();
1677     LIR_Address* lir_addr = left->as_address_ptr();
1678     Address addr = as_Address(lir_addr);
1679     switch (lir_addr->type()) {
1680       case T_INT:
1681         __ add2mem_32(addr, c, Z_R1_scratch);
1682         break;
1683       case T_LONG:
1684         __ add2mem_64(addr, c, Z_R1_scratch);
1685         break;
1686       default:
1687         ShouldNotReachHere();
1688     }
1689   } else {
1690     ShouldNotReachHere();
1691   }
1692 }
1693 
1694 void LIR_Assembler::fpop() {
1695   // do nothing
1696 }
1697 
1698 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1699   switch (code) {
1700     case lir_sqrt: {
1701       assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1702       FloatRegister src_reg = value->as_double_reg();
1703       FloatRegister dst_reg = dest->as_double_reg();
1704       __ z_sqdbr(dst_reg, src_reg);
1705       break;
1706     }
1707     case lir_abs: {
1708       assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1709       FloatRegister src_reg = value->as_double_reg();
1710       FloatRegister dst_reg = dest->as_double_reg();
1711       __ z_lpdbr(dst_reg, src_reg);
1712       break;
1713     }
1714     default: {
1715       ShouldNotReachHere();
1716       break;
1717     }
1718   }
1719 }
1720 
1721 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1722   if (left->is_single_cpu()) {
1723     Register reg = left->as_register();
1724     if (right->is_constant()) {
1725       int val = right->as_constant_ptr()->as_jint();
1726       switch (code) {
1727         case lir_logic_and: __ z_nilf(reg, val); break;
1728         case lir_logic_or:  __ z_oilf(reg, val); break;
1729         case lir_logic_xor: __ z_xilf(reg, val); break;
1730         default: ShouldNotReachHere();
1731       }
1732     } else if (right->is_stack()) {
1733       Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1734       switch (code) {
1735         case lir_logic_and: __ z_ny(reg, raddr); break;
1736         case lir_logic_or:  __ z_oy(reg, raddr); break;
1737         case lir_logic_xor: __ z_xy(reg, raddr); break;
1738         default: ShouldNotReachHere();
1739       }
1740     } else {
1741       Register rright = right->as_register();
1742       switch (code) {
1743         case lir_logic_and: __ z_nr(reg, rright); break;
1744         case lir_logic_or : __ z_or(reg, rright); break;
1745         case lir_logic_xor: __ z_xr(reg, rright); break;
1746         default: ShouldNotReachHere();
1747       }
1748     }
1749     move_regs(reg, dst->as_register());
1750   } else {
1751     Register l_lo = left->as_register_lo();
1752     if (right->is_constant()) {
1753       __ load_const_optimized(Z_R1_scratch, right->as_constant_ptr()->as_jlong());
1754       switch (code) {
1755         case lir_logic_and:
1756           __ z_ngr(l_lo, Z_R1_scratch);
1757           break;
1758         case lir_logic_or:
1759           __ z_ogr(l_lo, Z_R1_scratch);
1760           break;
1761         case lir_logic_xor:
1762           __ z_xgr(l_lo, Z_R1_scratch);
1763           break;
1764         default: ShouldNotReachHere();
1765       }
1766     } else {
1767       Register r_lo;
1768       if (is_reference_type(right->type())) {
1769         r_lo = right->as_register();
1770       } else {
1771         r_lo = right->as_register_lo();
1772       }
1773       switch (code) {
1774         case lir_logic_and:
1775           __ z_ngr(l_lo, r_lo);
1776           break;
1777         case lir_logic_or:
1778           __ z_ogr(l_lo, r_lo);
1779           break;
1780         case lir_logic_xor:
1781           __ z_xgr(l_lo, r_lo);
1782           break;
1783         default: ShouldNotReachHere();
1784       }
1785     }
1786 
1787     Register dst_lo = dst->as_register_lo();
1788 
1789     move_regs(l_lo, dst_lo);
1790   }
1791 }
1792 
1793 // See operand selection in LIRGenerator::do_ArithmeticOp_Int().
1794 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
1795   if (left->is_double_cpu()) {
1796     // 64 bit integer case
1797     assert(left->is_double_cpu(), "left must be register");
1798     assert(right->is_double_cpu() || is_power_of_2_long(right->as_jlong()),
1799            "right must be register or power of 2 constant");
1800     assert(result->is_double_cpu(), "result must be register");
1801 
1802     Register lreg = left->as_register_lo();
1803     Register dreg = result->as_register_lo();
1804 
1805     if (right->is_constant()) {
1806       // Convert division by a power of two into some shifts and logical operations.
1807       Register treg1 = Z_R0_scratch;
1808       Register treg2 = Z_R1_scratch;
1809       jlong divisor = right->as_jlong();
1810       jlong log_divisor = log2_long(right->as_jlong());
1811 
1812       if (divisor == min_jlong) {
1813         // Min_jlong is special. Result is '0' except for min_jlong/min_jlong = 1.
1814         if (dreg == lreg) {
1815           NearLabel done;
1816           __ load_const_optimized(treg2, min_jlong);
1817           __ z_cgr(lreg, treg2);
1818           __ z_lghi(dreg, 0);           // Preserves condition code.
1819           __ z_brne(done);
1820           __ z_lghi(dreg, 1);           // min_jlong / min_jlong = 1
1821           __ bind(done);
1822         } else {
1823           assert_different_registers(dreg, lreg);
1824           NearLabel done;
1825           __ z_lghi(dreg, 0);
1826           __ compare64_and_branch(lreg, min_jlong, Assembler::bcondNotEqual, done);
1827           __ z_lghi(dreg, 1);
1828           __ bind(done);
1829         }
1830         return;
1831       }
1832       __ move_reg_if_needed(dreg, T_LONG, lreg, T_LONG);
1833       if (divisor == 2) {
1834         __ z_srlg(treg2, dreg, 63);     // dividend < 0 ? 1 : 0
1835       } else {
1836         __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1837         __ and_imm(treg2, divisor - 1, treg1, true);
1838       }
1839       if (code == lir_idiv) {
1840         __ z_agr(dreg, treg2);
1841         __ z_srag(dreg, dreg, log_divisor);
1842       } else {
1843         assert(code == lir_irem, "check");
1844         __ z_agr(treg2, dreg);
1845         __ and_imm(treg2, ~(divisor - 1), treg1, true);
1846         __ z_sgr(dreg, treg2);
1847       }
1848       return;
1849     }
1850 
1851     // Divisor is not a power of 2 constant.
1852     Register rreg = right->as_register_lo();
1853     Register treg = temp->as_register_lo();
1854     assert(right->is_double_cpu(), "right must be register");
1855     assert(lreg == Z_R11, "see ldivInOpr()");
1856     assert(rreg != lreg, "right register must not be same as left register");
1857     assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10) ||
1858            (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see ldivInOpr(), ldivOutOpr(), lremOutOpr()");
1859 
1860     Register R1 = lreg->predecessor();
1861     Register R2 = rreg;
1862     assert(code != lir_idiv || lreg==dreg, "see code below");
1863     if (code == lir_idiv) {
1864       __ z_lcgr(lreg, lreg);
1865     } else {
1866       __ clear_reg(dreg, true, false);
1867     }
1868     NearLabel done;
1869     __ compare64_and_branch(R2, -1, Assembler::bcondEqual, done);
1870     if (code == lir_idiv) {
1871       __ z_lcgr(lreg, lreg); // Revert lcgr above.
1872     }
1873     if (ImplicitDiv0Checks) {
1874       // No debug info because the idiv won't trap.
1875       // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1876       // which is unnecessary, too.
1877       add_debug_info_for_div0(__ offset(), info);
1878     }
1879     __ z_dsgr(R1, R2);
1880     __ bind(done);
1881     return;
1882   }
1883 
1884   // 32 bit integer case
1885 
1886   assert(left->is_single_cpu(), "left must be register");
1887   assert(right->is_single_cpu() || is_power_of_2(right->as_jint()), "right must be register or power of 2 constant");
1888   assert(result->is_single_cpu(), "result must be register");
1889 
1890   Register lreg = left->as_register();
1891   Register dreg = result->as_register();
1892 
1893   if (right->is_constant()) {
1894     // Convert division by a power of two into some shifts and logical operations.
1895     Register treg1 = Z_R0_scratch;
1896     Register treg2 = Z_R1_scratch;
1897     jlong divisor = right->as_jint();
1898     jlong log_divisor = log2_long(right->as_jint());
1899     __ move_reg_if_needed(dreg, T_LONG, lreg, T_INT); // sign extend
1900     if (divisor == 2) {
1901       __ z_srlg(treg2, dreg, 63);     // dividend < 0 ?  1 : 0
1902     } else {
1903       __ z_srag(treg2, dreg, 63);     // dividend < 0 ? -1 : 0
1904       __ and_imm(treg2, divisor - 1, treg1, true);
1905     }
1906     if (code == lir_idiv) {
1907       __ z_agr(dreg, treg2);
1908       __ z_srag(dreg, dreg, log_divisor);
1909     } else {
1910       assert(code == lir_irem, "check");
1911       __ z_agr(treg2, dreg);
1912       __ and_imm(treg2, ~(divisor - 1), treg1, true);
1913       __ z_sgr(dreg, treg2);
1914     }
1915     return;
1916   }
1917 
1918   // Divisor is not a power of 2 constant.
1919   Register rreg = right->as_register();
1920   Register treg = temp->as_register();
1921   assert(right->is_single_cpu(), "right must be register");
1922   assert(lreg == Z_R11, "left register must be rax,");
1923   assert(rreg != lreg, "right register must not be same as left register");
1924   assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10)
1925       || (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see divInOpr(), divOutOpr(), remOutOpr()");
1926 
1927   Register R1 = lreg->predecessor();
1928   Register R2 = rreg;
1929   __ move_reg_if_needed(lreg, T_LONG, lreg, T_INT); // sign extend
1930   if (ImplicitDiv0Checks) {
1931     // No debug info because the idiv won't trap.
1932     // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1933     // which is unnecessary, too.
1934     add_debug_info_for_div0(__ offset(), info);
1935   }
1936   __ z_dsgfr(R1, R2);
1937 }
1938 
1939 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1940   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1941   assert(exceptionPC->as_register() == Z_EXC_PC, "should match");
1942 
1943   // Exception object is not added to oop map by LinearScan
1944   // (LinearScan assumes that no oops are in fixed registers).
1945   info->add_register_oop(exceptionOop);
1946 
1947   // Reuse the debug info from the safepoint poll for the throw op itself.
1948   __ get_PC(Z_EXC_PC);
1949   add_call_info(__ offset(), info); // for exception handler
1950   address stub = Runtime1::entry_for (compilation()->has_fpu_code() ? Runtime1::handle_exception_id
1951                                                                     : Runtime1::handle_exception_nofpu_id);
1952   emit_call_c(stub);
1953 }
1954 
1955 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1956   assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1957 
1958   __ branch_optimized(Assembler::bcondAlways, _unwind_handler_entry);
1959 }
1960 
1961 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1962   ciArrayKlass* default_type = op->expected_type();
1963   Register src = op->src()->as_register();
1964   Register dst = op->dst()->as_register();
1965   Register src_pos = op->src_pos()->as_register();
1966   Register dst_pos = op->dst_pos()->as_register();
1967   Register length  = op->length()->as_register();
1968   Register tmp = op->tmp()->as_register();
1969 
1970   CodeStub* stub = op->stub();
1971   int flags = op->flags();
1972   BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
1973   if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1974 
1975   // If we don't know anything, just go through the generic arraycopy.
1976   if (default_type == NULL) {
1977     address copyfunc_addr = StubRoutines::generic_arraycopy();
1978 
1979     if (copyfunc_addr == NULL) {
1980       // Take a slow path for generic arraycopy.
1981       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
1982       __ bind(*stub->continuation());
1983       return;
1984     }
1985 
1986     // Save outgoing arguments in callee saved registers (C convention) in case
1987     // a call to System.arraycopy is needed.
1988     Register callee_saved_src     = Z_R10;
1989     Register callee_saved_src_pos = Z_R11;
1990     Register callee_saved_dst     = Z_R12;
1991     Register callee_saved_dst_pos = Z_R13;
1992     Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
1993 
1994     __ lgr_if_needed(callee_saved_src, src);
1995     __ lgr_if_needed(callee_saved_src_pos, src_pos);
1996     __ lgr_if_needed(callee_saved_dst, dst);
1997     __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
1998     __ lgr_if_needed(callee_saved_length, length);
1999 
2000     // C function requires 64 bit values.
2001     __ z_lgfr(src_pos, src_pos);
2002     __ z_lgfr(dst_pos, dst_pos);
2003     __ z_lgfr(length, length);
2004 
2005     // Pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint.
2006 
2007     // The arguments are in the corresponding registers.
2008     assert(Z_ARG1 == src,     "assumption");
2009     assert(Z_ARG2 == src_pos, "assumption");
2010     assert(Z_ARG3 == dst,     "assumption");
2011     assert(Z_ARG4 == dst_pos, "assumption");
2012     assert(Z_ARG5 == length,  "assumption");
2013 #ifndef PRODUCT
2014     if (PrintC1Statistics) {
2015       __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_generic_arraycopystub_cnt);
2016       __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2017     }
2018 #endif
2019     emit_call_c(copyfunc_addr);
2020     CHECK_BAILOUT();
2021 
2022     __ compare32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2023 
2024     __ z_lgr(tmp, Z_RET);
2025     __ z_xilf(tmp, -1);
2026 
2027     // Restore values from callee saved registers so they are where the stub
2028     // expects them.
2029     __ lgr_if_needed(src, callee_saved_src);
2030     __ lgr_if_needed(src_pos, callee_saved_src_pos);
2031     __ lgr_if_needed(dst, callee_saved_dst);
2032     __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2033     __ lgr_if_needed(length, callee_saved_length);
2034 
2035     __ z_sr(length, tmp);
2036     __ z_ar(src_pos, tmp);
2037     __ z_ar(dst_pos, tmp);
2038     __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2039 
2040     __ bind(*stub->continuation());
2041     return;
2042   }
2043 
2044   assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
2045 
2046   int elem_size = type2aelembytes(basic_type);
2047   int shift_amount;
2048 
2049   switch (elem_size) {
2050     case 1 :
2051       shift_amount = 0;
2052       break;
2053     case 2 :
2054       shift_amount = 1;
2055       break;
2056     case 4 :
2057       shift_amount = 2;
2058       break;
2059     case 8 :
2060       shift_amount = 3;
2061       break;
2062     default:
2063       shift_amount = -1;
2064       ShouldNotReachHere();
2065   }
2066 
2067   Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
2068   Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
2069   Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
2070   Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
2071 
2072   // Length and pos's are all sign extended at this point on 64bit.
2073 
2074   // test for NULL
2075   if (flags & LIR_OpArrayCopy::src_null_check) {
2076     __ compareU64_and_branch(src, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2077   }
2078   if (flags & LIR_OpArrayCopy::dst_null_check) {
2079     __ compareU64_and_branch(dst, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2080   }
2081 
2082   // Check if negative.
2083   if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2084     __ compare32_and_branch(src_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2085   }
2086   if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2087     __ compare32_and_branch(dst_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2088   }
2089 
2090   // If the compiler was not able to prove that exact type of the source or the destination
2091   // of the arraycopy is an array type, check at runtime if the source or the destination is
2092   // an instance type.
2093   if (flags & LIR_OpArrayCopy::type_check) {
2094     assert(Klass::_lh_neutral_value == 0, "or replace z_lt instructions");
2095 
2096     if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2097       __ load_klass(tmp, dst);
2098       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2099       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2100     }
2101 
2102     if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2103       __ load_klass(tmp, src);
2104       __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2105       __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2106     }
2107   }
2108 
2109   if (flags & LIR_OpArrayCopy::src_range_check) {
2110     __ z_la(tmp, Address(src_pos, length));
2111     __ z_cl(tmp, src_length_addr);
2112     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2113   }
2114   if (flags & LIR_OpArrayCopy::dst_range_check) {
2115     __ z_la(tmp, Address(dst_pos, length));
2116     __ z_cl(tmp, dst_length_addr);
2117     __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2118   }
2119 
2120   if (flags & LIR_OpArrayCopy::length_positive_check) {
2121     __ z_ltr(length, length);
2122     __ branch_optimized(Assembler::bcondNegative, *stub->entry());
2123   }
2124 
2125   // Stubs require 64 bit values.
2126   __ z_lgfr(src_pos, src_pos); // int -> long
2127   __ z_lgfr(dst_pos, dst_pos); // int -> long
2128   __ z_lgfr(length, length);   // int -> long
2129 
2130   if (flags & LIR_OpArrayCopy::type_check) {
2131     // We don't know the array types are compatible.
2132     if (basic_type != T_OBJECT) {
2133       // Simple test for basic type arrays.
2134       if (UseCompressedClassPointers) {
2135         __ z_l(tmp, src_klass_addr);
2136         __ z_c(tmp, dst_klass_addr);
2137       } else {
2138         __ z_lg(tmp, src_klass_addr);
2139         __ z_cg(tmp, dst_klass_addr);
2140       }
2141       __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2142     } else {
2143       // For object arrays, if src is a sub class of dst then we can
2144       // safely do the copy.
2145       NearLabel cont, slow;
2146       Register src_klass = Z_R1_scratch;
2147       Register dst_klass = Z_R10;
2148 
2149       __ load_klass(src_klass, src);
2150       __ load_klass(dst_klass, dst);
2151 
2152       __ check_klass_subtype_fast_path(src_klass, dst_klass, tmp, &cont, &slow, NULL);
2153 
2154       store_parameter(src_klass, 0); // sub
2155       store_parameter(dst_klass, 1); // super
2156       emit_call_c(Runtime1::entry_for (Runtime1::slow_subtype_check_id));
2157       CHECK_BAILOUT2(cont, slow);
2158       // Sets condition code 0 for match (2 otherwise).
2159       __ branch_optimized(Assembler::bcondEqual, cont);
2160 
2161       __ bind(slow);
2162 
2163       address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2164       if (copyfunc_addr != NULL) { // use stub if available
2165         // Src is not a sub class of dst so we have to do a
2166         // per-element check.
2167 
2168         int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2169         if ((flags & mask) != mask) {
2170           // Check that at least both of them object arrays.
2171           assert(flags & mask, "one of the two should be known to be an object array");
2172 
2173           if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2174             __ load_klass(tmp, src);
2175           } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2176             __ load_klass(tmp, dst);
2177           }
2178           Address klass_lh_addr(tmp, Klass::layout_helper_offset());
2179           jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2180           __ load_const_optimized(Z_R1_scratch, objArray_lh);
2181           __ z_c(Z_R1_scratch, klass_lh_addr);
2182           __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2183         }
2184 
2185         // Save outgoing arguments in callee saved registers (C convention) in case
2186         // a call to System.arraycopy is needed.
2187         Register callee_saved_src     = Z_R10;
2188         Register callee_saved_src_pos = Z_R11;
2189         Register callee_saved_dst     = Z_R12;
2190         Register callee_saved_dst_pos = Z_R13;
2191         Register callee_saved_length  = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
2192 
2193         __ lgr_if_needed(callee_saved_src, src);
2194         __ lgr_if_needed(callee_saved_src_pos, src_pos);
2195         __ lgr_if_needed(callee_saved_dst, dst);
2196         __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
2197         __ lgr_if_needed(callee_saved_length, length);
2198 
2199         __ z_llgfr(length, length); // Higher 32bits must be null.
2200 
2201         __ z_sllg(Z_ARG1, src_pos, shift_amount); // index -> byte offset
2202         __ z_sllg(Z_ARG2, dst_pos, shift_amount); // index -> byte offset
2203 
2204         __ z_la(Z_ARG1, Address(src, Z_ARG1, arrayOopDesc::base_offset_in_bytes(basic_type)));
2205         assert_different_registers(Z_ARG1, dst, dst_pos, length);
2206         __ z_la(Z_ARG2, Address(dst, Z_ARG2, arrayOopDesc::base_offset_in_bytes(basic_type)));
2207         assert_different_registers(Z_ARG2, dst, length);
2208 
2209         __ z_lgr(Z_ARG3, length);
2210         assert_different_registers(Z_ARG3, dst);
2211 
2212         __ load_klass(Z_ARG5, dst);
2213         __ z_lg(Z_ARG5, Address(Z_ARG5, ObjArrayKlass::element_klass_offset()));
2214         __ z_lg(Z_ARG4, Address(Z_ARG5, Klass::super_check_offset_offset()));
2215         emit_call_c(copyfunc_addr);
2216         CHECK_BAILOUT2(cont, slow);
2217 
2218 #ifndef PRODUCT
2219         if (PrintC1Statistics) {
2220           NearLabel failed;
2221           __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, failed);
2222           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_cnt);
2223           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2224           __ bind(failed);
2225         }
2226 #endif
2227 
2228         __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2229 
2230 #ifndef PRODUCT
2231         if (PrintC1Statistics) {
2232           __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_attempt_cnt);
2233           __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2234         }
2235 #endif
2236 
2237         __ z_lgr(tmp, Z_RET);
2238         __ z_xilf(tmp, -1);
2239 
2240         // Restore previously spilled arguments
2241         __ lgr_if_needed(src, callee_saved_src);
2242         __ lgr_if_needed(src_pos, callee_saved_src_pos);
2243         __ lgr_if_needed(dst, callee_saved_dst);
2244         __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2245         __ lgr_if_needed(length, callee_saved_length);
2246 
2247         __ z_sr(length, tmp);
2248         __ z_ar(src_pos, tmp);
2249         __ z_ar(dst_pos, tmp);
2250       }
2251 
2252       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2253 
2254       __ bind(cont);
2255     }
2256   }
2257 
2258 #ifdef ASSERT
2259   if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2260     // Sanity check the known type with the incoming class. For the
2261     // primitive case the types must match exactly with src.klass and
2262     // dst.klass each exactly matching the default type. For the
2263     // object array case, if no type check is needed then either the
2264     // dst type is exactly the expected type and the src type is a
2265     // subtype which we can't check or src is the same array as dst
2266     // but not necessarily exactly of type default_type.
2267     NearLabel known_ok, halt;
2268     metadata2reg(default_type->constant_encoding(), tmp);
2269     if (UseCompressedClassPointers) {
2270       __ encode_klass_not_null(tmp);
2271     }
2272 
2273     if (basic_type != T_OBJECT) {
2274       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2275       else                                    { __ z_cg(tmp, dst_klass_addr); }
2276       __ branch_optimized(Assembler::bcondNotEqual, halt);
2277       if (UseCompressedClassPointers)         { __ z_c (tmp, src_klass_addr); }
2278       else                                    { __ z_cg(tmp, src_klass_addr); }
2279       __ branch_optimized(Assembler::bcondEqual, known_ok);
2280     } else {
2281       if (UseCompressedClassPointers)         { __ z_c (tmp, dst_klass_addr); }
2282       else                                    { __ z_cg(tmp, dst_klass_addr); }
2283       __ branch_optimized(Assembler::bcondEqual, known_ok);
2284       __ compareU64_and_branch(src, dst, Assembler::bcondEqual, known_ok);
2285     }
2286     __ bind(halt);
2287     __ stop("incorrect type information in arraycopy");
2288     __ bind(known_ok);
2289   }
2290 #endif
2291 
2292 #ifndef PRODUCT
2293   if (PrintC1Statistics) {
2294     __ load_const_optimized(Z_R1_scratch, Runtime1::arraycopy_count_address(basic_type));
2295     __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2296   }
2297 #endif
2298 
2299   __ z_sllg(tmp, src_pos, shift_amount); // index -> byte offset
2300   __ z_sllg(Z_R1_scratch, dst_pos, shift_amount); // index -> byte offset
2301 
2302   assert_different_registers(Z_ARG1, dst, dst_pos, length);
2303   __ z_la(Z_ARG1, Address(src, tmp, arrayOopDesc::base_offset_in_bytes(basic_type)));
2304   assert_different_registers(Z_ARG2, length);
2305   __ z_la(Z_ARG2, Address(dst, Z_R1_scratch, arrayOopDesc::base_offset_in_bytes(basic_type)));
2306   __ lgr_if_needed(Z_ARG3, length);
2307 
2308   bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2309   bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2310   const char *name;
2311   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2312   __ call_VM_leaf(entry);
2313 
2314   __ bind(*stub->continuation());
2315 }
2316 
2317 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2318   if (dest->is_single_cpu()) {
2319     if (left->type() == T_OBJECT) {
2320       switch (code) {
2321         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2322         case lir_shr:  __ z_srag (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2323         case lir_ushr: __ z_srlg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2324         default: ShouldNotReachHere();
2325       }
2326     } else {
2327       assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2328       Register masked_count = Z_R1_scratch;
2329       __ z_lr(masked_count, count->as_register());
2330       __ z_nill(masked_count, 31);
2331       switch (code) {
2332         case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), 0, masked_count); break;
2333         case lir_shr:  __ z_sra  (dest->as_register(), 0, masked_count); break;
2334         case lir_ushr: __ z_srl  (dest->as_register(), 0, masked_count); break;
2335         default: ShouldNotReachHere();
2336       }
2337     }
2338   } else {
2339     switch (code) {
2340       case lir_shl:  __ z_sllg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2341       case lir_shr:  __ z_srag (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2342       case lir_ushr: __ z_srlg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2343       default: ShouldNotReachHere();
2344     }
2345   }
2346 }
2347 
2348 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2349   if (left->type() == T_OBJECT) {
2350     count = count & 63;  // Shouldn't shift by more than sizeof(intptr_t).
2351     Register l = left->as_register();
2352     Register d = dest->as_register_lo();
2353     switch (code) {
2354       case lir_shl:  __ z_sllg (d, l, count); break;
2355       case lir_shr:  __ z_srag (d, l, count); break;
2356       case lir_ushr: __ z_srlg (d, l, count); break;
2357       default: ShouldNotReachHere();
2358     }
2359     return;
2360   }
2361   if (dest->is_single_cpu()) {
2362     assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2363     count = count & 0x1F; // Java spec
2364     switch (code) {
2365       case lir_shl:  __ z_sllg (dest->as_register(), left->as_register(), count); break;
2366       case lir_shr:  __ z_sra  (dest->as_register(), count); break;
2367       case lir_ushr: __ z_srl  (dest->as_register(), count); break;
2368       default: ShouldNotReachHere();
2369     }
2370   } else if (dest->is_double_cpu()) {
2371     count = count & 63; // Java spec
2372     Register l = left->as_pointer_register();
2373     Register d = dest->as_pointer_register();
2374     switch (code) {
2375       case lir_shl:  __ z_sllg (d, l, count); break;
2376       case lir_shr:  __ z_srag (d, l, count); break;
2377       case lir_ushr: __ z_srlg (d, l, count); break;
2378       default: ShouldNotReachHere();
2379     }
2380   } else {
2381     ShouldNotReachHere();
2382   }
2383 }
2384 
2385 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2386   if (op->init_check()) {
2387     // Make sure klass is initialized & doesn't have finalizer.
2388     const int state_offset = in_bytes(InstanceKlass::init_state_offset());
2389     Register iklass = op->klass()->as_register();
2390     add_debug_info_for_null_check_here(op->stub()->info());
2391     if (Immediate::is_uimm12(state_offset)) {
2392       __ z_cli(state_offset, iklass, InstanceKlass::fully_initialized);
2393     } else {
2394       __ z_cliy(state_offset, iklass, InstanceKlass::fully_initialized);
2395     }
2396     __ branch_optimized(Assembler::bcondNotEqual, *op->stub()->entry()); // Use long branch, because slow_case might be far.
2397   }
2398   __ allocate_object(op->obj()->as_register(),
2399                      op->tmp1()->as_register(),
2400                      op->tmp2()->as_register(),
2401                      op->header_size(),
2402                      op->object_size(),
2403                      op->klass()->as_register(),
2404                      *op->stub()->entry());
2405   __ bind(*op->stub()->continuation());
2406   __ verify_oop(op->obj()->as_register());
2407 }
2408 
2409 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2410   Register len = op->len()->as_register();
2411   __ move_reg_if_needed(len, T_LONG, len, T_INT); // sign extend
2412 
2413   if (UseSlowPath ||
2414       (!UseFastNewObjectArray && (is_reference_type(op->type()))) ||
2415       (!UseFastNewTypeArray   && (!is_reference_type(op->type())))) {
2416     __ z_brul(*op->stub()->entry());
2417   } else {
2418     __ allocate_array(op->obj()->as_register(),
2419                       op->len()->as_register(),
2420                       op->tmp1()->as_register(),
2421                       op->tmp2()->as_register(),
2422                       arrayOopDesc::header_size(op->type()),
2423                       type2aelembytes(op->type()),
2424                       op->klass()->as_register(),
2425                       *op->stub()->entry());
2426   }
2427   __ bind(*op->stub()->continuation());
2428 }
2429 
2430 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
2431                                         Register recv, Register tmp1, Label* update_done) {
2432   uint i;
2433   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2434     Label next_test;
2435     // See if the receiver is receiver[n].
2436     Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2437     __ z_cg(recv, receiver_addr);
2438     __ z_brne(next_test);
2439     Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
2440     __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2441     __ branch_optimized(Assembler::bcondAlways, *update_done);
2442     __ bind(next_test);
2443   }
2444 
2445   // Didn't find receiver; find next empty slot and fill it in.
2446   for (i = 0; i < VirtualCallData::row_limit(); i++) {
2447     Label next_test;
2448     Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2449     __ z_ltg(Z_R0_scratch, recv_addr);
2450     __ z_brne(next_test);
2451     __ z_stg(recv, recv_addr);
2452     __ load_const_optimized(tmp1, DataLayout::counter_increment);
2453     __ z_stg(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)), mdo);
2454     __ branch_optimized(Assembler::bcondAlways, *update_done);
2455     __ bind(next_test);
2456   }
2457 }
2458 
2459 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2460                                     ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2461   Unimplemented();
2462 }
2463 
2464 void LIR_Assembler::store_parameter(Register r, int param_num) {
2465   assert(param_num >= 0, "invalid num");
2466   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2467   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2468   __ z_stg(r, offset_in_bytes, Z_SP);
2469 }
2470 
2471 void LIR_Assembler::store_parameter(jint c, int param_num) {
2472   assert(param_num >= 0, "invalid num");
2473   int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2474   assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2475   __ store_const(Address(Z_SP, offset_in_bytes), c, Z_R1_scratch, true);
2476 }
2477 
2478 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2479   // We always need a stub for the failure case.
2480   CodeStub* stub = op->stub();
2481   Register obj = op->object()->as_register();
2482   Register k_RInfo = op->tmp1()->as_register();
2483   Register klass_RInfo = op->tmp2()->as_register();
2484   Register dst = op->result_opr()->as_register();
2485   Register Rtmp1 = Z_R1_scratch;
2486   ciKlass* k = op->klass();
2487 
2488   assert(!op->tmp3()->is_valid(), "tmp3's not needed");
2489 
2490   // Check if it needs to be profiled.
2491   ciMethodData* md = NULL;
2492   ciProfileData* data = NULL;
2493 
2494   if (op->should_profile()) {
2495     ciMethod* method = op->profiled_method();
2496     assert(method != NULL, "Should have method");
2497     int bci = op->profiled_bci();
2498     md = method->method_data_or_null();
2499     assert(md != NULL, "Sanity");
2500     data = md->bci_to_data(bci);
2501     assert(data != NULL,                "need data for type check");
2502     assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2503   }
2504 
2505   // Temp operands do not overlap with inputs, if this is their last
2506   // use (end of range is exclusive), so a register conflict is possible.
2507   if (obj == k_RInfo) {
2508     k_RInfo = dst;
2509   } else if (obj == klass_RInfo) {
2510     klass_RInfo = dst;
2511   }
2512   assert_different_registers(obj, k_RInfo, klass_RInfo);
2513 
2514   if (op->should_profile()) {
2515     NearLabel not_null;
2516     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2517     // Object is null; update MDO and exit.
2518     Register mdo = klass_RInfo;
2519     metadata2reg(md->constant_encoding(), mdo);
2520     Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2521     int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2522     __ or2mem_8(data_addr, header_bits);
2523     __ branch_optimized(Assembler::bcondAlways, *obj_is_null);
2524     __ bind(not_null);
2525   } else {
2526     __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondEqual, *obj_is_null);
2527   }
2528 
2529   NearLabel profile_cast_failure, profile_cast_success;
2530   Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2531   Label *success_target = op->should_profile() ? &profile_cast_success : success;
2532 
2533   // Patching may screw with our temporaries on sparc,
2534   // so let's do it before loading the class.
2535   if (k->is_loaded()) {
2536     metadata2reg(k->constant_encoding(), k_RInfo);
2537   } else {
2538     klass2reg_with_patching(k_RInfo, op->info_for_patch());
2539   }
2540   assert(obj != k_RInfo, "must be different");
2541 
2542   __ verify_oop(obj);
2543 
2544   // Get object class.
2545   // Not a safepoint as obj null check happens earlier.
2546   if (op->fast_check()) {
2547     if (UseCompressedClassPointers) {
2548       __ load_klass(klass_RInfo, obj);
2549       __ compareU64_and_branch(k_RInfo, klass_RInfo, Assembler::bcondNotEqual, *failure_target);
2550     } else {
2551       __ z_cg(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
2552       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2553     }
2554     // Successful cast, fall through to profile or jump.
2555   } else {
2556     bool need_slow_path = !k->is_loaded() ||
2557                           ((int) k->super_check_offset() == in_bytes(Klass::secondary_super_cache_offset()));
2558     intptr_t super_check_offset = k->is_loaded() ? k->super_check_offset() : -1L;
2559     __ load_klass(klass_RInfo, obj);
2560     // Perform the fast part of the checking logic.
2561     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1,
2562                                      (need_slow_path ? success_target : NULL),
2563                                      failure_target, NULL,
2564                                      RegisterOrConstant(super_check_offset));
2565     if (need_slow_path) {
2566       // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2567       address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2568       store_parameter(klass_RInfo, 0); // sub
2569       store_parameter(k_RInfo, 1);     // super
2570       emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2571       CHECK_BAILOUT2(profile_cast_failure, profile_cast_success);
2572       __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2573       // Fall through to success case.
2574     }
2575   }
2576 
2577   if (op->should_profile()) {
2578     Register mdo = klass_RInfo, recv = k_RInfo;
2579     assert_different_registers(obj, mdo, recv);
2580     __ bind(profile_cast_success);
2581     metadata2reg(md->constant_encoding(), mdo);
2582     __ load_klass(recv, obj);
2583     type_profile_helper(mdo, md, data, recv, Rtmp1, success);
2584     __ branch_optimized(Assembler::bcondAlways, *success);
2585 
2586     __ bind(profile_cast_failure);
2587     metadata2reg(md->constant_encoding(), mdo);
2588     __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2589     __ branch_optimized(Assembler::bcondAlways, *failure);
2590   } else {
2591     __ branch_optimized(Assembler::bcondAlways, *success);
2592   }
2593 }
2594 
2595 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2596   LIR_Code code = op->code();
2597   if (code == lir_store_check) {
2598     Register value = op->object()->as_register();
2599     Register array = op->array()->as_register();
2600     Register k_RInfo = op->tmp1()->as_register();
2601     Register klass_RInfo = op->tmp2()->as_register();
2602     Register Rtmp1 = Z_R1_scratch;
2603 
2604     CodeStub* stub = op->stub();
2605 
2606     // Check if it needs to be profiled.
2607     ciMethodData* md = NULL;
2608     ciProfileData* data = NULL;
2609 
2610     assert_different_registers(value, k_RInfo, klass_RInfo);
2611 
2612     if (op->should_profile()) {
2613       ciMethod* method = op->profiled_method();
2614       assert(method != NULL, "Should have method");
2615       int bci = op->profiled_bci();
2616       md = method->method_data_or_null();
2617       assert(md != NULL, "Sanity");
2618       data = md->bci_to_data(bci);
2619       assert(data != NULL,                "need data for type check");
2620       assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2621     }
2622     NearLabel profile_cast_success, profile_cast_failure, done;
2623     Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2624     Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2625 
2626     if (op->should_profile()) {
2627       NearLabel not_null;
2628       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2629       // Object is null; update MDO and exit.
2630       Register mdo = klass_RInfo;
2631       metadata2reg(md->constant_encoding(), mdo);
2632       Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2633       int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2634       __ or2mem_8(data_addr, header_bits);
2635       __ branch_optimized(Assembler::bcondAlways, done);
2636       __ bind(not_null);
2637     } else {
2638       __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondEqual, done);
2639     }
2640 
2641     add_debug_info_for_null_check_here(op->info_for_exception());
2642     __ load_klass(k_RInfo, array);
2643     __ load_klass(klass_RInfo, value);
2644 
2645     // Get instance klass (it's already uncompressed).
2646     __ z_lg(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
2647     // Perform the fast part of the checking logic.
2648     __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
2649     // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2650     address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2651     store_parameter(klass_RInfo, 0); // sub
2652     store_parameter(k_RInfo, 1);     // super
2653     emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2654     CHECK_BAILOUT3(profile_cast_success, profile_cast_failure, done);
2655     __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2656     // Fall through to success case.
2657 
2658     if (op->should_profile()) {
2659       Register mdo = klass_RInfo, recv = k_RInfo;
2660       assert_different_registers(value, mdo, recv);
2661       __ bind(profile_cast_success);
2662       metadata2reg(md->constant_encoding(), mdo);
2663       __ load_klass(recv, value);
2664       type_profile_helper(mdo, md, data, recv, Rtmp1, &done);
2665       __ branch_optimized(Assembler::bcondAlways, done);
2666 
2667       __ bind(profile_cast_failure);
2668       metadata2reg(md->constant_encoding(), mdo);
2669       __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2670       __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2671     }
2672 
2673     __ bind(done);
2674   } else {
2675     if (code == lir_checkcast) {
2676       Register obj = op->object()->as_register();
2677       Register dst = op->result_opr()->as_register();
2678       NearLabel success;
2679       emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2680       __ bind(success);
2681       __ lgr_if_needed(dst, obj);
2682     } else {
2683       if (code == lir_instanceof) {
2684         Register obj = op->object()->as_register();
2685         Register dst = op->result_opr()->as_register();
2686         NearLabel success, failure, done;
2687         emit_typecheck_helper(op, &success, &failure, &failure);
2688         __ bind(failure);
2689         __ clear_reg(dst);
2690         __ branch_optimized(Assembler::bcondAlways, done);
2691         __ bind(success);
2692         __ load_const_optimized(dst, 1);
2693         __ bind(done);
2694       } else {
2695         ShouldNotReachHere();
2696       }
2697     }
2698   }
2699 }
2700 
2701 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2702   Register addr = op->addr()->as_pointer_register();
2703   Register t1_cmp = Z_R1_scratch;
2704   if (op->code() == lir_cas_long) {
2705     assert(VM_Version::supports_cx8(), "wrong machine");
2706     Register cmp_value_lo = op->cmp_value()->as_register_lo();
2707     Register new_value_lo = op->new_value()->as_register_lo();
2708     __ z_lgr(t1_cmp, cmp_value_lo);
2709     // Perform the compare and swap operation.
2710     __ z_csg(t1_cmp, new_value_lo, 0, addr);
2711   } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2712     Register cmp_value = op->cmp_value()->as_register();
2713     Register new_value = op->new_value()->as_register();
2714     if (op->code() == lir_cas_obj) {
2715       if (UseCompressedOops) {
2716                  t1_cmp = op->tmp1()->as_register();
2717         Register t2_new = op->tmp2()->as_register();
2718         assert_different_registers(cmp_value, new_value, addr, t1_cmp, t2_new);
2719         __ oop_encoder(t1_cmp, cmp_value, true /*maybe null*/);
2720         __ oop_encoder(t2_new, new_value, true /*maybe null*/);
2721         __ z_cs(t1_cmp, t2_new, 0, addr);
2722       } else {
2723         __ z_lgr(t1_cmp, cmp_value);
2724         __ z_csg(t1_cmp, new_value, 0, addr);
2725       }
2726     } else {
2727       __ z_lr(t1_cmp, cmp_value);
2728       __ z_cs(t1_cmp, new_value, 0, addr);
2729     }
2730   } else {
2731     ShouldNotReachHere(); // new lir_cas_??
2732   }
2733 }
2734 
2735 void LIR_Assembler::set_24bit_FPU() {
2736   ShouldNotCallThis(); // x86 only
2737 }
2738 
2739 void LIR_Assembler::reset_FPU() {
2740   ShouldNotCallThis(); // x86 only
2741 }
2742 
2743 void LIR_Assembler::breakpoint() {
2744   Unimplemented();
2745   //  __ breakpoint_trap();
2746 }
2747 
2748 void LIR_Assembler::push(LIR_Opr opr) {
2749   ShouldNotCallThis(); // unused
2750 }
2751 
2752 void LIR_Assembler::pop(LIR_Opr opr) {
2753   ShouldNotCallThis(); // unused
2754 }
2755 
2756 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2757   Address addr = frame_map()->address_for_monitor_lock(monitor_no);
2758   __ add2reg(dst_opr->as_register(), addr.disp(), addr.base());
2759 }
2760 
2761 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2762   Register obj = op->obj_opr()->as_register();  // May not be an oop.
2763   Register hdr = op->hdr_opr()->as_register();
2764   Register lock = op->lock_opr()->as_register();
2765   if (!UseFastLocking) {
2766     __ branch_optimized(Assembler::bcondAlways, *op->stub()->entry());
2767   } else if (op->code() == lir_lock) {
2768     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2769     // Add debug info for NullPointerException only if one is possible.
2770     if (op->info() != NULL) {
2771       add_debug_info_for_null_check_here(op->info());
2772     }
2773     __ lock_object(hdr, obj, lock, *op->stub()->entry());
2774     // done
2775   } else if (op->code() == lir_unlock) {
2776     assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2777     __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2778   } else {
2779     ShouldNotReachHere();
2780   }
2781   __ bind(*op->stub()->continuation());
2782 }
2783 
2784 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2785   ciMethod* method = op->profiled_method();
2786   int bci          = op->profiled_bci();
2787   ciMethod* callee = op->profiled_callee();
2788 
2789   // Update counter for all call types.
2790   ciMethodData* md = method->method_data_or_null();
2791   assert(md != NULL, "Sanity");
2792   ciProfileData* data = md->bci_to_data(bci);
2793   assert(data != NULL && data->is_CounterData(), "need CounterData for calls");
2794   assert(op->mdo()->is_single_cpu(),  "mdo must be allocated");
2795   Register mdo  = op->mdo()->as_register();
2796   assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2797   Register tmp1 = op->tmp1()->as_register_lo();
2798   metadata2reg(md->constant_encoding(), mdo);
2799 
2800   Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2801   // Perform additional virtual call profiling for invokevirtual and
2802   // invokeinterface bytecodes
2803   if (op->should_profile_receiver_type()) {
2804     assert(op->recv()->is_single_cpu(), "recv must be allocated");
2805     Register recv = op->recv()->as_register();
2806     assert_different_registers(mdo, tmp1, recv);
2807     assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2808     ciKlass* known_klass = op->known_holder();
2809     if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2810       // We know the type that will be seen at this call site; we can
2811       // statically update the MethodData* rather than needing to do
2812       // dynamic tests on the receiver type.
2813 
2814       // NOTE: we should probably put a lock around this search to
2815       // avoid collisions by concurrent compilations.
2816       ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2817       uint i;
2818       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2819         ciKlass* receiver = vc_data->receiver(i);
2820         if (known_klass->equals(receiver)) {
2821           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2822           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2823           return;
2824         }
2825       }
2826 
2827       // Receiver type not found in profile data. Select an empty slot.
2828 
2829       // Note that this is less efficient than it should be because it
2830       // always does a write to the receiver part of the
2831       // VirtualCallData rather than just the first time.
2832       for (i = 0; i < VirtualCallData::row_limit(); i++) {
2833         ciKlass* receiver = vc_data->receiver(i);
2834         if (receiver == NULL) {
2835           Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
2836           metadata2reg(known_klass->constant_encoding(), tmp1);
2837           __ z_stg(tmp1, recv_addr);
2838           Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2839           __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2840           return;
2841         }
2842       }
2843     } else {
2844       __ load_klass(recv, recv);
2845       NearLabel update_done;
2846       type_profile_helper(mdo, md, data, recv, tmp1, &update_done);
2847       // Receiver did not match any saved receiver and there is no empty row for it.
2848       // Increment total counter to indicate polymorphic case.
2849       __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2850       __ bind(update_done);
2851     }
2852   } else {
2853     // static call
2854     __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2855   }
2856 }
2857 
2858 void LIR_Assembler::align_backward_branch_target() {
2859   __ align(OptoLoopAlignment);
2860 }
2861 
2862 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2863   ShouldNotCallThis(); // There are no delay slots on ZARCH_64.
2864 }
2865 
2866 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2867   // tmp must be unused
2868   assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2869   assert(left->is_register(), "can only handle registers");
2870 
2871   if (left->is_single_cpu()) {
2872     __ z_lcr(dest->as_register(), left->as_register());
2873   } else if (left->is_single_fpu()) {
2874     __ z_lcebr(dest->as_float_reg(), left->as_float_reg());
2875   } else if (left->is_double_fpu()) {
2876     __ z_lcdbr(dest->as_double_reg(), left->as_double_reg());
2877   } else {
2878     assert(left->is_double_cpu(), "Must be a long");
2879     __ z_lcgr(dest->as_register_lo(), left->as_register_lo());
2880   }
2881 }
2882 
2883 void LIR_Assembler::fxch(int i) {
2884   ShouldNotCallThis(); // x86 only
2885 }
2886 
2887 void LIR_Assembler::fld(int i) {
2888   ShouldNotCallThis(); // x86 only
2889 }
2890 
2891 void LIR_Assembler::ffree(int i) {
2892   ShouldNotCallThis(); // x86 only
2893 }
2894 
2895 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2896                             const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2897   assert(!tmp->is_valid(), "don't need temporary");
2898   emit_call_c(dest);
2899   CHECK_BAILOUT();
2900   if (info != NULL) {
2901     add_call_info_here(info);
2902   }
2903 }
2904 
2905 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2906   ShouldNotCallThis(); // not needed on ZARCH_64
2907 }
2908 
2909 void LIR_Assembler::membar() {
2910   __ z_fence();
2911 }
2912 
2913 void LIR_Assembler::membar_acquire() {
2914   __ z_acquire();
2915 }
2916 
2917 void LIR_Assembler::membar_release() {
2918   __ z_release();
2919 }
2920 
2921 void LIR_Assembler::membar_loadload() {
2922   __ z_acquire();
2923 }
2924 
2925 void LIR_Assembler::membar_storestore() {
2926   __ z_release();
2927 }
2928 
2929 void LIR_Assembler::membar_loadstore() {
2930   __ z_acquire();
2931 }
2932 
2933 void LIR_Assembler::membar_storeload() {
2934   __ z_fence();
2935 }
2936 
2937 void LIR_Assembler::on_spin_wait() {
2938   Unimplemented();
2939 }
2940 
2941 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2942   assert(patch_code == lir_patch_none, "Patch code not supported");
2943   LIR_Address* addr = addr_opr->as_address_ptr();
2944   assert(addr->scale() == LIR_Address::times_1, "scaling unsupported");
2945   __ load_address(dest->as_pointer_register(), as_Address(addr));
2946 }
2947 
2948 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2949   ShouldNotCallThis(); // unused
2950 }
2951 
2952 #ifdef ASSERT
2953 // Emit run-time assertion.
2954 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2955   Unimplemented();
2956 }
2957 #endif
2958 
2959 void LIR_Assembler::peephole(LIR_List*) {
2960   // Do nothing for now.
2961 }
2962 
2963 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2964   assert(code == lir_xadd, "lir_xchg not supported");
2965   Address src_addr = as_Address(src->as_address_ptr());
2966   Register base = src_addr.base();
2967   intptr_t disp = src_addr.disp();
2968   if (src_addr.index()->is_valid()) {
2969     // LAA and LAAG do not support index register.
2970     __ load_address(Z_R1_scratch, src_addr);
2971     base = Z_R1_scratch;
2972     disp = 0;
2973   }
2974   if (data->type() == T_INT) {
2975     __ z_laa(dest->as_register(), data->as_register(), disp, base);
2976   } else if (data->type() == T_LONG) {
2977     assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
2978     __ z_laag(dest->as_register_lo(), data->as_register_lo(), disp, base);
2979   } else {
2980     ShouldNotReachHere();
2981   }
2982 }
2983 
2984 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2985   Register obj = op->obj()->as_register();
2986   Register tmp1 = op->tmp()->as_pointer_register();
2987   Register tmp2 = Z_R1_scratch;
2988   Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2989   ciKlass* exact_klass = op->exact_klass();
2990   intptr_t current_klass = op->current_klass();
2991   bool not_null = op->not_null();
2992   bool no_conflict = op->no_conflict();
2993 
2994   Label update, next, none, null_seen, init_klass;
2995 
2996   bool do_null = !not_null;
2997   bool exact_klass_set = exact_klass != NULL && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2998   bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2999 
3000   assert(do_null || do_update, "why are we here?");
3001   assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
3002 
3003   __ verify_oop(obj);
3004 
3005   if (do_null || tmp1 != obj DEBUG_ONLY(|| true)) {
3006     __ z_ltgr(tmp1, obj);
3007   }
3008   if (do_null) {
3009     __ z_brnz(update);
3010     if (!TypeEntries::was_null_seen(current_klass)) {
3011       __ z_lg(tmp1, mdo_addr);
3012       __ z_oill(tmp1, TypeEntries::null_seen);
3013       __ z_stg(tmp1, mdo_addr);
3014     }
3015     if (do_update) {
3016       __ z_bru(next);
3017     }
3018   } else {
3019     __ asm_assert_ne("unexpect null obj", __LINE__);
3020   }
3021 
3022   __ bind(update);
3023 
3024   if (do_update) {
3025 #ifdef ASSERT
3026     if (exact_klass != NULL) {
3027       __ load_klass(tmp1, tmp1);
3028       metadata2reg(exact_klass->constant_encoding(), tmp2);
3029       __ z_cgr(tmp1, tmp2);
3030       __ asm_assert_eq("exact klass and actual klass differ", __LINE__);
3031     }
3032 #endif
3033 
3034     Label do_update;
3035     __ z_lg(tmp2, mdo_addr);
3036 
3037     if (!no_conflict) {
3038       if (exact_klass == NULL || TypeEntries::is_type_none(current_klass)) {
3039         if (exact_klass != NULL) {
3040           metadata2reg(exact_klass->constant_encoding(), tmp1);
3041         } else {
3042           __ load_klass(tmp1, tmp1);
3043         }
3044 
3045         // Klass seen before: nothing to do (regardless of unknown bit).
3046         __ z_lgr(Z_R0_scratch, tmp2);
3047         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3048         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3049         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3050 
3051         // Already unknown: Nothing to do anymore.
3052         __ z_tmll(tmp2, TypeEntries::type_unknown);
3053         __ z_brc(Assembler::bcondAllOne, next);
3054 
3055         if (TypeEntries::is_type_none(current_klass)) {
3056           __ z_lgr(Z_R0_scratch, tmp2);
3057           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3058           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3059           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, init_klass);
3060         }
3061       } else {
3062         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3063                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3064 
3065         // Already unknown: Nothing to do anymore.
3066         __ z_tmll(tmp2, TypeEntries::type_unknown);
3067         __ z_brc(Assembler::bcondAllOne, next);
3068       }
3069 
3070       // Different than before. Cannot keep accurate profile.
3071       __ z_oill(tmp2, TypeEntries::type_unknown);
3072       __ z_bru(do_update);
3073     } else {
3074       // There's a single possible klass at this profile point.
3075       assert(exact_klass != NULL, "should be");
3076       if (TypeEntries::is_type_none(current_klass)) {
3077         metadata2reg(exact_klass->constant_encoding(), tmp1);
3078         __ z_lgr(Z_R0_scratch, tmp2);
3079         assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3080         __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3081         __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3082 #ifdef ASSERT
3083         {
3084           Label ok;
3085           __ z_lgr(Z_R0_scratch, tmp2);
3086           assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3087           __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3088           __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, ok);
3089           __ stop("unexpected profiling mismatch");
3090           __ bind(ok);
3091         }
3092 #endif
3093 
3094       } else {
3095         assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3096                ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3097 
3098         // Already unknown: Nothing to do anymore.
3099         __ z_tmll(tmp2, TypeEntries::type_unknown);
3100         __ z_brc(Assembler::bcondAllOne, next);
3101         __ z_oill(tmp2, TypeEntries::type_unknown);
3102         __ z_bru(do_update);
3103       }
3104     }
3105 
3106     __ bind(init_klass);
3107     // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3108     __ z_ogr(tmp2, tmp1);
3109 
3110     __ bind(do_update);
3111     __ z_stg(tmp2, mdo_addr);
3112 
3113     __ bind(next);
3114   }
3115 }
3116 
3117 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3118   assert(op->crc()->is_single_cpu(), "crc must be register");
3119   assert(op->val()->is_single_cpu(), "byte value must be register");
3120   assert(op->result_opr()->is_single_cpu(), "result must be register");
3121   Register crc = op->crc()->as_register();
3122   Register val = op->val()->as_register();
3123   Register res = op->result_opr()->as_register();
3124 
3125   assert_different_registers(val, crc, res);
3126 
3127   __ load_const_optimized(res, StubRoutines::crc_table_addr());
3128   __ kernel_crc32_singleByteReg(crc, val, res, true);
3129   __ z_lgfr(res, crc);
3130 }
3131 
3132 #undef __